w». .

.9
+5

.;
5.. Emmﬂﬁv. a
1 I... .
31-...th: v
ﬁg.
.. ,, .P.

. 3.
#1135:

o. . z.
Pu. 55.5..

3:
.i ..

.5552.
.2” 5..

. .

”um?

.
. 3... .15
1.0.17

9:

. ~§.Bﬂdu .1 a‘
, m

4.? .2...
8... i.

4...".
..5u.¥.:$..2..

 

 

 

 

 

 

:54 .8.
i 1. .

f; »
.3... awry?
$.33}:

..
:1: i... .
i2..- .. .. .: ﬂat/an.
II (If!) 21‘ A ‘1I-.!.| IzIL
.. 3 V {3 5 3.3.53. §
‘ h“- t .43 hiotu.
Of 0 r. O I I‘ I
. 222.. 3;: .. :13...
. I‘. \‘¢§‘I-‘*.!K.‘

.. 7.3. K A £

., .aenﬂutdvﬁ. 3..
3.3i\f.-, . I‘3lei- .
nzaiﬁuﬁﬂeas
ﬁns}? ,
. ., .55 #1- o
.I‘ivﬁ.l}npuu"

33...... l .: stung...
3t... 6. 0: I»
» J!!»tL.JP. ‘np. I,
unlit: 32.1.9.
tub!- I. .
.I:btl 0
0!..."
.4‘5
5:
it‘d. (I:
. .

ti; \ ,thh 9.1.. in 2 ..
.. . I
tsz‘lzt‘! .232 3.5.5: at? I 1“?!
.\‘ «ﬁ‘l‘i'Zfﬁ-l gala-kl- 1"!!!
.»I’nrb.l‘."y. . \I)

 

 

 

Ox . 1.310 (It!
rﬁtio..‘i|l

s11 3130..|nsit.. . 1.1:?”
59;) .fIs-thtta. £8.75?!
.. .2. 5.3....83ii. .IKI:\
.‘ﬁ‘IQ-‘X 1"

 

 

VP.5...\.‘A~ ti. ..
1.32.549... l

 
    
   
   
 
   
   
 
    
 
   

 

30m
LIBRARY

Michigan State 1
University |

_ “w. 7

 

 

 
  

This is to certify that the

dissertation entitled

Spatial Organization in Soil Bacterial Communities

presented by

Christopher B. Blackwood

has been accepted towards fulﬁllment
of the requirements for

Ph. D. degreein Crop and Soil Sciences
Ecology, Evolutionary
Biology, and Behavior

fﬂﬂ/

Major professor

August 1, 2001
Date

MCI/haul” -' “ " "‘ ' ' ' ' 0-1277!

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

DATE DUE DATE DUE DATE DUE

 

6/01 cJCIHC/DaoDm.pBbo.15

 

 

,‘v- )1

 

SPATIAL ORGANIZATION IN SOIL BACTERIAL COMMUNITIES
By

Christopher B. Blackwood

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Crop and Soil Sciences
Ecology, Evolutionary Biology, and Behavior Program

2001

 

,
1r

 

 

 

SPATIAL ORGANIZATION IN SOIL BACTERIAL COMMUNITIES

By

Christopher B. Blackwood

AN ABSTRACT OF A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Crop and Soil Sciences
Ecology, Evolutionary Biology, and Behavior Program

2001

Professor Eldor A. Paul

 

 

 

 

ABSTRACT
SPATIAL ORGANIZATION IN SOIL BACTERIAL COMMUNITIES
By

Christopher B. Blackwood

Spatial variability of soil resources is high due to the heterogeneous arrangement
of pores and mineral and organic matter particles in soil, and due to the activity of plants
and soil fauna. The spatial organization of soil microbial communities was investigated
using terminal restriction fragment length polymorphism (T -RFLP) of eubacterial
ribosomal 168 genes, a molecular genetic technique independent of the cultivation of
microbes outside their natural environment. Methods for Optimal data processing and
multivariate statistical analysis of complex community T-RFLP proﬁles were
investigated using analytically-replicated datasets. Spatially-deﬁned microbial habitats
were detected by testing the signiﬁcance of the differences between eubacterial
community T-RFLP proﬁles. Light fraction and shoot residue (partially-decomposed
organic matter particles) were found to contain communities different from both the
rhizosphere and the soil heavy fraction (soil minerals with associated humiﬁed organic
matter). Communities in the external and internal portions of soil aggregates were found
to be slightly different, while those of different aggregate sizes were not different. The
establishment of soil fractions as distinct microbial habitats seemed to be dependent on
the differences in the organic matter contents of different fractions. However, different
cropping systems also caused divergence in communities which could not be explained
on the basis of organic matter contents of samples. Microscopic cell counts were shown

to be primarily sensitive to organic matter contents of samples, and not to qualitative

 

 

differences between cropping systems, in contrast to community composition. Number
of large cells (>O.18 um3) was signiﬁcantly affected by both soil fraction and cropping
system. In another approach to understanding spatial organization of soil eubacterial
communities, communities within soil samples of varying sizes and from differing
locations within replicated 1.5x2 m plots were assayed by T-RFLP. The spatial structure
of the communities was investigated using a generalized multivariate extension of
blocked-quadrat analysis and semivariance analyses, integrating analyses based on
varying sampling grain and extent. Signiﬁcant hierarchical spatial structure was detected
within the eubacterial communities, manifest by phase shifts in the relationship between
community variability and spatial scale. These studies indicate that signiﬁcant spatial
organization exists in eubacterial communities in soil, which implies that soil
communities are not randomly assembled and may be regulated by mechanisms

analogous to those for plant and animal communities.

 

 

ACKNOWLEDGEMENTS

I am extremely grateful to Dr. Eldor Paul, my graduate advisor, for the inspiration
and advice he has given me through my years at MSU. He gave me the freedom to
pursue my interests while ensuring that the goals were realistic and the science was
rigorous and exciting. I would also like to thank my committee members, Dr. Richard
Harwood, Dr. Kay Gross, and Dr. James Tiedje for their advice and support.

Dr. Terry Marsh graciously allowed me to perform much of the molecular genetic
work in this thesis in the Research On Microbial Evolution (R.O.M.E.) lab. The advice I
received from Dr. Marsh and other members of the R.O.M.E. lab, and from Dr. David
Treves, Hector Ayala del Rio, and Jim Stoddard was invaluable in my learning molecular
techniques. Dr. Al Smucker, Dr. Curtis Dell, and Dr. Shawel Haile—Mariam shared their
expertise of soil structure and helped perform some of the soil aggregate fractionation
work. Discussions with Dr. Sven Bohm, Dr. Sherri Morris, Dr. Ann Marie Fortuna, and
Helmut Stoyan also assisted in the formulation and accomplishment of this research. My
work was shaped by discussions with many individuals from the Center for Microbial
Ecology, the KBS Long Term Ecological Research site, the Living Field Lab, the
Ecology, Evolutionary Biology, and Behavior Program, and the Michigan State
Sustainable Agriculture Network. Financial support for my studies was provided by Dr.
Tiedje through the Center for Microbial Ecology, Dr. G. Philip Robertson through the
K88 LTER, and Dr. Smucker through USDA/NR1 grant 97-35107-4322.

Finally, I thank my wife Andria and sons Stewart and Eli for their continued
understanding and support, and my parents for providing me with the tools and

conﬁdence needed to accomplish my goals.

TABLE OF CONTENTS

List of Tables vi
List of Figures vii
Chapter 1: Introduction 1
Chapter 2: Methods of T-RFLP Data Analysis for Quantitative
Comparison of Microbial Communities 8
Abstract 9
Introduction 10
Methods 16
Results 22
Discussion 28
Conclusions 3 1

Chapter 3: Eubacterial Community Structure and Population
Size within the Soil Light Fraction, Rhizosphere, and Heavy

Fraction of Several Agricultural Cropping Systems 47
Abstract 48
Introduction 49
Methods 52
Results 60
Discussion 67

Chapter 4: Eubacterial Community Response to Position within

Soil Macroaggregates and Soil Management 89
Abstract 90
Introduction 90
Methods 94
Results 102
Discussion 108
Conclusions 1 10

Chapter 5: Use of Multivariate Statistics to Test Hierarchical
Structure within a Soil Eubacterial community Analyzed

By T-RFLP 126
Abstract 127
Introduction 128
Detection of Levels of Ecological
Organization: Statistical Approaches 132
Materials and Methods 143
Data Analysis Methods 148
Results ' 156
Discussion 161
Conclusions 168

 

‘66

 

 

 

 

LIST OF TABLES

Chapter 2

Table 1: Dendrogram characteristics for cluster analysis of manually-
aligned T-RFLP proﬁles of KBS soil fraction using a variety of data
processing methods and clustering algorithms. All proﬁles included

in the analysis.

Table 2: Dendrogram characteristics for cluster analyses of computer-
aligned T-RFLP proﬁles of KBS soil fractions using a variety of data
processing methods.

Table 3:Dendrogram characteristics for cluster analysis of manually-
aligned T-RFLP proﬁles of KBS soil fractions using a variety of data
processing methods and clustering algorithms. Only proﬁles with a
cumulative peak height of >10,000 included in analysis.

Table 3: Dendrogram characteristics for cluster analyses T-RFLP
proﬁles of bioreactor samples using a variety of data processing methods
and clustering algorithms.

Table 5: Redundancy analysis results testing the null hypothesis that
there is no difference between bioreactor sample T-RFLP proﬁles.
Table 6: Dendrogram characteristics and redundancy analysis results for
the set of alfalfa soil samples.

Chapter 3

Table 1: Biogeochamical characteristics of soil fractions.

Table 2: Numbers of bacterial cells and percent of cells in different cell
size classes.

Table 3: Redundancy analysis testing treatment effects on T-RFLP
proﬁles derived from different soil fractions and cropping systems.

Table 4: Terminal restriction fragments (T-RFs) associated with the same

treatment in 1998 and 1999.

Chapter 4

Table 1: Redundancy analysis of HF-l and LF-l isolated from different
aggregate size classes at KBS.

Table 2: Redundancy analysis of pooled aggregate layer samples from
KBS.

Table 3: Redundancy analysis of pooled aggregate layer samples from
Wooster.

Chapter 5

Table l: Grain-based regression analysis of T-RFLP proﬁle variability on
mean log soil weight.

Table 2: Distance function parameters ﬁt with nonlinear least-squares to
HV of 10 g sample T-RFLP proﬁles.

Table 3: Redundancy analysis and global Mantel tests of all ﬁfty 10 g
sample T-RFLP proﬁles per grid.

vi

37

38

39

40

41

41

78

79

80

81

117

118

119

178

178

179

LIST OF FIGURES

Chapter 2

Figure 1: Example of replicated T-RFLP electropherograms 42
Figure 2: Dendrogram of manually-aligned T-RFLP proﬁles of KBS soil
fractions constructed by Ward’s cluster analysis of raw T-RF peak

heights. 43
Figure 3: Dendrogram of manually-aligned T-RFLP proﬁles of KBS soil
fractions constructed by Ward’s cluster analysis of relative T-RF peak
heights. 44
Figure 4: Dendrogram of manually-aligned T-RFLP proﬁles of KBS soil
fractions constructed by UPGMA cluster analysis of relative T-RF peak
heights. 45
Figure 5: Canonical principal components plots derived from redundancy
analysis of bioreactor sample T-RFLP proﬁles (Rsal, MspI, and HhaI

proﬁles pooled). 46

Chapter 3
Figure 1: Number of DTAF—stained cells counted per g of soil fractions. 82
Figure 2: Regression of bacterial cells per g of soil fraction against percent

C of soil fraction. 83
Figure 3: Regression of percent of cells in the largest cell size class
(>O.18 um3) against percent C of the sample. 84
Figure 4: Partial canonical principal components plots derived from
redundancy analysis of 1998 RsaI T-RFLP proﬁles. 85

Figure 5: Dendrogram generated by Ward’s method of hierarchical cluster
analysis on 1999 Rsal T-RFLP proﬁles using Hellinger distance between
proﬁles. 87
Figure 6: Dendrogram generated by Ward’s method of hierarchical cluster
analysis on 1998 RsaI T-RFLP proﬁles using Hellinger distance between
proﬁles. 88

Chapter 4
Figure 1: Dendrogram of cluster analysis by Ward’s method of RsaI
T-RFLP proﬁles from aggregate layers of individual 4-6.3 mm KBS

aggregates. 120
Figure 2: Plots of principal coordinates of Jaccard distance between
T-RFLP proﬁles of layers of individual 4-6.3 mm KBS aggregates. 121

Figure 3: Dendrogram of cluster analysis by Ward’s method of R531

T-RFLP proﬁles from pooled KBS aggregate layers using Hellinger

distance. 122
Figure 4: Biplots of sample and T-RF scores for the three canonical

principal components (PCs) derived from redundancy analysis of

Hellinger distance between pooled aggregate layers from KBS. 123

vii

 

LIST OF FIGURES, continued

Figure 5: Aggregate layer and treatment interaction axes of canonical
principal components derived from distance-based redundancy analysis
of J accard distance between Wooster aggregate layer RsaI T-RFLP

proﬁles. 124
Figure 6: Principal coordinates derived from J accard distance between
Wooster aggregate layer RsaI T—RFLP proﬁles. 124

Figure 7: Dendrogram of cluster analysis by Ward’s method of RsaI
T-RFLP proﬁles from Wooster aggregate layers using J accard distance. 125

Chapter 5

Figure 1: Sampling map for grid 1. 180
Figure 2: MAD-t scores derived from DNA extraction experiment. 181
Figure 3: Grain-based analysis using three variability statistics. 182
Figure 4: Hellinger variance as a function of separation distance. 183
Figure 5: Integration of extent-based and grain-based analyses of Hellinger
variance. 185

Figure 6: Principal coordinates analysis of correlation super-matrix of
covariance matrices constructed from different distance classes in grid 2,

RsaI digest. 186
Figure 7: Cluster analysis correlation super-matrix describing covariance
matrices constructed from different distance classes in grid 2, RsaI digest. 187
Figure 8: Normalized Mantel correlogram of Hellinger and Jaccard

distances for all 10 g samples. 188
Appendix A: Total number of T-RFs detected in sets of samples at
multiple-grain sampling locations. 189

viii

Chapter 1

Introduction

 

 

 

 

 

 

 

 

 

The spatial structure of soil is a complex interconnected framework characterized

by pores of various sizes and tortuosity, and zones of varying bulk density and stability,
(Christensen 1996, Young and Ritz 2000). Among the mineral particles are
heterogeneously distributed particles of decomposing plant residue (Parkin 1993, Staricka
et al. 1991). Plant roots, fungal hyphae, and soil fauna grow through this matrix and
modify its spatial structure by forcing through narrow pores and entangling particles
(Oades 1993, Tisdall 1996). These organisms also process resources and energy, creating
gradients of organic matter, water, redox potential, and inorganic nutrients (Juma 1994,
Smucker 1993). Soil resources are heterogeneously distributed at larger scales as well,
from one to hundreds of meters (Boemer et al. 1998, Doberrnann et al. 1995, Robertson
et al. 1997, Stoyan et a1. 2000).

The major goal in this thesis is to investigate how soil bacterial communities
respond to spatial structure in their environment. The approach taken tests for patterns in
the distributions over all species in an environment. The patterns detected are thus
community-wide, and reflect the strength of interactions and similarities between
different species as well as their individual responses to the environment. This approach
required the use of multivariate statistical techniques. Multivariate hypothesis-testing
methods based on permutation tests were used where possible so that the interpretation of
results is not a subjective matter.

Microbial community composition was assayed using terminal restriction
fragment length polymorphism (T-RFLP) of eubacterial 16S ribosomal sequences
amplified by PCR from environmental genomic DNA extracts. This method is not

dependent on the growth of the bacteria in the laboratory, which is known to result in

severe biases in the species detected (Felske et al. 1999, Madsen 1996). There are
commonly occurring species in soil that have never been successfully isolated in the
laboratory, but can be detected via molecular methods (Head et al. 1998, Hugenholtz et
al. 1998).

In Chapter 2 of this thesis, several methods of processing T-RFLP data and
examining relationships between proﬁles using multivariate statistical methods are tested.
Four sets of T-RFLP proﬁles are used, representing a wide range in degree of divergence
between samples, from a set of soil samples taken in different regions of the United
States to a set of samples taken from within one meter of each other in an alfalfa ﬁeld.
Each set contained analytical replicates of the sample profiles. The data processing and
statistical methods were tested for their sensitivity to analytical variability present in T-
RFLP proﬁles. The methods found to be most robust are used in the remainder of the
chapters, including analysis of Hellinger distance and J accard distance between T-RFLP
proﬁles with the multivariate hypothesis—testing method redundancy analysis.

Chapter 3 is an investigation into the possibility that different microbial
communities are associated with different soil fractions, which could then be classiﬁed as
distinct soil habitats. The fractions investigated were hypothesized to contain different
communities due to differences in the age and types of organic matter found within them.
The fractions included the light and heavy fractions, shoot residue > 2 mm, and the
rhizosphere. Communities in all fractions were shown to be different from each other,
except for light fraction and shoot residue, which were essentially the same. The
rhizosphere is well-accepted as a habitat that has a community distinct from the bulk soil.

Detection of the light fraction/shoot residue habitat shows that organization of soil

 

 

”Wat

 

 

 

microbial communities around organic matter is more general than the rhizosphere — bulk
soil dichotomy. Differentiation of communities due to cropping system shows that other
aspects of the environment critically affect the community present, in addition to the total
amount of organic matter present. This is also shown for the relationship between
organic matter and total number of bacterial cells.

Hierarchical aggregate structure is currently the dominant model of soil structure
and has been shown to be useful in the study of organic matter turnover in soils. In
Chapter 4 the hypotheses that aggregates of different sizes and that layers within
aggregates contain different communities was tested. A new method of isolating the
outer portion of aggregates and its bacteria] community was used, resulting in knowledge
of the precise location of the sample within the aggregate. Analysis of individual
aggregates using a split-plot design was found to be the most satisfactory approach for
investigating whether the communities in different aggregate layers were different.
Aggregate layers had communities which were only slightly different. Aggregate-to-
aggregate variability was large. The effect of aggregate size on eubacterial communities
was quite weak. Organization of eubacterial communities due to position within and size
of soil aggregates was not as pronounced as organization by organic matter found in
Chapter 3. Organization by the particular tertiary structure of the soil (arrangement of
aggregates in relation to shoot residue, roots, macropores, etc.) is hypothesized to be
more important in the determination of the types of microbial communities present in
aggregates.

Spatial organization is a scale-dependent phenomenon. Hierarchical spatial

organization is one hypothetical way in which organization can change with scale,

incorporating the concept of levels of organization that operate over different scale
domains. Chapter 5 is a test of the hierarchical spatial organization of soil eubacterial
communities. Size and location of soil samples are varied, resulting in two datasets
measuring different aspects of how spatial structure of the communities changes with
scale. These are analyzed and integrated using a generalized multivariate extension of
paired-quadrat analysis and semivariance analysis that is called variability-scale analysis.
The variability-scale analysis is compared to Mantel methods and spatially-constrained
ordination. An approach called the correlation super-matrix is also proposed for
examining how species or T-RF covariance structure changes with scale. Soil eubacterial
communities within 1.5x2 m plots were found to be organized by three hierarchical levels

which are described.

 

References

Boemer, R.E.J., A.J. Scherzer, J .A. Brinkman. 1998. Spatial patterns of inorganic N, P
availability, and organic C in relation to soil disturbance: a chronosequence analysis.
Applied Soil Ecology 7: 159-177.

Christensen, ET 1996. Carbon in primary and secondary organomineral complexes. In
Structure and Organic Matter Storage in Agricultural Soils, edited by M. R. Carter and
B. A. Stewart. Boca Raton: CRC Press.

Doberrnann, A., P. Goovaerts, T. George. 1995. Sources of soil variation in an acid
Ultisol of the Philippines. Geoderma 68: 173-191.

Felske, A., A. Wolterink, R.v. Lis, W.M.d. Vos, A.D.L. Akkermans. 1999. Searching for
predominant soil bacteria: 16S rDNA cloning versus strain cultivation. FEMS Microbial
Ecology 30: 137-145.

Head, I.M., J .R. Saunders, R.W. Pickup. 1998. Microbial evolution, diversity, and
ecology: a decade of ribosomal RNA analysis of uncultivated microorganisms. Microbial
Ecology 35: 1-21.

Hugenholtz, P., B.M. Goebel, N.R. Pace. 1998. Impact of culture—independent studies on
the emerging phylogenetic view of bacterial diversity. Journal of Bacteriology 18024765-
4774.

Juma, NC. 1994. A conceptual framework to link carbon and nitrogen cycling to soil
structure formation. Agriculture, Ecosystem and Environment 51 :257-267.

Madsen, EL. 1996. A critical analysis of methods for determining the composition and
biogeochemical activities of soil microbial communities in situ. In Soil Biochemistry,
edited by G. Stotzky and J. Bollag. New York: Marcel Dekker.

Oades, J .M. 1993. The role of biology in the formation, stabilization, and degradation of
soil structure. Geoderma 56:377—400.

Parkin, TE. 1993. Spatial variability of microbial processes in soil - a review. Journal of
Environmental Quality 22:409-417.

Robertson, G.P., KM. Klingensmith, M.J. Klug, E.A. Paul, J .R. Crum, 3.6. Ellis. 1997.
Soil resources, microbial activity, and primary production across an agricultural
ecosystem. Ecological Applications 7: 158-170.

Smucker, A.J.M. 1993. Soil environmental modiﬁcations of root dynamics and
measurement. Annual Review of Phytopathology 3 1 : 191-216.

 

Staricka, J .A., R.R. Allmaras, W.W. Nelson. 1991. Spatial variation of crop residue
incorporated by tillage. Soil Science Society of America Journal 55: 1668-1674.

Stoyan, H., H. De-Polli, S. Bohm, G.P. Robertson, E.A. Paul. 2000. Spatial heterogeneity
of soil respiration and related properties at the plant scale. Plant and soil 222:203-214.

Tisdall, J .M. 1996. Formation of soil aggregates and accumulation of soil organic matter.
In Structure and Organic Matter Storage in Agricultural Soils, edited by M. R. Carter
and B. A. Stewart. Boca Raton: CRC Press.

Young, I.M., K. Ritz. 2000. Tillage, habitat space and function of soil microbes. Soil &
Tillage Research 53:201-213.

   
 

. «w v:rgu'012'.919'?“‘“!""'.'j'."3"9','

Chapter 2

Methods of T-RFLP Data Analysis for Quantitative
Comparison of Microbial Communities

 

—.-_.:r

 

 

 

 

 

 

aim 9-1—35-

 

 

 

Abstract

Terminal restriction fragment length polymorphism (T-RFLP) is a culture-
independent method of obtaining a ﬁngerprint of the composition of a microbial
community. T-RFLP proﬁles of complex eubacterial communities were generated
following PCR ampliﬁcation of the 16S ribosomal gene. Proﬁles of each sample were
replicated so that the relative utility of different methods of statistical analysis, T-RF
alignment, and computing the difference (or distance) between proﬁles could be
compared. Four sets of samples were used, representing a wide range in degree of
divergence between environmental conditions. These sets of samples included: soil
samples collected from three regions of the United States, different soil fractions derived
from three agronomic ﬁelds at one site, soil samples taken from within one meter of each
other within an alfalfa ﬁeld, and replicate laboratory bioreactors. All methods of analysis
could be used to differentiate between samples in the most divergent set, but only a
subset were successful on the more similar sets. Ward’s method of cluster analysis was
more effective at ﬁnding major groups within sets of proﬁles, while the average or
UPGMA method of cluster analysis had a slightly reduced error rate in clustering of
replicate proﬁles and was more sensitive to outliers. Analyses based on Euclidean
distance between proﬁles of raw T-RF peak heights did not result in clustering of
replicate proﬁles, while most replicate proﬁles were grouped together using Euclidean
distance between relative peak height proﬁles or Hellinger distance between proﬁles. It
was found that the multivariate hypothesis-testing technique redundancy analysis was
more effective at detecting differences between very similar samples than was cluster

analysis. Redundancy analysis using Hellinger distance was more sensitive than using

 

 

Euclidean distance between relative peak height proﬁles (Le. a lower p-value was

obtained in testing the null hypothesis that the profiles of different samples were the
same). Analysis of J accard distance between proﬁles, which only takes into account
presence/absence of a T-RF, was the most sensitive in redundancy analysis, and equally
sensitive in cluster analysis, compared to the other distance metrics, but only if all
proﬁles had cumulative peak heights greater than 10,000 ﬂuorescence units. It was
concluded T-RFLP was a very sensitive method of differentiating between microbial

communities when the appropriate statistical methods were used.

Introduction

Culture-independent methods of microbial community analysis involve the
analysis of signature biochemicals extracted directly from environmental samples.
Molecular genetic techniques, utilizing extracted genomic or ribosomal nucleic acids,
allow microbial community analysis to be coupled with a comprehensive phylogenetic
framework (Amann et al. 1995, Woese 1987). The use of such techniques has shown that
methods relying on growth of the organisms ex situ reveal a small fraction of the
diversity present in soil microbial communities (e.g. Torsvik et al. 1990, Ward et a1.
1992). This uncultured diversity includes both species that are closely related to cultured
organisms and species that represent virtually uncultured phylogenetic lineages (Head et
al. 1998, Hugenholtz et a1. 1998, Kuske et al. 1997).

A variety of molecular techniques have been developed for rapidly assaying
community structure. Most methods involve the separation of PCR amplicons based on

differences in DNA sequence of genes of functional or phylogenetic interest, often the

 

 

 

 

 

 

 

16S ribosomal RNA gene. These include denaturing gradient gel electrophoresis

(DGGE, Muyzer et al. 1993), ribosomal intergenic spacer analysis (RISA, Bomeman and
Triplett 1997), single strand conformation polymorphism (SSCP, Simon et a1. 1993),
ampliﬁed ribosomal DNA restriction analysis (ARDRA, Massol-Deya et al. 1995), and
terminal restriction fragment length polymorphism (T-RFLP, Bruce 1997, Liu et al.
1997). These methods do not reveal diversity per se unless the community is very
simple, since a fraction of the species revealed by DNA rehybridization rates or sequence
analysis of a clone library can be visualized on a gel (Dunbar et al. 2000, Nakatsu et al.
2000). However, these methods do provide a way to determine the relative abundance of
common species present in a sample, free of the constraint that the organisms must be
amenable to growth in the laboratory. They are valuable as rapid methods of ﬁnding
major differences between communities, and testing hypotheses based on a comparison
of samples.

T-RFLP has been shown to be effective at discriminating between microbial
communities in a wide range of environments and has several advantages over other
techniques (Tiedje et al. 1999). T-RFLP involves tagging one end of PCR amplicons
through the use of a ﬂuorescent molecule attached to a primer. The ampliﬁed product is
then cut with a restriction enzyme. Terminal restriction fragments (T-RFs) are separated
by electrophoresis and visualized by excitation of the ﬂuor. T-RFLP analysis provides
quantitative data about each T-RF detected, including size in base pairs (bp) and intensity
of ﬂuorescence (peak height). T-RF sizes can be directly compared to a database of

theoretical T-RFs derived from sequence information (e. g. Dunbar et al. 2001, Marsh et

 

al. 2000). In addition, T-RFLP has been found to detect a greater number of operational

taxonomic units compared to DGGE (Marsh et al. 1998, Moeseneder et al. 1999).

T-RFLP proﬁles have been shown to be relatively stable to variability in PCR
conditions (Osborn et al. 2000, Ramakrishnan et a1. 2000). Currently the least well-
deﬁned technical aspect of T-RFLP, as well as many other methods of microbial
community analysis, is the data processing and analysis of proﬁles. A wide range of
methods have been used in the literature. The goal of this study was to ﬁnd the optimal
procedure for use in comparing complex environmental T-RFLP proﬁles, resulting in the
lowest probability of type II errors (not ﬁnding differences between proﬁles when they
are actually different). The possibility of alignment of T-RFLP peaks using non-
hierarchical K-means cluster analysis was also examined.

The sensitivity of statistical methods to pre-analysis data processing or changes in
the statistical procedures is dependent not only on the analytical consistency of replicate
proﬁles, but also on the degree of divergence between proﬁles in the dataset. Therefore
four sets of samples were used, representing a wide range in biological complexity and
environmental differentiation. The most divergent set of samples were soil samples
collected from three regions of the United States, with differing levels of environmental
contamination and disturbance. Representing an intermediate level of divergence, soil
samples from one site were divided into chemically and physically distinct fractions.
Two sets of samples were characterized by quite low divergence in environmental
characteristics: a set of replicate laboratory bioreactors, and a set of soil samples taken
from within a 1.5x2 meter area of alfalfa It is expected that microbial communities will

diverge in species composition and relative abundance in accordance with the divergence

12

 

of their habitats. While this is supported by the analyses presented in this paper, the

hypothesis and the nature of the differences, with respect to spatial structure in soil, is
more fully explored in subsequent chapters.

The threshold for including an electropherogram peak as a valid T-RF is often left
unspeciﬁed, but has been set to 100 ﬂuorescence units (Franklin et al. 2001, Osborn et al.
2000), 50 units (Kerkhof et al. 2000), and 25 units (Dunbar et al. 2000, Dunbar et al.
2001). Other researchers have limited the number of peaks analyzed by rarefraction
(Hedrick et al. 2000) or by including only those comprising greater than one percent of
relative abundance in a sample (Derakshani et al. 2001, Lukow et al. 2000), or present in
all analytical replicates (Dunbar et al. 2000).

To compare T-RFLP proﬁles in cases where the community is very simple or the
primers targeted a small group of organisms, a simple inspection of electropherograms,
tables, or barcharts is normally used (e.g. Bernhard and Field 2000, Bruce 1997, Chin et
al. 1999, Flynn et al. 2000, Ramakrishnan et a1. 2000). While raw data is sometimes
relied upon to compare complex proﬁles (e. g. Derakshani et al. 2001, Gonzalez et a1.
2000, Kerkhof et al. 2000, Leser et al. 2000, Liidemann et al. 2000), quantitative
statistical approaches can also be employed. These have included cluster analysis
(Dunbar et al. 2000, Liu et al. 1997, Moeseneder et al. 1999, Scala and Kerkhof 2000),
MANOVA (Lukow et al. 2000), and principal components analysis (Clement et a1. 1998,
Franklin et al. 2001).

Statistical methods for assessing differences between T-RFLP profiles that were
tested in this study include redundancy analysis and the unweighted-pair group method

using arithmetic averages (UPGMA) and Ward’s method of hierarchical cluster analysis.

 

Redundancy analysis is equivalent to a principal components analysis of ﬁtted values

derived from a multiple linear regression of each variable (or T-RF). To test for
differences between experimental groups, the linear regression is performed with a set of
dummy variables indicating group identity for each sample (Legendre and Legendre
1998). The proportion of total variance in the original nonﬁtted dataset that the canonical
principal components can account for is then calculated, with an associated pseudo-F
statistic. The F statistic is then compared to a distribution generated by random
permutation of the data with respect to the dummy variables to obtain a p-value.
Calculation of a p-value by random permutation avoids the assumptions of
multinorrnality and restrictions on the number of variables that can be analyzed
(Legendre and Anderson 1999).

Cluster analysis is essentially a tool for exploratory data analysis, and should not
be used as ﬁnal proof in testing of hypotheses (Krzanowski and Marriott 1994). Cluster
analysis summarizes all the variability, and therefore all phenomena affecting the data,
within a dendrogram. The difference between the UPGMA and Ward’s methods of
cluster analysis is in how they calculate the distances between groups of samples (J obson
1992). The UPGMA method uses the average distance between samples calculated from
all possible pairs of samples from the two groups. Ward’s method uses the distance
between the group centroids, weighted by the total number of samples in both groups.
The distance between group centroids is equivalent to the serni-partial r-squared, or the
proportion of total variance that would be accounted for by joining the two groups.

The statistical methods discussed above must be based on some measurement of

the distance (or difference) between proﬁles. Cluster analysis or redundancy analysis

 

 

Mm"

 

 

 

 

 

 

using raw peak heights results is based on Euclidean distance. Several kinds of

transformations can be made to peak height before statistical analysis, resulting in
analyses based on other distance metrics. Use of relative peak height (or height divided
by the cumulative peak height of the proﬁle), is a common method (Bruce 1997, Chin et
al. 1999, Derakshani et al. 2001, Gonzalez et al. 2000, Liidemann et al. 2000,
Ramakrishnan et al. 2000). Legendre and Gallagher (in press) examined several data
transformations that could be performed prior to statistical analysis that result in analyses
based on ecologically appropriate distance metrics instead of Euclidean distance. They
found that use of relative abundance was an improvement but still unsatisfactory for
simulated community composition data. Analyses based on the Hellinger distance can be
performed by ﬁnding the Euclidean distance between proﬁles after the Hellinger
transformation, which is simply a square root transformation of relative abundance
(Legendre and Gallagher in press). Lukow et al. (2000) used this transformation in an
attempt to normalize peak height distribution.

Other studies have based comparison of T-RFLP proﬁles on binary indicators of
peak presence (Bernhard and Field 2000, Clement et al. 1998, Flynn et al. 2000, Franklin
et al. 2001, Kerkhof et al. 2000, Leser et al. 2000, Liu et al. 1997, Moeseneder et al.
1999, Scala and Kerkhof 2000) or binary indicators after deletion of peaks falling below
a height threshold due to standardization (Dunbar et al. 2000, Dunbar et al. 2001).
Several similarity coefﬁcients can be used to quantitatively compare proﬁles using binary
indicators; in this study we subtract J accard’s coefﬁcient from one to transform it into a
distance metric, and refer to this as J accard’s distance. Redundancy analysis using

J accard distance requires the alternate procedure, distance-based redundancy analysis, to

15

 

be employed. Any distance metric can be employed in distance-based redundancy

analysis because, prior to the redundancy analysis, principal coordinates analysis is used
to project the distances between proﬁles into a Euclidean space (Legendre and Anderson
1999). Distance-based redundancy analysis has the disadvantages of being somewhat
more complex and, unlike standard redundancy analysis, the impact of individual

variables (or T-RFs) on the results cannot be determined directly.

Methods
Sample collection

Four sets of samples were used in this study:
1. KBS Soil Fractions. Soil samples were collected from the agricultural ﬁeld plots of the
Long Term Ecological Research (LTER) site and the Living Field Lab (LFL) at the W.K.
Kellogg Biological Station in southwestern Michigan, USA. Three blocks of soil
weighing approximately 350 g were excavated and pooled from each plot. Sampling
depth was 10 cm. Field treatments included continuous alfalfa, conventionally managed
continuous corn, and organically managed ﬁrst year corn in a com—com-soybean-wheat
rotation with cover crops. Details of the ﬁeld treatments are described elsewhere (Jones
et a1. 1998, http://lter.kbs.msu.edu/Agronomics). The soil was fractionated into
rhizosphere, shoot residue, and light and heavy fractions from various sizes of soil
macroaggregates. Details of the soil fractionation and results from comparison of field
treatments and soil fractions are described in chapters 3 and 4.
2. Bioreactor Samples. Bioreactor samples were taken from a ﬂuidized-bed reactor with

activated carbon as the particulate carrier. The reactor was inoculated with an anaerobic

 

 

 

-" .
‘..r’

'_.

 

Mi

 

 

r. . 1

. . .3. . . s . '
LII-I N .IIJEIBI. .' s. burn-Cr

 

 

 

enrichment culture and fed continuously with ethanol and essential nutrients. Samples '

were removed, pelleted, and stored at -20°C until needed. Sample collection and T-
RFLP analysis as described below was performed by Dr. Terry Marsh, who kindly
provided the T-RFLP proﬁles for statistical analysis.
3. KBS Alfalfa Soil Samples. A ten g soil sample was collected from each of ﬁve
locations within a 1.5 x 2 m plot within an alfalfa field at the KBS LTER site. Samples
were collected from the layer of soil 2 to 4 cm deep. These samples were part of a larger
study of the spatial structure of soil microbial communities described in chapter 5.
4. Multi-Region Soil Samples. Soils were collected from Sault Saint Marie, MI, Milan,
TN, and Hawthorne NV. Approximateley 500 grams of soil were removed from the top
5 centimeters of soil, homogenized, and aliquoted to 100 ml specimen containers.
Samples were stored at -20°C until needed. Sample collection and T-RFLP analysis as
described below was performed by Dr. Terry Marsh, who kindly provided the T-RFLP
proﬁles for statistical analysis.
T-RFLP

Mixed community DNA was extracted from the KBS LFL and LTER soil
fractions (0.3 g) and alfalfa soil samples (10 g) using the standard or Large—Scale
Ultraclean Soil DNA extraction kit, respectively, following the manufacturers
instructions (Mo Bio Laboratories, Solana Beach, CA). Genomic DNA was found to be
of sufﬁcient purity to be used directly in PCR reactions. PCR was performed using a
standard reaction mixture of 160 M of each deoxynucleoside triphosphate, 3 mM
MgC12, 0.05 U/uL Taq DNA polymerase and the appropriate volume of accompanying

10X PCR buffer (Gibco BRL, Gaithersburg, MD), and 0.2 ug/mL bovine serum albumin

 

(Boehringer Mannheim Biochemicals, Indianapolis, IN). Primers used were the general

eubacterial primer 8-27F (AGAG'I'I'I‘GATCCTGGCTCAG, E. coli numbering, Amann
et al. 1995, Integrated DNA Technologies, Coralville, IA) and the universal primer 1392—
1406R (ACGGGCGGTGTGTACA) amplifying the 16S ribosomal gene. PCR reactions
were optimized for each sample of genomic DNA using a master mix with primer
concentrations of 0.4 uM. PCR was performed in a Perkin-Elmer 9600 thermocycler
using an initial denaturation step of 95°C followed by 22 cycles of the following
program: denaturation at 94°C for 30 sec., primer annealing at 55°C for 30 sec., and
extension at 72°C for 30 sec. A modiﬁed hot start procedure was used where PCR tubes
were not placed in the therrnocycler until the block temperature had reached 80°C. A
ﬁnal extension at 72°C for 7 min. was performed after the programmed number of cycles
was complete. PCR product concentration and speciﬁcity was checked by
electrophoresis on a 1% agarose gel, followed by staining with ethidium bromide.
Optimizations were performed by adjusting the amount of genomic DNA extract used
(0.4 to 2 uL) in order to obtain a strong hand without visible non-speciﬁc product.

PCR reactions (50-75 uL) were performed in triplicate for each sample using the
optimal conditions found previously. These reactions were performed using the same
PCR master mix and program described above except that the forward primer was 0.6
uM hexachloroﬂuorscein (hex)-labeled 8—27F (Integrated DNA Technologies). PCR
replicates were then pooled and puriﬁed using the Promega PCR Preps Wizard Kit as
directed by the supplier, except that elution was performed with 19 [LL of sterile water
heated to 55-65°C. Five uL of puriﬁed PCR product was mixed with 5 uL of restriction

enzyme master mix containing 1.5 U/uL of restriction enzyme and one 11L of the

18

“ff: .VV.

hp“-

 

 

 

 

 

 

 

accompanying reaction buffer (Gibco). Restriction reactions were incubated for three

hours at 37°C, followed by 16 min. at 65°C to denature the restriction enzyme. Three 11L
of the restricted PCR product was mixed with one uL of 2500 TAMRA size standard
(Applied Biosystems Instruments, Foster City, CA). DNA fragments were separated by
size by electrophoresis at 1800 V for 14 hours on an ABI 373 automated DNA sequencer
at Michigan State University’s DNA Sequencing Facility. The 5' terminal fragments
were visualized by excitation of the hex molecule attached to the forward primer. The
gel image was captured and analyzed using Genescan Analysis Software 3.1. A peak
height threshold of 50 ﬂuorescence units was used in the initial analysis of the
electropherogram. Negative controls (no genomic DNA) were conducted with every
PCR and run on several Genescan gels. Contamination in PCR reactions was not
detected. Small peaks occasionally appeared in negative control lanes on Genescan gels,
but the cumulative peak height was always below 1000 units.

Ampliﬁcation of bioreactor and multi-region soil samples were performed as
described above with the following modiﬁcations. l6S rRNA genes were ampliﬁed from
community DNAs using primers 8-27F and 1492R (5’-GGTI‘ACC'I'I‘G'I'I‘ACGACTI‘-
3’). The PCR reaction mixture (100 til) contained 1X PCR buffer, 0.2 mM dNTPs, 1.5
mM MgC12, 0.4 M of 8-27F primer, 0.2 11M of 1492R primer, 0.1 ug BSA/u], 0.2 ng
template DNA/til, 0.05 U Taq polymerase/til (PE Amplitaq). Thermocycling was
performed in a GeneAmp 2400 PCR System thermal cycler (Perkin Elmer, Norwalk, CT)
at 94°C for 5 min followed by 30 cycles of 94°C for 50 sec, 55°C for 50 sec, 72°C for 1
min 30 sec, and a ﬁnal extension step at 72°C for 7 min. Ampliﬁcations that checked

positively on a 1.2% agarose gel were cleaned and concentrated using a Microcon YM-

 

100 (Millipore Corp. Bedford, MA). DNA concentrations were determined

spectrophotometrically. Puriﬁed PCR products (600 ng/reaction) were digested with
HhaI, MspI, or RsaI restriction endonuclease at 37°C for 3-4 hrs.
Replication experiments

For the set of KBS soil fraction samples, 32 samples were run on two Genescan
gels. Several other samples were also run once. For the remaining sets of samples, two
sets of PCR reactions were performed and pooled separately for each sample. Each of
these were then restricted and run twice on the same Genescan gel, for a total of four
replicate runs per sample.
Proﬁle alignment and statistical analsyses

Peak (or T-RF) heights in ﬂuorescence units and sizes in base pairs (bp) were
transferred to an Excel ﬁle. T-RFs were aligned between samples either by inspection of
the electropherogram and manual grouping of peaks into categories, or by use of a
custom algorithm written in SAS (Version 8 IML and Stat components) based on K-
means cluster analysis of the T-RF sizes. The algorithm also separated T-RFs when
multiple T-RFs from the same sample were clustered together. In this event, a new
cluster was generated and any T-RFs from samples with two or more T-RFs present in
the original cluster were placed into the new cluster. The process was then repeated for
the new cluster. The algorithm also allowed a maximum range in bp to be set for all
clusters. Alignments were generated using this algorithm with maximum ranges of from
1 to 10 bp. Computer alignment was only tested on the set of KBS soil fraction samples.
Alignment of peaks by manual inspection was done in Excel and was based primarily on

the size of peaks in bp. The pattern of peaks was also used to determine their alignment

20

 

when groups of overlapping peaks were found between samples. The identities of

samples were concealed during manual alignment.

Hierarchical cluster analysis was performed on aligned T-RFLP proﬁles using
SAS Version 8. Clustering was performed using the unweighted-pair group method
using arithmetic averages (UPGMA) and Ward’s method (SAS Institute Inc., Jobson
1992). Several methods of data processing were compared. The baseline for including a
peak was set to 50, 100, or 200 ﬂuorescence units. Rarefraction was also used as a
method of determining which small peaks should be included in the analysis, following
the procedure of Hedrick et al. (2000). Clustering was then based on Euclidean distance
between raw or relative peak height, Hellinger distance, or one minus J accard’s
coefﬁcient (referred to as J accard distance). Hellinger distance is equivalent to the
Euclidean distance between proﬁles after square-root transformation of relative peak
heights (Legendre and Gallagher in press). J accard’s coefﬁcient is based on binary
variables of peak presence and is equal to the ratio of the number of T-RFs present in
both proﬁles being compared to the total number of T-RFs present in either proﬁle.
Clustering was performed both with and without samples whose peak heights summed to
greater than 10,000 ﬂuorescence units. The cophenetic correlation was calculated for
dendrograms using an algorithm written in SAS IML.

Statistical signiﬁcance of the difference between samples, and as a corollary the
similarity of replicate proﬁles, was tested using redundancy analysis with the computer
software Canoco (Microcomputer Power, Ithaca, NY). This analysis compares a pseudo-
F statistic, calculated from the proportion of the total variance explained by sample

identity, to the values of F of 9999 random permutations of the sample identities of the

21

 

proﬁles (Legendre and Legendre 1998). Distance-based redundancy analysis was used to

determine significance using J accard distance (Legendre and Anderson 1999). The
J accard distance matrix and its principal coordinates were calculated using an algorithm
written in SAS IML, adapted from original code provided by Dr. Carl Ramm at Michigan

State University.

Results

An example of several T-RFLP electropherograms is shown in Figure 1.
Occasionally the baseline ﬂuorescence of the T-RFLP electropherograms was elevated
(i.e. ﬂuorescence did not reach zero between widely-spaced peaks). If the value of the
baseline could be ascertained, the baseline was subtracted from peak height in that
region. If the baseline varied inconsistently the sample was re-run. Evaluation of I
clustering errors was performed using dendrograms showing the hierarchical
relationships between T-RFLP proﬁles as found by the clustering procedure (see Figures
2-4 for examples). The number of clusters chosen to be examined was equal to the
number required to explain 50 percent of the variance in the entire dataset. An error was
counted when two replicate T-RFLP proﬁles (i.e. proﬁles derived from the same DNA
extract) were clustered into different groups. Errors in the dendrograms in Figures 2-4
occur where lines joining analytical replicates cross group divisions. Group colors were
exchanged for the proﬁles where errors occur in an attempt to allow the degree of error in

the dendrogram to be assessed intuitively.

22

 

I. KBS LTER soil fraction samples

In the ﬁrst set of samples there were 32 replicated proﬁles (see Table 1 and
Figures 2-4). Clustering using raw, unstandardized peak height consistently resulted in
the greatest number of errors. Use of relative peak height (peak height divided by the
cumulative peak height of the given sample) resulted in the fewest number of errors.
Clustering using binary variables (J accard distance) had an error rate higher than for
relative peak height, but still much lower than for raw peak heights (see Table 1).
Deleting peaks with heights less than 100 ﬂuorescence units, less than one percent of the
cumulative peak height for a given sample, or by rarefraction did not result in
improvement of clustering on any variable, compared to use of all peaks with heights
greater than 50. Deleting all peaks with heights less than 200 increased the number of
errors (see Table 1). Use of the Hellinger transformation resulted in an increase of
between zero and two errors over the analogous dendrogram based on relative peak
height. UPGMA clustering normally contained one to two fewer errors than clustering
by Ward’s method, and also resulted in a higher cophenetic correlation (or correlation
between elements of the original distance matrix and a distance matrix constructed from
the results of the cluster analysis). However, clustering by Ward’s method required
fewer groups to explain 50 percent of the variance in the dataset.

After alignment of T-RFLP proﬁles by the computer algorithm described in the
Methods, the minimum number of errors present in a dendrogram equaled four (using
Ward’s method; see Table 2). The computer alignment resulting in this dendrogram had
a maximum range of 4 bp and a mean range of 1.3 bp for groups of peaks (i.e. peaks in

different samples considered to represent the same T-RF). The manual alignment had a

23

 

maximum range of 4.4 bp and a mean range of 1.4 bp for groups of peaks. Deletion of '

samples with cumulative peak height less than 10,000 ﬂuorescence units resulted in 22
replicated proﬁles (see Table 3; the minimum of the full dataset was 4650). Using the
manual alignment of proﬁles, the number of errors was zero or one for all combinations
of clustering methods and distance metrics (raw peak height was not analyzed). The
number of errors resulting from Ward’s cluster analysis of the computer-aligned proﬁles
was found to be fewest using a maximum range of 2 bp for groups of peaks (mean = 0.9
bp). However, the number of errors was still higher than that for manually-aligned
proﬁles.
2. Bioreactor samples

The various methods of data processing and analysis did not greatly affect the
error rate of clustering of bioreactor replicates, and no method appeared to perform the
best (see Table 4). This was consistent for R531, MspI, and Hhal digests analyzed
separately, as well as for analysis of all three digests simultaneously. Since there were
four replicate proﬁles per sample, the error rate of nine out of 18 potential errors, which
is typical for this set of samples, could be generated by a variety of dendrograms. These
could include 1. the clustering of three replicates of each sample into distinct groups, with
the fourth of each sample an outlier, or 2. the perfect clustering of replicates of one
sample, a second sample being split evenly between two groups, and a third with no
replicates in the same group.

The signiﬁcance of the differences between sample proﬁles was tested using
redundancy analysis of standardized peak heights, Hellinger-transformed peak heights,

and principal coordinates of J accard distance. The sample identities were found to

24

 

explain a signiﬁcant amount of variance in the dataset for the RsaI and Hhal proﬁles

analyzed separately and for the RsaI, MspI, and HhaI profiles analyzed together (see
Table 5). Sample identities were signiﬁcant for the MspI profiles only for J accard
distance. P-values for all digests were lowest for analysis of J accard distance, and
highest for standardized peak heights, except for the RsaI proﬁle where analysis of
J accard distance had the highest p-value. The RsaI dataset included several proﬁles
where the cumulative peak height was less than 10,000 ﬂuorescence units. Plots of the
ﬁrst two canonical principal components generated by the redundancy analyses are
shown in Figure 5.
3. Alfalfa soil samples

While clustering J accard distance resulted in the lowest number of errors for the
alfalfa soil samples, all dendrograms had error rates that were quite high (see Table 6).
The errors could again be generated by a variety of dendrogram topologies since there
were four replicates per sample. Redundancy analysis detected signiﬁcant differences
between community proﬁles when analyzing Hellinger-transformed variables and
principal coordinates of J accard distance, but not relative peak height variables.
4. Multi-Regional Soil Samples

No errors were observed in cluster analysis of replicate proﬁles from the multi-
regional set of soil samples using any method of processing and analysis (data not
shown). This result was consistent for RsaI, MspI, and HhaI digests. In general, two
groups were required to explain 50 percent of the variance in the dataset. This grouping

divided all replicate proﬁles of one sample from the eight replicate proﬁles of the

25

 

remaining two samples. Three groups explained 75 to 85 percent of the total variance in

the dataset, with each group being made up by all of one sample’s replicate profiles.
Comparison of Statistical Methods

The number of errors found in cluster dendrograms was dependent on the set of
samples being analyzed. The error rate was lower using UPGMA compared to Ward’s
method if some proﬁles had cumulative peak height less than 10,000 ﬂuorescence units
(see Table 1). However, the two methods had essentially equivalent error rates if all
samples had cumulative peak heights greater than 10,000 (see Table 3). UPGMA
clustering was more true to the original distances between samples, as indicated by higher
cophenetic correlation (Jobson 1992). This also leads, however, to an increase in the
number of groups required to explain 50 percent of the total variance in the datasets, and
a greater inﬂuence of outliers on dendrogram topology (see Figures 3 and 4).

Redundancy analysis was able to detect signiﬁcant differences between samples,
and as a corollary significant similarities between analytical replicates, for all datasets
tested. The cause of signiﬁcant differences can be assessed using plots of canonical
principal components (see Figure 5), although it should be noted that these plots are
biased to show group differences, compared to standard principal components analysis,
due to the nature of redundancy analysis.
Proﬁle Alignment and Peak Height Baseline

Alignment of proﬁles was shown to be best performed manually, aided by
patterns within the peak positions as well as the putative fragment sizes calculated using
the internal lane standard. Deletion of the smallest peaks using any of a variety of

algorithms was shown to have relatively little effect on the error rate of analyses, except

26

 

that an increase in the number of errors was observed where larger peaks started to be

deleted as well (i.e. peaks with heights between 100 and 200 ﬂuorescence units).
Comparison of Community Distance Metrics

Cluster analyses based on Euclidean distance calculated from raw peak heights
resulted in an unacceptable number of errors for all sets of samples except the most
divergent (for example, compare ﬁgures 2 and 3). The Hellinger transformation had
inconsistent results on the number of clustering errors compared to clustering based on
relative peak height, sometimes decreasing and sometimes increasing the error rate. The
Hellinger transformation did consistently result in a greater number of groups being
required to account for 50 percent of the variance in the dataset, although other changes
in the topologies of the dendrograms were minor. However, the Hellinger transformation
also consistently resulted in a lower p-value when redundancy analysis was performed to
test the signiﬁcance of the differences between samples. This reduction in p-value was
dramatic for the set of closely-spaced alfalfa soil samples. Other data transformations
examined by Legendre and Gallagher (in press) were either discarded in preliminary
analyses (i.e. the chord distance) or because of the heavy weighting of rare T-RFs (e.g.
the chi-square distance). The latter property makes a distance metric inappropriate for T-
RFLP data since rarity of a T-RF might not reﬂect rarity of the associated genotype, but
simply sampling variability in detection of less-abundant (i.e. low population size or
mean peak heights), but still common, species (see Legendre and Gallagher in press, and
Legendre and Legendre 1998 for discussions on weighting of rare species).

Use of J accard distance when some proﬁles had low cumulative peak height

(<10,000 ﬂuorescence units) resulted in a large number of errors in cluster analysis and a

27

 

higher p-value in redundancy analysis. When all proﬁles analyzed had cumulative peak

height >10,000, clustering using Jaccard distance was as good as relative or Hellinger-
transforrned peak height, and redundancy analysis was more sensitive (with a lower p-

value).

Discussion

A primary weakness in many microbial community analysis studies is the lack of
quantitative statistical comparison of community proﬁles. This may be appropriate if the
methodology used only allows for qualitative comparisons, but in such cases conclusions
may be affected by subjective interpretation. This study is designed to aid in the choice
of methods to process and analyze quantitative T-RFLP data. The results may also be
useful for other methods of community analysis such as DGGE or RISA. The results are
based on analytical replicates with the goal of ﬁnding a method of data analysis such that
further analyses will not need to be replicated at the analytical level. Analytical
replication may be prohibitive in molecular work if there are a large number of samples
due to the relatively high cost and time involved. Analysis of treatment or ﬁeld replicates
is always a necessity.

The results of the analyses are different, in absolute terms, for each of the sets of
samples examined because the sets represent a wide range in degree of sample
divergence. The trends in sensitivity of data analysis methods are consistent across this
gradient and resulted in some methods being clearly preferable. As expected, the multi-
regional set of soil samples resulted in the most divergent set of T-RFLP proﬁles, and the

bioreactor and alfalfa soil samples resulted in the least divergent T-RFLP proﬁles. This

28

 

gradient was useful in showing the relative sensitivities of the various methods to the

analytical variability present in T-RFLP proﬁles.

Ward’s method of hierarchical cluster analysis seems to sacriﬁce some precision
in clustering for the ability to more efficiently identify major groups within T-RFLP
datasets. This efﬁciency results in Ward’s analyses accounting for a larger proportion of
the total variance with fewer groups compared to UPGMA. Also the scale of the
dendrogram plot is more heterogeneous across different levels in the hierarchy (see
Figures 3 and 4), resulting in the ability to more easily choose the number of major
groups in the dendrogram. Dendrograms constructed from UPGMA analyses could be
plotted using the semi-partial r-squared scale (and perhaps they should be when looking
for natural groups), but inversions will be present. The UPGMA method may be
considered a more conservative method of ﬁnding natural groups and outliers within sets
of T-RFLP proﬁles, which is in agreement with comments by Jobson (1992).

The problems associated with choosing the number of important groups that are
present when comparing dendrograms with differing scales were avoided in the current
study by examining whatever number of groups was required to account for 50 percent of
the total variance. This was done in order to compare error rates of dendrograms on an
equal basis, but is not recommended for applications of cluster analysis other than
assessing error rates since the 50 percent level may not correspond to a biologically
meaningful number of groups.

Redundancy analysis and cluster analysis are fundamentally different types of
methods, and are hence not directly comparable. The goal of redundancy analysis is not

to summarize all variability within the dataset, but to explicitly test whether that

29

 

variability that can be attributed to the differences between experimental groups is

signiﬁcant. In that case, it would be expected that redundancy analysis detected
signiﬁcant differences between samples in the less divergent sample sets, while cluster
analysis often failed to show them, because these differences accounted for only 25-40%
of the variance in the datasets.

The failure of our T-RFLP proﬁle alignment algorithm compared to alignment by
hand was a disappointment. Future work should focus on development of computer
algorithms which can take these patterns into account, since at this point manual
alignment of proﬁles represents the least standardized or objective step of quantitative T-
RFLP analysis.

The fact that deletion of the smallest peaks had relatively little effect on
dendrogram error rate is somewhat surprising because small peaks tend to be the
“noisiest”, with inconsistent presence in replicates of a single sample. An effect of
deletion of small peaks was particularly expected to occur for the analysis of J accard
distance since, in that case, small peaks are given equal weight to large peaks. However,
J accard distance does weight “common” (or frequently-present) T-RFs more than “rare”
(infrequently-present) T-RFs, and noisy peaks are by definition rare. These results imply
there were a number of T-RFs with generally low peak heights that were important for
distinguishing between samples.

Problems associated with Euclidean distance between proﬁles (raw peak height)
are likely associated with variability in the amount of DNA loaded into each gel lane.
This can result in variability in absolute peak height and presence of some of the smaller

peaks. Analysis of relative peak height led, in general, to optimal performance in cluster

30

 

analysis (equal to Hellinger distance), but not redundancy analysis. Use of the Hellinger

distance for T-RFLP data is superior based both on the present empirical results and the
theoretical considerations previously discussed (Legendre and Galagher in press).

J accard distance performed well in cluster analysis and superior to other distance metrics
in redundancy analysis only when DNA load was uniformly high. The difference in
performance with changes in cumulative peak height may be due to increased stability of

small peaks when DNA load is high.

Conclusions

T-RFLP has been shown to be a valuable method of assaying microbial
community structure. Given statistical analyses that were sensitive enough (low
probabilities of type II error), it was possible to use T-RFLP to reject the null hypotheses
that communities were identical in replicate bioreactors, or in soil samples collected
within two meters of each other. With this level of sensitivity, the utility of T-RFLP in
quantitative comparison of microbial communities is obvious. If the experimental design
is such that appropriate hypotheses can be formulated, redundancy analysis of Hellinger-
transformed peak height and/or distance-based redundancy analysis of J accard distance
(if all profiles have a cumulative peak height greater than 10,000) are recommended as
the most sensitive methods to distinguish between groups of profiles. If the goal of
analysis is exploratory data analysis, clustering using both Ward’s method, to ﬁnd natural
groups, and UPGMA, to identify potential outliers, is recommended. The validity of
clustering results are basically equivalent using relative peak height, Hellinger-

transforrned peak height, or J accard distance (if all samples have cumulative peak height

31

 

 

 

greater than 10,000). This study was not an exhaustive examination of all multivariate
statistical methods that could be used for T-RFLP data. Future work could examine the
potential of other distance metrics and other methods of data analysis, such as
correspondence analysis, principal coordinates plots, and use of artiﬁcial neural
networks. The use of quantitative statistical analysis coupled with molecular methods
creates new opportunities for addressing applied and ecological problems in microbial

community analysis.

32

 

 

 

References

Amann, R.I., W. Ludwig, K. Schleifer. 1995. Phylogenetic identiﬁcation and in situ
detection of individual microbial cells without cultivation. Microbiological Reviews
59:143-169.

Bernhard, A.E., K.G. Field. 2000. Identiﬁcation of nonpoint sources of fecal pollution in
coastal waters by using host-speciﬁc 16S ribosomal DNA genetic markers from fecal
anaerobes. Applied and Environmental Microbiology 66: 1587-1594.

Bomeman, J ., E.W. Triplett. 1997. Molecular microbial diversity in soils from Eastern
Amazonia: evidence for unusual microorganisms and microbial population shifts
associated with deforestation. Applied and Environmental Microbiology 63:2647-2653.

Bruce, K.D. 1997. Analysis of mer gene subclasses wthin bacterial communities in soils
and sediments resolved by ﬂuorescent-PCR-restriction fragment length polymorphism
proﬁling. Applied and Environmental Microbiology 63:4914-4919.

Chin, K., T. Lukow, R. Conrad. 1999. Effect of temperature on structure and function of
the methanogenic archaeal community in an anoxic rice ﬁeld soil. Applied and
Environmental Microbiology 65:2341-2349.

Clement, B.G., L.E. Kehl, K.L. DeBord, C.L. Kitts. 1998. Terminal restriction fragment
patterns (TRFPs), a rapid, PCR-based method for the comparison of complex bacterial
communities. Journal of Microbiological Methods 31: 135- 142.

Derakshani, M., T. Lukow, W. Liesack. 2001. Novel bacterial lineages at the
(sub)division level as detected by signature nucleotide-targeted recovery of 16S rRNA
genes from bulk soil and rice roots of ﬂooded rice microcosms. Applied and
Environmental Microbiology 67:623-631.

Dunbar, J ., L.O. Ticknor, C.R. Kuske. 2000. Assessment of microbial diversity in four
southwestern United States soils by 16S rRNA gene terminal restriction fragment
analysis. Applied and Environmental Microbiology 66:2943-2950.

Dunbar, J ., L.O. Ticknor, C.R. Kuske. 2001. Phylogenetic speciﬁcity and reproducibility
and new method for analysis of terminal restriction fragment proﬁles of 16S rRNA genes
from bacterial communities. Applied and Environmental Microbiology 67: 190-197.

Flynn, S.J., F.E. Loffler, J .M. Tiedje. 2000. Microbial community changes associated
with a shift from reductive dechlorination of PCE to reductive dechlorination of cis-DCE
and VC. Environmental Science and Technology 34: 1056-1061 .

Franklin, F.B., J .L. Garland, C.H. Bolster, A.L. Mills. 2001. Impact of dilution of
microbial community structure and functional potential: comparison of numerical

33

 

 

simulations and batch culture experiments. Applied and Environmental Micorbiology
67:702-712.

Gonzalez, J .M., R. Simo, R. Massana, J .S. Covert, E.O. Casamayor, C. Pedros-Alio,
M.A. Moran. 2000. Bacterial community structure associated with a
dimethylsulfonioproprionate-producing North Alantic algal bloom. Applied and
Environmental Microbiology 66:4237-4246.

Head, I.M., J .R. Saunders, R.W. Pickup. 1998. Microbial evolution, diversity, and
ecology: a decade of ribosomal RNA analysis of uncultivated microorganisms. Microbial
Ecology 35:1-21.

Hedrick, D.B., A. Peacock, J .R. Stephen, S.J. Macnaughton, J. Briiggman, D.C. White.
2000. Measuring soil microbial community diversity using polar lipid fatty acid and

denaturing gradient gel electrophoresis data. Journal of Microbiological Methods 41 :235-
248.

Hugenholtz, P., B.M. Goebel, N.R. Pace. 1998. Impact of culture-independent studies on

the emerging phylogenetic view of bacterial diversity. Journal of Bacteriology 180:4765-
4774.

J obson, J .D. 1992. Applied Multivariate Data Analysis Volume II: Categorical and
Multivariate Methods. New York: Springer-Verlag.

Jones, M.E., R.R. Harwood, N.C. Dehne, J. Smeenk, B. Parker. 1998. Enhancing soil
nitrogen mineralization and corn yield with overseeded cover crops. Soil and Water
Conservation 53:245-249.

Kerkhof, L., M. Santoro, J. Garland. 2000. Response of soybean rhizosphere
communities to human hygiene water addition as determined by community level
physiological proﬁling (CLPP) and terminal restriction fragment length polymorphism
(T-RFLP) analysis. FEMS Microbiology Letters 184:95-101.

Krzanowski, W.J., F.H.C. Marriott. 1994. Multivariate Analysis Part 2: Classiﬁcation,
Covariance Structures and Repeated Measurements. London: Arnold.

Kuske, C.R., S.M. Barns, J .D. Busch. 1997. Diverse uncultivated bacterial groups from
soils of the arid southwestern United States that are present in many geographic regions.
Applied and Environmental Microbiology 63:3614-3621.

Legendre, P., MJ. Anderson. 1999. Distance-based redundancy analysis: testing
multispecies responses in multifactorial ecological experiments. Ecological Monographs
69: 1-24.

Legendre, P., E.D. Gallagher. in press. Ecologically meaningful transformations for
ordination of species data. Oecologia.

34

 

Legendre, P., L. Legendre. 1998. Numerical Ecology. 2 ed. Amsterdam: Elsevier.

Leser, T.D., R.H. Lindecrona, T.K. Jensen, B.B. Jensen, K. Moller. 2000. Changes in
bacterial community structure in the colon of pigs fed different experimental diets and

after infection with Brachyspira hyodysenteriae. Applied and Environmental
Microbiology 66:3290—3296.

Liu, W.-T., T.L. Marsh, H. Cheng, L.J. Fomey. 1997. Characterization of microbial
diversity by determining terminal restriction fragment length polymorphisms of genes
encoding 1S rRNA. Applied and Environmental Microbiology 63:4516-4522.

Liidemann, H., I. Arth, W. Liesack. 2000. Spatial changes in the bacterial community
structure along a vertical oxygen gradient in ﬂooded paddy soil cores. Applied and
Environmental Microbiology 66:754-762.

Lukow, T., P.F. Dunﬁeld, W. Liesack. 2000. Use of the T-RFLP technique to assess
spatial and temporal changes in the bacterial community structure within an agricultural

soil planted with transgenic and non-transgenic potato plants. FEMS Microbiology
Ecology 32:241-247.

Marsh, T.L., W. Liu, L.J. Fomey, H. Cheng. 1998. Beginning a molecular analysis of
eukaryal community in activated sludge. Water Science and Technology 37:455-460.

Marsh, T.L., P. Saxman, J. Cole, J. Tiedje. 2000. Terminal restriction fragment length
polymorphism analysis program, a web-based research tool for microbial community
analysis. Applied and Environmental Microbiology 66:3616-3620.

Massol-Deya, A.A., D.A. Odelson, R.F. Hickey, J .M. Tiedje. 1995. Bacterial community
ﬁngerprinting of ampliﬁed 16S and 16-23S ribosomal DNA gene sequences and
restriction endonuclease analysis (ARDRA). In Molecular Microbial Ecology Manual,
edited by A. D. L. Akkermans, J. D. V. Elsas and F. J. D. Bruin. Dordrecht: Kluwer
Academic Publishers.

Moeseneder, M.M., J .M. Arrieta, G. Muyzer, C. Winter, 6.]. Hemdl. 1999. Optimization
of terminal-restriction fragment length polymorphism analysis for complex marine
bacterioplankton communities and comparison with denaturing gradient gel
electrophoresis. Applied and Environmental Microbiology 65:3518-3525.

Muyzer, G.A., E.C.d. Waal, A.G. Uitterlinden. 1993. Proﬁling of complex microbial
populations by denaturing gradient gel electrophoresis analysis of polymerase chain

reaction-ampliﬁed genes coding for 16S rRNA. Applied and Environmental
Microbiology 59:695-700.

35

 

 

Nakatsu, C.H., V. Torsvik, L. ereas. 2000. Soil community analysis using DGGE of
16S rDNA polymerase chain reaction products. Soil Science Society of America Journal
64: 1382-1388.

Osborn, A.M., E.R.B. Moore, K.N. Timmis. 2000. An evaluation of terminal-restriction
fragment length polymorphism (T -RFLP) analysis for the study of microbial structure
and dynamics. Environmental Microbiology 2:39-50.

Ramakrishnan, B., T. Lueders, R. Conrad, M. Friedrich. 2000. Effect of soil aggregate
size on methanogenesis and archaeal community structure in anoxic rice ﬁeld soil. FEMS
Microbiology Ecology 32:261-270.

Scala, D.J., L.J. Kerkhof. 2000. Horizontal heterogeneity of denitrifying bacterial
communities in marine sediments by terminal restriction fragment length polymorphism
analysis. Applied and Environmental Microbiology 66:1980—1986.

Simon, L., R.C. Lévesque, M. Lalonde. 1993. Identiﬁcation of endomycorrhizal fungi
colonizing roots by ﬂuorescent single-strand conformation polymorphism-polymerase
chain reaction. Applied and Environmental Microbiology 59:421 1-4215.

Tiedje, J .M., S. Asuming-Brempong, K. Nﬁsslein, T.L. Marsh, 8.]. Flynn. 1999. Opening
the black box of soil microbial diversity. Applied Soil Ecology 13:109-122.

Torsvik, V., J. Goksoyr, F.L. Daae. 1990. High diversity of DNA of soil bacteria. Applied
and Environmental Microbiology 56:782-787.

Ward, D.M., M.M. Bateson, R. Weller, A.L. Ruff-Roberts. 1992. Ribosomal RNA
analysis of microorganisms as they occur in nature. In Advances in Microbial Ecology,
edited by K. C. Marshall. New York: Plenum Press.

Woese, CR. 1987. Bacterial evolution. Microbiological Reviews 51:221-271.

36

 

 

 

 

 

 

Table l: Dendrogram characteristics for cluster analyses of manually-aligned T-RFLP
proﬁles of KBS soil fractions using a variety of data processing methods and clustering
algorithms. All proﬁles included in analysis.

 

 

 

 

 

 

Ward’s UPGMA
Variable Baseline Errors Errors
out of 32 # Groups out of 32 # Groups
50 14 6 12 9
Height 100 14 6 12 9
200 18 6 12 9
50 2 6 1 10
100 2 6 1 8
Relative Height 200 3 6 5 7
Rarefraction 2 6 1 9
1% 2 6 1 10
50 2 11 1 13
100 2 1 1 1 13
Hellinger-Transformed 200 7 1 1 5 12
Rarefraction 4 11 2 12
1% 3 11 2 12
50 6 6 5 8
100 11 5 5 7
J accard distance 200 8 4 9 4
Rarefraction 7 7 3 9
1% 5 7 4 7

 

37

   

 

,-,,__——‘—-

 

 

 

Table 2: Dendrogram characteristics for cluster analyses of computer-aligned T-RFLP
proﬁles of KBS soil fractions using a variety of data processing methods.

 

 

 

 

 

. . . Errors
Variable Baseline Maximum Range Mean Range out of 32
All Samples
Height 50 2 1.1 15
50 2 1.1 5
1% 2 1.1 5
. . 1% 3 1.5 11
Relative Height 1% 4 2.1
1% 7 4 10
1% 10 5.8 12
50 2 1.1
1% 1 0.5 11
1% 2 1.1 6
J accard distance 1% 3 1.5 10
1% 4 2.1 8
1% 7 4 14
1% 10 5.8 14
Samples with cumulative peak height > 10,000
1% 2 1.1 2
Relative Height 1% 3 1.5 6
1% 4 2 7

 

38

 

 

 

 

 

 

 

_ wd a L «DC w _ mend 85:6 Emouﬂ.
:3 a o who 0 o on 8:ng E82
£3 2 _ who 2 o a? uoéoacﬁéwéam
”no N. _ and o. c on BEe§E%a=£
Rd w o Gd 5 _ and ease cease
Rd w o Gd 5 _ on swam 2:33

5:28.80 8:80 u an .8 So coca—88o 8:80 u mm «o So

ozone—300 mcobm 0:05:80 88cm 05—31mm 2255;

<22: pea?

 

29:28 E @3205 25.2 A we Ems: xaom 332255 a 5:5 moan i=0 .mﬁﬁtowﬁ 9583—0 was $052: wEmmoooa 53 no
bomb; a was: £838: :8 mmvm do 3an man—m? 8:33-3—«555 do mama—mam Loan—o .8 85338820 E83659 ”m 03.;

39

 

 

 

 

 

 

 

bad v c wad v m Engaged 853% 9863.
3d v w End v a OS 8536 2803.
had m w _ wd m .2 eosoﬁaemm doctotcmb.._om:=_om
had m w dwd m 2 d2 doctodwcabdowcﬁom
wad m 3 mad v I couoacoam 2E0: gun—om
mad m 3 Rd m 2 dd— Ewﬁm 9:33“
ad 4 e a; v 2 8. imam
55%.: dosh—«5w 8E9:— uasm can .134— .132 .5.“ mucmsﬁm
vad m.m as mad m.m ms ectomcosm 853% €892
wad Em md wad Wm ha dd_ 853% 9862.
dad v a mwd Nam Em cocombucmm danceameabeowE—E:
dad ad a Nwd v a Q: Betcha—82-83503
wad Em 5a and Wm md :osomcocmm Emmi age—om
3o 2 3 23 S 2 8. 23.5 2:23.
mad m.m md_ :wd m.m Ea d2 Emmi
ESP—anon comm—«.3 859:— ~u=l 93 Jam: 4de new 8535..“ anaemia.
core—oboe £580 a E do So eons—286 3:80 a 3 do So
6:28.300 muotm ococozaou mcotm 25on 29mg;
<20n5 £53

 

.mEEtomE 9:33.20 98 3052:

wEmmouoE 8% a0 but? a mean 838% 883.53 do moan Add? “8 month—ace .5326 .8 85:883.? Emdwedcoﬂ ”v ozah.

4O

 

 

#.*-_7

.¢-.¢\d‘-:~=

> .12.< :3: .L.‘

 

 

58d 8 emm 86 m a mod 4 e 008:6 e82:

 

maddd a. _ aa _ m mad m .1 end m _m dogotcabdowE—E:

va_ _d d; oadm dad v 2 mwd m m _ 2303 038.0%
:28 8.80 0 So :28 2.80 o So

023A. ”— Bums—Sm 8:885 oichnaou 8:80 # deEm osachsaou 8:20 a Omar—m 038:;

 

mama—£2 8:30:28”— <20n5 £08?
.8388 :8 £3.20 do 8m 05 8a 338.— 838“ 8:30:88 new 82808820 Same—Eva 6 035,—.

 

 

 

 

 

mdddd cg oadm 858:0 E803.

dedd M: dawn doanmcmbaomeﬁom Has: 43% 4%:
good dd oaomu Emmi 0320M

mdddd mg oaam 858% E802

Good Nd camm coccemeabeowE—E: =2:
mmmdd Nd camm 2E0: 038.0%

emddd m4 sag 8.83:0 E802.

$43 2 $8 88885858: as:
m~a_ d m; oamm £303 03.30%

a_ 5d ed oamm 853% E803

mdddd ed oamm doctoamcﬁzowﬁzom Ema
:ddd w.m exude £30m 038—3—

0:_m>-m m 852me comtomoi 0335; 058d

 

.8503
“ELM—L. 29:8 88885 80258 082856 0: m_ 085 :2: £85an :2. 05 9.58 3:52 ﬂax—men 858.60% mm 035.

41

 

 

1. .Q:::..:,: ,.~

.9 :N‘h‘

 

cozuﬁu brow: - 2.22 .290 - n:

n l ‘ 4 1.1 ¢ no
noon
wooo

 

 

 

 

 

 

 

.ooo
c308“. bio: - 2.22 .200 - a:

« «<1‘ \LJIWP,.' 11‘ 4 1114,:. we
wean
wooo
”one

 

 

338m :35 - E8 55o .28 - Bo

III ‘1 I he
wean
mono
wooa

328m .85 . ES .25 .58 - 8m

«‘1 - ‘ -‘ﬁJqu¢Js‘|Juq Ho
noon
.ooo
mooa

oov on. ocv owv one com com onm can com on“ com cew wwm wow won on. o.. o~_. oo.. . ‘x .«J

 

 

 

 

 

 

 

 

 

 

V

 

V

 

 

 

 

 

 

.zwmv as 2: .3 5&3 33238 303328 £389: a 3:33“ 33 oEoE scam 56:85 85820::
2a £38 32:? .33 83 E 3:05ch mo uni m_ 33 3.858: .mguwohonmobooﬁ E? 628298 mo oEEaxm Q 95w?—

42.

 

 

.u a Jr a:
I 4W www.cthQB

 

IIWUBILO‘

 

 

.m_.o

wol-ILQLU-N'

.oué

 

 

 

 

.mN.o

:03.» 5 ﬂ EQKQR downs—#96 803 30:53

82: 3 328 2: 33on 880wa 8.: @8833 203 83032 :25» .860 a Bzmmmmn v.95 oocutg 38 2: mo $3 :398

8 3:38 gnaw scam on: a 3 ©8358 oh 28 08a: 0589 05 mo «En a»: on“ 3 8&6 moaoa egg—mom .memo: #89 mm
-H 38 mo magma 5620 MESS 3 383880 828mb =8 wmvzo 8&qu ASA. teammwizusgﬁ mo $38209 ”N PEER

43

 

vo—.o

 

CIMUDOLDt

.m—.o

row.o

”DI-ILILU‘G—

-mw.o

 

 

 

 

non.o

.339 5 ﬂ E‘suﬁﬂﬁ .vomqﬁoxo 203 9828a

805 3 £28 05 3:8» Bahama 85 “.8836 203 830%»: :23 «5.8 a gamma“ 33 ooﬁtg 38 2: mo gem 5.298

8 BEES 95E scam 6:: a 3 380580 8a 23 08m: 0305 came “Eu as 0513 “ohm. 880E 382.5% .meE: 389 ”EA.
0338 .«o $9323 .5820 F983 3 383.280 338$ :8 $530 838a ASL. wonmﬁéﬁncaﬁ mo Suhmoécom ”n van—mam

 

 

000300: U-DIUDL.

 

.09.—

 

-m~.—

 

 

¢>ULOGO O-IOQCUU

 

tom.—

:38 E a mgnmwxﬁﬁ .uowcﬁoxo 803 80:63

805 an $28 2: $3on “cohobmu BE “.2836 203 333%: no; .830 a 3:93“ 33 025.3» :38 2: mo $8 5398

8 3:38 .3on comm on: a 3 380950 03 was 0:8: 058m 2: mo 3mm“. as 2: .8 3th $305 Snowmen .mﬁwmoa axon ”EA.
03:28 mo $9493 .6320 420% 3 383.338 228.3 :8 mg mo 8an mummé vocmﬁ$=a=§ mo Swamp—ESQ a. 959m

45

 

 

 

  
 

‘2!!\'!‘

 

Figure 5: Canonical principal components plots derived from redundancy analysis of

bioreactor sample T-RFLP proﬁles (RsaI, MspI, and HhaI proﬁles pooled).
A = bioreactor sample 27; O = bioreactor sample 12; I = bioreactor sample 19

Canonical P02

Canonical PC2

Canonical P02

 

 

 

 

 

 

 

 

 

 

 

 

46

1.0
Relative Peak Height
0.5 « .
I I
0.0 « ‘
A
o
A o
-o.s - ‘
-1.0 . . . 1 .
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Canonical P01
1.0
Hellinger-transformed Peak Height
0.5 «
. 'L
0.0 - A C.
A
o
o
-0.5 1 ‘
A
-1.0 r T , r v
~1.5 -1.0 0.5 0.0 0.5 1.0 1.5
Canonical P01
1.0
Jaccard's Coefficient
A
0.5 4 ‘
A
A
0.0 ~ ~
0.
I
-0.5 ~ ‘
-1.0 v 1 v . .
-1.5 ~1.0 «0.5 0.0 0.5 1.0 1.5
Canonical P01

  

In".Mowutﬂuhuuurgogi .

 

 

Chapter 3
Eubacterial Community Structure and Population Size within

the Soil Light Fraction, Rhizosphere, and Heavy Fraction of
Several Agricultural Cropping Systems

47

 

 

 

 

 

Abstract

Soil fractions were hypothesized to be distinct microbial habitats on the basis of
differences in the age and types of organic matter they contain. This was tested in several
agronomic cropping systems by comparison of the eubacterial community composition
and numbers of bacterial cells present. Terminal restriction fragment length
polymorphism (T-RFLP) proﬁles were significantly affected by both soil fraction and
cropping system, accounting for 35-50% of the variability in the proﬁles. There was a
major difference between heavy fraction of soil, which includes the mineral particles and
associated humiﬁed organic matter, and the rhizosphere and light fraction/shoot residue
in soil, which includes soil organic matter particles. Differences were not based on
organic carbon content of fractions alone, however, since T-RFLP proﬁles were also
signiﬁcantly differentiated by cropping system and by rhizosphere versus light
fraction/shoot residue. Heavy fraction communities were found have the least amount of
random variability in T-RFLP proﬁles, resulting in the clearest cropping system effects,
while rhizosphere communities were the most variable. Proﬁles from organically-
managed corn soil were more variable than for either conventionally-managed corn or
alfalfa. The percentage of cells >O.l8 um3 was also better explained by treatment effects
than by organic carbon content of samples. In contrast, total number of bacterial cells/g
fraction was not explained better by treatment differences than by carbon content of
samples. The results show that habitat diversity in soil, related both to the amounts and
types of organic matter, as well as other potential factors, may be important in

maintaining the high soil bacterial species diversity and evenness that is found in soil.

48

 

 

Introduction

Particles of organic matter isolated by ﬂotation in liquids of varying density are
known as light fraction (LF) and have been shown to be useful predictors of soil
respiration and nitrogen mineralization (Biederbeck et al. 1994, Hassink 1995, Janzen et
al. 1992). Light fraction is sensitive to soil management regimes (Bremer et al. 1994,
Cambardella and Elliott 1993) and has been proposed as an important component of soil
quality (Gregorich et al. 1994, Yakovchenko et al. 1998). It has also been shown to be
critical in the formation and stabilization of soil aggregates (Golchin et al. 1994a). The
LP is composed of recently deposited organic matter particles (primarily plant residues),
with greater rates of turnover than other soil organic matter fractions (Buyanovsky et al.
1994, Gregorich et al. 1995), higher carbohydrate contents, and higher C:N ratios
(Christensen 1996). This is particularly true for the “free” LF isolated from outside of
soil aggregates (Golchin et al. 1994b, 1995).

The presence of a spatially-distinct component of soil with easily-utilizable
substrates raises the question of whether this fraction is colonized and inhabited by a
community of microorganisms different from other soil fractions. Such habitat diversity
in soil could explain the high microbial species richness and evenness (or non-dominance
by any species) detected by ribosomal sequencing (Bomeman and Triplett 1997, Nakatsu
et al. 2000, Niisslein and Tiedje 1998) and DNA rehybridization rate studies (Sandaa et
al. 1999, Torsvik et al. 1990). Competitive exclusion will not occur if species remain in
different habitat types and do not interact (Huston 1999). When competing species do

colonize the same habitats, patchiness of the habitat can allow for the coexistence of the

49

 

species through differential dispersal ability (fugitive species), or independently spatially-
aggregated distributions coupled with limited dispersal (Hanski 1995, Pacala and Levin
1997). Plant residues are known to be patchily distributed in soil (Staricka et al. 1991).
Differing limiting resources in patches can lead to differences in local competitive
equilibria, resulting in coexistence of competing species by spatial resource partitioning
(Tilman 1982). Nitrogen limits microbial growth in LP, resulting in N immobilization
during LF decomposition (J anzen et al. 1992, Yakovchenko et al. 1998), whereas
available C generally limits microbial growth in bulk soil (Smith and Paul 1990). These
mechanisms of enhancing species coexistence are not mutually-exclusive. All of them,
except for the spatial-aggregation hypothesis, require some difference in the ability of
species to proliferate in the LF patches compared to other soil fractions. Therefore they
are not supported if the community in LF are not signiﬁcantly different from those in
other soil fractions. If the LF microbial community is different from that in other soil
fractions, the LF represents a distinct soil habitat that should be considered to contain
potentially unique ecological dynamics.

The rhizosphere, or the region of soil adjacent to and under direct influence of
plant roots, is a well-recognized soil habitat (Bolton et a1. 1993). Enhanced microbial
growth is supported by root exudates and sloughed cells (Breland and Bakken 1991,
Rouatt et al. 1960, deerberg and Baath 1998). The rhizosphere is colonized by and
supports a community different from that in the “bulk soil” (Maloney et al. 1997,
Marilley and Aragno 1999, Ringelberg et al. 1997). Other ecological dynamics
distinctive of the rhizosphere (compared to the bulk soil) include interactions with other

plant symbionts (Budi et al. 1999, Christensen and Jakobsen 1993, Denton et al. 1999),

50

reduced allocation of C to storage molecules (Tunlid et al. 1985), enhanced predation
(Badalucco et al. 1996), increased proportion of species with copiotrophic growth
strategies (Mahaffee and Kloepper 1997), and unique spatial structure (Bowen and
Rovira 1991).

Some microbiological studies have supported the hypothesis that LP or particulate
plant residues are habitats with unique effects on the microbial community. The large
increases in the number and activity of bacterial cells in soil microsites containing
decomposing plant residues has been recognized for a long time (e. g. Thom 1935).
Kanazawa and Filip (1986) found that the LF contained 27-42% of the total numbers of
culturable microorganisms, depending on growth strategy, and 67% of the total soil ATP.
This was similar to the proportion of the total organic C stored in the LP, 42%. Ahmed
and Oades (1984) reported only 0.2-0.4% of the total soil ATP was contained in the LF,
whereas it contained 11-12% of the total soil organic C. This conﬂict between reports
remains unresolved (Gregorich and J anzen 1996). Experimental additions of plant
residue to soil microcosms has induced localized increases in dehydrogenase activity
(Gaillard et al. 1999, Ronn et al. 1996), microbial biomass (Chotte et al. 1998), soil
microfauna (Gaillard et al. 1999), and bacterial conjugation (Sengelov et a1. 2000).

We tested the hypothesis that the LF, HF, and rhizosphere are distinct soil
microbial habitats containing differing eubacterial communities. We also examined the
macroscopic shoot residue that is present in soil and is normally excluded from analysis
because it is larger than 2 mm in diameter. Microbial community analysis was conducted
using terminal restriction fragment length polymorphism (T -RFLP) of the 16S ribosomal

gene which was PCR-ampliﬁed from eubacteria. Bacterial number within soil fractions

51

 

 

 

was assessed to contribute to the issue outlined above. The responses of the communities

within fractions to differing cropping systems was examined to determine whether
communities in different micro—habitats respond differently to environmental conditions

varying at a larger scale.

Methods
Field Sites and Sample Collection

Samples were collected from two ﬁeld experiments at the W.K. Kellogg
Biological Station in Southwestern Michigan. Soils at the site are Typic Hapludalfs and
approximately 43% sand and 40% silt (Robertson et al. 1997). Two ﬁeld treatments of
the Living Field Laboratory site were sampled: conventionally-managed continuous corn
and organically-managed ﬁrst-year corn. The conventionally—managed continuous corn
receives synthetic fertilizer and pesticides on a regular basis. The organically-managed
ﬁrst-year corn is in a com-com-soybean-wheat rotation, with cover crops planted after
corn and wheat. Organically-managed corn receives compost (dairy manure and
deciduous tree leaves) at the beginning of each growing season. The experiment includes
rotation entry point plots in a randomized block design (with four replicate blocks), so
first-year corn can be sampled every year. Treatments were begun in 1993. Secondary
tillage is used to control weeds in organically-managed corn. Details of the management
of this site can be found in Jones et al. (1998). Samples were collected from within rows,
between corn plants in each of the four replicate ﬁelds per treatment.

Samples were also collected from the perennial alfalfa treatment of an adjacent

experiment, the Kellogg Biological Station Long Term Ecological Research site. Alfalfa

52

ﬁelds were established in 1989, and plants were killed every ﬁve years with herbicide
and replanted to maintain stand vigor. This occurred in the spring of 1999, between dates
sampled in this study. Alfalfa plants receive occasional lime amendments and pesticide
application (http://lter.kbs.msu.edu/Agronomics). Samples were collected in September
of 1998 and August of 1999 nearby permanent sampling locations in four of the replicate
ﬁelds. Samples were 10 cm deep. In 1998, nine replicate soil cores were collected with a
1.9 cm diameter soil probe from each ﬁeld. These were pooled in the ﬁeld and stored in
Whirlpak bags for transportation to the lab. In 1999, three 350 g soil blocks were
excavated per ﬁeld, and were transferred intact to glass jars for transportation. Samples
were stored at 4°C until fractionation was complete, which was within 3 weeks in 1998
and 8 weeks in 1999.

Soil Fractionation

Density separations are performed in liquids of two different densities in this
study, 1 and 1.7 g/cc. To distinguish between the fractions, the density at which the
fraction was isolated will follow its abbreviation; hence LF isolated using water will be
designated LF-l.

Soil cores from 1998 were mixed by hand and a whole-soil subsample was
removed for weighing and drying at 65°C. Samples were forced through a 2 mm sieve
by gently breaking aggregates along planes of weakness. Roots and shoot residue > 2
mm were removed with forceps. Layers of soil < 1 mm thickness clinging to the roots
and shoot residue were included with those fractions. Density separation was performed
using a procedure modiﬁed from Golchin et al. (1994b) to isolate free or inter-aggregate

LF. Bulk soil samples (with roots and shoot residue removed) weighing 50 to 75 g were

53

placed in sterile centrifuge bottles. Sterile water was added to a total volume of < 200
mL. Tubes were then inverted by hand 10 times and material clinging to sides of tubes
and caps was washed into the suspension with sterile water. The total volume within the
bottle was brought to 200 mL, and particles were allowed to settle 30 minutes. Bottles
were then centrifuged in a swinging-bucket centrifuge at 3000 rpm for 30 minutes. The
water and ﬂoating particles were isolated by aspiration into a ﬁltration ﬂask. This
suspension was rinsed on a 20 um sieve, and then LF-l particles were transferred to ﬁlter
paper and collected with forceps. All fractions (roots/rhizosphere, shoot residue, LF-l
and HF-l) were divided into three subsamples. These were: 1. frozen for DNA extraction
2. stored in 4.9% formaldehyde 3. weighed before and after drying at 65°C. Density
separation was also performed on bulk soil samples using a solution of sodium
polytungstate adjusted to a density of 1.7 g/mL.

Samples collected in 1999 were fractionated in essentially the same way except
that whole soil blocks were gently separated along planes of weakness over a nest of
sieves with mesh sizes of 6.3, 4, and 2 mm. Soil clods were broken down until they ﬁt
through the 6.3 mm sieve. The nest of sieves was then shaken by hand until only stable
aggregates that would not ﬁt through the mesh were left on the 4 and 2 mm sieves. This
resulted in isolation of aggregates of sizes 4-6.3, 2-4, and 0-2 mm. The 4-6.3 and 2-4 mm
macroaggregate fractions were subsampled for analysis of aggregate layers, described in
chapter 4. Density separation was performed as described above after macroaggregates
had been forced through a 2 mm sieve. This resulted in a LF-l and HF-l for each

aggregate size class. Isolation of roots/rhizosphere and shoot residue was performed

54

W
. ,_ 77,93; x
-S.

during sieving steps. Fractions and whole-soil were subsampled and stored as described
above.
Carbon and Nitrogen Analyses

Carbon and nitrogen contents of dried, ground 1998 samples were obtained using
a Carlo-Erba NAISOO series 2 Nitrogen-Carbon-Sulfur Analyzer.
Direct Microscopy

Samples preserved with formaldehyde were dispersed by blending in a Waring
blender (HF and whole soil), or by vortexing a sample diluted with 5 mL of water for 5
minutes with approximately 1 mL of 1 mm diameter glass beads (LF. rhizosphere, shoot
residue). Bacterial cell numbers were determined microscopically in samples from 1998
following Paul et al. (1999). Brieﬂy, 4 uL of diluted fraction was placed in each of ﬁve 6
mm diameter wells of an analytical microscope slide (Cel-Line Associates, Newﬁeld, NJ)
and allowed to dry overnight. Dried sample smears were then stained with 5-(4,6-
dichlorotriazin-Z-yl) aminoﬂuoroscein (DTAF) for 40 minutes, followed by rinsing in
phosphate buffer (30 minutes X 3 rinses) and water (30 minutes). Wells were ﬂooded
with type FF immersion oil and then a coverslip was glued in place. Bacteria were
observed with a Leitz Orthoplan 2 microscope at IOOOX magniﬁcation using a 63X
objective, 10X eyepiece, and 1.6X zoom. Digital images of bacteria were obtained using
a Princeton Instruments CCD microscope camera. Cells were counted and measured by a
script written in the image analysis program IPLab (Princeton Instruments, Trenton, NJ).

Calculations of cell biovolume were based on formulas in Paul and Clark (1996).

55

T—RF LP Analysis

T-RFLP was performed essentially as described in Liu (1997) and in Chapter 2.
Community DNA was extracted from samples using the standard Ultraclean Soil DNA
extraction kit (Mo Bio Laboratories, Solana Beach, CA). Genomic DNA was found to be
of sufﬁcient purity to be used directly in PCR reactions. PCR was performed using a
standard reaction mixture of 160 M of each deoxynucleoside triphosphate, 3 mM
MgC12, 0.05 U/uL Taq DNA polymerase and the appropriate volume of accompanying
10X PCR buffer (Gibco BRL, Gaithersburg, MD), and 0.2 ug/mL bovine serum albumin
(Boehringer Mannheim Biochemicals, Indianapolis, IN). PCR mastermix, without
primers, and PCR reaction tubes were sterilized for 14 minutes with direct ultraviolet
radiation in a Cleanspot PCR/UV workstation. Primers used were the general eubacterial
primer 8-27F (AGAGTITGATCCTGGCTCAG, E. coli numbering, Amann et al. 1995,
Integrated DNA Technologies, Coralville, IA) and the universal primer 139241406R

(ACGGGCGGTGTGTACA). PCR reactions were optimized for each sample of
genomic DNA using a master mix with primer concentrations of 0.4 uM. Optimizations
were performed by adjusting the amount of genomic DNA extract used (0.4 to 7.5 le50
[LL reaction) and the number of PCR cycles run (22 or 28) to obtain a strong band without
visible non-speciﬁc product. PCR was performed in a Perkin-Elmer 9600 thermocycler

using an initial denaturation step of 95°C followed by 22-28 cycles of the following
program: denaturation at 94°C for 30 sec., primer annealing at 55°C for 30 sec., and
extension at 72°C for 30 sec. A modiﬁed hot start procedure was used where PCR tubes
were not placed in the thermocycler until the block temperature had reached 80°C. A

ﬁnal extension at 72°C for 7 min. was performed after the programmed number of cycles

56

was complete. PCR product concentration and speciﬁcity was checked by
electrophoresis on a 1% agarose gel, followed by staining with ethidium bromide.

PCR reactions (50-75 11L) were performed in triplicate for each sample using the
optimal conditions found previously. These reactions were performed using the same
PCR master mix and program described above except that the forward primer was 0.6
uM hexachloroﬂuorscein (hex)-labeled 8-27F (Integrated DNA Technologies). PCR
replicates were pooled and puriﬁed using the Promega PCR Preps Wizard Kit as directed
by the supplier, except that elution was performed with 19 [AL of sterile water heated to
55-65°C. Five uL of puriﬁed PCR product was mixed with 5 [LL of restriction enzyme
master mix containing 1.5 U/uL of restriction enzyme and one uL of the accompanying
reaction buffer (Gibco). Restriction reactions were incubated for three hours at 37°C,
followed by 16 min. at 65°C to denature the restriction enzyme. Three til. of the

restricted PCR product was mixed with one [.11. of 2500 TAMRA size standard (Applied
Biosystems Instruments, Foster City, CA). DNA fragments were separated by size by
electrophoresis at 1800 V for 14 hours on an ABI 373 automated DNA sequencer at
Michigan State University’s DNA Sequencing Facility. The 5’ terminal fragments (T -
RFs) were visualized by excitation of the hex molecule attached to the forward primer.
The gel image was captured and analyzed using Genescan Analysis Software 3.1. A peak
height threshold of 50 ﬂuorescence units was used in the initial analysis of the
electropherogram. Negative controls (no genomic DNA) were conducted with every
PCR and run on several Genescan gels. Contamination in PCR reactions was not

detected. Small peaks occasionally appeared in negative control lanes on Genescan gels,

57

, L

 

 

but the cumulative peak height was always below 1000 units. Samples were re-run if the
cumulative peak height was below 9500 ﬂuorescence units.

T-RFLP proﬁles for all samples were generated using the restriction enzyme RsaI.
MspI was used to generate additional proﬁles for replicates 1,2, and 3 of 1998
rhizosphere, LF-l, and I-IF-l, and in 1999 for all rhizosphere replicates and replicates 3
and 4 of 0-2 mm and 4-6.3 mm aggregate LF—l and HF-l. T-RFs of sizes between 50
and 500 bp were aligned against a previously-deﬁned database with identities of samples
concealed.

Statistical Analysis

Most statistical analyses were performed using SAS Version 8 Stat and [ML
components, Sigmastat, and Excel. Redundancy analysis was performed using Canoco.
All analyses were performed taking into account blocking of replicates in the ﬁeld and
lab, which was not found to be signiﬁcant.

Percent C and N was analyzed using one-way analysis of variance (ANOVA) to
test for a signiﬁcant effect of cropping system. Total numbers of cells and percent of
cells in the largest and smallest size classes were analyzed by two-way ANOVA with
cropping system and soil fraction as factors, as well as by regression against C and N
contents of samples. The signiﬁcance of the increased ﬁt by more complex ANOVA
models compared to regression models was assessed with a partial F-test.

Relationships between T-RFLP proﬁles were examined using Hellinger distance
and Jaccard distance (see chapter 2 for a discussion of data analysis of T-RFLP proﬁles).
Differentiation of eubacterial T-RFLP proﬁles was tested using redundancy analysis (or

distance-based redundancy analysis in the case of J accard’s distance) with dummy

58

“HM I...l..l. .u .7...—

 

 

variables coding for cropping system, soil fraction, and interaction terms. Coding
followed the method described in (Legendre and Anderson 1999). Distributions of
partial pseudo-F statistics were generated with 9999 random permutations of the
identities of proﬁles in the software Canoco. To test for fraction effects in 1999, four
HF-l and LF-l proﬁles per cropping system needed to be used, instead of 12, to be
balanced with the number of root/rhizosphere and shoot residue replicates. Aggregate
size class was shown to not be a signiﬁcant inﬂuence on I-IF-l or LF—l proﬁles in this
dataset (see chapter 4), so 4-6.3 mm aggregates were randomly chosen to represent the
different aggregate size classes in 1999. A 1999 alfalfa shoot residue RsaI proﬁle and a
1999 conventional corn rhizosphere MspI proﬁle were missing due to inability to obtain
adequately strong T-RFLP proﬁles. These were replaced by the mean vector generated
from the other three replicates of the respective treatment, following the method of
Legendre and Anderson (1999). Relationships between proﬁles were also examined
using Ward’s method of hierarchical cluster analysis, principal components analysis, and
canonical principal components plots obtained from the redundancy analysis. Percent C
and N were also tested for signiﬁcant effects on T-RFLP proﬁles using redundancy
analysis.

The association of individual T-RFs with cropping systems and soil fractions was
assessed using percentages of variability explained by the different factors in redundancy
analysis, followed by examination of indicator values (IndVal) calculated according to
Dufréne and Legendre (1997). These were calculated for each T—RF-treatment
combination, with and without interaction between treatments. IndVal is equal to the

proportion of samples within the treatment where the T-RF is present times the mean

59

(Hellinger-transformed) T-RF abundance in that treatment divided by the sum of the
mean abundances across all treatments. It is maximum when all occurrences of a T-RF
are within the treatment being examined, and is sensitive to changes in both abundance
and frequency between treatments.

T-RFs that were found to be consistently affected by treatment in 1998 and 1999
were compared to sequences in the Ribosomal Database Project version 8.1

(http://rdp.cme.msu.edu) using the online TAP-TRFLP software (Marsh et al. 2000).

Results
Fraction Mass and C and N Content

The proportion of the total C and N stored in the root/rhizosphere, shoot residue
and LF-l fractions is disproportionate (1-5% in each fraction) compared to the proportion
of the total soil mass they take up (<0.5% in each fraction, see Table 1). The light
fraction isolated using liquid with a speciﬁc gravity of 1.0 g/cc (LF-l) was 5 to 10 times
less than that isolated using sodium polytungstate at a density of 1.7 g/cc (LF-l.7).
Percent organic C in LF-1.7 was slightly less than in LF-l. Cropping system had a
signiﬁcant effect on the proportion of the total soil mass, percent C and N, and proportion
of total C and N present in most soil fractions in 1998, tested using one-way AN OVA
(see Table 1). Alfalfa soil consistently contained the lowest amounts of root/rhizosphere,
shoot residue and LF-l, and conventional corn soil contained the greatest amounts.
Proportion of total soil C and N stored in these fractions followed the same trends. The
organic corn soil contained the greatest amount of LF-l .7, as well as C and N stored in

that fraction. Percent C and N within shoot residue, LF-l, and LF-1.7 increased in the

60

order of alfalfa, organic corn, and conventional corn (see Table 1). Percent C and N was
lower in conventional corn I-IF-l than in the other cropping systems. C to N ratio was
approximately 18 in all fractions except HF-l, where it was 10-11. C to N ratio did not
vary significantly due to cropping system. Proportion of total soil weight in the different
soil fractions was similar for 1999 samples (data not shown), although more variable
within cropping system treatments.
Microscopic Bacterial Cell Counts

Two-way AN OVA found that bacterial cell numbers expressed as cells/g fraction
were signiﬁcantly different between all fractions (p<0.0001), with cells/g LF-l being the
highest and cells/g HF-l being the lowest (see Figure 1). There were no signiﬁcant
effects of cropping system or interaction effects between fraction and cropping system on
cell numbers. HF-l contained 94-98% of all bacteria detected in soil, which was
calculated as the weighted sum of the bacteria detected in all soil fractions (see Table 2).
Number of cells detected in bulk soil was almost the same as that in HE], indicating that
fractionation did not result in signiﬁcant loss or destruction of bacteria.

Linear regression of cells/g fraction against percent C was signiﬁcant at the
0.0001 signiﬁcance level (R2=0.61, see Figure 2). Variability of cells/g fraction
increased with percent C, with two strong outliers present. Analysis without these
outliers did not change the signiﬁcance of the regression, but increased the R2 to 0.76.
Both outliers were alfalfa LF-l samples. A partial F-test did not ﬁnd that the AN OVA
model with signiﬁcant effects (soil fraction) ﬁt cells/g fraction significantly better than
the regression against percent C (F=1.5, d.o.f.=3,58, p=0.20). Nitrogen content of

samples had no signiﬁcant effects on any microscopic measurements.

61

L

 

Despite the overall linear relationship between number of cells per g fraction and
amount of C per g of fraction, the number of cells/pg C was signiﬁcantly affected by
fraction (p<0.0001) and fraction-cropping system interaction (p=0.0246) in two-way
ANOVA. The number of cells/11g C is signiﬁcantly greater in HF] than in other
fractions, and within HF-l the number of cells/jig C is signiﬁcantly greater in alfalfa than
in either corn system (see Table 2). This can be seen in Figure 2 as a slightly increased
rate of change of numbers of cells with C content in the HF] range of C content (0—5%)
compared to the rate of change at greater C contents.

The percentage of cells in the largest size class (>O.18 um3) was signiﬁcantly
affected by both soil fraction (p<0.0001) and cropping system (p<0.01), but there was no
interaction effect (R2=0.53, see Table 2). Alfalfa samples contained a signiﬁcantly larger
proportion of large cells than either corn system (see Table 2). Percentage large cells
increased with C content of fraction (see Figure 3), although the ﬁt of the two-way
AN OVA model was signiﬁcantly better (F=3.4, d.o.f.=5,53, p<0.01). The percentage of
cells in the smallest size class (<0.065 um3) followed the inverse of the trends in the
largest size class, except that there was no signiﬁcant cropping system effect (see Table
2). While cells in the smallest size class constituted the majority of cells in all samples
(50 to 70%), they made up only 10 to 15% of the soil microbial biomass.

Eubacterial Community T-RFLP Analysis

The effects of cropping system and soil fraction were examined in four T-RFLP

datasets: RsaI and MspI proﬁles of sets of samples collected in 1998 and 1999. There

were 120 T-RFs detected in proﬁles generated by the restriction enzyme RsaI in 1998,

62

     

and 104 in 1999. There were 124 T-RFs detected in proﬁles generated by the restriction
enzyme MspI in 1998, and 114 in 1999.

Redundancy analysis indicated that there were signiﬁcant effects of cropping
system and soil fraction on both RsaI and MspI T-RFLP proﬁles in both 1998 and 1999
(see Table 3). Interaction effects were also signiﬁcant, indicating that the effects of soil
fraction were not uniform for different cropping systems, and/or the effects of cropping
systems were not uniform for different soil fractions. Treatment effects together
accounted for 40 to 55% of the total variance in Hellinger-transformed proﬁles, and 35 to
38% of the total variance in Jaccard distances between proﬁles (see Table 3). Percent C
and N, and C to N ratio of samples accounted for a signiﬁcant amount of variation in the
Hellinger-transformed T-RFLP proﬁles (p=0.0001, 15.5%, data not shown). Effects of
nutrient concentration per se were not analyzed further because the variation between
samples in C and N contents was mainly due to fraction and cropping system differences,
and these treatments accounted for greater proportions of the total variability than did
nutrient contents.

Figure 4 shows partial canonical principal components plots derived from the
redundancy analysis of 1998 RsaI proﬁles, which are typical of all the datasets. These
plots are efﬁcient ways to examine why the redundancy analysis found overall signiﬁcant
treatment effects. In Figure 4a the effects of cropping system and the interaction between
cropping system and soil fraction are partialled out, and the separation of samples due to
soil fraction is maximized. A division between I-IF-l and other fractions is the major
cause of variability from soil fraction, and is captured in the ﬁrst canonical axis.

Rhizosphere samples are separated from LF-l and shoot residue on the second axis. In

63

Figure 4b the interaction effects have been shown in addition to the variability due to soil
fraction, with only effects of cropping system partialled out. While the same overall
topology is present, interaction effects cause alfalfa samples to separate from the other
cropping systems, and conventional corn rhizosphere samples become undifferentiated
from other com LF-l and shoot residue samples.

Figure 4c shows the variability due only to cropping system, resulting in
separation of alfalfa and conventional com samples by the ﬁrst canonical axis and
separation of organic corn samples from other cropping systems by the second axis.
Including interaction effects in the cropping systems plot had little effect on the grouping
of samples and is not shown. Including all treatment effects results in Figure 4d, where a
division between l-IF-l of alfalfa and conventional corn from the other samples is
captured on the ﬁrst axis, along with separation of organic corn LF-l, HF-l, and shoot
residue from other cropping system within each soil fraction. The second axis divides
alfalfa from conventional corn samples in LF-l, I-IF-l, and shoot residue. Note that other
divisions among the treatments that were evident in plots 4a,b,c are also included in the
ordination including all treatment effects, but are not shown because they occur in higher-
order canonical principal components. The percentage of the total variability accounted
for by individual axes in Figure 4 is relatively low because of the large number of
treatments involved in the dataset, requiring a large number of canonical principal
components to account for all the variability due to treatment effects (see Table 3).

The 1998 MspI, 1999 RsaI, and 1999 MspI clustering dendrograms had the same
general topology as the canonical ordination plots. The major division of samples

evident was between an I-IF-l group (including whole soil, analyzed in 1999) and another

L

 

 

group including LF-l , shoot residue and rhizosphere (see Figure 5). Samples from
different cropping systems are separated within each of the major soil fraction groups.
Alfalfa and conventional corn samples are always separated, with organic corn samples
sometimes forming a third group and sometimes being split between groups characterized
by the other cropping systems. Among the fractions, rhizosphere samples were most
often outliers from a cropping system group.

The exception to this pattern is the 1998 RsaI dendrograms (see Figure 6), where
alfalfa and conventional corn are divided ﬁrst, with organic corn soil samples split
between the groups. Limited grouping of soil fraction and cropping system samples
occurs in smaller clusters.

Twenty RsaI T-RFs in 1998, and 13 in 1999, had at least 20% of the variability in
their Hellinger-transformed peak heights explained by soil fraction in the redundancy
analysis, accounting for variability due to cropping system. Of these, 9 were T-RFs of
the same size, potentially representing the same organisms in both years. Six of these
were associated with the same fraction in each year, as indicated by IndVal scores (see
Table 4). Out of 22 MspI T-RFs in 1998 and 27 T-RFs in 1999 that were associated with
particular soil fractions, 3 MspI fragments were affected by soil fraction in both years in
the same way. No sequence was found in the RDP database that would produce both an
RsaI and an MspI T-RF that was found in this study to be consistently affected by soil
fraction in the same way across years. One MspI fragment of size 405408 bp that was
found to be associated predominantly with roots/rhizosphere is close in size to the MspI
fragment that would be generated from the garden pea chloroplast sequence in the

database (404 bp). Hence the 405-408 bp MspI fragment detected in the rhizosphere may

65

“.4”

wrmrf v. V't

 

 

be due to the presence of plant DNA in the genomic DNA extract from those samples.
The overall importance of any particular T-RF to the statistical tests and community
ordination described above is small, however, due to the large number of fragments that
were found to be important in separating different fractions (>20 MspI fragments in each
year). RsaI fragments generated from the garden pea chloroplast sequences are outside
the range analyzed (i.e. >500 bp).

Sixteen RsaI T-RFs in 1998, and 18 in 1999, had at least 20% of the variability in
their Hellinger-transformed peak heights explained by cropping system in the redundancy
analysis, after accounting for the variability due to soil fraction. Six of these were the
same size T-RFs and were associated with the same cropping systems across sampling
years (see Table 4). Four MspI T-RFs were associated with cropping system in the same
way both years, out of 35 in 1998 and 11 in 1999. No sequences were present in the
database that would generate both an RsaI and an MspI T-RF that was found to be
consistently affected by cropping system across years.

Twenty-four RsaI T-RFs in 1998, and 23 in 1999, had at least 20% of the
variability in their Hellinger-transformed peak heights explained by interaction terms, or
after correcting for the variability due to cropping system and soil fraction. Seven of
these T-RFs were the same size. IndVal scores were low overall for these T-RFs when
calculated for cropping system-soil fraction treatments, and the T—RFs were not strongly
associated with the same treatments in different years. The case was quite different for
MspI digests, where 28 T-RFs in 1998 and 27 in 1999 had at least 20% of the variability

in their Hellinger-transformed peak heights explained by interaction terms, or by an

66

interaction term and cropping system or soil fraction. Seven of these fragments were the

same size and were similarly affected in 1998 and 1999 (see Table 4).

Discussion

The light fraction has been shown to be a physically-deﬁned soil fraction that
harbors a eubacterial community that is distinct from that found in the majority of the
soil, as assayed by T-RFLP. The LF-l community is also different from the rhizosphere
community, although these communities are more similar to each other than to HF-l.
This is not surprising: LF-l and the rhizosphere are both characterized by high amounts
of labile C, increased bacterial growth rates and increased predator populations. They
differ by the formation of mutualistic associations between roots and certain other
organisms (e. g. mycorrhizal fungi, Rhizobium). Also the availability of complex
carbohydrates and lignin contained within plant cell walls is probably much greater in
LF, presenting very different catabolic enzyme requirements to the microbial community.
LF-l, shoot residue, and the rhizosphere had C to N ratios of approximately 18, lower
than that found in plant tissue (60-80), and higher than that in bulk soil or HF-l (10-13).
For LF-l the intermediate C to N ratio is evidence of the microbial biomass present and
the partially decomposed state of the organic particles, since all mineral soil particles are
washed away. For the rhizosphere the increased C to N ratio represents the balance
between the C to N ratios of the soil particles clinging to the roots and the roots
themselves, which were included in rhizosphere in this study.

The shoot residue > 2 mm was the soil fraction with the eubacterial community

that was most like the LF-l ’5. Shoot residue and LF-l are soil fractions that are

67

chemically and physically very similar, differing primarily by particle size and artifacts
associated with their isolation. These include time and method of isolation, and the
washing away of mineral soil particles clinging to the organic particles of LF-l but not
shoot residue. The LF-l/shoot residue effect cannot be attributed simply to the steps
involved in fractionation because the communities of these fractions are so similar.
Cropping system also had a large effect on T-RFLP proﬁles. The increased
variability of the organically-managed corn system, at the ﬁeld replicate level, was
expected since it involved a plant community that was temporally more diverse than
either of the other two cropping systems. The organically-managed ﬁelds also received
composted manure and leaves and more frequent tillage, which may affect the soil
bacterial communities. Previous work at the KBS Living Field Lab has shown that
addition of compost increases the content of particulate (>53 tun) organic matter (Wilson
et al. 2001). It is also probable that the addition of compost increases LF-l in the
organically-managed plots. This may be the cause for the differences in T-RFLP proﬁles
in organic compared to conventionally-managed soils, as well as the increased variability
in the organically-managed soil. These ﬁndings support previous reports of an effect of
cropping system or other agricultural soil management on microbial community structure
at other sites (Bossio et al. 1998, Lukow et al. 2000, Zelles et al. 1995). In contrast,
Buckley (2000) used 168 ribosomal RNA oligonucleotide probing and did not ﬁnd shifts
in the composition of the bacterial community due to agronomic treatment at the KBS
LTER site. This difference may be due to the fact that the probes used by Buckley
(2000) detect shifts in relative abundance of phylogenetically very broad groups, whereas

the T-RFLP procedure used in this study can potentially detect shifts in populations of

68

‘1.“ M'— a- .

 

individual species. No changes in nitriﬁer community composition were detected due to
agronomic treatments at the KBS LTER using DGGE of the nitriﬁer l6S ribosomal genes
(Phillips et al. 2000). Increased ectomycorrhizal fungal species diversity has been
observed in Poplar plots compared to annual crop plots (K. Kosola and EA. Paul, in
preparation). In a comparison of the more divergent environments at KBS, native never-
tilled deciduous forest and agronomic or historically-tilled plots, changes in the relative
abundance of phylogenetically broad bacterial groups and nitriﬁer species has been
detected (Buckley 2000, Bruns 1999).

The stability of grouping soil fraction samples in dendrograms and ordinations,
and of grouping cropping system within each soil fraction, decreased in the order I-[F- l,
LF-l/shoot residue, and rhizosphere. Rhizosphere samples were never differentiated as a
fraction-group in dendrograms, and were at times split between the HF-l and LF-l
groups. Rhizosphere samples frequently caused noise in the grouping of cropping
systems. This was the opposite of the expected trend, since the rhizosphere is frequently
cited as exerting a selective pressure on microbial communities. The results here indicate
that there is a group of organisms that are well—adapted to the HF-l habitat, and changes
in this habitat (e.g. due to cropping system) can result in quite stable shifts in community
composition. The temporal scale of the cropping system treatments may be important in
determining the stability of changes within fractions. In this study the treatments had
been in place 6 to 10 years, whereas studies ﬁnding greater effects of plant species on the
rhizosphere are often in place one year (e.g. Latour et al. 1996, Maloney et al. 1997).
The rhizosphere community is consistently different from other habitats, but may be

assembled more randomly, perhaps due to proliferation of randomly-encountered

69

 

dormant bacteria in HF-l that have the life history traits enabling them to respond to high

amounts of labile C. The same line of reasoning could apply to the LF-l/shoot residue
habitat, except that it displayed increased stability, and may require more specialized
organisms since the substrate is generally more complex.

Attempts to identify particular T-RFs that were found to be consistently important
in deﬁning differences between treatments were hindered by our use of very broad PCR
primers. These primers amplify the 168 gene from most eubacterial organisms (Amman
1995), resulting in an analysis that broadly assays the community for major structural
changes. The number of sequences that can produce any individual T-RF is potentially
large. We attempted to work around this problem by performing digests with two
restriction enzymes, but no sequences could be found in the database that would match
similarly-acting T-RFs from the different digests. This could be due to the importance of
organisms with no sequences currently represented in the database. It may be necessary
to use more speciﬁc primers if the goal is to obtain phylogenetic information about the
very complex communities, as has also been pointed out by Dunbar et al. (2001).

Expressed as a proportion of total numbers of cells in the soil, the bacterial
populations present in LF and rhizosphere in this study may seem low; however they are
very close to the proportion of the total amount of C found in the fractions (see Tables 1
and 2). This proportional relationship between bacterial cells and organic C is similar to
what was found by Kanazawa and Filip (1986). The relationship was further generalized
in this study because it was found that percent C could explain the majority of the

variation in cell numbers across cropping systems and fractions. The slightly increased

70

 

 

 

ratio of cells to organic C present in HF-l is probably due to the accumulation of inactive
cells, as indicated by the larger percentage of small cells in this fraction.

These results contrast with those of Ahmed and Oades (1984), who found that
0.2-0.4% of the total soil ATP and 11-12% of the total C was contained in the LF. The
liquid used for density separation by Ahmed and Oades (1984) was ZnBrz at a density of
1.6 g/cc, whereas in the present study and in Kanazawa and Filip (1986) water at a
density of 1.0 g/cc was used. Osmotic pressure or other toxicity exerted by concentrated
ZnBrz may have caused cell lysis and lead to destruction of ATP preferentially in the LP
of Ahmed and Oades (1984), while a large portion of HF cells could have been protected
inside microaggregates. Preliminary experiments with the ﬂuorescent redox indicator 5-
cyano-2,3-ditolyl tetrazolium chloride (CT C) indicated that sodium polytungstate
solution with density of 1.7 g/cc inactivated cells; we therefore recommend the use of
water for isolation of LF for microbiological work until a more complete analysis of the
effects of dense liquids on microorganisms is carried out. The dispersion methods used
by Ahmed and Oades (1984) prior to density separation were also more vigorous than
those used in this study or by Kanazawa and Filip (1986). This could have resulted in the
redistribution of bacterial cells prior to analysis. The alternative explanation is that the
different dispersion technique and density used resulted in the isolation of a soil fraction
with very different microbiological properties than what was isolated here.

Increased growth rate of bacteria in LF and decomposing shoot residue is
supported by the ﬁnding in this study that there was an increase in the percentage of large
cells in those fractions (Bath 1994), which was also explained in part by increased C

concentration. While C content explained changes in cell numbers across soil fractions

71

 

 

and cropping systems, changes in eubacterial T-RFLP profiles and in the percentage of

cells in the largest size class were not explained as well by C content of samples as by
their treatment identities. Hence while C content is important in deﬁning habitats in soil,
other factors must also play a role. These could include the quality of the organic matter,
physical disruption, or species interactions.

The hypothesis that habitat diversity enhances soil microbial diversity is
supported by the detection of unique communities in different soil fractions. Future
studies could examine the individual mechanisms of habitat diversity enhancing
microbial diversity that were outlined in the Introduction. It is necessary to learn how
microbial communities organize themselves to be able to manage the communities to the
degree that is sometimes called for (e.g. Beare 1997, Kennedy 1999, Smith and Paul
1990). If communities are organized into separate habitats, these should be taken into
consideration when looking for management effects because effects may be present in
some habitats and not others. Different habitats may also represent opportunities for
management, such as refuges from competition or locations of enhanced nutrient

availability to be exploited by beneﬁcial organisms.

72

 

 

References

Ahmed, M., J .M. Oades. 1984. Distribution of organic matter and adenosine triphosphate
after fractionation of soils by physical procedures. Soil Biology and Biochemistry 16:465-
470.

Amann, R.I., W. Ludwig, K. Schleifer. 1995. Phylogenetic identiﬁcation and in situ

detection of individual microbial cells without cultivation. Microbiological Reviews
59:143-169.

Baath, E. 1994. Thymidine and leucine incorporation in soil bacteria with different cell
size. Microbial Ecology 27:267-278.

Badalucco, L., R]. Kuikman, P. Nannipieri. 1996. Protease and dearninase activities in

wheat rhizosphere and their relation to bacterial and protozoan populations. Biology and
Fertility of Soils 23:99-104.

Beare, M.H. 1997. Fungal and bacterial pathways of organic matter decomposition and
nitrogen mineralization in arable soils. In Soil Ecology in Sustainable Agricultural
Systems, edited by L. Brussaard and R. Ferrara-Cerrato. Boca Raton: CRC Press.

Biederbeck, V.O., H.H. Janzen, C.A. Campbell, R.P. Zentner. 1994.1.abile soil organic
matter as inﬂuenced by cropping practices in an arid environment. Soil Biology and
Biochemistry 26: 1647-1656.

Bolton, H., Jr, J.K. Frederickson, L.F. Elliott. 1993. Microbial ecology of the rhizosphere.
In Soil Microbial Ecology: Applications in Agricultural and Environmental Management,
edited by F. B. Metting, J r. New York: Marcel Dekker.

Bomeman, J ., E.W. Triplett. 1997. Molecular rrricrobial diversity in soils from Eastern
Amazonia: evidence for unusual microorganisms and microbial population shifts
associated with deforestation. Applied and Environmental Microbiology 63:2647-2653.

Bossio, D.A., K.M. Scow, N. Gunapala, K.J. Graham. 1998. Determinants of soil
microbial communities: effects of agricultural management, season, and soil type on
phospholipid fatty acid proﬁles. Microbial Ecology 36: 1-12.

Bowen, G.D., A.D. Rovira. 1991. The rhizosphere: the hidden half of the hidden half. In
Plant Roots: The Hidden Half, edited by Y. Waisel, A. Eshel and U. Kafkaﬁ. New York:
Marcel Dekker.

Breland, T.A., L.R. Bakken. 1991. Microbial growth and nitrogen immobilization in the

root zone of barley (Hordeum vulgare L.), Italian ryegrass (Lolium multiﬂorum L.), and
white clover (Trifolium repens L.). Biology and F ertility of Soils 12:154-160.

73

Bremer, E., H.H. Janzen, A.M. Johnston. 1994. Sensitivity of total, light fraction and

mineralizable organic matter to management practices in a Lethbridge soil. Canadian
Journal of Soil Science 74: 131-138.

Bruns, M.A., J .R. Stephen, G.A. Kowalchuk, J.I. Prosser, E.A. Paul. 1999. Comparative
diversity of ammonia oxidizer 16S rRNA gene sequences in native, tilled, and
successional soils. Applied and Environmental Microbiology 65:2994-3000.

Buckley, D. H. 2000. The Diversity and Dynamics of Microbial Groups in Soils from
Agroecosystems. PhD Dissertation. Michigan State University, East Lasing.

Budi, S.W., D.V. Tuinen, G. Martinotti, S. Gianinazzi. 1999. Isolation from the Sorghum
bicolor mycorrhizosphere of a bacterium compatible with arbuscular mycorrhiza
development and antaganostic towards soilbome fungal pathogens. Applied and
Environmental Microbiology 65:5 148-5 150.

Buyanovsky, G.A., M. Aslam, G.H. Wagner. 1994. Carbon turnover in soil physical
fractions. Soil Science Society of America Journal 58:1167-1173.

Cambardella, C.A., E.T. Elliott. 1993. Carbon and nitrogen in aggregates from cultivated
and native grassland soils. Soil Science Society of America Journal 57:1071-1076.

Chotte, J .L., J .N. Ladd, M. Amato. 1998. Sites of microbial assimilation, and turnover of
soluble and particulate 14C-labelled substrates decomposing in a clay soil. Soil Biology
and Biochemistry 30:205-218.

Christensen, ET 1996. Carbon in primary and secondary organomineral complexes. In
Structure and Organic Matter Storage in Agricultural Soils, edited by M. R. Carter and
B. A. Stewart. Boca Raton: CRC Press.

Christensen, H., I. Jakobsen. 1993. Reduction of bacterial growth by a vesicular-
arbuscular mycorrhizal fungus in the rhizosphere of cucumber (Cucumis sativus L.).
Biology and Fertility of Soils 15:253-258.

Denton, C.S., R.D. Bardgett, R. Cook, PJ. Hobbs. 1999. Low amounts of root herbivory
positively inﬂuence the rhizosphere microbial community in a temperate grassland soil.
Soil Biology and Biochemistry 31: 155-165.

Dufréne, M., P. Legendre. 1997. Species assemblages and indicator species: the need for
a ﬂexible assymetric approach. Ecological Monographs 67:345-366.

Dunbar, J ., L.O. Ticknor, C.R. Kuske. 2001. Phylogenetic speciﬁcity and reproducibility

and new method for analysis of terminal restriction fragment proﬁles of 16S rRNA genes
from bacterial communities. Applied and Environmental Microbiology 67:190-197.

74

Gaillard, V., C. Chenu, S. Recous, G. Richard. 1999. Carbon, nitrogen and microbial
gradients induced by plant residues decomposing in soil. European Journal of Soil
Science 50:567-578.

Golchin, A., J.M. Oades, J .O. Skjemstad, P. Clarke. 1994a. Soil structure and carbon
cycling. Australian Journal of Soil Research 32: 1043-1068.

Golchin, A., J .M. Oades, J .O. Skjemstad, P. Clarke. 1994b. Study of free and occluded

particulate organic matter in soils by solid state 13C CP/MAS NMR spectroscopy and
scanning electron microscopy. Australian Journal of Soil Research 32:285-309.

Golchin, A., J.M. Oades, J.O. Skjemstad, P. Clarke. 1995. Structural and dynamic

properties of soil organic matter as reﬂected by 13

spectrometry and solid-state 13C NMR spectroscopy in density fractions of an Oxisol
under forest and pasture. Australian Journal of Soil Research 33:59-76.

C natural abundance, pyrolysis mass

Gregorich, E.G., M.R. Carter, D.A. Angers, C.M. Monreal, B.H. Ellert. 1994. Towards a
minimum data set to assess soil organic matter quality in agricultural soils. Canadian
Journal of Soil Science 74:367-385.

Gregorich, E.G., B.H. Ellert, C.M. Monreal. 1995. Turnover of soil organic matter and

storage of corn residue carbon estimated from natural 13C abundance. Canadian Journal
of Soil Science 75:161-167.

Gregorich, E.G., H.H. Janzen. 1996. Storage of soil carbon in the light fraction and
macroorganic matter. In Structure and Organic Matter Storage in Agricultural Soils,
edited by M. R. Carter and B. A. Stewart. Boca Raton: CRC Press.

Hanski, I. 1995. Effects of landscape pattern on competitive interactions. In Mosaic
Landscapes and Ecological Processes, edited by L. Hansson, L. Fahrig and G. Merriam.
London: Chapman & Hall.

Hassink, J. 1995. Density fractions of soil macroorganic matter and microbial biomass as
predictors of C and N mineralization. Soil Biology and Biochemistry 27: 1099-1 108.

Huston, M.A. 1999. Local processes and regional patterns: appropriate scales for
understanding variation in the diversity of plants and animals. Oikos 86:393-401.

Janzen, H.H., C.A. Campbell, SA. Brandt, G.P. Lafond, L. Townley-Smith. 1992. Light-
fraction organic matter in soils from long-terrn crop rotations. Soil Science Society of
America Journal 56: 1799-1806.

Jones, M.E., R.R. Harwood, N.C. Dehne, J. Smeenk, B. Parker. 1998. Enhancing soil

nitrogen mineralization and corn yield with overseeded cover crops. Soil and Water
Conservation 53:245-249.

75

 

 

Kanazawa, 8., Z. Filip. 1986. Distribution of microrganisms, total biomass, and enzyme
activities in different particles of brown soil. Microbial Ecology 12:205-215.

Kennedy, AC. 1999. Microbial diversity in agroecosystem quality. In Biodiversity in
Agroecosystems, edited by W. W. Collins and C. O. Qualset. Boca Raton: CRC Press.

Latour, X., T. Corberand, G. Laguerre, F. Allard, P. Lemanceau. 1996. The composition
of ﬂuorescent Pseudomonad populations associated with roots as inﬂuenced by plant and
soil type. Applied and Environmental Microbiology 62:2449-2456.

Legendre, P., M.J. Anderson. 1999. Distance-based redundancy analysis: testing
multispecies responses in multifactorial ecological experiments. Ecological Monographs
69:1-24.

Lukow, T., P.F. Dunﬁeld, W. Liesack. 2000. Use of the T-RFLP technique to assess
spatial and temporal changes in the bacterial community structure within an agricultural

soil planted with transgenic and non-transgenic potato plants. FEMS Microbiology
Ecology 32:241-247.

Mahaffee, W.F., J .W. Kloepper. 1997. Temporal changes in the bacterial communities of
soil, rhizosphere, and endorhiza associated with ﬁeld-grown cucumber (Cucumis sativus
L.). Microbial Ecology 34:210-223.

Maloney, P.E., A.H.C.v. Bruggen, S. Hu. 1997. Bacterial community structure in relation
to the carbon environments in lettuce and tomato rhizospheres and in bulk soil. Microbial
Ecology 34:109-117.

Marilley, L., M. Aragno. 1999. Phylogenetic diversity of bacterial communities differing
in degree of proximity of Lolium perenne and Trifolium repens roots. Applied Soil
Ecology 13:127-136.

Marsh, T.L., P. Saxman, J. Cole, J. Tiedje. 2000. Terminal restriction fragment length
polymorphism analysis program, a web-based research tool for microbial community
analysis. Applied and Environmental Microbiology 66:3616-3620.

Nakatsu, C.H., V. Torsvik, L. Ovreas. 2000. Soil community analysis using DGGE of
16S rDNA polymerase chain reaction products. Soil Science Society of America Journal
64:1382-1388.

Niisslein, K., J .M. Tiedje. 1998. Characterization of the dominant and rare members of a
young Hawaiian soil bacterial community with small-subunit ribosomal DNA ampliﬁed
from DNA fractionated on the basis of its guanine and cytosine composition. Applied and
Environmental Microbiology 64: 1283- 1289.

76

 

Pacala, S.W., S.A. Levin. 1997. Biologically generated spatial pattern and the
coexistence of competing species. In Spatial Ecology: The Role of Space in Population
Dynamics and Interspeciﬁc Interactions, edited by D. Tilman and P. Kareiva. Princeton:
Princeton University Press.

Paul, E.A., F.E. Clark. 1996. Soil Microbiology and Biochemistry. 2 ed. San Diego:
Academic Press.

Paul, E.A., D. Harris, M.J. Klug, R.W. Ruess. 1999. The determination of microbial
biomass. In Standard Soil Methods for Long Term Ecological Research, edited by G. P.
Robertson, D. C. Coleman, C. S. Bledsoe and P. Sollins. New York: Oxford University
Press. ‘

Phillips, C.J., D. Harris, S.L. Dollhopf, K.L. Gross, J .I. Prosser, E.A. Paul. 2000. Effects
of agronomic treatment on structure and function of ammonia-oxidizing communities.
Applied and Environmental Microbiology 66:5410-5418.

Ringelberg, D.B., J.O. Stair, J. Almeida, R.J. Norby, E.G. O'Neill, D.C. White. 1997.
Consequences of rising atmospheric carbon dioxide levels for the belowground
microbiota associated with the white oak. Journal of Environmental Quality 26:495-503.

Robertson, G.P., K.M. Klingensmith, M.J. Klug, E.A. Paul, J .R. Crum, B.G. Ellis. 1997.
Soil resources, microbial activity, and primary production across an agricultural
ecosystem. Ecological Applications 72158-170.

Ronn, R., B.S. Grifﬁths, F. Ekelund, S. Christensen. 1996. Spatial distribution and
successional pattern of microbial activity and micro-fauna] populations on decomposing
barley roots. Journal of Applied Ecology 33:662-667.

Rouatt, J .W., H. Katznelson, T.M.B. Payne. 1960. Statistical evaluation of the
rhizosphere effect. Soil Science Society of America Proceedings 24:271-273.

Sandaa, R.A., V. Torsvik, Q). Enger, F.L. Daae, T. Castberg, D. Hahn. 1999. Analysis of
bacterial communities in heavy metal-contaminated soils at different levels of resolution.
FEMS Microbiology Ecology 30:237-251.

Sengelev, G., G.A. Kowalchuk, SJ. Sorensen. 2000. Inﬂuence of fungal-bacterial
interactions on bacterial conjugation in the residuesphere. FEMS Microbiology Ecology
31:39-45.

Smith, J .L., BA. Paul. 1990. The signiﬁcance of soil microbial biomass estimations. In
Soil Biochemistry, edited by J. Bollag and G. Stotzky. New York: Marcel Dekker.

deerberg, K.H., E. Biath. 1998. Bacterial activity along a young barley root measured

by the thymidine and leucine incorporation techniques. Soil Biology and Biochemistry
30:1259-1268.

77

Staricka, J .A., R.R. Allmaras, W.W. Nelson. 1991. Spatial variation of crop residue
incorporated by tillage. Soil Science Society of America Journal 55: 1668-1674.

Thom, C. 1935. Micropopulations correlated to decomposition processes. Transactions of
the Third International Congress of Soil Science 1:160-163.

Tilman, D. 1982. Resource Competition and Community Structure. Princeton, NJ:
Princeton University Press.

Torsvik, V., J. Goksoyr, F.L. Daae. 1990. High diversity of DNA of soil bacteria. Applied
and Environmental Microbiology 56:782-787.

Tunlid, A., B.H. Baird, M.B. Trexler, S. Olsson, R.H. Findlay, G. Odham, D.C. White.
1985. Determination of phospholipid ester-linked fatty acids and poly B-hydroxybutyrate
for the estimation of bacterial biomass and activity in the rhizosphere of the rape plant
Brassica napus (L.). Canadian Journal of Microbiology 31:1113-1119.

Willson, T.C., E.A. Paul, R.R. Harwood. 2001. Biologically active soil organic matter
fractions in sustainable cropping systems. Applied Soil Ecology 16:63-76.

Yakovchenko, V.P., L.J. Sikora, P.D. Millner. 1998. Carbon and nitrogen mineralization
of added particulate and macroorganic matter. Soil Biology and Biochemistry 30:2139-
2146.

Zelles, L., R. Rackwitz, Q.Y. Bai, T. Beck, F. Beese. 1995. Discrimination of microbial

diversity by fatty acid proﬁles of phospholipids and lipopolysaccharides in differently
cultivated soils. Plant and Soil 170:115-122.

78

 

 

 

 

 

 

 

 

 

mZ 330.0 33 0.0 m2 m2 300.0 0
28 w— 88 0.02 3.38 v.3 3.8 m2 3.: 3.03 3.8 30.2 500 .80
38 w— 38 06 3.8 3.: 3.8 2 8.8 0.: 30.8 00.0 EoU .280
20 32 3.30 v.3 3.: 23 3.8 m2 3.: v.23 30.8 00.0 «032 0.2.":
m2 3300.0 0000.0 0 2 00.0 3000.0 2000.0v 0
28 S 2.8 mm 3.8 30 300.8 2.0 2.8 3.. 60.8 2.00 E00 .80
:0 : 8. 0 Va 3.: am $00.9 8.0 80.8 0.0 A 2 .8 0.8 500 .Eou
2v 0_ 2.: 3 3.8 mm 20.8 _.0 2.8 3.2 2.8 0.00 «0302 2.”?—
m2 3000.0 3000.0 2300.0 810.0 300.0 0
20 M: 3.8 2.2 3.8 >2 2.8 0‘.— 3.8 3.33 20.8 02.0 500 .30
38 32 3.8 5.2 38 0.3 30.8 3.2 6.38 m._3 30.8 8 .0 :80 .EeU
20 M: 2.8 0.0 3.8 0.. 30.8 0.2 2.8 3.0.3 20.8 m0.0 £302 Tm..—
mZ mz 230.0 0300.0 mZ 0000.0 0
30 M: 3.8 m3 8.: 06 2 .8 0.0 3.38 0.: 20.8 33.0 500 .30
38 33 3.8 0.3 3.: 3.m 30.8 3.0 3.801: 3— .8 3.3.0 E00 .280 SE 3A
38 32 3.8 3.2 3.8 2.3 3.8 3.2 2.8 323 30.8 8.0 £32 2663— Beam
mZ 02 0030.0 3000.0 0000.0 3300.0 0
20 02 3.8 3.2 3.8 N..— 308 3.0 3.8 03 30.8 3V0 E00 .30
20 M: 3.8 N..— 38 3.3 60.8 0.0 3.: v.0 32.8 3.0 E00 .280 222083252
28 C 3.8 2.2 3.8 N..— 608 0.0 3.: 3.0— 30.8 33.0 £002 360%
. Z 0 33203 8293
2.0 :8 .88 0o 8 =8 :39 he 00 Z axe U 8 3 =8 8 9.0080 000680
.vuZ ”228300

0.8058 8e 885530 E 83.3 ._o>o_ m0.0n0 “a 0:88ch Ho: 8893 350080 8:95 mZ .<>OZ< 33-25 a E .808: 20
£523 8893 35000.8 .20 80000 20 .20 3:38:33 20 83:2: mos—«>0 £50680 :8 .3 85:28.85 388280305 ”3 £059.

79

 

 

 

 

 

 

 

m7. 37. 0300.0 0330.0 0

2 .8 0.00 3.: 3.0. 3.8 0.00 8.8 0... E00 .300

3.8 3.30 8.30 3.3. 3.30 ..00 3.8 3.0 0000 .0000

3.30 ..00 2.0 0.0. 3.8 ..30 8.8 3.3 00.002 ..m:
37. 37. 00.0.0 32 0

3.8 3.33 3... 3.3. 3.8 3.. 3.8 3.3 0000 .300

3.8 3.33 8.: 3.0. 3.8 3.. 3.8 0.3 0.000 .0000

.08 ..03 3.30 0.0. 3.8 3.0 3.30 0.3 00.00.43 ..01.
m7. 37. 0030.0 37. 0

3.8 ...0 6.3 0.3. 3.8 3.3 3.: 3.3 0000 .300

3.8 ...0 3.3V 3.3. 3.8 3.3 2.8 0.3 0000 .0000 88 3A

3.30 3.33 3.30 3.3. 3.80.0 3.8 ..3 00.00.40. 00.0.03. .0003
32 37. 3330.0 37. 0

0.0 «:8 8.0 mm. 030 0.0 3.: 3. :80 .05

3.30 3.30 3.8 0.0. 2.: ..3 3... 3.0 E00 .0000 0000000003

38 3.30 3.30 3.3. 3.8 3.0 3.: 0.3 00.002 .300”.

... n0:. 300.0 v * m00073.0 A 0.000000 .., 00. 0. 8000.33 000000”.
0:00 00 0:00 00 :8 .000. 00 00 0 312.00 .000000m 30.00000 .

 

3.00.00 < >0 Z< 303.03 0000.03? 00 00.00.0800 00.. 0x00 000 - * .vuz $0000.50 0000003 20 0000500000

0. m00.0> ..0>0. 30.0u0 .0 “0000.036 .00 E00030 30.00000 00:00.. 32 .<>OZ< 303-000 0 0. 000000.. 05 0.5.3 00.030 30.0000
..0 000000 000 00 000000.036 000 0000.00. 000.0070 0030.0 00.0 ..00 000000.06 0. 0:00 00 0000000 .000 0:00 0.000000 00 8000.02 ”3 030,—.

80

     

 

 

0.33

0.3.

3.0

3.3.

0000.00> .0.0. 00 c0

 

 

 

 

 

 

 

 

58.0 88.0 58.0 0 .002 80.
. . . . o
0 00 .wow... 00.0.0.0 .g... 8.009, “.032 .o .0 .000 090
. . . . o
0 03 3M03m..0 20.0.“...0 23000.0 0000.:0> a..0 .0. ..0 0 .00.). 000.
w 33 23030.. 0 .w0ﬁ0 20.0."...0 0000.00.» ...0 .0. ..0 00 .000 300.
0000.05 000000..
. . . . o
0 0. $000.00 .moﬁo .woﬂo 8500., ”.5200 0 .002 000.
. . . . > 0 o
0 0.. .moﬁo .wowo .30.... 80000 “.00. Co 0 .000 000.
. . . 0000:0> 0 0 0
0 00 30$... .wowo .mowmo . ”.0 z 0. .002 000.
0 3v 20.0.“...0 23000.0 20000.0 0000.00> ”.0 .0. ..0 00 .000 000.
0000.05 00000.02
50m 00.80.20. 0.0.0.3 w0.0000U 00:00.... :00

 

.3HZ 00003 .00.). .3000 VH7. ”0.0.0 0... 00 0000050000 80.0000 0000 .3 00.00000w 0.03 000.0>-0 0.00.0.3
w0.00000 .000 000000... ..00 3.00.0... 0.0.0 00300.. 00.0.0.0 0.0.0-... 00 0.00.00 .0000000. M0000. 0.00.000 000000003. ”3 02.0...

81

 

 

 

 

 

03.0.0.0... 0.3053090 .E:.§.:8.U 3 30-03 0 .000 0 030-030

5.000.000 3 3 3-03 0 300-300

0 03-00 0 .000 ..n... 333-033

00.000.001.000350 0.9 6.00.300000 .35. 0.00.03}. 3 30.03 < .000 .0... 303-003

50.280830... _ «IN 0 a: 80-80

«20.000002303003000: 0 .303 < 05 0.00 80-00

3300850..§=t..0§0§0§0..,-388230020 m. 3.00 0 .2. 05 .00 02-00.

00.003.000.00 .00-0.050.000.0003 .00-00000.30 00000000000 . . 33-33 .-.-... . 3 . -03.

23300300000003300306003000.000003.20.030.030. 3 3003 < 33.33.
$8003.000.05.00.00.08000eta-.002 2 30-00 .00 .8 05 < .2. 00-00
00500.3300000000302000003030300}.003353.00. 3 03-03 < .00 .000 .0... 03-33

00303.3“.3 05.30.333.300..00000000000m 3 . 3 O 3 3-0 3 .00.).

0.20.30.33qu..,-0.30.03.30.00.63.0032030.0.3.000 0. 00-.0 ..n... 000-.00

S§§§§3§G.3088330é003800 0. 00-: 0 $0.00..

52.00.030.80000.00003: aka-5500.9 0.0.3.000 33 03-03 .0... 030-030

0008003030....00..,-00052000....003800 0 :10 0 .00-000

00000000000300.6330003000003 000000003000 6300.00.30 3 0 3-03 ..0... 0. 3-. . 3

00.000503500002000 0 00-00 < 80.0%

30008000§§§§0€300.08 0 00-00 0 0:0:

050000003 .0500.20300Q.0..030.¥.Eﬁﬁtgrv 0.3.0.8000. 3 33-03 D 03 . - . 3.
50.00.00000033-0 620.00.00.33 6333003000000 3 30.33 0 30-00
000.20.00.0200.000.03.030Q.000.-00.000.003.000.52....03303000 0 33-03 0 33-33
00.00.000.00«0.000.003.0003..00-000.00.00.30.E:.20.0000030< 3 . 33-. 3 . --.-. 03-03

00.000.00.k.§...000203 ..,-0000000000.... 0 33.3 .0: 33-33 .000
000 0.000w 00..0.00> 0.. 00.3000
-... 0. 30500000000 ..3 03.0 0. 000000000 5.3 0.0000 80.03 .0 0 30 008.000... .0N.0 00-... 00.8.0000

 

E00 0.00%0HO .0..0...0H< .0000 .000..00>000HU 00000000....u0 .000 3A 00.0.00. .00—.0um 00.80... .0w..u.-n.-.

000000.. 3030003-"... .0>.00.0x0-3..00.00. .00 0.0 000-... .000....0 00-. 0000-. 0.0003 05 0000.. ”000.0 0000...... .0 000- ... 0. 30.0000.
0.000w 0500 05 0.0... 000000.000 3005 5.3 0.0003 .00 00000000 000 00 30.. 00 0. 00.0 00 300. 0000.00 05 0. 00.0 0.0-... 0 30.00.06
000003 ..0 0 00... .0003 0030 0 0. 000-... 00.00000. .000. 05 3000000000 .0 .003 00>.3 0 0.5.3 00.00000. 00.0-... 05 :0 000.00.

.00 000.0 0.00. 0.5.05 0.07. .000. .000 300. 0. 005.00.. 00.00 05 5.3 00.0.8000 2000--C 0.000.000. 02.0.0000 .0580... “0 0.00...

82

02.00.. :00
Tu... .mmm.cm .33.”. 7...... =0@ .55

 

 

 

 

 

 

‘ O
m 8
E
-o.o_.XN m.
e
u.
0|.
P.-
M
I,
J
3..
- wam n...
O.- O
u
D
22.6.

00.8 0.0.00.0 .000.-000. 0.00 .000 00.80.. ..00 .2. 3.u7. 3.000.0qu .00. 0.000-000.002-.00..0.m
05 0.> .0080... 300000.030 0.0 0.0.8. .020..... 5.3 0.03. 00.80... ..00 ..0 3 .2. 00.0000 0.80 000.00.02.33. .0 85002 H. 0.53...

83

scrum: E conhmo oxo

mm. on mm ON 9. o— m o
P . — — _ P — .l o

 

120:6

 

-:o E:

525.

uouoeu ﬁ/suea leuemeg

 

 

 

:SéN

.ocuz .zom isnum gnawed—tum .2662 80:me .Tnﬁud .Tmmum .GduNm ”.26— 0050556 good
05 “a “:85ch 2 538%»: 05. £2.23 :8 no 0 =50qu amp—Ema .83an :8 mo w .89 230 E363 mo co_wm2mom ”N 9.53

84

20250 o\o

 

 

 

 

 

mm on mm cm 3 or m o o
p p — p O/
_ F _ m
0
II,
I '9 I._
m m m m m.
m m 1E mvl
3 w
8 O
m
m_. U
«I?
m - <
._ m om .0
Ir
._ -Nm 8
.n
vm w
8

.couZ dew V==num .Eonmmoﬁﬁnm .0523. Hoonmum .75“.— .Tnmmum Smanm ”—90— 586 2t “a “ESE—um?
ﬂ =23“:th och .2953 05 we U :8ch grams AME: 36A mam—o 0N6 =8 “womb: on. 5 £8 mo “:08?” he :oEmeoM ”n 0.5me

85

 

 

 

Figure 4: Partial canonical principal components plots derived from redundancy analysis
of 1998 RsaI T-RFLP proﬁles. Circles=HF-1, triangles up=LF-1, triangles down=shoot
residue, squares=rhizosphere, white=alfalfa, black=conventional corn, grey=organic corn.

Figure 4a: Soil fraction axes (2 of 3) with cropping system and interaction effects

 

 

 

 

partialled out.

1.0

o\°

CO

- 0.5

V: qI a I p

N [:1

8 ‘A O .
00 o o 0'

TD [3% ‘A év 2

g V

C -O.5 a v

(U

0
-1.0 . r

-0.5 0.0 0.5

Canonical PC1, 10.1%

1.0

Figure 4b: Soil fraction and interaction axes (2 of 9) with cropping systems effects

partialled out.
1.0 4 v
o\°
0?
L0 05 V
G? ' v Q
0 v o o
‘L . ‘3’!’ ‘V ' '
a 0-0 ‘ ‘AA I ‘ O
o
.2 AA I I. o
8 D I o
C -o 5 d D I
(U . C] [j
0
-1.0 . v .
-1.0 -O.5 0.0 0.5

 

 

 

Canonical PC1, 10.6%

86

 

1.0

 

     

Figure 4, continued

Figure 4c: Cropping system axes (2 of 2) with effects of soil fraction and interactions
partialled out.

1.0

 

o

01

h
0

Q7 0
VOACSEI VVA L'I

Canonical PC2, 7.0%

 

 

 

-1.o . .
-1.o -o.5 0.0 0.5 1.0

Canonical PC1, 9.2%

1

Figure 4d: All-treatment axes (2 of 1 l).

 

 

 

 

1.0
V v
o\°
(D
' .5 - A
2 0 ‘ v A V A
.. 0 V
O. 0.0 ‘ ‘ D O
_ A v - .
CO v . C
.2 ‘L ' o
C A v I I
O -0 5 .
C ' ‘v
cu ‘ '
0
-1.0 . ‘ '
-1.0 -O.5 0.0 0.5 1.0

Canonical PC1, 12.0%

87

 

 

 

Lucas—u Lo €035,525 .0 0602

MVN— —nnva—~Nma~—annnv--— —VV—N~nnn—~vnV— —~— —vvvwvn~—vv~v1_annwu— —v~v~v—~vv_va~vvnnn~nawnnnnwvw —VV— ——
mustdadddamm—EwmdddmdmdujdA.Emwmdlﬁdgug44.3542:.EG:j4.3::Exizzzzzizz—Ezmizizzizizzizizzizi

 

 

 

 

 

 

 

 

 

..3-ETowwuullvouuvé33v-I-o$6vuuouvlolN-I$53-$33.6Tuwou-NNTETTvoouvwlwovoovuovouv
l t llllllll I . . .JaJJJJJJJJa; . Itt lllllll 1 :53: 34.34.544.34...-
.23...
-39...
B
_ o
L
Q
. q.-
-omo.o a
m
z
. _
[at o a
_
a
L
U
-92... .—
.
I
0
m
-37..
.346

 

35:5: 88:3:u3o: 550m ":8 20:3"? .2osmmoﬁﬁnm

.328: Hoonmum Jim—HA . _ .quE .38 EEH ”moumwohwwa EE N-ono £033:me EE YNHN .moﬁonma EE 004‘“: .32 280m
"Eco oEsmSHO .ﬂ_£_mn< .Eoo _m:o::o>:oou0 .38 aha “w:_3o=8 on. .3 :Bagcﬁ 8:352 295m .3an 503:0: 8:82:
hag—om me: 8E8: LE? 3mm 83 :o max—«Sm hams—o 783853: no 850E Pub“? .3 335:0» EEwo:::oQ ”m 95»:—

88

Lupus—U Lo co.»o>&oaao no 0202

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

vvmnm.v«~.vwm.vm~_m~v_v~v~vmvm_~a_-_.vnv~_m_~m.

muammmugmmmamamammzzzzzzmmmmzzzzzzmqmqmdooammmm: :
ocoaccccooccccccccccccuuouooaooouuaooouuuuuuuuuu o
. L
r: _ _ F42 .ooo o m
_ _ . q.-
U
_ i m
.890:
_ A l_ _
- M
_
.SQo»
L.
M
a
_
. whoé E
a
m

.8ﬁo

 

8:85: 38:8:n38 82h 8onamonﬁﬁm .0328: Hoosmum .7,—AHA .Tn—mnm .38 :5on
”Eco 2:aw:on0 .a.:£_an< .Eoo _a:o=:o>:oouU .38 mop. ”$232.8 of .3 882:2 82:82 0388 .8an 8038: 3:82:
882:0: wEm: 8an m5? 2mm mg— :o 29:25 85:? 202882: “0 228:: 9:83 .3 888:8 882825 no «harm

89

 

   

Chapter 4

Eubacterial Community Response to Position within Soil
Macroaggregates and Soil Management

90

Abstract

Hierarchical aggregate structure is currently the dominant model of soil structure
and has been shown to be useful in the study of organic matter turnover in soils. The
hypotheses that I . aggregates of different sizes and 2. layers within aggregates contain
different bacterial communities were tested using terminal restriction fragment length
polymorphism (T-RFLP) of the 16S ribosomal gene of eubacteria. Soil aggregate erosion
was used to isolate layers of aggregates with their resident bacteria, resulting in
knowledge of the precise location of the sample within the aggregate. Analysis of
individual aggregates using a split-plot design was found to be the most satisfactory
approach for investigating whether the communities in different aggregate layers were
different. Marginally signiﬁcant differences between eubacterial T-RFLP proﬁles of
different aggregate layers were found. Aggregate-to-aggregate variability was large.
Effects of macroaggregate size on eubacterial T-RFLP proﬁles were not detected. The
major factor affecting T-RFLP proﬁles was cropping system (continuous conventional
corn, organic corn in a crop rotation, and continuous alfalfa) or land management
(conventional or no-till corn and successional vegetation) at ﬁeld experiments in
Michigan and Ohio, respectively. Organization by the tertiary structure of the soil
(arrangement of aggregates in relation to shoot residue, roots, macrOpores, etc.) is
hypothesized to be more important than aggregate layer or size in the determination of

the types of microbial communities present in aggregates.

Introduction

The most common model of the spatial structure of soil particles at millimeter and

91

smaller scales is that of a hierarchical arrangement of soil aggregates (Edwards and
Bremner 1967, Tisdall and Oades 1982, Elliott and Coleman 1988, Ladd et al. 1996,
Christensen 1996). The process of aggregation is dynamic, with microaggregates being
bound together by plant roots, fungi, and plant residues to form macroaggregates, deﬁned
as >250 um in diameter (Elliott 1986, Miller and J astrow 1990). These organic particles
decompose, leading to formation of new microaggregates and disintegration of the
macroaggregates (Beare et al. 1994b, Golchin et al. 1994). Larger macroaggregates
generally contain more labile organic matter (Beare et al. 1994a, Elliott 1986, Gupta and
Gerrnida 1988) with a more rapid turnover time (Buyanovsky et al. 1994, Monreal et al.
1997).

Numerous studies have been conducted to determine the effects of aggregate size
class on soil microorganisms, with varying results. These studies generally involve initial
dispersion steps designed to isolate microaggregates. Using various combinations of
chloroform incubation, plate-counting, ATP determination, and enzyme assays, microbial
populations were found to be greatest in < 0.05 mm aggregates by Kanazawa and Filip
(1986), 0.05-0.25 m aggregates by Monreal and Kodama (1997), and > 0.25 mm
aggregates by Gupta and Gerrnida (1988). Mendes et al. (1999, 1998) found the trends in
microbial biomass and number of Rhizobium leguminosarum across aggregate size
classes were not consistent between sampling dates, although number of bacterial cells
was consistent. Differentiation of the microbial community in different aggregate size
classes was not detected in phospholipid fatty acid proﬁles by Peterson et a1. (1997), or in
archaeal 168 ribosomal terminal restriction fragment length polymorphism (T-RFLP)

proﬁles by Ramakrishnan et al. (2000). Poly et al. (2000) found some differences in

92

restriction fragment patterns of PCR-ampliﬁed nifH gene sequences from different
aggregate size classes, but the signiﬁcance of the results is difficult to assess because the
soil fraction data presented is not replicated.

Another common model of intra-aggregate structure is the division of aggregates
into central or internal portions and external portions. Macroaggregates can develop
anaerobic cores due to microbial respiration (Lefelaar 1993, Priesack and Kisser-Priesack
1993, Zausig et al. 1993), resulting in denitriﬁcation (Hojberg et al. 1994, Sexstone et al.
1985). Exterior and interior layers of macroaggregates have also been found to differ in
age of organic matter, and in contents of organic C and N, phosphorus, cations, and
inorganic N (Santos et al. 1997, Smucker, in preparation). It is also thought that there
may be protection of soil microbes from predators such as nematodes and protozoans in
the centers of aggregates (Elliott et a1. 1980, Elliott and Coleman 1988).

Several studies have addressed the question of whether there are changes in the
microbial community due to the differences in the environments in the external and
internal layers of soil aggregates. Hattori’s (1988) washing-sonication method involves
suspension of soil aggregates in water to separate bacteria located on the outer portions of
aggregates, followed by sonication to isolate those bacteria in the inner portion. Hattori
(1988) showed that bacteria in the outer portions of soil aggregates isolated by this
method were more sensitive to air-drying, HgClz, ethylene dibromide, and protozoan
predation, and were more responsive to addition of substrate. Bacteria introduced into
sterile aggregates accumulated more rapidly in the outer portion. Populations of Gram
negative bacteria were greater than Gram positives in the inner portion of the aggregates,

whereas the reverse was true for the outer portions. Drazkiewicz (1994) found most

93

   

 

types of cultivable bacteria had increased populations in the internal portion of aggregates
in a heavy loam, but only a few had signiﬁcantly increased populations in a silt loam.
Dabek-Szreniawska (1993) examined the effects of keratin-carbamide fertilization on
cultivable bacterial populations, and found no consistent trends in the effects due to
location within the aggregate. Ribosomal intergenic spacer analysis was used by Ranjard
et al. (2000b) to examine the eubacterial community response to Hg(II) Contamination in
outer and inner portions of soil aggregates; however it is difﬁcult to assess the
signiﬁcance of the results with respect to soil fraction because replicated data were not
presented. Ranjard et al. (2000a) have identiﬁed two ribosomal intergenic spacer
sequences that are only present in the external portion of aggregates and the < 2 pm soil
particle fractions after incubation with Hg(II).

The drawback of the washing-sonication method used in previous studies is that
differential adhesion of bacteria to soil particles and the number and length of washes
will affect which cells are washed out of the soil and counted as being in the outer
portions of the aggregate. It is not clear if this method could be standardized such that
the locations of the communities isolated are less vague. A different method of dividing
the community of an aggregate into internal and external portions is used in the study
presented here. The method is based on the physical erosion of individual aggregates and
has been proven useful in studies of the location of soil organic matter and nutrients
(Santos et al. 1997). Erosion is stopped when the desired proportion of the total
aggregate has been isolated. It has the additional advantage that the soil environment is
isolated with the community, so other aspects such as chemical characteristics can be

studied.

94

 

   

In this study we test the hypothesis that eubacterial communities differ based on
their position within macroaggregates using both layers from individual aggregates and
pooled aggregate layer samples. The hypothesis that communities differ based on
macroaggregate size class is also tested. T-RFLP of the 16S ribosomal gene is used to
test these hypotheses with respect to the entire eubacterial community. The use of
molecular methods avoids many of the biases and random variability associated with
cultivation-based techniques (Madsen 1996, Muyzer 1998). The ﬁnding that eubacterial
communities differ between physical soil fractions would imply that these fractions
comprise spatially-distinct habitats, with implications for the ecology and diversity of soil

microorganisms as discussed in chapter 3.

Methods
Field Sites and Sample Collection

Samples were collected in 1999 from three ﬁeld treatments at the Kellogg
Biological Station (KBS) as described in chapter 3. Soils at the site are Typic Hapludalfs
and approximately 43% sand and 40% silt (Robertson et al. 1997). Field treatments
included: conventionally-managed continuous corn, organically-managed ﬁrst-year corn
(from the Living Field Laboratory, established in 1993), and continuous alfalfa (from the
Long Term Ecological Research site, established in 1988). See chapter 3, Jones et al.
(1998) and http://1ter.kbs.msu.edu/Agronomics for descriptions of the management of
these ﬁeld treatments. The organically-managed ﬁrst-year corn is in a com-com-
soybean-wheat rotation, with cover crops planted after corn and wheat. Samples were

collected from within rows, between plants, in corn ﬁelds, and nearby permanent

95

 

sampling stations in alfalfa ﬁelds. Four replicate fields per treatment were sampled.
Three 350 g soil blocks 10 cm deep were excavated per ﬁeld, and were transferred intact
to glass jars for transportation. Samples were stored at 4°C until fractionation was
complete.

Samples were collected from three ﬁeld treatments in Wooster, OH, in 2000.
Treatments included no-till continuous corn, conventionally-tilled continuous corn, and
native successional vegetation. Treatments were started in 1962. Corn plots have been
managed using standard agronomic practices in the region. Soils at the site are silt-loam.
Three replicate plots per treatment were sampled by excavating a soil block 10 cm deep
and storing in a plastic container at 4°C until fractionation was complete.

Soil Fractionation

Whole soil blocks from KBS were gently separated along planes of weakness
over a nest of sieves with mesh sizes of 6.3, 4, and 2 mm, and roots and shoot residue >2
mm were removed. Soil aggregates were separated until they ﬁt through the 6.3 mm
sieve, following recommendations by J astrow and Miller (1991). The nest of sieves was
then shaken by hand until only stable aggregates that would not ﬁt through the mesh were
left on the 4 and 2 mm sieves. This resulted in isolation of aggregates of sizes 4-6.3, 2-4,
and 0-2 mm. The 4-6.3 and 2—4 mm macroaggregate fractions were subsampled for
analysis of aggregate layers. Density separation was performed on all KBS aggregate
size classes with water, resulting in the isolation of a heavy fraction (HF-1) and light
fraction (LF-l) as described in chapter 3.

Soil samples from Wooster were allowed to air dry completely before

fractionation. Soil blocks were then gently separated along planes of weakness over a

96

nest of sieves with mesh sizes of 9.5, 6.3, 4, and 2 mm, and roots and shoot residue were
removed. The sieves were shaken by hand, and 4-6.3 and 2-4 mm aggregates were
isolated for this study.

Four 4-6.3 mm aggregates and four (KBS) or eight (Wooster) 2-4 mm aggregates
from each plot were weighed and placed individually in the upper compartments of
meso-soil aggregate erosion chambers (A. Smucker, personal communication). The
removable upper compartments have textured walls and a screen bottom. Material that is
eroded due to rubbing of the aggregate against the walls falls through the screen and is
trapped in a lower compartment. Chambers were shaken on a horizontal rotary shaker
and upper compartments with aggregates were weighed periodically to determine the
proportion of the aggregate that had been eroded. Weighing was performed as frequently
as every 10 minutes and shaking was at 150 to 250 rpm, depending on the rate of erosion.
After the weight of the aggregate had decreased by 33%, the material trapped in the
bottom of the chamber (the exterior layer of the aggregate) was removed. Erosion was
continued until the weight of the aggregate was reduced by another 33%. The uneroded
portion of the aggregate in the upper part of the erosion chamber (the aggregate interior)
was then isolated. Aggregates that broke into multiple large fragments during erosion
were replaced.

The layers of four 4-6.3 mm aggregates from one plot (replicate 4) of each KBS
treatment were analyzed individually. Aggregate replicates from the remaining plots and
for all 24 mm aggregates were pooled for analysis. Samples were divided into three
subsamples. These were I . frozen for DNA extraction 2. stored in 4.9% formaldehyde 3.

weighed before and after drying at 65°C.

97

 

   

Fractionation of the 4-6.3 mm KBS aggregates from ﬁeld replicate 4, to be
analyzed individually without pooling of aggregate layers from the same treatment, was
completed 7 weeks after sampling. Fractionation of the remainder of the KBS aggregates
was completed approximately 17 weeks after sampling. To test for an effect of storage
over an extended period, heavy fraction samples were isolated using water for density
separation (HF-1) from whole aggregates after 20 weeks of storage, for comparison to
those isolated after 8 weeks described in chapter 3. Fractionation of Wooster aggregates
was completed within three weeks after sampling.

T-RF LP Analysis

T—RFLP was performed essentially as described in Liu (1997) and in chapter 2.
Community DNA was extracted from samples using the standard Ultraclean Soil DNA
extraction kit (Mo Bio Laboratories, Solana Beach, CA). Genomic DNA was found to be
of sufﬁcient purity to be used directly in PCR reactions. PCR was performed using a
standard reaction mixture of 160 11M of each deoxynucleoside triphosphate, 3 mM
MgC12, 0.05 U/uL Taq DNA polymerase and the appropriate volume of accompanying
10X PCR buffer (Gibco BRL, Gaithersburg, MD), and 0.2 ug/mL bovine serum albumin
(Boehringer Mannheim Biochemicals, Indianapolis, IN). PCR mastermix, without
primers, and PCR reaction tubes were sterilized for 14 minutes with direct ultraviolet
radiation in a Cleanspot PCR/UV workstation. Primers used were the general eubacterial
primer 8-27F (AGAG’I'I'I‘GATCCTGGCTCAG, E. coli numbering, Amann et al. 1995,
Integrated DNA Technologies, Coralville, IA) and the universal primer 1392-1406R

(ACGGGCGGTGTGTACA). PCR reactions were optimized for each sample of

genomic DNA using a master mix with primer concentrations of 0.4 uM. Optimizations

98

3— '-— -— —-- w- "w“: "J inf”

a”!

run“

 

 

were performed by adjusting the amount of genomic DNA extract used (0.4 to 7.5 111150
uL reaction) and the number of PCR cycles run (28 to 33) to obtain a strong band without
visible non-speciﬁc product. PCR was performed in a Perkin-Elmer 9600 thermocycler
using an initial denaturation step of 95°C followed by 28—33 cycles of the following
program: denaturation at 94°C for 30 sec., primer annealing at 55°C for 30 sec., and
extension at 72°C for 30 sec. A modified hot start procedure was used where PCR tubes
were not placed in the thermocycler until the block temperature had reached 80°C. A
final extension at 72°C for 7 nrin. was performed after the programmed number of cycles

was complete. PCR product concentration and speciﬁcity was checked by
electrophoresis on a 1% agarose gel, followed by staining with ethidium bromide.

PCR reactions (50-75 pl.) were performed in triplicate for each sample using the
optimal conditions found previously. These reactions were performed using the same
PCR master mix and program described above except that the forward primer was 0.6
pM hexachloroﬂuorscein (hex)-labeled 8-27F (Integrated DNA Technologies). PCR
replicates were pooled. The PCR product from aggregates analyzed individually was

puriﬁed by washing three times with sterile water in Microcon 100 concentrators, and
then isolated in 17 uL sterile water. PCR product from pooled aggregates samples was
puriﬁed using the Promega PCR Preps Wizard Kit as directed by the supplier, except that
elution was performed with 19 LLL of sterile water heated to 55-65°C. Five 11L of puriﬁed
PCR product was mixed with 5 uL of restriction enzyme master mix containing 1.5 U/uL
of restriction enzyme and one uL of the accompanying reaction buffer (Gibco).
Restriction reactions were incubated for three hours at 37°C, followed by 16 min. at 65°C

to denature the restriction enzyme. Three pl. of the restricted PCR product was mixed

99

 

 

 

with one uL of 2500 TAMRA size standard (Applied Biosystems Instruments, Foster .
City, CA). DNA fragments were separated by size by electrophoresis at 1800 V for 14
hours on an ABI 37 3 automated DNA sequencer at Michigan State University’s DNA
Sequencing Facility. The 5’ terminal fragments (T -RFs) were visualized by excitation of
the hex molecule attached to the forward primer. The gel image was captured and
analyzed using Genescan Analysis Software 3.1. A peak height threshold of 50
ﬂuorescence units was used in the initial analysis of the electropherogram. T-RFLP
proﬁles for all samples were generated using the restriction enzyme RsaI. MspI was used
to generate additional proﬁles for two replicates of each cropping system/aggregate
layer/aggregate size class combination for the pooled KBS samples. Negative controls
(no genomic DNA) were conducted with every PCR and run on several Genescan gels.
Contamination in PCR reactions was not detected. Small peaks occasionally appeared in
negative control lanes on Genescan gels, but the cumulative peak height was always
below 1000 units. Samples were re-run if the cumulative peak height was below 9500
fluorescence units, accept as noted.
Direct Microscopy

Cells were dispersed in formaldehyde-ﬁxed individual aggregate layer samples by
diluting to 2 mL with water and vortexing for 5 min. with 0.5 mL of 1 mm glass beads.
Bacterial cell numbers were quantiﬁed in the individual aggregate samples following the
procedure of Paul et al. (1999). Brieﬂy, 4 [.11. of diluted fraction that had been ﬁxed with
formaldehyde was placed in each of ﬁve 6 mm diameter wells of an analytical
microscope slide (Cel-Line Associates, Newﬁeld, NJ) and allowed to dry overnight.

Dried sample smears were then stained with 5-(4,6-dichlorotriazin-2—yl)

100

 

 

aminoﬂuoroscein (DTAF) for 40 minutes, followed by rinsing in phosphate buffer (30
minutes X 3 rinses) and water (30 minutes). Wells were flooded with type FF immersion
oil and then a coverslip was glued in place. Bacteria were observed at 100X
magniﬁcation using a 63X objective and 1.6X zoom with a Leitz Orthoplan 2
microscope. Digital images of bacteria were obtained using a Princeton Instruments
digital microscope camera. Cells were counted and measured by a script written in the
image analysis program IPLab (Princeton Instruments, Trenton, NJ).

Carbon and Nitrogen Analyses

Carbon and nitrogen contents of dried, ground samples were obtained using a
Carlo-Erba NA1500 series 2 Nitrogen-Carbon-Sulfur Analyzer.

Statistical Analysis

Statistical analyses were performed using SAS version 8 (Stat and IML
components) and Canoco. All analyses were performed taking into account blocking of
replicates in the ﬁeld and lab, which was not found to be signiﬁcant.

Percent C and N of pooled aggregate layers was analyzed using three-way
analysis of variance (ANOVA) to test for signiﬁcant effects of aggregate layer, aggregate
size class, cropping system, and interaction effects. Percent C and N and microscopic
cell counts in individual aggregate layers were analyzed as a split-plot experimental
design, with each aggregate as a whole plot and aggregate layer as the split plot.
Regression analysis was used to test for an effect of percent C and N on bacterial cell
counts.

T—RLFP proﬁles were aligned against a database of T-RFs with sample identities

concealed. Cumulative peak height was standardized to 10,000 ﬂuorescence units, with

101

 

   

..E

 

peaks deleted if their height was less than 50 after standardization. Relationships
between T-RFLP proﬁles were examined using Hellinger distance and J accard distance
(see chapter 2 for a complete discussion of data analysis of T-RFLP proﬁles).
Differentiation of eubacterial T-RFLP proﬁles was tested using redundancy analysis (or
distance-based redundancy analysis in the case of J accard’s distance) with dummy
variables coding for cropping system, soil fraction, and interaction terms. Coding
followed the method described in (Legendre and Anderson 1999). Distributions of
partial pseudo-F statistics were generated with 9999 random permutations of the
identities of proﬁles in the software Canoco. To test for aggregate layer effects on
individual aggregates, proﬁles were permuted within individual aggregates as in a split-
plot experimental design. This method was also compared to unrestricted permutation
for pooled aggregate layer proﬁles since layers of only 4 to 8 aggregates were pooled
together.

Adequately strong T-RFLP proﬁles were not obtained for the following samples:
an alfalfa whole soil sample, an organic corn whole soil sample, an organic corn 0-2 mm
HF sample, and a conventional com 24 mm LF sample, from KBS, and one successional
vegetation 4-6.3 mm internal aggregate sample from Wooster. Where necessary these
were replaced by the mean vector generated from the other three replicates of the
respective treatment, following the method of Legendre and Anderson (1999). In other
datasets some treatments could not simultaneously tested due to presence of other weak
proﬁles (see Results).

Relationships between proﬁles were also examined using Ward’s method of

hierarchical cluster analysis, principal components analysis (with Hellinger distance),

102

principal coordinates analysis (with J accard distance), and canonical principal
components plots obtained from the redundancy analyses. Percent C and N were tested

for signiﬁcant effects on T-RFLP proﬁles using redundancy analysis.

Rams
KBS Aggregate Size Classes and Density Fractions

One aggregate size class made up approximately 38% of the soil mass, but this
was different for each cropping system. In organic corn, 0-2 mm aggregates made up
38% of the soil weight, in conventional corn it was the 2-4 mm aggregates, and in alfalfa
it was the 4-6.3 mm aggregates. Mean proportion of the total soil for other size class-
cropping system combinations was 29-33%. These differences were signiﬁcant in the 2-
4 mm size class, where alfalfa had signiﬁcantly increased mass (F=12.6, d.o.f=2,9,
p=0.002).

Aggregate size class did not have a signiﬁcant effect on T-RFLP proﬁles in
redundancy analysis of either LF-l or HF-l, but cropping system was signiﬁcant (see
Table 1). HF-l samples were analyzed with whole soil samples to test the hypothesis that
whole soil sample T—RFLP proﬁles are equivalent to HF proﬁles. This was hypothesized
because HF bacteria make up greater than 90% of the bacteria in the soil (see chapter 3).
Results were consistent in analyses performed including whole soil proﬁles with HF
aggregate size classes, without whole soils (data not shown), and with only the 2-4 and 4-
6.3 mm macroaggregate size classes. Interaction terms were signiﬁcant when whole soils
were analyzed with aggregate HF-l samples using J accard distance. Examination of the

ﬁrst four canonical principal components showed that this was primarily due to increased

103

”'“Vw~ v‘V -".""' S'“I".'".,'~‘ .ro‘.’ 1; .»

 

 

separation of cropping systems in the whole soil and 0—2 mm aggregate HF-l profiles
(data not shown).

Redundancy analysis of HF-l proﬁles from samples that had and had not been
stored at 4°C for an additional 12 weeks showed a marginally signiﬁcant storage effect
when analyzed with Hellinger distance (p=0.0556), accounting for a small amount of the
total variance in the dataset (5.8%). Two T-RFs (114-115 bp and 443-446 bp) had
greater than 20% of the variation in their Hellinger-transformed abundance explained by
storage, after accounting for effects of cropping system. The variation that storage
accounted for when analyzing J accard distance (4.7%) was not signiﬁcant (p=0.2700).
The twentieth principal coordinate had 19% of its variance explained by storage, which
was the only principal coordinate with greater than 12% explained, after accounting for
cropping system effects. Further analyses of aggregate layer T-RFLP proﬁles from'KBS
were conducted after deletion of the two T-RFs identiﬁed as being affected by storage in
the analysis of Hellinger distance.

Layers of Individual KBS Aggregates

Several individual 4-6.3 mm aggregate layers did not produce adequately strong
T-RFLP proﬁles for analysis. Limiting samples to those with a cumulative T-RF peak
height greater than 7500 ﬂuorescence units resulted in 18 proﬁles, with 7 aggregates fully
represented by both layers. Three of these were from alfalfa soil, three were from
organic corn, and one was from conventional corn. Redundancy analysis on Hellinger
distance, using each aggregate as a block for purposes of permutation as in a split-plot
design, found that aggregate layer was marginally signiﬁcant (p=0.0771) and accounted

for 12.5% of the variation in the dataset. Greater than 20% of the variability in Hellinger-

104

transformed abundance was explained by aggregate layer for 12 T-RFs. Divergences in
proﬁles due to the individual aggregate of origin accounted for a large proportion of the
variance in the dataset (47%), but this was not statistically signiﬁcant due to the large
number of parameters involved (p=0.3583).

Limiting samples to those with cumulative peak height greater than 10,000
ﬂuorescence units for analysis by J accard distance (see chapter 2) resulted in 13
acceptable proﬁles. The set was heavily weighted toward alfalfa samples, and only 4
aggregates were represented by both layers, so exploratory data analysis of J accard
distance was used. Figure 1 shows the Ward’s clustering dendrogram based on J accard
distance between these 13 samples. While clustering based on cropping system (or plot
in the case of this set of aggregates) is not evident, four of six extemal-layer samples are
separated from the other samples, accounting for 23% of the variance in the dataset. The
same four extemal-layer samples are “grouped” on the upper and right edges of the data
cloud in the plot of the ﬁrst two principal coordinates (see Figure 2a). They are grouped
much more clearly, and with an additional extemal-layer sample, in the upper-left of the
plot of the third and fourth principal coordinates (which account for 23% of the variance
in the dataset, see Figure 2b). The extemal-layer sample that never groups with the other
extemal-layer samples is consistently grouped with its complementary intemal-layer
sample of the same aggregate. Other samples are not grouped by aggregate.

There were no signiﬁcant effects of aggregate layer or cropping system on percent
C or N, bacterial cells per g fraction, or percentage of cells in the smallest (<0.065 um3)
size class in the individual 4-6.3 mm aggregates at the 0.05 signiﬁcance level (data not

shown). The effects of aggregate layer were signiﬁcant for the percentage of cells in the

105

   

 

 

largest size class (>0.18 um3, F=5.3, d.o.f.=1,9, p=0.047). Proportion of cells >0.18 m3

was 15.4% in aggregate exteriors and 16.7% in interiors. Cropping system had a
signiﬁcant effect on percentage of cells in the largest size class in the aggregate interiors
(F=4.29, d.o.f.=2,9, p=0.049), with 13.9% of cells in this class in conventional corn
aggregate interiors, 17.9% in organic corn, and 18.2% in alfalfa. Cropping system effects
were not signiﬁcant in the aggregate exteriors (mean=15.4%) or overall between
aggregates.

Regression analysis showed that the relationship between percent of cells in the
largest size class and percent organic C in the sample was signiﬁcant (F=6.8, d.o.f=1,22,
p=0.016, R2=24%). The regression of total numbers of bacterial cells on percent C was
not signiﬁcant (F=1.56, d.o.f.=1,22, p=0.225). Percent C and N did not explain a
signiﬁcant amount of the variation in the T-RFLP proﬁles of the individual 4-6.3 mm
aggregates (p=0.3494).

KBS Pooled Aggregate Layers

Effects of aggregate layer and cropping system were tested using all T-RFLP
proﬁles generated from pooled aggregate layers from KBS. The effect of aggregate size
class was tested after reducing the number of replicates in the 2-4 mm size class to three,
to be balanced with the 4—6.3 mm size class, where the layers of aggregates of one ﬁeld
replicate were not pooled. Aggregate layer was found to have a signiﬁcant effect on
MspI T-RFLP proﬁles using Jaccard distance if permutations were restricted to the layers
of corresponding aggregate samples, as in a split-plot design. The variability explained
by aggregate layer was 5.5% of the total variability in Jaccard distances (see Table 2).

The effect of aggregate layer was also marginally signiﬁcant in the case of Hellinger

106

distances between MspI proﬁles using the same permutation scheme, but was not
signiﬁcant for RsaI proﬁles. Three T-RFs had greater than 20% of the variability in their
Hellinger-transformed abundance explained by aggregate layer in the MspI proﬁles after
accounting for effects of aggregate size, cropping system, and treatment interactions:
435-437 bp (23% explained, found only in external layers from corn soil), 439-442 bp
(43%, found only in external layers), and 450-452 bp (25%, found in six internal layer
samples and two alfalfa external layer samples).

Aggregate size class and cropping system were consistently found signiﬁcant in
both RsaI and MspI proﬁles in pooled aggregate layer samples, while interaction effects
were only signiﬁcant in MspI proﬁles (see Table 2). Clustering by treatments was weak
in dendrograms such as Figure 3. Figure 4 shows the separation of samples by aggregate
size class and cropping system in the canonical principal components ordination from the
RsaI Hellinger distance redundancy analysis. While cropping systems are completely
separated in the canonical cropping systems ordination (see Figure 4a), there is some
overlap of aggregate size classes in the size class ordination (see Figure 4b). Treatment
explained greater than 20% of the variation in 21 T-RFs. Their relationships to the
treatments are shown by plotting their canonical principal component scores with the
sample scores in Figure 4.

Percent C in pooled KBS aggregate layers was not signiﬁcantly affected by
aggregate layer, size class, or interaction effects, as indicated three-way ANOVA. The
AN OVA model was signiﬁcant overall due to cropping system effects (F=8.3,
d.o.f.=4,37, p<0.0001). Percent C was 1.8% in organic corn, 1.3% in alfalfa, and 0.9% in

conventional corn aggregate layers. Treatment effects on percent N were also signiﬁcant

107

-2 1"

rm" r-r

.“q

N’V’Z~‘C ’3': v—r—. " .

   

 

 

 

overall (F=3.5, d.o.f.=4,37, p=0.016), with trends due to cropping system following the

same trend as for percent C (p=0.049). Percent N was also signiﬁcantly increased in the
2-4 mm (0.24%) compared to 4-6.3 mm aggregates (0.11%, p=0.017).

Percent C explained a small but signiﬁcant proportion of the total variability in T—
RFLP proﬁles of the pooled aggregate layers from KBS in both RsaI (p=0.0004,
variability explained=6%) and MspI (p=0.0087, variability explained=9%) proﬁles.
Percent C was not signiﬁcant if treatment effects were ﬁrst partialled out. Effects of
percent N were not signiﬁcant (p=0.727 for R831 and 0.4279 for MspI).
Wooster Pooled Aggregate Layers

To achieve balanced replicate numbers of proﬁles between treatments, two
datasets were used to test different treatment effects on T-RFLP proﬁles with redundancy
analysis. The 4-6.3 mm aggregate size class was used to test the effects of aggregate
layer and soil management. Aggregate layer (with or without restricted permutations)
and interaction effects were signiﬁcant when analyzing principal coordinates of J accard
distance between proﬁles, and soil management was marginally-signiﬁcant (see Table 3).
The effects of aggregate layer (using restricted permutation) was the only signiﬁcant
treatment effect when analyzing Hellinger distance between the same proﬁles. The
signiﬁcant effects can be seen in the canonical principal components plot derived from
the redundancy analysis of aggregate layer and treatment interaction effects in Figure 5.
The aggregate layers are separated, and all sucessional vegetation samples are separated
from the continuous corn samples. In plots of non-canonical analyses, such as the
J accard distance principal coordinates plot in Figure 6, separation of samples due to

management is not evident. Two-thirds of the aggregate interiors group apart from the

108

   

 

aggregate exteriors. Less structure is evident in Figure 6, the Ward’s clustering

dendrogram of J accard distances, the main feature of which is a successional vegetation
aggregate interior as an outlier.

The signiﬁcance of aggregate size on Wooster T-RFLP proﬁles was tested with
no-till and successional vegetation aggregate layers. Signiﬁcant treatment effects were
only observed when analyzing J accard distance. Soil management and treatment
interaction effects were significant and aggregate size was marginally signiﬁcant (see
Table 3). The absence of any aggregate layer effect implies that 2-4 mm aggregates are

much more uniform than the 4-6.3 mm aggregates.

Discussion

The strongest inﬂuence on the T-RFLP proﬁles detected in this study was from
land management regimes, accounting for from 10 to 40% of the variance in patterns of
T-RF abundance, depending on the dataset and method of analysis. This is in agreement
with the analyses of other of soil fractions described in chapter 3.

Signiﬁcant effects of aggregate size at KBS were detected in pooled aggregate
layer proﬁles, but not in proﬁles from HF-l or LF-l from differing aggregate size classes
from subsamples of the same soil. Different communities in aggregate layers may
confound effects of aggregate size in the I-IF-l and LF-l, where aggregate layers are
combined by default. It cannot be discounted that the differences in T-RFLP profiles due
to aggregate size class in the pooled aggregate layer samples may be due to differential
effects of storage and erosion on aggregate size class since 1. the effects of aggregate

layer were relatively weak, 2. grouping by layer was not observed in the canonical

109

 

 

ordination plots, and 3. the overall variability accounted for by size class was only 5%. A
small but signiﬁcant storage effect was detected in HF-l proﬁles from samples that had
and had not been stored an additional 12 weeks, and the RsaI T-RFs involved were
deleted in other analyses. Aggregate size class was not found to be a signiﬁcant factor in
proﬁles from Wooster aggregate layer samples, where length of storage of aggregates
was not as great.

Effects of aggregate layer were detected in RsaI proﬁles from layers of individual
KBS aggregates using exploratory data analysis of strong proﬁles (see Chapter 2 for a
discussion of exploratory data analysis of T-RFLP proﬁles). Within pooled aggregate
layers, effects of aggregate layer on T-RFLP proﬁles were detected only in MspI digests,
which were not corrected for storage. The proportion of the total variability accounted
for by layer in the MspI digests was similar to the amount accounted for by the storage
effect. Therefore the effects of aggregate layer were probably not signiﬁcant for these
samples, as indicated by the pooled aggregate layer RsaI proﬁles.

Aggregate layer effects may have been detected in individual aggregate layers and
not pooled aggregate layers due to high aggregate-to-aggregate variability. Aggregate—to-
aggregate variability was taken into account in the signiﬁcance testing of individual
aggregate layers by permuting layers only within individual aggregates. This form of
permutation was also required because aggregate layer was nested within cropping
system treatment for individual aggregates. Pooling of aggregate layers was an attempt
to average out this variability. Success in the ability to average out the variability,
however, depends on its strength. If very strong aggregate-to—aggregate variability exists,

pooling of layers from only four aggregates may confound any differences between

110

 

 

 

layers, rather than enhancing the ability to detect them. While permuting within ﬁeld '

replicate of pooled aggregate layers is not strictly correct, it is probable that not enough
aggregates were pooled to average out the aggregate-to-aggregate variability or result in
independence of aggregate layer and cropping system treatments. Note that while the
split-plot design alters the permutation procedure it does not alter the assumption of
redundancy analysis of a linear relationship between T-RFs and treatment. Hence the
effects of aggregate layer must still be consistent across aggregates (corrected for other
treatment and interaction effects) for signiﬁcant differences to be found.

Permuting using the split-plot design resulted in signiﬁcant differences between
aggregate layers for the 4—6.3 mm Wooster aggregates, but not when 2-4 mm aggregates
were included in the analysis. This is not surprising since it is likely that differences in
the environments of external and internal aggregate layers are greater when aggregates
are larger. Differences between aggregate proﬁles based on tillage were not found at
Wooster, although the T-RFLP proﬁles from successional vegetation were quite different
from those of the continuous corn ﬁelds. This result is not consistent with the differences
between proﬁles due to cropping system at KBS presented here and in chapter 3. The
cropping systems studied at KBS may be more divergent than the tilled and no-till
continuous corn plots at Wooster because many agronomic variables at KBS are varied
besides tillage, including the type and diversity of crops and addition of fertilizers and
compost.

Conclusions
There is high variability in the literature concerning which aggregate size class

contains the greatest numbers and activities of soil microorganisms, as noted in the

111

 

 

Introduction. Soil aggregate dynamics are dependent on particular site characteristics
such as clay and organic matter content and soil management (Ladd et al. 1996).
Seasonal variability in soil aggregation is also likely important in the ﬁeld (Mendes et al.
1999, Mendes and Bottomley 1998), and variability in methods of soil fractionation are
likely the cause of many differences between studies (Christensen 1996, J astrow and
Miller 1991). The results obtained here agree in general with other culture-independent
studies which have found that aggregate size class has small effects on microbial
communities, if any (Peterson et al. 1997, Ramakrishnan et al. 2000).

The effects of the position of a community within a soil aggregate may be greater
than size class, but is dependent on statistical comparisons within individual aggregates.
Tertiary soil structure, such as proximity of sites to macropores or decomposing shoot
residue, may be essential in determining the composition and activity of microbial
communities within different aggregates and sites within aggregates (Parkin 1993, Young
and Ritz 2000). While some consistent environmental differences may exist between
macroaggregate size classes and layers within aggregates, their turnover time of up to 10
years (Buyanovsky et al. 1994, Monreal et al. 1997) may be too rapid relative to the low
rates of microbial growth in soil to allow much differentiation of communities to occur
(approximately 3 generations per year, Harris and Paul 1994). More rapid growth in the
rhizosphere, LF, and shoot residue, on the other hand, allows for extensive changes in the
makeup of the community due to differential success of species under different
conditions (see Chapter 3). Dispersal rates relative to rates of habitat turnover and
growth will also affect the differentiation of communities in different soil fractions.

Microbial dispersal is generally viewed as passive and dependent on the soil water status,

112

 

 

 

although more study of this phenomenon is greatly needed (Murphy and Tate 1996). If
the structure of soil aggregates is truly hierarchical, differences between microaggregate
size classes, which turn over much more slowly than macroaggregates, may also be
important in determining community composition. Environmental heterogeneity at a
broader scale was found to be important since communities were consistently
differentiated by cropping system or soil management. This shows that the numerically
dominant bacterial community in agricultural soil, those in heavy fraction or aggregates,

is sensitive to long term changes in their environment.

113

“.2.‘ ”a. 1- .

 

 

 

 

References

Beare, M.H., M.L. Cabrera, P.F. Hendrix, D.C. Coleman. 1994a. Aggregate-protected
and unprotected organic matter pools in conventional and no-tillage soils. Soil Science
Society of America Journal 58:787-795.

Beare, M.H., P.F. Hendrix, D.C. Coleman. 1994b. Water-stable aggregates and organic
matter fractions in conventional- and no-tillage soils. Soil Science Society of America
Journal 58:777-786.

Buyanovsky, G.A., M. Aslam, G.H. Wagner. 1994. Carbon turnover in soil physical
fractions. Soil Science Society of America Journal 58:1 167-1173.

Christensen, ET 1996. Carbon in primary and secondary organomineral complexes. In
Structure and Organic Matter Storage in Agricultural Soils , edited by M. R. Carter and
B. A. Stewart. Boca Raton: CRC Press.

Dabek-Szreniawska, M. 1993. Effect of keratin-carbamide fertilization on
microorganisms in soil aggregates. Polish Journal of Soil Science 26:49-57.

Drazkiewicz, M. 1994. Distribution of microorganisms in soil aggregates: effect of
aggregate size. F olia Microbiologica 39:276-282.

Edwards, A.P., J .P. Bremner. 1967. Microaggregates in soils. Journal of Soil Science
18:64-73.

Elliott, ET. 1986. Aggregate structure and carbon, nitrogen, and phosphorus in native
and cultivated soils. Soil Science Society of America Journal 50:627-633.

Elliott, E.T., R.V. Anderson, D.C. Coleman, C.V. Cole. 1980. Habitable pore space and
microbial trophic interactions. Oikos 35:327-335.

Elliott, E.T., D.C. Coleman. 1988. Let the soil work for us. Ecological Bulletins 39:23-
32.

Golchin, A., J .M. Oades, J .O. Skjemstad, P. Clarke. 1994. Soil structure and carbon
cycling. Australian Journal of Soil Research 32:1043-1068.

Gupta, V.V.S.R., J .J . Germida. 1988. Distribution of microbial biomass and its activity in
different soil aggregate size classes as affected by cultivation. Soil Biology and
Biochemistry 20:777-786.

Harris, D., E.A. Paul. 1994. Measurement of bacterial growth rates in soil. Applied Soil
Ecology 1:277-290.

114

 

 

 

Hattori, T. 1988. Soil aggregates as microhabitats of microorganisms. Report of the
Institute for Agricultural Research, Tohoku University 37:23-36.

Hojberg, 0., NR Revsbech, J .M. Tiedje. 1994. Denitriﬁcation in soil aggregates
analyzed with microsensors for nitrous oxide and oxygen. Soil Science Society of
America Journal 58: 1691-1698.

J astrow, J .D., R.M. Miller. 1991. Methods assessing the effects of biota on soil structure.
Agriculture, Ecosystems and Environment 34:279-303.

Jones, M.E., R.R. Harwood, N.C. Dehne, J. Smeenk, B. Parker. 1998. Enhancing soil
nitrogen mineralization and corn yield with overseeded cover crops. Soil and Water
Conservation 53:245-249.

Kanazawa, 8., Z. Filip. 1986. Distribution of microrganisms, total biomass, and enzyme
activities in different particles of brown soil. Microbial Ecology 12:205-215.

Ladd, J .N., R.C. Foster, P. Nannipieri, J .M. Oades. 1996. Soil structure and biological
activity. In Soil Biochemistry , edited by G. Stotzky and J. Bollag. New York: Marcel
Dekker.

Lefelaar, RA. 1993. Water movement, oxygen supply and biological processes at the
aggregate scale. Geoderma 57:143-165.

Legendre, P., M.J. Anderson. 1999. Distance-based redundancy analysis: testing

multispecies responses in multifactorial ecological experiments. Ecological Monographs
69:1-24.

Madsen, EL. 1996. A critical analysis of methods for determining the composition and
biogeochemical activities of soil microbial communities in situ. In Soil Biochemistry ,
edited by G. Stotzky and J. Bollag. New York: Marcel Dekker.

Mendes, I.C., A.K. Bandick, R.P. Dick, P.J. Bottomley. 1999. Microbial biomass and
activities in soil aggregates affected by winter cover crops. Soil Science Society of
America Journal 63:873-881.

Mendes, I.C., P.J. Bottomley. 1998. Distribution of a population of Rhizobium
leguminosasrum bv. trifolii among different size classes of soil aggregates. Applied and

Environmental Microbiology 64:970-975.

Miller, R.M., J .D. J astrow. 1990. Hierarchy of root and mycorrhizal fungal interactions
with soil aggregation. Soil Biology and Biochemistry 22:579-584.

Monreal, C.M., H. Kodoma. 1997. Inﬂuence of aggregate architecture and minerals on
living habitats and soil organic matter. Canadian Journal of Soil Science 77:367-377.

115

 

 

 

 

 

Monreal, C.M., H.-R. Schulten, H. Kodoma. 1997. Age, turnover and molecular diversity

of soil organic matter in aggregates of a Gleysol. Canadian Journal of Soil Science
77:379-388.

Murphy, S.L., R.L. Tate, III. 1996. Bacterial movement through soil. In Soil
Biochemistry, edited by G. Stotzky and J. Bollag. New York: Marcel Dekker.

Muyzer, G. 1998. Structure, function and dynamics of microbial communities: the
molecular biological approach. In Advances in Molecular Ecology , edited by G. R.
Carvalho: IOS Press.

Parkin, TE. 1993. Spatial variability of microbial processes in soil - a review. Journal of
Environmental Quality 22:409-417.

Paul, E.A., D. Harris, M.J. Klug, R.W. Ruess. 1999. The determination of microbial
biomass. In Standard Soil Methods for Long Term Ecological Research, edited by G. P.
Robertson, D. C. Coleman, C. S. Bledsoe and P. Sollins. New York: Oxford University
Press.

Peterson, 8.0., K. Debosz, P. Schojonning, B.T. Christensen, S. Elmholt. 1997.
Phospholipid fatty acid proﬁles and C availability in wet-stable macroaggregates from
conventionally and organically farmed soils. Geoderma 78:181-196.

Poly, F., L. Ranjard, S. Nazaret, F. Gourbiere, L.J. Monrozier. 2000. Comparison of nifH
gene pools in soils and soil microenvironments with contrasting properties. Applied and
Environmental Microbiology 67:2255-2262.

Priesack, E., G.M. Kisser-Priesack. 1993. Modelling diffusion and microbial uptake of 13
C-glucose in soil aggregates. Geoderma 56:561-573.

Ramakrishnan, B., T. Lueders, R. Conrad, M. Friedrich. 2000. Effect of soil aggregate
size on methanogenesis and archaeal community structure in anoxic rice ﬁeld soil. FEMS
Microbiology Ecology 32:261-270.

Ranjard, L., E. Brothier, S. Nazaret. 2000a. Sequencing bands of ribosomal intergenic
spacer analysis ﬁngerprints for characterization and microscale distribution of soil
bacterium populations responding to mercury spiking. Applied and Environmental
Microbiology 66:5334-5339.

Ranjard, L., S. Nazaret, F. Gourbiére, J. Thioulouse, P. Linet, A. Richaume. 2000b. A
soil microscale study to reveal the heterogeneity of Hg(II) impact on indigenous bacteria
by quantiﬁcation of adapted phenotypes and analysis of community DNA ﬁngerprints,
FEMS Microbiology Ecology 31: 107-1 15.

116

 

 

 

 

Robertson, G.P., K.M. Klingensmith, M.J. Klug, E.A. Paul, J.R. Crum, B.G. Ellis. 1997.
Soil resources, microbial activity, and primary production across an agricultural
ecosystem. Ecological Applications 7: 158-170.

Santos, D., S.L.S. Murphy, H. Taubner, A.J.M. Smucker, R. Horn. 1997. Uniform
separation of concentric surface layers from soil aggregates. Soil Science Society of
America Journal 61:720-724.

Sexstone, A.J., N.P. Revsbech, T.B. Parkin, J.M. Tiedje. 1985. Direct measurement of
oxygen proﬁles and denitriﬁcation rates in soil aggregates. Soil Science Society of
America Journal 49:645.

Tisdall, J .M., J .M. Oades. 1982. Organic matter and water-stable aggregates in soils.
Journal of Soil Science 33: 141-163.

Young, I.M., K. Ritz. 2000. Tillage, habitat space and function of soil microbes. Soil &
Tillage Research 53:201-213.

Zausig, J ., W. Stepniewski, R. Horn. 1993. Oxygen concentration and redox potential
gradients in unsaturated model soil aggregates. Soil Science Society of America Journal
57:908-916.

117

Table l: Redundancy analysis of HF-l and LF-l isolated from different aggregate size
classes at KBS. Analyses of “all” fractions included 0-2 mm HF-l and whole soil
samples for HF-l analysis, and 0-2 mm LF-l for LF-l analysis. N=4 replicates per
fractioanropping system treatment.

 

 

 

 

 

 

Soil Cropping Interaction Sum
Fraction System
Hellinger distance
P 0.4731 0.0001 0.1348
HF'I’ a" % of total variance 20.6 20.6
HF-l, 2-4 P 0.5893 0.0001 0.8249
and 4-6.3 % of total variance 23.8 23.8
P 0.6239 0.0001 0.4269
LF'I’ 3" % of total variance 31.4 31.4
LF-l, 2-4 P 0.4445 0.0001 0.3449
and 4-6.3 % of total variance 34.7 34.7
J accard distance
P 0.2697 0.0001 0.01 16
HF '1’ a“ % of total variance 15.4 12.7 28.1
HF-l, 2-4 P 0.7966 0.0081 0.9512
and 4-6.3 % of total variance 10.5 10.5

 

118

 

 

 

 

 

 

 

 

 

Table 2: Redundancy analysis of pooled aggregate layer samples from KBS. P-values
generated by 9999 random permutations of sample identity. N=3 replicates per
layerXaggregate sizeXcrop treatment in testing cropping effects and N=4 replicates per
treatment when testing other effects on RsaI proﬁles. N=2 replicates per treatment for
MspI proﬁles.

 

 

 

 

 

Layer Size Crop Interaction Sum
Hellinger Distance
P (unrestricted) 0. 1893
RsaI P (split-plot) 0.1998 0.0003 0.0001 0.0563
% of total variability 5 14.1 19.1
P (unrestricted) 0.5300
MspI P (spli H3101) 0.0648 0.0004 0.0001 0.0338
% of total variability 9.2 18.5 10.9 38.6
J accard Distance
P (unrestricted) 0.5203
RsaI P (split-plot) 0.1417 0.0046 0.0001 0.0800
% of total variability 4.4 10.2 14.2
P (unrestricted) 0.2254
MspI P (split-plot) 0.0096 0.0317 0.0006 0.0358
% of total variability 5.5 6.3 13.3 11.3 36.3

 

119

 

 

 

 

 

 

 

 

 

 

 

 

 

cam _ .2 me 2:85., Bo. do ca 8:32
53 ago 88¢ 33 92953 a cacao
cameo 838225 a c a
5:58? .92 cc as
$55 .9.de .23 233 923:3 a “Wehmww
3:3 328.823 8 .
:ouﬁowo.’ 3:28.83 :5 88 2.62
in a: .3. 5:58? .32 do a. 8:22:
$.86 unwed <2 $8 0 9295.5 .— .82..
3:3 $382353 a c H
E 2. gear? .39 .6 ca
2 cad 8m .0 <2 £85 c2953 a ccccmcﬂww
a 823 828225 a .
83w8wwa 88 m8:
Esm :28825 28:82:82 36 8.34

 

4:258: EoEIoWaSWEXoNB Bmwommﬁwxuozﬁ 8m 83238 mnz 5:82 29:3 .8 823588 88:8
mama .3 888:8 823d 8880? 88.: 8383 8.8— Bawouwwa :28: .6 £835 35:52.2 ”m 2:5.

120

 

 

 

 

 

 

Shaun-c Eshmluom

000.0 mN0.0 0m0.0 m~0.0 00—.0 mN—.0 0m—.0 m~—.0 00~.0 mNN.0 0mN.0
F p _ p p _ _ _ _ . \\L

ml-—Iv¢uh

 

 

wl—mlvcwh

 

 

 

 

 

 

wIN—Im—v

 

mlo—I~0v

 

wlmalfcwhllllllA
.Imalvmahllllllg

_Im~—I~0v

 

 

 

 

 

 

 

_Imm|a—v

 

 

mlvmlm—v

 

 

 

.I~N—Iv¢wh

 

 

 

.I—alvcuh

 

 

 

_Ima—Ivmuh

 

 

 

 

 

.-o_n~ov

.A3EBEHH .3E8xoumv :93— Bmwouwmméonﬁs: cots—303:2: oumwouwwmAEoo 2:330

ﬂ 8v :80 3:239:00 mm 2v .8303 m_ 36.5 58:5: 8E ”w:_>»o=£ on: 2 @6583 3:3: 03 $383 .moumwﬁwwa mmM

EE mbé 3:232: :8 $992 Samamwm Soc 8an :59 3mm 00 0052: 9:33 3 $335 8320 00 EEwohEoQ A 9.53%

121

 

 

Figure 2: Plots of principal coordinates of J accard distance between T-RFLP proﬁles of
layers of individual 4-6.3 mm KBS aggregates. Lines connect layers of the same
aggregate. Circles=intemal layer, Triangles=extemal layer, White=alfalfa,
Black=conventional corn, Grey=organic corn

Figure 2a: First and second principal coordinates

 

 

 

 

0.4
o\°
E 0.3 -
_ A
N 0.2 -
an
‘5
c 0.1- O :
'9 0.0 ~
8
0 '0.1 ‘ .
"if: d A0
.9- '0.2 .
O
.E -o.3 .
L—
‘L -o.4 , , , 1
-o.4 -o.2 0.0 0.2 0.4 0.6

Principal Coordinate 1, 16%

Figure 2b: Third and fourth principal coordinates

 

 

 

 

0.5
o\°
1-
‘- 0.4 ‘ ‘
<F 0.3 a
a)
E? 02 - A
..C. ' \o
'9 0.1u A
8 \
0 0.0 - A
g -o.1 . A . O\A
'5
g -o.2 - O
8- O
n. -o.3 , .

-0.3 -0.2 -O.1 0.0 0.1 0.2 0.3 0.4
Principal Coordinate 3, 12%

122

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Bug—U Lo 8.0093 50 0302
Nuvva-V——mna—VN—N~navnw—.nnNNNNfN———N—
..w...U.N_.MW-m_--m.ww_m.m_um_wwm_wuw
NNNNNNNvvwvvwNNNVvaNNNNNVVVVVVNNNVVV v
aceGUCCCCCCCUUCUUUUUOOOODOOCCOOUUCOOU 0

_ll_ -000.0
U

o

_ ll rm~0.0L
o

3

U

m

C

.—
:0m000
.

a

L

o

n.

_
1m~0.0l
o

m
r00—.0

 

 

 

 

.BEBEHH ..mEonum .38 REP ”moumwoﬁw:
EE Tana .moﬁwoaw: EE mdénv .38 280m ”Eco omﬁanO .ﬂEEH/x .Eoo _::oc:o>:oouu .38 “BE 6058:: :ow:_=om
$5: 29»:— oﬁwEwm: mm! @2000 880 3E8: m5»: Ema 00 @0508 99:3 3 mam—«S: 55:—o Ho EEonEVQ ”m 0.53%

123

 

 

 

 

 

 

Figure 4: Biplots of sample and T-RF scores for the three canonical principal
components (PCs) derived from redundancy analysis of Hellinger distance between
pooled aggregate layers from KBS. T-RFs shown had 20-76% of variability explained by
canonical ordination. Axes are constrained to maximize variation due to cropping system
and aggregate size. Circles=2-4 mm aggregates, Triangles=4-6.3 mm aggregates,
White=alfalfa, Black=conventional corn, Grey=organic corn.

Figure 4a: Aggregate layer/cropping system axes 1 and 2.

Canonical PC2, 6%

1.0

 

.0
01
l

.0
o

.C'>
01
l

 

-1.0

 

 

 

o
O 80bp
482bp 8
' 467bp
A A 1b
0 0 am 2‘ p
0 0
128b 0b”twat. lb 56b 1‘193‘bp 20 3% ”p
90.; 441bp p ‘ 9ﬁ8bp
430bap . ‘ 449bp
3”“ A 104th 33713:)
110bp
1 Mb? —[ ‘1
-O.5 0.0 0.5 1.0

Canonical PC1, 9%

124

 

f

 

 

 

Figure 4, continued

Figure 4b: Aggregate layer/cropping system axes 2 and 3.

 

 

 

 

1.0
56bp
o
'd' 0.5 - b
“ 168688 p
8 ‘43“) 311k)A %%
o. ‘p O
307mg 3 A
TO 0 O 4 ‘ 2 g” 467bp 00
0 83bp 0 song
'5 o . o 0
O 110bp 422””
C -0 5424bp 11955.0”.
(6 ' 271bp .
0 65bp
441bp
'1 .0 I T I
-1.0 -O.5 0.0 0.5

Canonical P02, 6%

125

 

 

Figure 5: Aggregate layer and treatment interaction axes of canonical principal
components derived from distance—based redundancy analysis of J accard distance
between Wooster aggregate layer RsaI T-RFLP proﬁles. Circles=aggregate interior,
Triangles=aggregate exterior, White=successional vegetation, Black=conventional—tillage

corn, Grey=no-till corn.

 

 

 

 

1.0
o\°
N
' 0.5 *
°°, 9 o
N . A
O . A A
E 0.0 1

O O

.23 o M
C ‘ 4
8 cs
(U
0

-1.0 v v v

-1.0 -0.5 0.0 0.5

Canonical P01, 9.8%

1.0

Figure 6: Principal coordinates derived from Jaccard distance between Wooster
aggregate layer RsaI T-RFLP proﬁles. Circles=aggregate interior, Triangles=aggregate
exterior, White=successional vegetation, Black=conventiona1-tillage corn, Grey=no-till

 

 

 

 

corn.
°\° 0.50
“I
1-
1-
of 0.25 -
(D
4—0
(B
.E
'0 0.00 -
h
0
O
O
a -0.25W
.9-
o
.E
h
r; -O.50
-0.50

i

-0.25 0.00 0.25 0.50 0.75

Principal Coordinate 1, 13.5%

126

1.00

 

 

   

 

 

Leas—u Lo 8.00:025 ..o 0:02

2m ~_z 0.0 M.,. 0! mm: hwm vwm 0.0 In «.0 0.2 08 emu wwz —wm nwu

r _ _ _

_

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

—.0
'

 

 

 

 

CIQUDDLDﬁ

1N0

ODI-ULOLH‘I—

 

 

 

 

..0.0

23:3: :3: a 8:352 2:3: 29:3
some E 59:3: 3:0 2: .3E8:_u~ .3E8xonm .EOU =:-o:uZ £2383.» 3:2mmooosmum .Eoo ow3__:-3:o::o>:oouo :85sz
0383 33: ﬂab: Samewwm .8803 ES... 3an nan—M-,: 3mm .8 0050:: $033 .3 Ema—3:3 85:3 .3 883655 ”b 953%

127

     

Chapter 5

Use of Multivariate Spatial Statistics to Test Hierarchical Structure within 3 Soil
Eubacterial Community Analyzed by T-RFLP

128

Abstract

The presence of hierarchical spatial organization was investigated in soil
eubacterial communities within three 1.5x2 meter plots. Hierarchical spatial structure is
indicated by changes in spatial organization with changes in scale of sampling. Ten
samples of various sizes, or sampling grains, were collected from each plot, with a
minimum weight of approximately 10 mg. Fifty 10 g samples were also taken in each
plot. Eubacterial community composition was assayed using terminal restriction
fragment length polymorphism (T-RFLP) of the 16S ribosomal gene. Paired-quadrat
analysis and semivariance analysis were adapted to multivariate problems in a
generalized method called variability-scale analysis, where the changes in variability
statistics or community distance metrics due to spatial scale is examined. Analyses based
on changes in sampling grain and extent were integrated. Three hierarchical levels-of
organization in soil eubacterial communities were separated by phase shifts in spatial
organization, where there were signiﬁcant changes in the slope of the relationship
between variability statistics and spatial scale. These included one level, operating at a
scale of 10 to 100 g or 2 to 25 cm, with minimal change in community variability with
spatial scale. Experimental designs with treatments blocked within these scales should
not violate assumptions of standard (non-spatial) statistics. The extent-based scale-
variability analysis of spatial structure in soil microbial communities was also compared
to Mantel methods and spatially-constrained ordination, with similar results. A method
of examining the variability in covariance structure across spatial scales is proposed
called the correlation super-matrix, based on the correlation between covariance matrices

constructed from samples at different scales. Changes in covariance structure due to

129

 

 

 

spatial scale are then investigated using principal coordinates analysis and cluster
analysis of the correlation super-matrix. This method veriﬁed conclusions based on the
extent—based analyses of other methods, but was not as useful in the grain-based analysis
because the covariance structures of communities sampled at different grain sizes was
extremely diverse. Soil eubacterial communities displayed signiﬁcant hierarchical spatial

structure, suggesting they are regulated by mechanisms analogous to those for plant and

animal communities.

Introduction

The complexity of soil microbial communities is reﬂected in the enormous
amount of microbial diversity that can be detected in soil (Hugenholtz et al. 1998,
Torsvik et al. 1990). Biological, chemical, and physical properties are generally
heterogeneous in soil (Harris 1994, Parkin 1993, Stark 1994). A key question for our
understanding of soil microbial communities is: How much of the heterogeneity in
microbial community composition has a mechanistic ecological basis, and how much is
truly random variability that is historical artifact? More generally, this can be seen as
“the central question of community ecology,” (Roughgarden 1989). There are strong
arguments to be made for either hypothesis. Hraber and Milne (1997) found that the
manifestation of community assembly rules is dependent on the rate of colonization by
new species relative to the rate of operation of the rules. Random variability in soil
bacterial community structure may be likely because it has been shown that bacterial
growth (Harris and Paul 1994, Smith and Paul 1990) and dispersal (Murphy and Tate

1996) are generally low in soil. The major determinant of soil bacterial community

130

 

'w,

 

structure may be the environments serving as sources for bacterial colonizers. However
even this signal could be rapidly diluted since the time frame for a soil bacterial
community to reach equilibrium number of species is less than 200 years (Nﬁsslein and
Tiedje 1998), while the lifespan of initial colonizers, as well as subsequent arrivals, may
extend to millions of years (Greenblatt et al. 1999). On the other hand, numerous
mechanisms, or assembly rules, can be hypothesized to affect soil bacterial community
composition and diversity in a predictable fashion. Traditional ecology provides a
plethora of potential ecological mechanisms, including competition (Leibold 1995,
MacArthur and Levins 1967), trophic interaction (Holt 1984, Schmitt 1987), disturbance
(Bormann and Likens 1979, Connell 1978), and resource distribution (Grime 1994,
Harrison 1997). In controlled experiments, these mechanisms have all been shown to
have the potential to organize soil microbial communities (Minamisawa and Mitsui 2000,
Grifﬁths and Bardgett 1997, Beare et al. 1992, and chapters 3 and 4, respectively). It is
unknown, however, how strongly these mechanisms act in the ﬁeld, or, if they do have
signiﬁcant impact in the ﬁeld, whether their interactions result in a general coherent
structure. This lack of basic knowledge is due to the inadequacy of our understanding of
the spatial structure of soil microbial communities, which is a direct reﬂection of those
mechanisms, if any, organizing a community.

Recently the heterogeneity of soils has been highlighted, and the spatial structure
quantiﬁed, through the use of semivariance analyses of soil nutrients (Boemer et al. 1998,
Dobermann et al. 1995, Schlesinger et al. 1996), potential rates of biological processes
(Robertson et al. 1993, 1997, Stoyan et al. 2000) and numbers of soil organisms

(Klironomos et al. 1999, Morris 1999, Robertson and Freckman 1995, Rossi et al. 1997).

131

f.

 

 

Cavigelli et al. (1995) used semivariance analysis to quantify spatial structure of 56
microbial fatty acids across separation distances from 2 cm to 80 m. They found no
spatial autocorrelation of microbial fatty acids at ranges greater than 20 cm, and for most
fatty acids found no spatial autocorrelation at any scale at all. At the other extreme,
Felske and Akkerrnans (1998) concluded that 168 rDNA DGGE patterns were
qualitatively homogeneous over ranges of 1 to 40 m.

A complete understanding of the spatial structure of ecological phenomena must
involve consideration of both spatial pattern and scale (Levin 1992). Spatial patterns in
community composition are an outcome of the interactions between organisms and their
abiotic and biotic environment. Spatial scale can be used to denote both spatial extent
(maximum size) and grain or resolution (minimum size) of any object or phenomenon
(Schneider 1994). Variability of an environmental characteristic, and patterns caused by
it, will normally be found at some spatial scales but not at others. Hence the spatial scale
of observation will affect which mechanisms regulating community structure are detected
(Levin 1992, Schneider 1994). Note that scientiﬁc observation itself is a biological
phenomenon with a particular spatial scale which will often not coincide with the scale of
the object or phenomena under study (i.e. the scale of a soil core is equal to the
measurement grain; it need not be the characteristic scale of a forest, microbial
population, or any natural phenomenon).

An important distinction must be drawn between a spatial scale, deﬁned above,
and a level of organization, which is a range of conditions, including spatial scales, over
which phenomena affecting the parameter of interest are homogeneous. To deal with the

complexity generated through the interaction of many phenomena operating across a

132

   

wide variety of spatial scales, it has been suggested that mechanisms regulating
communities may form a hierarchy, resulting in spatially nested levels of organization
that operate at particular scales (Allen and Starr 1982, O'Neill et al. 1986). In formal
hierarchy theory, hierarchical structure was attributed to the separation of ecological
processes by their rates, and the constraining of dynamics at lower levels in the hierarchy
by higher levels. Other treatments of the topic have stressed the separation of ecological
processes by effects on species (Parker and Pickett 1998), or spatial extent (Wu and
Loucks 1995). The detection of such organization would be a powerful method of
summarizing the outcomes of complex phenomena. This approach has proven to be
successful in several communities (see next section).

Ecological studies are often described as global, regional, landscape, or local.
These terms connote both particular spatial scales (i.e. the “regional scale” is about the
size of a large state) and levels of organization (community structure at the “regional
level” is organized by climate and geology). The design of ecological studies around pre-
conceived hierarchies without reference to patterns intrinsic to the organisms is now
under question (Hoekstra et al. 1991, O'Neill et al. 1986). The hierarchy described above
has been shown to be useful in understanding species endemicity in some soil bacterial
culture collections (Cho and T iedje 2000, Fulthorpe et al. 1998), but not for genetic
structure in Rhizobium leguminosarum (Hagen and Hamrick 1996, Strain et al. 1995, but
see also Souza et a1. 1994). For microbial communities, it seems likely that there are
levels of organization that operate at spatial scales smaller than what is normally
considered the local (ﬁeld or plot) scale (see Beare et al. 1995), but this has not been

explicitly tested.

133

   

 

 

 

The goal of the current study is to explicitly and quantitatively consider scale and
test for hierarchical structure in an investigation of spatial structure of soil bacterial
communities. This was accomplished using multivariate spatial statistics coupled with
community-level terminal restriction fragment length polymorphism (T-RFLP) of the
16S ribosomal gene. Molecular genetic techniques such as T-RFLP have become well-
established methods to assay in situ community structure of soil bacteria, avoiding the
problems associated with elective enrichment culture (Muyzer 1998, Tiedje et al. 1999).
The speciﬁc hypothesis tested is that a soil microbial community is structured by levels
of organization operating at different spatial scales. This is a holistic method of testing
the overall importance of ecological phenomena in community assembly, as opposed to
random assortment as discussed previously. Several new approaches to multivariate
spatial statistics were developed in conjunction with this study, so the theoretical and
historical framework of hierarchy theory and multivariate spatial statistics are briefly

presented prior to a description of the experiment.

Detection of Levels of Ecological Organization: Statistical Approaches
Grain-Based Analysis of Spatial Structure in Community Ecology

Hierarchy theory is one of the few conceptual approaches in ecology that fully
incorporates the concept of scale (Peterson and Parker 1998). This results in an intuitive
method of recognizing organization in the face of complexity and heterogeneity. As such
it has proved useful as a conceptual basis for ecological mapping (O'Connor et al. 1996),
modeling (Maurer 1990, Palmer 1992), new research methods (Johnson and Gage 1997 ,
Quinones 1994), and strategies of land management (Eswaran et al. 2000, Hoekstra and

Flather 1987, Jennings and Reganold 1991, Lee and Grant 1995, Palik et al. 2000). In

134

 

 

 

 

these contexts hierarchy theory is normally used as justiﬁcation for assuming that one
structure or process is constrained by another (usually spatially—constrained). Empirical
tests of hierarchy theory have focused exclusively on structural components of
ecosystems, namely abundances of species. Several studies have looked for hierarchical
structure among traits of species in a community. These have tested predictions from
hierarchy theory using empirically-suggested (W altho and Kolasa 1994) or pre-conceived
(Gaedke 1998) levels of organization among aquatic species.

Other studies have investigated the presence of hierarchical spatial structure, and
therefore distinct levels of organization regulating species abundance, in plant
communities. The hierarchical spatial structure is detected by discontinuities in spatial
pattern across. a range of spatial scales. The changes in spatial pattern have most often
been detected using variance analysis. If a quantity is randomly distributed, a plot of log
variance versus log sampling grain has a slope of -1 (Levin 1992, Wiens 1989). If there
is spatial structure, however, the slope will be greater than —1, and will depend on the
particular pattern and strength of autocorrelation. Discontinuities in this slope indicate
phase shifts in spatial structure, and hence changes in organization. O’Neill et al. (1991a)
calculated the variance in percent cover of vegetation across 32 transects radiating from a
single position. Using this method they were able to detect hierarchical structure of
varying degrees in all six landscapes examined. Several other studies have looked at
distributions of single species within communities (Cullinan et al. 1997, Kotliar 1996,
O'Neill et al. 1991b). Where abundance of more than one species was recorded, these
were simply analyzed separately. In these studies, various forms of blocked-quadrat-

variance analysis (BQV) were used to examine effects of sampling grain on spatial

135

 

:’.;...'v;

 

 

structure. The BQV approach is basically the same as in O’Neill (1991a), except that

only contiguous blocks of quadrats within a single transect or grid are compared. This
method of generating quadrats of different scales confounds sampling grain with
separation distance between quadrat centers, resulting in an initial increase in variance
with scale rather than a continuous decrease. The BQV methods have a long history in
plant ecology and differ based on how quadrats are blocked and paired for estimation of
variance (Grieg-Smith 1952, Hill 1973, Galiano 1983). The placement of this line of
research into the conceptual context of hierarchy theory represents a leap forward in our
ability to interpret the spatial patterns that have been investigated by ecologists for so
long.

Multivariate studies have been conducted to examine spatial structure
encompassing the entire community. Community structure is dependent on the existence
of non-random associations between species. The debate over causes of such
associations has a long history, ranging from common responses of species to the
environment (Gleason 1939) to the existence of coevolved species complexes (Clements
1936). It is probable that many phenomena are involved (Ricklefs 1987, Roughgarden
1989, Belyea and Lancaster 1999, Wilson 1999); hence holistic methods such as the
study of spatial structure and hierarchy theory are particularly useful.

The simplest form of multivariate analysis of community spatial structure has
consisted of an ordination to derive new variables (ordination axes) that summarize the
majority of the variance in species distributions. Site scores of the ﬁrst few variables are
then plotted along the length of a transect (Bouxin and Gautier 1982, Shmida and

Whittaker 1981). The variance of these derived variables has also been examined using

136

techniques related to BQV (Galiano 1983, Milne 1991). Multiscale ordination is another
method that has been proposed as a way to partition variance of an ordination between
scales (Hoef and Glenn-Lewin 1989, Noy—Meir and Anderson 1971). Covariance
matrices derived from different plot sizes are summed, followed by a principal
components analysis. The variation of each eigenvector can then be partitioned into the
contributions from each scale. However, Wackemagel (1998) writes that if principal
components analysis is performed on a covariance matrix that incorporates dynamics of
mechanisms operating at different scales, the analysis will be invalid because the
orthogonality of principal components will be scale-dependent (i.e. they will be
correlated at some scales). Hence multiple scales of spatial pattern and other sources of
variation may interact to confound the ordination, obscuring dynamics present at any
given scale. In addition, it is assumed that spatial structure will be the primary source of
variation in the dataset and will therefore be reﬂected in the ﬁrst few ordination axes.
Depending on the ordination, the majority of the variability in the dataset may be left
unexamined, and weaker spatial patterns will be missed. This problem arises because the
null hypothesis of no spatial structure is not explicitly tested on the entire dataset.
Finally, the ordination axes analyzed are dependent on a particular dataset. This causes
difﬁculty in the comparison with other samples since it requires the assumption that the
same phenomena are affecting both datasets in the same way, and hence would result in
identical ordinations and comparable axes.

The variance test of Schluter (1984) tests for non-random associations between
species by comparing the variance of total numbers of individuals within samples to the

sum of the variances of each species. If these variances are not equal, then the sum of the

137

. +———“=—.r—-. 3'11.

 

 

 

 

 

covariances between species must be different from zero, implying there is some overall

pattern present in the species associations. Yoshioka and Yoshioka (1989) adapted this by
computing variances for each block size (sampling grain) separately. The departure from
random association between species was tested for each scale using nested ANOVA.
This test is invalid if the species abundance data are transformed to relative abundances
prior to analysis, as is often recommended for ecological data and is required if the
variance in total numbers of individuals is not meaningful (i.e. due to analytical
variability or sampling effort). However, the interpretation of the variance and
covariance statistics after transformation is dependent only on variability in community
composition, having removed the variability due strictly to site productivity. Hence
plotting the sum (or mean) of the species variances against sampling grain will reﬂect
how the heterogeneity in relative abundance across all species (i.e. the heterogeneity in
community composition) changes with spatial scale. The sum of species covariances
measures the same property of the dataset as the sum of the species variances after
transformation to relative abundances, namely heterogeneity of community composition.
Other methods are required to examine the overall homogeneity of elements within the
covariance matrix across spatial scales (see next section).

The utility of mean variance in describing multivariate patterns becomes obvious
if it is considered that the mean species variance is equal to one half of the mean pairwise
Euclidean distance between samples, which is commonly used as the basis for
community ordination, clustering, and classiﬁcation methods. Euclidean distance itself
has been recognized as inappropriate for these applications (Legendre and Legendre

1998), but alternative distance metrics can be derived by using an appropriate

138

 

 

 

 

transformation prior to calculating Euclidean distance. The transformation to relative '
abundances is an improvement, but several further transformations have been shown to
be superior by Legendre and Gallagher (in press). In this paper we use the Hellinger
distance because it does not heavily weight rare species and is simple to calculate, being
equal to the Euclidean distance between the square-root of relative abundances. The use
of ecologically-appropriate data transformations and the focus on mean distance (or mean
variance) greatly enhances the ﬂexibility of the types of data and analyses that can be
performed in grain-based approaches to assessing spatial structure. Signiﬁcance tests can
still be performed using randomization procedures (see below). The generalized concept
of comparing a variability statistic to spatial scale is further developed below, and will be
referred to as variability-scale analysis.
Extent-Based Analysis of Spatial Structure in Geostatistics

A rich set of methods for analyzing spatial structure based on separation distance
between samples rather than sample grain has been developed within the ﬁeld of
geostatistics. Examination of the change in variance due to changes in separation
distance is known as semivariance analysis or variography. It is equivalent
systematically varying sampling extent at all locations. Inferences about sampling grain
are also possible (Bellehumeur et al. 1997, Schneider 1994). Identical methods were
independently discovered in community ecology and were known as paired—quadrat-
variance analysis (PQV, Goodall 1974). Comparisons of the extent-based versus grain-
based methods have generally found that the extent-based methods are more sensitive, at
least in part because the BQV variants that have been tested confound extent with grain

(Carpenter and Chaney 1983, Ludwig and Goodall 1979), although this was also noted by

139

 

 

 

 

 

Schneider (1994). In geostatistics it is standard to ﬁt semivariance to combinations of
variogram functions (Robertson and Gross 1994, Rossi et al. 1992). Typically these
include an intercept called nugget variance. Nugget variance is due to analytical
variability and variability at scales smaller than the sampling grain and minimum
separation distance. Other variance functions of separation distance may increase
unbounded (e.g. the linear model) or reach an asymptote or sill (e.g. the exponential
model). The separation distance at which a sill is reached is the range of inﬂuence of the
particular phenomenon causing pattern at the scale. While not often recognized as such,
semivariance analyses are tests of hierarchy theory. Hierarchical patterns are often found
and modeled in the form of nested functions, where variance reaches a sill but then starts
to increase again with further increases in separation distance.

Attempts at combining extent-based methods with multivariate analysis often
involve semivariance analysis of the ﬁrst few axes of an ordination (Cash and Breen
1992, Jonsson and Moen 1998, Rossi et al. 1997) or the equivalent multiscale ordination
(Schaefer and Messier 1994). This approach has the same shortcomings as when used
with BQV. Multivariate factorial kriging analysis (FKA) is another procedure that has
been developed by geostatisticians to incorporate spatial structure into principal
components analysis (W ackernagel 1998). This involves ﬁtting each variance and
covariance (or cross-variogram in geostatistical jargon) in a multivariate dataset to a
common basic variogram model, which would include the same functions and ranges.
The ﬁtted model coefﬁcients (nugget, sills) are then used to construct a variance-
covariance matrix, or coregionalization matrix, for each function/range modeled. Hence

a coregionalization matrix represents the complete dynamics, at a given spatial scale, of a

140

 

 

 

 

set of variables, independent of dynamics at other scales. Principal components analysis
can then be performed on coregionalization matrices for different scales separately. The
goal of the analysis is to assess relationships between complex sets of variables without
averaging over multiple spatial scales. Such averaging causes samples to not be
independent, resulting in principal components that are correlated at some scales
(W ackemagel 1998). It is not unusual to ﬁnd that variables important in ordination at
one scale are unimportant at another, or that relationships between variables are reversed
at different scales (e.g. Dobermann et al. 1995, Goovaerts 1994, Monestiez et al. 1994).
Multivariate factorial kriging analysis is a highly reductionistic approach to
assessing the degree of spatial organization in communities. First the importance of
spatial scale is examined variable by variable (or species by species) and covariance by
covariance. Then the overall importance to the entire community is built up from each of
these individual analyses, ﬁnally expressed as qualitative differences between principal
components analyses. On the other hand, use of a multivariate distance coefﬁcient in a
variability-scale analysis is a holistic method of accomplishing the same goal.
Variability-scale analysis can easily be extended to the extent-based sampling design;
samples at the same grain are taken at varying distances apart, and mean distance
coefﬁcients are calculated over pairs of samples in each category of separation distance.
Again, the approach is not limited to Euclidean distance since transformations can be
performed to result in more ecologically-relevant distances. The choice of distance
coefﬁcient is even more ﬂexible than previously mentioned since any ecological distance
coefﬁcient can be calculated and plotted as a function of sampling grain or extent.

Whittaker (1960) plotted J accard’s coefﬁcient as a function of separation distance and

141

 

found that similarity of communities decreased at a steady rate as separation distance

increased until a threshold was reached, at which point the similarity between
communities declined precipitously. The relationship to hierarchy theory is clear: two
levels of organization operated simultaneously but at different spatial scales to organize
the communities. Each distance coefﬁcient has unique advantages and disadvantages
(Legendre and Legendre 1998); in this study we use both Hellinger distance and one
minus J accard’s similarity coefﬁcient (hereafter referred to as Jaccard’s distance).
While it may be desirable to use FKA to examine the homogeneity of structure
within the covariance matrix, across spatial scales, instead of the overall strength of
covariance, in many ecological applications it is not practical. Ecological communities
normally contain many species, sometimes numbering in the hundreds. Hundreds to
thousands of variances and covariances would therefore have to be separately modeled
using nonlinear regression to construct the coregionalization matrices of FKA. It is also
unlikely that the majority of these would be adequately ﬁt by a common model, as
required in FKA. We propose a new method, called the correlation super-matrix, to
adapt the goals of FKA to large ecological datasets. As in FKA, variance and cross-
variograrns are calculated for all variables (species) for each distance class separately.
Hence a different variance-covariance matrix can be constructed from empirical data for
each distance class. Rather than model each element of these matrices, however, the
matrices can also be compared using a correlation coefﬁcient. This method of comparing
distance or similarity matrices is known as the cophenetic correlation (J obson 1992) or
“normalized” Mantel statistic (Legendre and Legendre 1998). By calculating the

correlation for each pair of covariance matrices, a new “super”-matrix can be constructed

142

 

 

 

 

 

 

containing the similarities of the Species covariance structures for each pair of distance
classes. The method can be applied to the grain-based methods as well. The signiﬁcance
of these pairwaise correlations could each be tested using permutation or a normal
distribution approximation, although this could easily result in an excessive number of
pairwise tests. Structure within the correlation matrix could also be investigated using
multivariate exploratory data analysis techniques such as cluster analysis or principal
coordinates analysis. The results would summarize the similarities in species covariances
across the different spatial scales, albeit without explicit hypothesis-testing. An
additional advantage of the correlation super-matrix approach is that it does allow the
differences in relationships to be quantitatively assessed in a single analysis, rather than
qualitatively in separate principal components analyses as in FKA. The correlation
super-matrix approach does not result in the ability to attribute differences between scales
to particular variables; however this can easily be accomplished by carrying out a
principal components analysis on the individual covariance matrix for any particular
scale of interest.
Mantel Procedures and Constrained Ordination

Other holistic methods of assessing the signiﬁcance of spatial pattern in
community composition include the Mantel test and constrained ordination. These are
both based on separation distance between samples (extent). The Mantel test can be used
to determine whether two distance matrices, one containing spatial distance and the other
calculated using a community distance coefﬁcient, are signiﬁcantly correlated (Legendre
and Fortin 1989). A positive result implies that communities are more different the

farther apart they are. The normalized Mantel statistic (rM) can also be calculated using a

143

~.—~-.__nh.

 

series of matrices indicating which sample pairs belong to different separation distance
classes. Plotted against separation distance, this Mantel statistic indicates regions of
positive or negative spatial autocorrelation. It can be interpreted in much the same way
as the semivariogram, where changes in slope represent a change in the spatial pattern
detected. One complicating factor is that the value of rM for each distance class is relative
to the contrasting structure in the remaining distance classes; i.e. changing a value outside
the distance class will change rM.

Community ordination procedures are commonly used to extract major gradients
within community composition from a group of sites. The amount of variability in
community composition that can be correlated with independent measurements or
treatments at the same sites can be determined by constraining the ordination axes to be
linear combinations of environmental variables. If the environmental variables consist of
geographical coordinates, and other polynomial terms derived from them, then the
inﬂuence of spatial position on community composition can be determined (Borcard et al.
1992). If redundancy analysis, the constrained version of principal components analysis,
is used then the procedure is analogous to a multivariate trend-surface-analysis. Another
approach is the use of canonical correspondence analysis. The constrained ordination
approach is limited by the ability of the geographical variables to capture the spatial
patterns present. The extrapolation to other scales of sampling grain is also more
difﬁcult.

In presentations of Mantel procedures and constrained ordination, several
characteristics of the procedures have been stressed: the null hypothesis of no spatial

pattern can be explicitly tested using randomization tests, any community distance or

144

 

*1»; 1.

 

 

similarity metric can be used, and the analyses can be used to partition variability in
community composition among various sources (Borcard et al. 1992, Legendre and
Anderson 1999, Legendre and Fortin 1989). The other methods of spatial analysis
described share these characteristics, if adapted in the same way that Mantel procedures

and constrained ordination have been.

Materials and Methods
Sampling Design

Sampling was performed in three 1 ha alfalfa ﬁeld replicates at the W.K. Kellogg
Biological Station Long Term Ecological Research Site in June of 1999. Plots had been
in agronomically—managed perennial alfalfa for 10 years at time of sampling. Alfalfa was
killed with herbicide every ﬁve years and resown to maintain stand vigor.
Approximately two months prior to sampling, 5 year-old stands of alfalfa had been killed
with 3.5 Uha of the herbicide glyphosphate (Roundup). Plots were then fertilized with
2.9 Mg/ha dolomitic lime and re-planted to alfalfa at a rate of 10 kg/ha for each of two
perpendicular passes. Details of the management of this site can be found elsewhere
(http://lter.kbs.msu.edu/Agronomics).

Within each of the three ﬁeld replicates, a sampling grid with adjacent samples
being 39.6 cm apart was placed over an approximately 1.6 x 2 m area. Ten locations
within each grid were randomly chosen for multiple-grain sampling (see Figure 1). Five
soil samples differing in soil weight were collected at each of these locations to obtain a
set of samples representing a wide range of sampling scales. Approximately 1 and 10 g

samples were collected by pushing sterile sample tubes (5 mL Falcon specimen tubes and

145

 

 

 

 

 

50 mL Corning centrifuge tubes, respectively) into the soil. Approximately 100 g
samples were collected by pushing the tip of a 6.4 cm hydraulic Giddens tube (soaked in
bleach between samples) into the soil and then excavating the sample with a sterile
laboratory weighing spatula. 100 g samples were then placed in sterile Whirlpack
specimen bags. Smaller samples (10 and 100 mg) were collected using sterile weighing
spatulas and placed in sterile 1.5 mL microcentrifuge tubes. All samples were from the
layer of soil 2-4 cm deep to avoid increased variability at the soil surface.

Two additional 10 g samples were taken 8.5 cm from each of the ten multiple-
grain sampling locations in each grid described above. These were on opposite sides of
the multiple-grain sampling locations. The orientation of these samples within the
artiﬁcial coordinate system was determined randomly. If the multiple-grain sampling
location was at the edge of a grid and the orientation of the additional samples was such
that one of them would be placed outside of the grid, this sample was instead located
within the grid 8.5 cm from the other additional sample for that location (see Figure 1).
Twenty additional 10 g samples were collected at each ﬁeld site to ﬁll in the sampling
grids described above. All 10 g samples were collected as described above for the
multiple-grain sampling locations. The location of the center of each sample was
recorded with respect to the artiﬁcial coordinate system constructed.

This sampling scheme resulted in each of the three 1.6 x 2 m sampling grids being
characterized by two sets of samples: I. ten samples at each of ﬁve sampling scales (10
mg to 100 g), 2. ﬁfty 10 g samples spaced 8.5 to 200 cm apart. All samples were

immediately frozen and stored with dry ice while being transported to the laboratory,

where they were transferred to -20°C storage. The actual wet weight of all samples was

146

-., i...

 

 

 

recorded in the laboratory and used in statistical analyses. References made hereafter to

sampling scales such as the “10 mg” scale are for convenience.
Eubacterial Community Characterization by T-RFLP

T-RFLP was performed essentially as described in Liu (1997) and in chapter 1.
Community DNA was extracted from 10 mg to 1 g soil samples using the standard
Ultraclean Soil DNA extraction kit (Mo Bio Laboratories, Solana Beach, CA).
Preliminary experiments were performed to determine whether slurrying and
homogenizing 10 and 100 g samples with sterile water had any effect on the variability of
T-RFLP proﬁles of a set of samples: 1. Eight 10 g samples were slurried with 10 m1. of
sterile water and then homogenized by shaking on a vortexer. DNA was then extracted
from 0.5 mL subsamples (approximately 5% of sample) using the standard Ultraclean
Soil DNA extraction kit, and from the remainder of each sample using the Large-Scale
Ultraclean kit. 2. Seven 100 g samples were slurried and homogenized with 50 mL
sterile water. DNA was extracted from 0.5 mL subsamples (approximately 1% of
sample) using the standard Ultraclean kit, from 15 mL subsamples (approximately 20%
of 'sample) using the Large-scale Ultraclean kit, and from the remainder of each sample
using four large-scale Ultraclean kits which were then pooled together. T-RFLP was
performed and the variability of T-RFLP proﬁles for each set of samples was assessed as
described below. Based on the results of this experiment (see Results below), remaining
100 g samples were slurried and DNA was extracted from 15 mL subsamples, DNA was
extracted from entire 10 g samples at multiple-grain sampling locations, and 10 g samples
at the remaining 40 locations per grid were slurried and DNA was extracted from 0.5 mL

subsamples.

147

..A‘ --—_-:--n—— .

 

 

" 5““ if?! i‘!:‘“"i§f$.‘\*i‘(l§.f§§t9'?’é"'

 

Genomic DNA was found to be of sufﬁcient purity to be used directly in PCR
reactions. PCR was performed using a standard reaction mixture of 160 M of each
deoxynucleoside triphosphate, 3 mM MgC12, 0.05 U/uL Taq DNA polymerase and the
appropriate volume of accompanying 10X PCR buffer (Gibco BRL, Gaithersburg, MD),
and 0.2 ug/mL bovine serum albumin (Boehringer Mannheim Biochemicals,
Indianapolis, IN). PCR mastermix, without primers, and PCR reaction tubes were
sterilized for 14 minutes with direct ultraviolet radiation in a Cleanspot PCR/UV
workstation. Primers used were the general eubacterial primer 8-27F
(AGAG'ITI‘GATCCTGGCTCAG, E. coli numbering, Amann et al. 1995, Integrated
DNA Technologies, Coralville, IA) and the universal primer 1392-1406R
(ACGGGCGGTGTGTACA). PCR reactions were optimized for each sample of
genomic DNA using a master mix with primer concentrations of 0.4 pM. Optimizations
were performed by adjusting the amount of genomic DNA extract used (0.4 to 15 ulJSO
uL reaction) and the number of PCR cycles run (18, 22, 28, or 33) to obtain a strong band
without visible non-speciﬁc product. PCR was performed in a Perkin-Elmer 9600
thermocycler using an initial denaturation step of 95°C followed by 18-33 cycles of the
following program: denaturation at 94°C for 30 sec., primer annealing at 55°C for 30
sec., and extension at 72°C for 30 see. A modiﬁed hot start procedure was used where
PCR tubes were not placed in the thermocycler until the block temperature had reached
80°C. A ﬁnal extension at 72°C for 7 min. was performed after the programmed number
of cycles was complete. PCR product was checked by electrophoresis on a 1% agarose

gel stained with ethidium bromide.

148

 

 

 

 

PCR reactions (50-75 uL) were performed in triplicate for each sample using the

optimal conditions found previously. These reactions were performed using the same
PCR master mix and program described above except that the forward primer was 0.6
uM hexachloroﬂuorscein (hex)—labeled 8-27F (Integrated DNA Technologies). Negative
controls (no genomic DNA) were conducted with every PCR and run on several
Genescan gels. Contamination in PCR reactions was not detected. Small peaks
occasionally appeared in negative control lanes on Genescan gels, but the cumulative
peak height was always below 1000 units. Samples were re-run if the cumulative peak
height was below 9500 ﬂuorescence units.

PCR replicates were pooled and puriﬁed using the Promega PCR Preps Wizard
Kit as directed by the supplier, except that elution was performed with 19 [L of sterile
water heated to 55-65°C. Five uL of puriﬁed PCR product was mixed with 5 [L of
restriction enzyme master mix containing 1.5 U/uL of restriction enzyme and one pl. of
the accompanying reaction buffer (Gibco). Restriction reactions were incubated for three
hours at 37°C, followed by 16 min. at 65°C to denature the restriction enzyme. Three uL
of the restricted PCR product was mixed with one “L of 2500 TAMRA size standard
(Applied Biosystems Instruments, Foster City, CA). DNA fragments were separated by
size by electrophoresis at 1800 V for 14 hours on an AB1373 automated DNA sequencer
at Michigan State University’s DNA Sequencing Facility. The 5’ terminal fragments (T—
RFs) were visualized by excitation of the hex molecule attached to the forward primer.
The gel image was captured and analyzed using Genescan Analysis Software 3.1. A peak
height threshold of 50 ﬂuorescence units was used in the initial analysis of the

electropherogram. T-RFLP proﬁles for all samples were generated using the restriction

149

 

 

enzyme RsaI. MspI was used to generate additional proﬁles for all samples in grid 2, and

Hhal and HaeIII proﬁles were also generated for several samples in grid 2.

The effect of the number of cycles of PCR on variability of the T-RFLP proﬁles
was assessed using ﬁve samples for which optimal PCR conditions had been determined
as 22 PCR cycles and 1.2 uL genomic DNA per 75 uL reaction. Optimal PCR conditions
for 1:6 dilutions of these samples was determined as 28 cycles and 0.6 uL genomic DNA
per 75 [IL reaction. When 1:6 diluted samples were PCR-ampliﬁed using 22 cycles and

0.6 uL genomic DNA per 75 1.1L reaction, the PCR product was approximately 1/8 the
strength of an optimal reaction. Three analytical PCR replications were then performed
for each of these diluted samples at 28 cycles and 24 replications were performed at 22
cycles using the same master mix. The pooled replicates then had equivalent amounts of
DNA in each sample, differing only by the number of cycles and analytical replicates
used. The 24-replicate pooled samples were concentrated using a Speed-Vac centrifuge.
These test samples were then analyzed by T-RFLP as described above. Variability was

compared between the sets of samples run at different numbers of PCR cycles.

Data Analysis Methods

The details of T-RFLP data processing have been described elsewhere (chapter 2).
Brieﬂy, all T-RFLP proﬁles were standardized to a cumulative peak height of 10,000
ﬂuorescence units, except for the few proﬁles that were used that had a cumulative peak
height less than 10,000 ﬂuorescence units. T-RFs with a peak height of less than 50
ﬂuorescence units after standardization were thrown out. This is equivalent to using

relative abundance of T-RFs that were strong enough to have been detected if the

150

 

 

 

 

 

 

cumulative peak height had equaled 10,000 ﬂuorescence units in all samples. T-RFs

were aligned in an Excel spreadsheet against a set of categories previously deﬁned for
proﬁles of the particular restriction enzyme used. Sample identities were concealed
during alignment.

All data analyses were performed using the computer software Microsoft Excel or
SAS version 8 (principally ML and Stat software components) except as otherwise
noted.
Variability Statistics

Statistical analyses were based on several scores designed to describe the
variability or heterogeneity of a set of T-RFLP proﬁles. Mean variance, or one half of
mean Euclidean distance, was calculated from relative abundances because of its
precedence in studies on spatial structure. Other statistics were used because they have
been shown to be superior to mean variance in comparison of proﬁles of multi-species
communities. Mean Hellinger variance (HV) was adopted based on the recommendation
of Hellinger distance in Legendre and Gallagher (in press) for use in analysis of species
abundances. Mean Hellinger variance was determined by calculating the mean variance

after a Hellinger-transformation, resulting in:

2
1 1 " ’yr _. _ 1" yr-
HV=— _ —L-y. , y .=— "
pj=l n-1;[ yi+ HI] ”I n; yk+

where p is the total number of T-RFs, n is the number of proﬁles in a set, y is T-RF peak

 

height, and y,,. is the sum of all peak heights in proﬁle i. This is also equivalent to one

half of the mean Hellinger distance among samples, which is determined by calculating

151

 

 

 

 

 

 

mean Euclidean distance after Hellinger-transformation (Legendre and Gallagher in

press).

The J accard similarity coefficient takes into account presence-absence data and
only uses information from a variable (T -RF) when it is present in at least one of the two
proﬁles being compared. It is calculated:

a
5,:
a+b+c

 

where a is the number of T-RFs present in both proﬁles being compared, and b and c are
the number unique to each proﬁle. The distance complement of the J accard coefﬁcient
(JD=1-S J) was calculated and will be referred to here as Jaccard distance. Mean Jaccard
distance is simply the mean value from a set of pairwise comparisons.

Heterogeneity among T-RFLP proﬁles was also compared using the mean
absolute deviation of standardized t-scores (MAD—t scores), an index derived from

recommendations in Edgington (1995), and calculated:

yij_yj
Si

where y]. is the mean peak height and Sj is the standard deviation of peak height for T-RF

M.M.: [1:

p j=l n i=1

 

j in the n proﬁles being compared. A MAD-t score equals the mean number of standard
deviations any given peak height is from the mean height for that T-RF. The overall
MAD-t score is calculated by averaging MAD-t scores calculated for each T-RF. Hence
rare and common T-RFs are equally weighted and it is a true index of data “evenness”.
Unlike HV and JD, calculation of MAD-t scores must be based on an entire set of

samples, and not sets of sample pairs.

152

   

 

DNA Extraction Experiment

For the preliminary DNA extraction experiment conducted on 10 and 100 g soil
samples, mean variance, HV, mean JD, and MAD-t scores were calculated for each set of
DNA extractions. The signiﬁcance of the differences found in MAD-t scores were
determined using distribution—free random permutation tests as described by Edgington
(1995). The null hypothesis was that the difference in MAD-t scores for a pair of DNA
extraction treatments was not different from that which would be found if the proﬁles
were randomly associated with the treatments. Empirical differences in MAD—t scores
were compared to a distribution of differences generated by 9999 random permutations of
proﬁles with respect to their identities in the two treatments being compared. The p-
value for such a test is the proportion of all values, including the empirical value, that are
equal to or greater than the empirical value.

Multiple-Grain Analysis

Variability-scale analysis was performed by calculating mean JD, log HV, and
MAD-t scores for each set of T-RFLP proﬁles representing samples taken at a given scale
from a given grid. Scores were plotted as a function of log mean soil weight, which
represents the sampling grain. Least-squares regression models of the variability
statistics as a function of log soil weight were generated. MAD-t scores for the MspI
proﬁles were shifted above scores for the RsaI proﬁles; therefore residuals about the
means of MspI and RsaI MAD-t scores were used in regression. The ﬁt of simple linear
regression (without hierarchical structure) was compared to more complex hybrid models

(indicating hierarchical structure) using partial F-tests.

153

 

 

 

 

 

T-RF covariance matrices were also constructed for each set of samples from the

Hellinger-transformed peak heights. These were then compared by calculating the
Pearson correlation coefﬁcient between elements of the matrices (also called the
normalized Mantel statistic or cophenetic correlation). A correlation super-matrix was
constructed by pairwise comparison of all covariance matrices. This was then analyzed
by principal coordinates analysis and cluster analysis using average linkage (UPGMA)
and Ward’s linkage methods.
Extent-Based or Separation Distance Analysis

For each set of ﬁfty 10 g samples, the signiﬁcance of differences in T-RFLP
proﬁles due to subsets being run on different gels was tested using redundancy analysis in
the software Canoco (Microcomputer Power, Ithaca, NY). This test compares the
amount of variability in the proﬁles that can be accounted for by external data (such as
gel identity) to a distribution generated by random permutations of the proﬁle identities
(Legendre and Legendre 1998). To perform redundancy analysis on ID, principal
coordinates analysis was ﬁrst used to derive new variables from the pairwise JD matrix
(Legendre and Anderson 1999). Euclidean distance calculated from these variables
preserves the Jaccard distances of the original matrix. The proportion of variability that
could be accounted for in the T-RFLP proﬁles was signiﬁcant for several grids (p=0.0001
to 0.37, 9999 random permutations). Therefore all further extent-based analyses of the
10 g samples were performed on residuals from multiple linear regression accounting for
the gel-effect. Regressions were performed directly on Hellinger-transformed peak

height or on ID principal coordinates. For some grids this had a large effect on the

154

.4“’.__..

MYE‘.

 

 

 

 

 

analysis, and for some it did not (usually accounting for approximately 10% of the total
variance in the datasets, but for as low as 4% and as high as 22%).

Various different sizes of distance classes (lag distances) were investigated for
grouping pairs of samples. A lag distance of 20 cm was chosen so that there would be
greater than 30 pairs of samples in all distance classes. The distance classes with the two
greatest lag distances were excluded from analyses to avoid edge effects from classes
containing only pairs where both samples were from the edges of the grid. Variability-
scale analysis was conducted by calculating HV and mean JD values for each distance
class and plotting as a function of mean geographic distance. Both isotropic and
anisotropic pairings of samples were used to look for directional differences. Isotropic
values were used in subsequent analyses.

The signiﬁcance of the spatial autocorrelation within each distance class was
tested by comparison of HV and mean JD to distributions created by 9999 random
permutations of the geographic coordinates of samples. The speciﬁc null hypothesis
tested was that the statistic from a distance class was not more different from the mean
value over all sample pairs than would be expected due to chance (i.e. no spatial
autocorrelation, Legendre and Legendre 1998).

Exponential or linear models, with nugget effects, were then ﬁtted to the distance
plots using weighted least-squares regression (Cressie 1985). The forms of these models
are, respectively:

V = C0 + Cum) (1 - e""’“’) and V = C0 + gum}:

where V is HV or mean JD, h is the distance between samples, a is one-third of the

“practical” range of spatial autocorrelation for exponential models, Co is the nugget (or

155

 

 

~—' .41 W"

 

 

 

 

 

unaccounted for variability), and C1 is the sill (or variability associated with spatial

structure) in the exponential model and the change in variability with separation distance
between samples in the linear model.

New distance functions for HV were constructed for alternative sampling grains
from the ﬁtted exponential models using the principals described in Bellehumeur et al.
(1997). Brieﬂy, a theoretical point model was formed by assuming: ap=(3a-L)/3 and
C1(¢,p)p=C1/(1-F), where L is the distance that the edges of samples were brought closer
together by taking real samples instead of “point samples” (i.e. the diameter of the
samples), and F is the proportion of the structural variability that is masked by taking real
samples. F is calculated by stochastic integration of an exponential model (without
nugget) with sill equal to one and range equal to ap. The integration was accomplished
by generating 100,000 random lags (h) between zero and L, constraining the distribution
by the geometry of the sampling area. The exponential portion of the distance model for
an alternative sampling grain was then constructed by ﬁnding L and F for the new grain
and rearranging the above equations to solve for C1 and a. The unexplained variability Co
was not discounted as a function of the ratio between the new and original sampling
grains since this procedure assumes that there is no analytical variability contributing to
Co; in fact, preliminary experiments showed analytical variability may be equal to Co
(data not shown), so the Co was not adjusted with sampling grain.

The distance functions for alternative sampling grains were used to estimate HV
of the set of multiple-grain sample locations had they been sampled at the alternative
grains. This was done by ﬁnding the number of pairs of samples within this set that fell

into each of the 20 cm lag distance classes. The mean separation distance within each

156

 

 

 

 

 

class was used to calculate an HV for that class using the new distance function, and the
overall HV for the entire set of multiple-grain samples was calculated by ﬁnding the
weighted mean HV of all classes. These estimates were then plotted as a function of log
sampling grain with the empirical multiple-grain location data described above.

T-RF covariance matrices were calculated from Hellinger—transformed peak
heights for each lag distance class. A correlation super-matrix was constructed and
analyzed using principal coordinates analysis and cluster analysis, as described above.

Redundancy analysis was also used to perform multivariate trend-surface analysis
using the computer software Canoco. The external data used were the geographical x,y
coordinates of each sample, and combinations of the coordinates to the third power (x2,
y2, xy, x3, y3, xzy, xyz) as suggested by Legendre and Legendre (1998). The signiﬁcance
of the variability explained by these geographical variables was tested with 9999 random
permutations. Forward selection of external variables was also used to check the
signiﬁcance of each geographical term after accounting for the terms that explained more
variability.

Mantel correlograms were constructed from pairwise T-RFLP distance matrices
and a series of similarity matrices indicating presence of pairs in the distance classes
(Legendre and Fortin 1989, Legendre and Legendre 1998). Therefore a negative
normalized Mantel statistic for a distance class indicates positive spatial autocorrelation
at that scale. The normalized Mantel statistic was also calculated between the T—RFLP
and geographic distance matrices to test for an overall effect of separation distance across
all scales (with the opposite interpretation of the sign of the statistic). Mantel statistics

were tested for signiﬁcance using 9999 random permutations of the geographic

157

 

 

 

 

 

 

 

 

coordinates of the samples. The null hypothesis tested is therefore that the empirical
Mantel statistics are not more different from zero than would be expected by random

arrangement of the samples (i.e. no spatial structure).

Results
DNA Extraction Experiment

In the DNA extraction experiment, HV, mean JD (data not shown) and MAD-t
scores (Figure 2) were higher for the set of DNA extractions derived from 5%
subsamples of the slurried 10 g samples when compared to the set of 95% subsample
extracts. Random permutation testing of the signiﬁcance of the difference in MAD-t
scores resulted in a marginally-signiﬁcant p-value (0.089). It was therefore decided to
extract DNA from whole samples for the 10 g samples from multiple-grain locations.
There were ten such samples from each of the three sampling grids. The remaining forty
10 g samples from each grid were slurried and DNA was extracted from 5% subsamples
to reduce the cost of DNA extraction. It was felt this was justiﬁable due to the increased
number of samples in this analysis and the relatively weak effect of subsarnpling.

For the 100 g samples, variability statistics decreased in the order of 1%, 80%,
and 20% subsamples (see Figure 2). Random permutation testing found that the
difference in MAD-t scores between the 1% and 20% DNA extracts were marginally-
signiﬁcant (p=0.069), while neither score was signiﬁcantly different from that for the set
of 80% DNA extracts. These somewhat contradictory results led to the decision that 100

g samples would be slurried and DNA would be extracted from 20% subsamples.

158

(918£9'{>v,fl '(I'.:g"i‘€ti‘ffygi ‘

 

 

 

 

 

 

PCR Cycles Experiment

The effect of number of cycles on heterogeneity of a set of samples was
contradictory to the hypothesis based on the theory that increased numbers of PCR cycles
increases the number of nucleotide mismatches due to random polymerase mistakes. The
HV was higher for the set of samples ampliﬁed using 22 PCR cycles, diluted genomic
DNA and 24 PCR replicates than either the set from 28 PCR cycles, diluted genomic
DNA and 3 PCR replicates or the set from 22 PCR cycles, undiluted genomic DNA, and
3 PCR replicates. The HV of the latter two sets were approximately equal (data not
shown). Pooling of PCR replicates is performed to reduce the signiﬁcance of PCR
artifacts; however these results indicate that this may also increase the overall
heterogeneity of PCR amplicons.

Grain-Based Analyses

The overall trend for all variability statistics was a decrease with increasing scale
of sampling grain (see Figure 3). The slope of a simple linear regression of log HV on
log soil weight is much closer to zero than -—1 (although it is also signiﬁcantly different
from zero, p=0.00005). The slope for variance is almost the same as for HV (data not
shown). While this could be taken to imply that there is strong autocorrelation in the
peak heights themselves (i.e. a small-grain sample proﬁle is almost the same as its
corresponding larger-grain sample proﬁle from the same location), that is not the case for
this set of T-RFLP proﬁles. Based on the fairly noisy proﬁles we obtained, we suggest
that instead this result implies that there are self-similar (or fractal) patterns across these
scales. However, the fact that there is a signiﬁcantly negative slope for HV as well as

mean JD and MAD-t implies that pattern from some phenomena (and its accompanying

159

 

 

 

 

 

 

 

variability) are not completely fractal and are not being detected at the larger sampling
grains.

While simple linear regression explained a signiﬁcant portion of the relationship
between the variability statistics and sampling grain, distinct zones which are
characterized by unique slopes are apparent from the data plots, indicating hierarchical
structure. Partial F-tests indicated that models which include a change in slope (which
will be referred to as hierarchical models) had signiﬁcantly reduced residual sums of
squares for HV and MAD-t when compared to simple linear regression (see Table 1 and
Figure 3). The hierarchical models were constructed by separate linear regressions in
regions where the slopes appeared to differ. The slope of a regression was discarded
from the model if it was not signiﬁcantly different from zero. The phase-shift region
where the slope changes is undeﬁned in these models since there is no data describing the
relationship there. The hierarchical model for mean JD was not signiﬁcantly superior to
the linear model, although it did have a lower residual sum of squares. The HV and mean
JD plots had changes in slope between sampling scales different from where the slope
changed for MAD-t scores. This is not overly surprising since MAD-t scores weight rare
T-RFs much more heavily than HV or JD. Potential outliers were not removed in the
testing of these models. Removal of outliers would cause the hierarchical model to ﬁt the
data signiﬁcantly better than the linear model for mean JD, remove the hierarchical
relationship for MAD-t, and have no effect on HV.

The correlation between Hellinger covariance matrices from different sampling
grains was uniformly low (ranging from —0.03 to 0.31, mean 0.08), resulting in

uninformative principal coordinates and cluster analyses of the correlation super-matrix.

160

 

 

 

In general, comparison of covariance structures was not as useful as mean variability of

T-RF peak heights for this dataset.
Extent-Based Analyses

As expected, HV and mean JD increase as separation distance between samples
increases, as shown for HV in Figure 4. Mean JD plots looked almost identical to HV
plots, except for the scale of the y-axis (data not shown). Hellinger variance for grids 2
(RsaI and MspI digests) and 3 (RsaI) reach an asymptote between approximately 50 and
75 cm, while the HV plot for grid 1 (RsaI) rises linearly across the entire range of
separation distances. Permutation testing indicated that one or two of the ﬁrst few
distance classes had HV signiﬁcantly below the mean HV, indicating signiﬁcantly
positive spatial autocorrelation at that scale (see Figure 4).

Exponential models were ﬁt to HV plots for grids 2 and 3 and a linear model was
ﬁt for grid 1 (see Figure 4 and Table 2). Spatial structure accounted for 18 to 23% of the
total variance in each grid dataset. Estimates of HV for the multiple-grain sampling
locations based on the ﬁtted exponential models are shown in Figure 5. The range over
which the estimation was made overlaps with two sampling grains for which samples are
available (10 and 100 g). The 10 g samples at multiple-grain locations went into
construction of the models, along with the forty other 10 g samples per grid, and the
empirical HV values are very close to the estimated values. The model estimates for the
100 g samples are quite close to the empirical values for grid 2 digests, but are well-
above the empirical value for grid 3. This may be due to noise within the empirical
values, which are based on fewer samples than the sets from which the models were

constructed (10 versus 50 samples). Between 10 and 1000 g the HV estimates are

161

"P

 

 

 

 

 

 

 

 

relatively ﬂat, in agreement with the hierarchical least-squares model ﬁtted to the
empirical values in the grain-based analysis (see Figure 3). At scales greater than 1000 g
the slopes change, becoming noticeably steeper. There may be some ﬂattening of the
curve again as soil weight approaches 105 g, although this transition is not completely
represented within the range shown.

Inferences from the extent-based models to other sampling grains outside the
range estimated over in Figure 5 cannot be made since they would involve either I .
smaller sampling grains than that which the model was based on or 2. sampling grains
larger than the area sampled. It should be noted that the estimates for samples larger than
10‘1 g are theoretical since samples taken at such a scale would overlap. However, the
spatial structure reﬂected in HV estimates, even over theoretical sampling grains, is
certainly real, being based on a model derived from empirical extent-based data. The
construction of estimates and theoretical samples is a convenient way to integrate the
grain-based and extent-based analyses. Hierarchical spatial structure is evident from the
HV estimates since there are two well-deﬁned regions with differing slopes. Hierarchical
structure is not evident and estimates cannot be made, however, from the grid 1 extent-
based analysis since the relationship between HV and separation distance for grid 1 is
linear.

Analysis of the correlation super-matrix describing the relationship between
Hellinger covariance matrices deﬁned by distance classes found that the ﬁrst distance
class (0 to 20 cm separation distance) had a unique T-RF covariance structure in all four
grids (see Figures 6 and 7 for examples). Cluster analysis also resulted in distance

classes 8, 9, and 10 being clustered separately from classes 2 through 6, with distance

162

 

 

 

 

 

 

class 7 being in either cluster depending on the grid. Principal coordinate analysis

showed distance classes 2 through 10 in a gradient without strong hierarchical structure,
although classes 9 and 10 were always somewhat separate from the rest of the gradient.
When covariance matrices from the grain-based analysis were included with those
derived from distance classes, matrices derived from samples taken at alternative
sampling grains were identiﬁed as outliers (data not shown).

Redundancy analysis found linear combinations of geographical terms did
account for a signiﬁcant portion of the variability in Hellinger and J accard distances in all
cases except one (see Table 3). When the analysis was limited to those two or three
geographical terms found to be statistically signiﬁcant using forward selection of
variables, the amount of variation explained dropped from 20-25% to 5-13%.
Geographical terms consistently explained a greater portion of the variability in Hellinger
distances than in J accard distances.

Global Mantel tests found weak but signiﬁcant correlation between geographic
distance and Hellinger distance between T-RFLP proﬁles for all grids (see Table 3). The
same trend was present but not as strong for J accard distance. Mantel correlograms
(where negative values imply positive spatial autocorrelation within a distance class,
relative to the rest of the dataset) indicated that the statistically signiﬁcant spatial
structure was short-range autocorrelation (<50 cm; see Figure 8). The relationship
between rM and separation distance is essentially linear for the HD-based Mantel
correlograms, as well as the grid 1 JD-based Mantel correlogram. Jaccard distance-based

Mantel correlograms for the other grids reach an asymptote after approximately 70 cm.

163

 

 

 

- -—- -ox._——a—'_

 
 

 

       

orfw'u ‘ t?"‘f'.~’.??l»‘.‘£'. 11".!!er$Hsbl§ifif1f§$¥ffil :- .

Discussion
Hierarchical Spatial Structure Detected within the Soil Microbial Community

The soil bacterial community was shown to have signiﬁcant spatial structure in an
area that is relatively small and homogenous from the perspective of the plant
community: the layer of soil at a depth of from 2 to 4 cm, over a 1.5 x 2 m area of soil,
under perennial alfalfa. Furthermore, this spatial structure was shown to be organized by
at least three hierarchical levels. The levels of organization are hierarchical since they
operate within distinct ranges of spatial scale, or scale domains (W iens 1989), and hence
variability within the lower levels are affected by higher levels, but not vice versa. Note
that while mechanisms operating at low levels within the hierarchy will not affect
variability at scales larger than the level’s scale domain, this does not imply that those
mechanisms will not affect other aspects of phenomena at larger scales. In particular,
mechanisms causing patchiness in T-RF abundance at lower levels may determine mean
abundance in samples taken at larger scales.

While this study was not designed to determine the causes of the hierarchical
spatial structure, it is possible to hypothesize what they may be. The level operating at
smaller scales, 10 mg to 1 g, or 1 to 13 mm, may be regulated by resource heterogeneity
at that scale, such as proximity of a site to individual roots, pieces of decomposing shoot
residue, or macropores. These create gradients of water, organic matter, terminal electron
acceptors, and other organisms (see chapter 3). Aggregate structure and disturbance
history (such as by soil fauna or aggregate disintegration) may also affect communities at
this scale (see chapter 4). The scale domain of the highest level detected (1000 to 105 g

or 25 to 220 cm) has been intensively investigated recently (e.g. Boemer et a1. 1998,

164

     

 

 

 

Schlesinger et al. 1996, Stoyan et al. 2000). Therefore it is well known that there is

considerable resource heterogeneity at these scales, a large portion of which is generated
by the plants themselves. We suspect heterogeneity of plant density, organic matter,
phosphorus, pH, etc. regulates eubacterial community structure at this level.

The intermediate level (10 to 1000 g or 2 to 25 cm) is interesting because there is
virtually no change in variance with scale. This implies that there are few important
mechanisms regulating eubacterial community structure that operate at these scales. If
strong mechanisms did exist at these scales, variance would decrease with increasing
sampling grain since the spatial patterns generated by the mechanisms would become
increasingly diluted. Instead it appears that the scale is not yet large enough to begin to
decrease the variability due to spatial structure generated at the higher level, but it is large
enough to mask all variability due to spatial structure generated at the lower level. The
implications of this for designing future sampling is discussed below.

Results of this study indicate greater spatial structure in the soil microbial
community than was found previously by Cavigelli et al. (1995) or Felske and
Akkermans (1998). One difference is that sampling in this study was extended to much
smaller spatial scales. This has also been advocated for more detailed analyses in soil
microbial diversity studies (Grundmann and Gourbiere 1999). The above studies also
skipped quantitative comparison of community proﬁles altogether (Felske and
Akkermans 1998) or used univariate geostatistics without testing for structure within the
community as a whole (Cavigelli et al. 1995). Hopefully the statistical methods outlined
here will contribute to the ability to perform true tests of the hypothesis of spatial

structure within a community, microbial or otherwise.

165

 

Comparison of Statistical Methods

The hierarchical structure was detected by extending previous work on univariate
spatial structure to the use of community distance coefﬁcients under the general name of
variability-scale analysis. It was shown that extent- and grain-based approaches can be
fully integrated if they are based on the same statistic. This was accomplished by
modeling the spatial structure of the statistic in the extent-based approach and then using
the method of Bellehumeur et al. (1997) to derive new models at different sampling
grains. The new models are then applied to the same locations that were sampled using
the grain-based approach. This integration of methods is an important development
because grain- and extent-based methods have the same goal but are practical over
different spatial scales. For example, it would have been unrealistic to use the grain-
based approach at scales much larger than 100 g because of the difﬁculties in obtaining a
representative DNA extract from large samples. Likewise, using the extent-based
approach to try to examine small-scale structure would have either resulted in an
unmanageable number of samples or much less sensitivity at all scales.

The approach taken here of testing for changes in slope of the variability-scale
plots using regression and partial F—statistics may be questioned since the datapoints are
derived from essentially the same sets of locations and are therefore not independent.
However, as pointed out by O’Neill et al. (1991a), this bias will make the analysis
conservative since successive points will tend to be similar and reduce the likelihood of
ﬁnding changes in slope.

Several other extent-based approaches were also used for comparison to the

extent-based variability-scale analysis. Redundancy analysis found that 20-25% of the

166

 

 

 

 

 

 

 

 

 

 

variability in the 10 g samples was due to spatial structure, in agreement with the extent-

based variability-scale analysis (see Tables 2 and 3). A disadvantage of redundancy
analysis is that it requires many geographical terms to detect complex spatial structures,
causing over-parameterization of the analysis and inﬂating p-values. However, forward
selection of a few terms resulted in a subset that was not adequate to account for as much
variability as the extent-based variability-scale analysis. Maps of the sampling area could
be constructed from canonical principal components scores derived from the redundancy
analysis; however these must be interpreted with caution since they are “best-case”
scenarios, dependent on the geographical terms present, and each map would show only
the portion of the total spatial structure present that its principal component had captured.
Hence the form of the spatial structure is more difﬁcult to ascertain for complex datasets
when using redundancy analysis. Global Mantel tests also showed there was a signiﬁcant
correlation between geographic distance and Hellinger distance (see Table 3). As has
been pointed out in the past (Legendre and Fortin 1989), this test is most sensitive to
gradients across the sampled area or other simple structures. Mantel correlograms
showed similar forms of spatial structure to those found by extent-based variability-scale
analysis, as well as signiﬁcant spatial autocorrelation in the same distance classes (see
Figure 8). The strength of this spatial structure (i.e. the amount of variability accounted
for), however, is not readily apparent from the Mantel correlograms for reasons discussed
previously. Neither Mantel correlograms nor redundancy analysis can be fully integrated
with a grain-based approach.

Another statistical method proposed here that can integrate extent- and grain-

based approaches is the exploratory data analysis of a correlation super-matrix. This

167

t re'Its:9rrrF(51‘!szrrsrvgetgsqchrtﬁ$frtf

 

method is loosely based on Mantel procedures since correlation coefﬁcients between
matrices are used. The pairwise correlations from a set of matrices are used to create a
correlation super-matrix which summarizes the relationships between all matrices. The
matrices used here were T-RF covariance matrices. Cluster analysis and principal
coordinates analysis were then performed on the correlation super-matrix to summarize
these relationships. The result is an analysis of the similarities in the covariance
structures for each scale separately analyzed. Scales from both extent- and grain-based
approaches can be used. Analysis of the correlation super-matrix from the extent-based
approach reinforced the general results found in variability-scale analysis. These results
are more general, however, implying not only that communities are more similar when
they are closer together, but also that the interactions between T-RFs are organized
differently at different spatial scales. This was particularly true for the transition between
the 0-20 cm separation distance scale and other scales, but also true for the 140-200 cm
scale compared to other scales (see Figures 6 and 7). The former transition between
covariance structures is at the same scale as the transition between levels of organization
detected by variance-scale analysis at approximately 1000 g soil.

The attempt to integrate the extent- and grain-based approaches using the
correlation super-matrix analysis was not fruitful for the current dataset. This was
because the covariance matrices from different sampling grains were widely different
from one another and from the covariance matrices from the extent-based approach
(except for the 10 g samples). It may be that dynamics at the sampling grains used were

so diverse that the covariance structures were completely different, or that there was

168

   

 

signiﬁcantly greater noise in the covariance matrices from the grain-based approach,

being based on fewer samples.
Interpretation of Molecular Community Fingerprint Data

Structure detected using MAD-t scores was somewhat different from what was
detected using HV, with a phase transition between 10 and 100 mg. MAD-t scores
analyze the T-RFLP proﬁles from a different perspective, weighting rare (i.e. less-
frequently occurring) T-RFs equally with common ones, unlike either HV or mean JD.
With each transition between sampling grains, higher MAD-t scores were obtained,
indicating more variability and more rare T-RFs being detected. The jump in MAD-t
scores between 10 and 100 mg was disproportionate, however, to the increases between
all other scales (see Figure 3). It could be hypothesized that the size of many micro-
habitats that support populations of specialized eubacteria are approximately 10 mg in
weight, or a mm in diameter. However, it must be noted that T-RFLP shares with all
other PCR-based techniques the potential problems of methodological artifacts such as
chimera formation and point mutations (W intzingerode et al. 1997). Artifacts are known
to increase with PCR cycle number, although variability actually decreased with cycle
number in a preliminary experiment presented here.

Mean JD, based on presence of T—RFs only and containing no information from
the peak heights, portrayed the same trends as HV but generally evidence of structure
was weaker. Utilization of peak height data enhanced the detail with which the
community could be viewed using T-RFLP.

Molecular methods were used to avoid the severe biases in community

composition known to occur with culture-based methods. DNA extraction and PCR

169

 

 

 

"HF 5141-! \.H-vetv-.-liet§5rifﬂt."'" "

methods have their own biases, and so the structure detected in this study does not
include those organisms not amenable to DNA extraction and PCR ampliﬁcation using

our methods. Great precaution was taken to ensure that biases would not occur due to the

scale or position of samples.

Conclusions

This study is a snapshot of the microbial community structure, which is dynamic
due to the seasonality of root growth, alfalfa harvests, and abiotic factors such as
temperature, moisture, etc. It is a holistic snapshot, both from the viewpoint that the
strength of many ecological phenomena were tested simultaneously, and that the structure
was detected from the distributions of all T-RFs and their interactions. While signiﬁcant
random variability did exist (not associated with the spatial structure), the results
strengthen the theory that soil eubacterial communities are not simply random
assemblages formed by historical events. They display structure that suggests that at
least some species are regulated by mechanisms analogous to those for macroorganism
distributions.

The appropriate scale of sampling will depend on the goals of the analysis. Often
the goal is to assess the effects of some ﬁeld treatment. If the hypothesized effect is on
variability or spatial structure, a multiple-scale approach such as the one here should be
taken. If, on the other hand, the goal is to test for changes in the average community
composition without regard to spatial structure, the sort of “null level” detected in this
study (between 10 and 1000 g) presents a unique opportunity. If treatments are blocked

within the scale domain of such a level, and samples are also taken at a grain within the

170

 

 

 

 

 

 

   
  

A

 

. . . r. u < o u nyu 124;?" «trance?! stwstutitqtqﬁti€‘-§Ff'€'

scale domain, then increased variability, decreased sensitivity, and non-independence of
replicates due to spatial structure is avoided. Otherwise, many samples must be taken in
order to characterize the spatial variability, or they can be composited in an attempt
(probably in vain) to overcome the spatial variability. Likewise, we do not feel there is a
unique “scale of a community”, in part because communities are affected in some way by
mechanisms operating at all scales from molecular to global. Community composition is

ultimately affected by which species the ecologist chooses to include, and the particular

 

scales chosen for investigation. Therefore it is more useful to focus efforts on examining

how organization changes with scale than on ﬁnding “the scale” (Levin 1992).

171

 

 

 

References

Allen, T.F.H., T.B. Starr. 1982. Hierarchy: Perspectives for Ecological Complexity .
Chicago: University of Chicago Press.

Amann, R.I., W. Ludwig, K. Schleifer. 1995. Phylogenetic identiﬁcation and in situ

detection of individual microbial cells without cultivation. Microbiological Reviews
59:143-169.

Beare, M.H., D.C. Coleman, D.A. Crossley Jr, P.F. Hendrix, E.P. Odum. 1995. A
hierarchical approach to evaluating the signiﬁcance of soil biodiversity to
biogeochemical cycling. In The Signiﬁcance and Regulation of Soil Biodiversity, edited
by H. P. Collins, G. P. Robertson and M. J. Klug. Netherlands: Kluwer Academic
Publishers.

Beare, M.H., R.W. Parmelee, P.F. Hendrix, W. Cheng. 1992. Microbial and fauna]
interactions and effects on litter nitrogen and decomposition in agroecosystems.
Ecological Monographs 62:569-591.

Bellehumeur, C., P. Legendre, D. Marcotte. 1997. Variance and sampling scales in a
tropical rain forest: changing the size of sampling units. Plant Ecology 130:89-98.

Belyea, L.R., J. Lancaster. 1999. Assembly rules within a contingent ecology. Oikos
86:402-416.

Boemer, R.E.J., A.J. Scherzer, J.A. Brinkman. 1998. Spatial patterns of inorganic N, P
availability, and organic C in relation to soil disturbance: a chronosequence analysis.
Applied Soil Ecology 7: 159-177.

Borcard, D., P. Legendre, P. Drapeau. 1992. Partialling out the spatial component of
ecological variation. Ecology 73: 1045-1055.

Borrnann, F.H., G.E. Likens. 1979. Catastrophic disturbance and the steady state in
northern hardwood forests. American Scientist 67 2660-669.

Bouxin, G., N. Gautier. 1982. Pattern analysis in Belgian limestone grasslands. Vegetatio
49:65-83.

Carpenter, S.R., J .E. Chaney. 1983. Scale of spatial pattern: four methods compared.
Vegetatio 53:153-160.

Cash, G.G., J.J. Breen. 1992. Principal component analysis and spatial correlation:
environmental analytical software tools. Chemosphere 24: 1607-1623.

172

 

_. m w‘r

 

 

 

 

 

 

Cavigelli, M.A., G.P. Robertson, M.J. Klug. 1995. Fatty acid methyl ester (FAME)
proﬁles as measures of soil microbial community structure. Plant and Soil 170:99-113.

Cho, J., J .M. Tiedje. 2000. Biogeography and degree of endemicity of ﬂuorescent
Pseudomonas strains in soil. Applied and Environmental Microbiology 66:5448-5456.

Clements, RE. 1936. Nature and structure of the climax. Journal of Ecology 24:252-284.

Connell, J.H. 1978. Diversity in tropical rain forests and coral reefs. Science 199: 1302-
13 10.

Cressie, N. 1985. Fitting variogram models by weighted least squares. Mathematical
Geology 17:563-586.

Cullinan, V.I., M.A. Simmons, J .M. Thomas. 1997. A Bayesian test of hierarchy theory:

scaling up variability in plant cover from ﬁeld to remotely sensed data. Landscape
Ecology 12:273-285.

Dobermann, A., P. Goovaerts, T. George. 1995. Sources of soil variation in an acid
Ultisol of the Philippines. Geoderma 68:173-191.

Edgington, ES. 1995. Randomization Tests, 3 ed. New York: Marcel Dekker.

Eswaran, H., F.H. Beinroth, S.M. Virmani. 2000. Resource management domains: A

biophysical unit for assessing and monitoring land quality. Agriculture Ecosystems and
Environment 81:155-162.

Felske, A., A.D.L. Akkermans. 1998. Spatial homogeneity of abundant bacterial 16S
rRNA molecules in grassland soils. Microbial Ecology 36:31-36.

Fulthorpe, R.R., A.N. Rhodes, J .M. Tiedje. 1998. High levels of endemicity of 3-

chlorobenzoate-degrading soil bacteria. Applied and Environmental Microbiology
64:1620-1627.

Gaedke, U. 1998. The response of the pelagic food web to re-oligotrophication of a large
and deep lake (L. Constance): evidence for scale-dependent hierarchical patterns?
Ergebnisse der Limnologie 53:317-333.

Galiano, ER 1983. Detection of multi-species patterns in plant populations. Vegetatio
53:129-138.

Gleason, HA. 1939. The individualistic concept of the plant association. American
Midland Naturalist 21 :92-1 10.

Goodall, D.W. 1974. A new method for the analysis of spatial pattern by random pairing
of quadrats. Vegetatio 29:135-146.

173

 

 

w..-

 

 

Goovaerts, P. 1994. Study of the relationships between two sets of variables using
multivariate geostatistics. Geoderma 62:93-107.

Greenblatt, CL, A. Davis, B.G. Clement, C.L. Kitts, T. Cox, R.J. Cano. 1999. Diversity
of microorganisms isolated from amber. Microbial Ecology 38:58-68.

Grieg-Smith, P. 1952. The use of random and contiguous quadrats in the study of the
structure of plant communities. Annals of Botany New Series 16:293-316.

Grifﬁths, B.S., R.D. Bardgett. 1997. Interactions between microbe-feeding invertebrates
and soil microorganisms. In Modern Soil Microbiology , edited by J. D. v. Elsas, J. T.
Trevors and E. M. H. Wellington. New York: Marcel Dekker.

Grime, LP. 1994. The role of plasticity in exploiting environmental heterogeneity. In
Exploitation of Environmental Heterogeneity by Plants: Ec0physiological Processes
Above- and Belowground, edited by M. M. Caldwell and R. W. Pearcy. San Diego:
Academic Press.

Grundmann, L.G., F. Gourbiere. 1999. A micro-sampling approach to improve the
inventory of bacterial diversity in soil. Applied Soil Ecology 13: 123-126.

Hagen, M.J., J .L. Hamrick. 1996. A hierarchical analysis of population genetic structure
in Rhizobium leguminosarum bv. trifolii. Molecular Ecology 5: 177-186.

Harris, D., E.A. Paul. 1994. Measurement of bacterial growth rates in soil. Applied Soil
Ecology 1:277-290.

Harris, RI. 1994. Consequences of the spatial distribution of microbial communities in
soil. In Beyond the Biomass, edited by K. Ritz, J. Dighton and K. E. Giller.

Harrison, S. 1997. How natural habitat patchiness affects the distribution of diversity in
Californian serpentine Chaparral. Ecology 78:1898-1906.

Hill, MO. 1973. The intensity of spatial pattern in plant communities. Journal of Ecology
61:225-235.

Hoef, J .M.V., D.C. Glenn-Lewin. 1989. Multiscale ordination: a method for detecting
pattern at several scales. Vegetatio 82:59-67.

Hoekstra, T.W., T.F.H. Allen, C.H. Flather. 1991. Implicit scaling in ecological research.
BioScience 41:148-154.

Hoekstra, T.W., C.H. Flather. 1987. Theoretical basis for integrating wildlife in
renewable resource inventories. Journal of Environmental Management 24:95-110.

174

 
  

I
. Iatrle . cv-u- - :4wrt~-s~_:~-.w .v-asnstcz-r-turwutw-ﬁf("i’F’IYV

 

“.-

 

 

 

Holt, RD. 1984. Spatial heterogeneity, indirect interactions, and the coexistence of prey
species. American Naturalist 1242377-406.

Hraber, P.T., B.T. Milne. 1997. Community assembly in a model ecosystem. Ecological
Modelling 104:267-285.

Hugenholtz, P., B.M. Goebel, N .R. Pace. 1998. Impact of culture-independent studies on

the emerging phylogenetic view of bacterial diversity. Journal of Bacteriology
1 80:4765-4774.

Jennings, M.D., J .P. Reganold. 1991. Hierarchy and subsidy-stress as a theoretical basis
for managing environmentally sensitive areas. Landscape and Urban Planning 21:31-46.

J obson, J .D. 1992. Applied Multivariate Data Analysis Volume II: Categorical and
Multivariate Methods. New York: Springer-Verlag.

Johnson, L.B., S.H. Gage. 1997. Landscape approaches to the analysis of aquatic
ecosystems. Freshwater Biology 37: 1 13-132.

Jonsson, B.G., J. Moen. 1998. Patterns in species associations in plant communities: the
importance of scale. Journal of Vegetation Science 9:327-332.

Klironomos, J .N., M.C. Rillig, M.F. Allen. 1999. Designing belowground ﬁeld

experiments with the help of semi-variance and power analyses. Applied Soil Ecology
12:227-238.

Kotliar, N.B. 1996. Scale dependency and the expression of hierarchical structure in
Delphinium patches. Vegetatio 127:117-128.

Lee, D.C., W.E. Grant. 1995. A hierarchical approach to ﬁsheries planning and modeling
in the Columbia River basin. Environmental Management 19: l7-25.

Legendre, P., M.J. Anderson. 1999. Distance-based redundancy analysis: testing
multispecies responses in multifactorial ecological experiments. Ecological Monographs
69:1-24.

Legendre, P., M. Fortin. 1989. Spatial pattern and ecological analysis. Vegetatio 80: 107-
138.

Legendre, P., E.D. Gallagher. in press. Ecologically meaningful transformations for
ordination of species data. Oecologia .

Legendre, P., L. Legendre. 1998. Numerical Ecology, 2 ed. Amsterdam: Elsevier.

Leibold, M.A. 1995. The niche concept revisited: mechanistic models and community
context. Ecology 76: 1371-1382.

175

 

 

 

 

Levin, SA. 1992. The problem of pattern and scale in ecology. Ecology 73: 1943-1967.

Liu, W.-T., T.L. Marsh, H. Cheng, L.J. Fomey. 1997. Characterization of microbial
diversity by determining terminal restriction fragment length polymorphisms of genes
encoding IS rRNA. Applied and Environmental Microbiology 63:4516-4522.

Ludwig, J .A., D.W. Goodall. 1979. A comparison of paired- with blocked-quadrat
variance methods for the analysis of spatial pattern. Vegetatio 38:49-59.

MacArthur, R., R. Levins. 1967. The limiting similarity, convergence, and divergence of
coexisting species. The American Naturalist 101 :377-385.

Maurer, BA. 1990. Dipodomys populations as energy-processing systems: regulation,
competition, and hierarchical organization. Ecological Modelling 50:157-176.

Milne, B.T. 1991. Heterogeneity as a multiscale characteristic of landscapes. In
Ecological Heterogeneity, edited by J. Kolasa and S. T. A. Pickett. New York: Springer-
Verlag.

Minamisawa, K., H. Mitsui. 2000. Genetic ecology of soybean bradyrhizobia. In Soil
Biochemistry, edited by J. Bollag and G. Stotzky. New York: Marcel Dekker.

Monestiez, P., M. Goulard, G. Charmet. 1994. Geostatistics for Spatial genetic structures:

study of wild populations of perennial ryegrass. Theoretical and Applied Genetics 88:33-
41.

Morris. 1999. Spatial distribution of fungal and bacterial biomass in southern Ohio
hardwood forest soils: ﬁne scale variability and microscale patterns. Soil Biology and
Biochemistry 31 : 1375-1386.

Murphy, S.L., R.L. Tate, III. 1996. Bacterial movement through soil. In Soil
Biochemistry, edited by G. Stotzky and J. Bollag. New York: Marcel Dekker.

Muyzer, G. 1998. Structure, function and dynamics of microbial communities: the
molecular biological approach. In Advances in Molecular Ecology, edited by G. R.
Carvalho: IOS Press.

Noy-Meir, 1., D]. Anderson. 1971. Multiple pattern analysis, or multiscale ordination:
towards a vegetation hologram? In Many Species Populations, Ecosystems, and Systems
Analysis, edited by G. P. Patil, E. C. Pielou and W. E. Waters. University Park PA: The
Pennsylvania State University Press.

Ntisslein, K., J .M. Tiedje. 1998. Characterization of the dominant and rare members of a
young Hawaiian soil bacterial community with small-subunit ribosomal DNA ampliﬁed

176

 

 

from DNA fractionated on the basis of its guanine and cytosine composition. Applied and
Environmental Microbiology 64:1283-1289.

O'Connor, R.J., M.T. Jones, D. White, C. Hunsaker, T. Loveland, B. Jones, E. Preston.
1996. Spatial partitioning of environmental correlates of avian biodiversity in the
conterrninous United States. Biodiversity Letters 3:97-110.

O'Neill, R.V., D.L. DeAngelis, J.B. Waide, T.F.H. Allen. 1986. A Hierarchical Concept
of Ecosystems. Princeton, NJ: Princeton University Press.

O'Neill, R.V., R.H. Gardner, B.T. Milne, M.G. Turner, B. Jackson. 1991a. Heterogeneity
and spatial hierarchies. In Ecological Heterogeneity, edited by J. Kolasa and S. T. A.
Pickett. New York: Springer-Verlag.

O'Neill, R.V., S.J. Turner, V.I. Cullinan, D.P. Cofﬁn, T. Cook, W. Conley, J. Brunt, J .M.
Thomas, M.R. Conley, J. Gosz. 1991b. Multiple landscape scales: an intersite
comparison. Landscape Ecology 5: 137-144.

Palik, B.J., P.C. Goebel, L.K. Kirkman, L. West. 2000. Using landscape hierarchies to
guide restoration of disturbed ecosystems. Ecological Applications 10:189-202.

Palmer, J .B. 1992. Hierarchical and concurrent individual based modeling. In Individual-

Based Models and Approaches in Ecology, edited by D. L. DeAngelis and L. J. Gross.
New York: Chapman & Hall.

Parker, V.T., S.T.A. Pickett. 1998. Historical contingency and multiple scales of
dynamics within plant communities. In Ecological Scale: Theory and Applications,
edited by D. L. Peterson and V. T. Parker. New York: Columbia University Press.

Parkin, T.B. 1993. Spatial variability of microbial processes in soil - a review. Journal of
Environmental Quality 22:409-417.

Peterson, D.L., V.T. Parker. 1998. Dimensions of scale in ecology, resource
management, and society. In Ecological Scale: Theory and Applications, edited by D. L.
Peterson and V. T. Parker. New York: Columbia University Press.

Quinones, RA. 1994. A comment on the use of allometry in the study of pelagic
ecosystem processes. Scientia Marina 58:11-16.

Ricklefs, RE. 1987. Community diversity: relative roles of local and regional processes.
Science 235:167-171.

Robertson, G.P., J.R. Crum, B.G. Ellis. 1993. The spatial variability of soil resources
following long-term disturbance. Oecologia 96:451-456.

177

 

 

 

 

 

 

vmr--'

 

 

 

« v u "u H.H.." it?{21?1‘(1'11“.$0.99‘rlcrﬂbi'S-P"(5'§".i’ '

Robertson, G.P., D.W. Freckman. 1995. The spatial distribution of nematode trophic
groups across a cultivated ecosystem. Ecology 76:1425-1432.

Robertson, G.P., K.L. Gross. 1994. Assessing the heterogeneity of belowground
resources: quantifying pattern and scale. In Exploitation of Environmental Heterogeneity
by Plants: Ecophysiological Processes Above- and Belowground, edited by M. M.
Caldwell and R. W. Pearcy. San Diego: Academic Press.

Robertson, G.P., K.M. Klingensmith, M.J. Klug, E.A. Paul, J .R. Crum, B.G. Ellis. 1997.
Soil resources, microbial activity, and primary production across an agricultural
ecosystem. Ecological Applications 7: 158-170.

Rossi, J., P. Lavelle, A. Albrecht. 1997. Relationships between spatial pattern of the
endogeic earthworm Polypheretima elongata and soil heterogeneity. Soil Biology and
Biochemistry 29:485-488.

 

Rossi, R.E., D.J. Mulla, A.G. Joumel, B.H. Franz. 1992. Geostatistical tools for modeling
and interpreting ecological spatial dependence. Ecological Monographs 62:277-314.

Roughgarden, J. 1989. The structure and assembly of communities. In Perspectives in

Ecological Theory, edited by J. Roughgarden, R. M. May and S. A. Levin. Princeton, NJ:
Princeton University Press.

Schaefer, J.A., F. Messier. 1994. A paired-quadrat method for use in multiscale
ordination. Vegetatio 113:9-1 l.

Schlesinger, W.H., J.A. Raikes, A.E. Hartley, A.E. Cross. 1996. On the spatial pattern of
soil nutrients in desert ecosystems. Ecology 77:364-374.

Schluter, D. 1984. A variance test for detecting species associations, with some example
applications. Ecology 65:998-1005.

Schmitt, RI. 1987. Indirect interactions between prey: apparent competition, predator
aggregation, and habitat segregation. Ecology 68:1887-1897.

Schneider, DC. 1994. Quantitative Ecology. San Diego, CA: Academic Press.

Shmida, A., R.A. Whittaker. 1981. Pattern and biological microsite effects in two shrub
communities, Southern California. Ecology 62:234-251.

Smith, J.L., E.A. Paul. 1990. The signiﬁcance of soil microbial biomass estimations. In
Soil Biochemistry, edited by J. Bollag and G. Stotzky. New York: Marcel Dekker.

Souza, V., L. Eguiarte, G. Avila, R. Capello, C. Gallardo, J. Montoya, D. Pinero. 1994.
Genetic structure of Rhizobium etli biovar phaseoli associated with wild and cultivated

178

 

 

 

 

 

 

 

 

 

bean plants (Phaseolus vulgaris and Phaseolus coccineus ) in Morelos, Mexico. Applied
and Environmental Microbiology 60: 1260-1268.

Stark, J .M. 1994. Causes of soil nutrient heterogeneity at different scales. In Exploitation
of Environmental Heterogeneity by Plants: Ecophysiological Processes Above- and
Belowground, edited by M. M. Caldwell and R. W. Pearcy. San Diego: Academic Press.

Stoyan, H., H. De-Polli, S. Bohm, G.P. Robertson, E.A. Paul. 2000. Spatial heterogeneity
of soil respiration and related properties at the plant scale. Plant and soil 222:203-214.

Strain, S.R., T.S. Whittam, P.J. Bottomley. 1995. Analysis of genetic structure in soil
populations of Rhizobium leguminosarum recovered from the USA and the UK.
Molecular Ecology 4: 105-1 14.

Tiedje, J .M., S. Asuming-Brempong, K. Niisslein, T.L. Marsh, S.J. Flynn. 1999. Opening
the black box of soil microbial diversity. Applied Soil Ecology 13:109-122.

Torsvik, V., J. Goksoyr, F.L. Daae. 1990. High diversity of DNA of soil bacteria. Applied
and Environmental Microbiology 56:782-787.

Wackemagel, H. 1998. Multivariate geostatistics: an introduction with applications .
Berlin: Springer.

Waltho, N., J. Kolasa. 1994. Organization of instabilities in multispecies systems, a test
of hierarchy theory. Proceedings of the National Academy of Sciences, USA 91:1682—
1685.

Whittaker, RH. 1960. Vegetation of the Siskiyou Mountains, Oregon and California.
Ecological Monographs 30:279-338.

Wiens, J.A. 1989. Spatial scaling in ecology. Functional Ecology 3:385-397.

Wilson, J.B. 1999. Assembly rules in plant communities. In Ecological Assembly Rules:
Perspectives, Advances, Retreats, edited by E. Weiher and P. Keddy. Cambridge:
Cambridge Univertsity Press.

Wintzingerode, F.v., U.B. G6bel, E. Stackebrandt. 1997. Determination of microbial
diversity in environmental samples: pitfalls of PCR-based rRNA analysis. FEMS
Microbiology Reviews 21:213-229.

Wu, J ., O.L. Loucks. 1995. From balance of nature to hierarchical patch dynamics: a
paradigm shift in ecology. The Quarterly Review of Biology 70:439-466.

Yoshioka, P.M., B.B. Yoshioka. 1989. A multispecies, multiscale analysis of spatial

pattern and its application to a shallow-water gorgonian community. Marine Ecology
Progress Series 54:257-264.

179

 

3.3.

 

 

 

 

 

 

6 S MN: 358%.: as: N
E 3 MN: 3.8830 Em m
2 :2 NS 3:285 Ex N
No - NS a2: Ham _
es .NM Eu .mm coEomoE 3:32 :CO

 

 

..o:oE 3:5:an 2: E 232:3 33% 2: :0 0:5: 30:22: 2: 2 am 40:08 Soc: 2: Lo: >E E=E_§E\Aoo->: E25525
:cm .m_o:oE 3:55:38 2: :8 AoU+_Uv\_U .3 :23528 :2020: 222:: :5QO 2: 3 BEN—98 2.53:; :39 Co :oEomoa
2: 2 :oEomoE .moEoE L‘EMF 29:3 w 9 :0 >5 2 85:3:32 32:—co: £3» E 22258: 5:25: 8:920 ”N czar—r

 

 

 

 

 

 

L: a N _ ._ ms 8% 0.2 s t .N so: So. we N w _ ._ No.0 use :2:
. . . o . . . . . . . 8:32:
E o 2 2 N 8.- N80 2 N ovoo 8o- Soc mi ago 288282
. . . o . . . . . 8§a>
8 o 2 _ a a. 8d- a: a t N m8 o 8 o- we as M: _ :5 o 52:3:
m :6 a seem a :3 mm: 2% a :5 mm: 223%
:oﬂEmEoU 625 3:02 3058222 6:02 52:4 bzﬁatm>

 

 

 

232i 2: 5:» 3360me E032: Co 82?: 2: 2 :: £223?

:0 2:3 15:82 2: 2 mmm ._o:oE 52:: 2: :2: 5:2: bEmocEwi 8a: 2: E Lo: 80: 38:. 325222: 2: :2: 2852;:
:3: 2: :0 72¢ Etna a; 2 20:2: 2: :o :OowmEoo 625 6:2: :23 E :23: 2: :0 32$ 2: :5, :2383: 2a 20:2:
03: 2: 5:5. :22. 85:3: 2; .EmBB :8 wo_ :moE :o b:5mtm> 2:05 LEM? Co mibmcm 558qu 893-590 Q Baum.

180

 

.-uwr v' 1.‘
_. -

 

 

 

 

 

 

 

 

 

28: N33 98: 2.55 N 85:5
8: 3o 2: 2 .o E 383 Ea :25:
£85 285 38: £85 a 85:5 Es:
Ed 2: 25 N; E 25:0:
. . . . 8:393
2 o 8 o 8 o 8 o 2529 s 8:2
xx S x . x a . x x .x n» .x . wax—om. 2-x 820m 8535
N N N Kr N N .
28.0 t w _ .o : :3 88¢ 2 3082
E: 98 2:8 .32
N: a: _N.o Nd any“; as s =<
. . . . 85293
2 o o. o N_ o N_ o 5:252 s 85
an S . NA. . NA . NA. . .. . .Auxunuo—om
N x Nx Nx Nx x N x Nx m Nx x match 3522;
885 2 8.: 285 as: m 25:»:
. . . . :22298 252 2-x :<
N o NN o N o NN a $532 s
N 26 m 25 N 25 _ 2.6 223%
E22 :32 32:33.8“

 

 

 

225 252 2-x 82cm

.2 .ovmv 332:; he 5:023. 2228 E Ewe—mam mm :2028

.:_._w .22 3an as»: 29:8 w 2 PE =: no 2m2 .2522 Eco—w :5 29:3: 35:—5:02 ”m 2:5.

181

 

f...
e.,
O
.9.
'2
IL
(,
’
‘
t.
I...
l,
u‘ .
.0
...
..l
\
-
t
5..
.p
D.
1‘.
\.
9..
.- .
O. ,
..,
\
I.
U
‘-

 

mem 60

 

 

CON omw 00F 0.1 ONF cow ow om 0v ON 0
.3 p .— F p n “p! p .p h Op..- 0
O.
.- ON
o 00 o o O o o 0: 0.?
I 00 O
p . w
o o o o o o o 0 at N
o. . o 0.r 8 0
U.
. 02 u.
o o o 00 o o 001 ONF
o I OVF
5 On p n n L. om?

 

 

28m 9 boas—Hickman

Ea 82856 was 3.95% .8363 w OS can 2 05 2 Enema—um 3886 :83 803 550% 8: 8.8m 3:953 350 4.83
v.8 $383 w 2 850 .833 Ed £8332 wEEEwm 58w 29:38 “a moaﬁam w OS new 0. ._ 2% .8 m2: mam—95m ”u «azur—

182

 

 

Figure 2: MAD-t scores derived from DNA extraction experiment. Bars with different
letters within a weight category are signiﬁcantly different at the p=0.1 signiﬁcance level,
=7 for 100 g samples and 8 for 10 g samples.

MAD-t

 

     

 

 

 

logsamples
A
0.25 . 100 9 samples
A
AB
B
0.20- I I
, I .
/ /
// //
0.00 M‘
1% 20% 80% 5% 95%

Approximate proportion of whole sample analyzed

183

 

" E'- EI'L". 5!}! 7| Ixil‘oilttlhtﬁtf'h'

   

Figure 3: Grain-based analysis using three variability statistics. Colors indicate the three
diﬂ‘erent grid replicates (red=grid 1, blue=grid 2, green=gn'd 3). Circles are sets of RsaI
digests, triangles are sets of MspI digests. Lines represent the hierarchical models which
ﬁt the data signiﬁcantly better than linear regression models, with undeﬁned regions
shown as dashed lines. 7711's ﬁgure is in color.

Mean Jaccard Distance Log Hellinger Variance

MAD-t score residuals

 

 

 

 

 

 

 

 

 

 

 

 

1.4
O
O
1.3 l
A . 0
O
1.2 ‘ .‘ i. .
1 1 d . ‘ :
. .
O
1.0 < o
O
0.9 . T . . .
-3 -2 -1 0 1 2
Log Mean Soil Weight
0.65
O
0.60 -
0.55 « ‘
1 t
0.50 1
O A
0.45 1 A A
*° :
0.40 - o
. O
0.35 4 o
0.30 . . .
-3 -2 -1 0 1 2
Log Mean Soil Weight
0.20
0.15 . .
0.10 «
A I
0.05 4 . A
0.00. ' ° 0
N
0.05 ‘ C ' ‘
O
-0.10 . a e .
-3 -2 -1 0 1 2

Log Mean Soil Weight

184

 

 

 

 

 

 

 

Figure 4: Hellinger variance as a function of separation distance. Hollow triangles
indicate values signiﬁcantly different from overall variance using the progressive
Bonforroni correction for spatial analysis (initial signiﬁcance value p=0.05). Hollow
circle indicates a value signiﬁcantly different from overall variance at the 0.05 level.
Horizontal dashed lines indicate overall variance. Vertical dashed lines indicate the
practical range of spatial structure found by weighted least squares modeling.

A. Grid 1, RsaI

13

 

Hellinger Semivariance

 

 

 

 

T

0 50 100 150 200
Separation Distance, cm

B. Grid 2, RsaI
15

 

Hellinger Semivariance

 

 

11

 

 

0 50 100 150 200
Separation Distance, cm

185

 

 

 

. flung“?! .

 

 

Figure 4, continued:

C. Grid 3, RsaI
11

Hellinger Semivariance

D. Grid 2, MspI
13

Hellinger Semivariance

 

 

 

 

 

 

0 50

7 I

100 150

Separation Distance, cm

 

 

 

........................................................................................................................................

 

 

0 50

T

150
Separation Distance, cm

100

186

 
  

. r... t .. 7.1”:- ~19 12!}! ‘23“!1m9t110i9$95-gE-+(t?f39'

 

 

200

 

200

 

 

 

 

Figure 5: Integration of extent-based and grain-based analyses of Hellinger variance.
Lines show estimates of log HV at larger grains based on application of the extent-based
models to the multiple-grain sampling locations. Colors indicate the three different grid
replicates (red=grid l, blue=gﬁd 2, green=grid 3). Solid lines and circles are sets of RsaI
digests, dashed line and triangles are sets of MspI digests. This ﬁgure is in color.

 

1.4
O

8 O

1.3 ~
g . . °
.g . .

q o

> 1.2 ‘ ‘
5 o
O) 11 . o O
'5 ' —- —1—
ﬁ \
I 1.04 o \ \
O)
3 ~\

0.9 1

 

 

 

-3-2-1012345

Log Mean Soil Weight

187

 

 

 

 

 

 

 

Figure 6: Principal coordinates analysis of correlation super-matrix of covariance

matrices constructed from different distance classes in grid 2, RsaI digest. Numbers in
the datapoint name represent the distance class from which each covariance matrix is

derived (i.e. d1 is derived from distance class 1, 0-20 cm).

Principal Coordinate 2 (27%)

95.63
com

 

.0
01

0.4 .
0.3 1
0.2 -
0.1 1

0.0 ~

c':
2..

l

 

d1

d3

d4
d5

d2

d6

d9

d7

d8

d10

 

 

-0.4

I

-O.3

-O.2 -0.1

I

0.0

I

0.1

I

0.2

I

0.3

Principal Coordinate 1 (32%)

188

0.4

 

‘ "”Hhiruuiﬂ '

 

 

 

 

 

O

. I?
Q.
'1
‘.»
.10
'
i i
Du.
.I.
t .
I.
r.

of

vs

a.

I

t

ctﬁrz‘iii"

Loans—U Lo 5.00:3 ‘0 0'0:

.k‘sluttdotlzretl$te

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o :5 no: on: So m8 :5 an: «8 NS .8
r _ V _ F _ o o
- _ _ _
M - ..o
. .0
. .
r l L
o
a
a
m
.~.o .
a
.
o
.
u
C L
.n O o
m
_
‘
o
w
..T.o
um.o

 

.AEo omé ._ mam—o 8.336 Ea: potent
ﬂ :mO 6.5 Burrow m_ anE 005950 :80 :03? Eoc mam—o 8536 05 882%: 0:8: “538% 2: E 83:52 .umowmn 3mm .N
2% E 88.20 RENE—u 220:6 Soc 3836.50 mooEaE oocetgoo wcmntomou waELodsm cozﬂoccoo mix—«SW Bums—U ”h 0.5me

 

189

 

 

 

 

Figure 8: Normalized Mantel correlogram of Hellinger and Jaccard distances for all 10 g
samples. Colors indicate the three different grid replicates (red=grid 1, blue=grid 2,
green=grid 3). Circles are sets of RsaI digests, triangles are sets of MspI digests. Circled
datapoints are signiﬁcantly different from zero using the progressive Bonferroni
correction and boxed datapoints are signiﬁcant using uncorrected tests (p<0.05). This

ﬁgure is in color.

r,HD

rM, JD

0.10

 

 

0.08 1
0.06 1
0.04 1
0.02 1
0.00 1
-0.02 1
.004 1
-0.06 1
-0.08 1

 

O
A

$30

0 OD.

CDC.
0} O
0.
0.>

E A
©

 

 

-0.10

0 20 40 60 80 100 120 140 160 180 200 220

Separation Distance, cm

 

0.12
0.10 1
0.08 1
0.06 1
0.04 1
0.02 1
0.00 1
-0.02 1
-0.04 1
-0.06 1
-0.08 1

 

O
O A:
@ E]
(9 ©

.0.

‘A
a

CD.

’0

El

 

 

-O.10

0

20 40 60 80 100 120 140 160 180 200 220

Separation Distance, cm

190

 

 

 

 

 

' .. ’ 36116101510

 

Appendix A: Total number of T-RFs detected in sets of samples at multiple-grain
sampling locations. Note that this is not the average number of T-RFs per sample.
Colors indicate the three different gn'd replicates (red=grid 1, blue=grid 2, green=grid 3).
Circles are sets of RsaI digests, triangles are sets of MspI digests. This ﬁgure is in color.

 

 

 

 

 

8 110

m o .

D 1 0

i” 0

LI. 90 .

m1 0 ‘ ‘

'_ o

q- 80 '1 A
0 Q

g 70 1 . . .
:E: ' o

.79

o 50 4 .

'— -3 -2 -1 0 1 2 3

Log Mean Soil Weight

191