E

:

3...

5...,

. Uumhwuhi
,SRN p51... .

«Hz-I:
u r

E... xv

..I Ara
1‘

:54 L .

. «91.... a. Rafi

t} .
is III:
a». .«rfllruhm
5.1.7.1" . .I? I
.! . 13v .lehuanxk 1%”,
I :9. .5
4.1 III .mhr’f a
.3505 .. .1. 4.7.5.
I fur. I
my 3,
5 2, -
...b.x....¢..., .
‘.;;?vtmn.
25.1.1... .01.}

u . .5.” A. k. , . I “4,4”,1‘iy Ill. Ivunfi
.I. a“ man. IVEfiXEflw . . . ,fi...w~1w.nu .8 an»? .afr figfigflfi . I mu... 3.45ka
W: :3.Ivs......1...r(\ ." I A . .., ,. I I V . ., . . I I . n.

3 1:03.43id53fl

 

WW3 . LIPHAHY
M'Chigan State
(. 4 ”I I University

TV

This is to certify that the
dissertation entitled

TRANSLATIONAL POWER DIFFERS BETWEEN BACTERIA
PURSUING DIFFERENT ECOLOGICAL STRATEGIES

presented by

Les Dethlefsen

has been accepted towards fulfillment
of the requirements for the

Doctorate Microbiology and Molecular Genetics
of in .
Philosophy Ecology, Evolutionary Biology,

and Behavior

Wéssor’s SignatUFe

v/aq/oy
Date

 

MSU is an Affirmative Action/Equal Opportunity Institution

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 www.mm

TRANSLATIONAL POWER DIFFERS BETWEEN BACTERIA
PURSUING DIFFERENT ECOLOGICAL STRATEGIES
by

Les Dethlefsen

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Microbiology and Molecular Genetics
Program in Ecology, Evolutionary Biology and Behavior

2004

 

 

ABSTRACT
TRANSLATIONAL POWER DIFFERS BETWEEN BACTERIA
PURSUING DIFFERENT ECOLOGICAL STRATEGIES
by

Les Dethlefsen

Translation, the polymerization of amino acids into protein, consumes more energy than
any other process in the bacterial cell. The translational apparatus, including ribosomes,
translation factors, tRNA molecules, tRNA synthetases, and other less abundant
components, accounts for a substantial fraction of bacterial cell mass. Hence, selection
for optimal performance of the translational apparatus is likely to be strong. Nonetheless,
premature translational termination events, known as processivity errors, are not rare in
laboratory-adapted Escherichia coli, the only organism for which data exist. Genetic,
biochemical and physiological evidence, as well as models of translation, suggest the
existence of an evolutionary tradeoff between translational power (the net rate of protein
synthesis per mass invested in the translational apparatus) and translational yield (the net
mass of protein synthesized per energy consumed). Microorganisms selected for fast
maximal growth rates and a rapid response to resource abundance may favor high
translational power to permit rapid protein synthesis and rapid growth, at the expense of
more frequent processivity errors that waste energy by generating truncated polypeptides
that are subsequently degraded. On the other hand, microorganisms adapted to exploit
small fluctuations and limited availability of resources may favor high translational yield

to minimize resource thresholds for grth and survival, at the expense of a reduced rate

of protein synthesis for a given investment in the translational apparatus. This hypothesis
is tested using a collection of recent soil isolates containing bacteria of contrasting
ecological strategies in each of several diverse phylogenetic groups, as well as one well-
characterized representative of each ecological strategy. The specific growth rate, cell
density and cell volume were measured for each of these 10 bacterial strains in batch
culture in two media; a novel protocol was developed for measurement of the DNA,
RNA and protein content of small samples of bacterial biomass. As predicted,
translational power is higher in bacteria capable of a rapid growth response to abundant
resources. Codon usage analysis can provide a comparison of the relative strength of
selection for translational power between strains, so the relationship between codon bias
and ecological strategy was examined in a large number of bacteria with fully sequenced
genomes, using the number of copies of the ribosomal RNA operon per genome as an
index of ecological strategy. The observed pattern of stronger translational selection in
organisms with more copies of the rRNA operon is consistent with expectations based on
macromolecular measurements. This pattern is better explained by a cost associated with
translational power rather than the absence of a benefit of translational power among
strains adapted to small resource fluctuations. Because codon bias directly affects
translational power, we investigated whether variation in codon bias among organisms
could explain the observed variation in translational power. However, the degree to
which codon bias accelerates translation in E. coli is too small to explain the observed
variation in translational power between strains. Instead, differences in translational
power among microbes must be explained by differences in the performance of the

translational apparatus itself.

This work is dedicated to my wife, Michelle, and my sons, Conrad and Arik.

iv

ACKNOWLEDGMENTS

I am grateful for many sources of financial support that have allowed me to complete the
work reported in this dissertation: a Michigan State University Distinguished Graduate
Fellowship, the Center for Microbial Ecology at Michigan State University, a STAR
Fellowship from the US. Environmental Protection Agency, a Center for Biological
Modeling/Quantitative Biology Interdisciplinary Research Award, and the US. National
Science Foundation in various guises. I am thankful as well for additional support
provided by the Department of Microbiology and Molecular Genetics and the Graduate

School at Michigan State University.

I give my sincere thanks to the many individuals who have enriched my education during
graduate school, only a few of whom can be named here. Committee members, all of
whom have made intellectual contributions to this work: Dr. John Breznak, Dr. Frank
Dazzo, Dr. Michael Klug, Dr. Richard Lenski and Dr. Alan Tessier. Professors who
deserve the status of honorary committee members: Dr. Julius Jackson and Dr. James
Tiedje. All my colleagues, past and present, in the Schmidt and Breznak labs,
particularly Dr. Bradley Stevenson and Dr. Joel Klappenbach. Many students and faculty
colleagues in the Department of Microbiology and Molecular Genetics, the Center for

Microbial Ecology and the Ecology, Evolutionary Biology and Behavior Program.

Finally, I express my deepest appreciation and sincere respect for my mentor,

Dr. Thomas M. Schmidt.

TABLE OF CONTENTS

List of Tables ....................................................................................................... viii
List of Figures ........................................................................................................ ix
Key to Symbols and Abbreviations ........................................................................ x
Chapter 1. Introduction: Unexplained variability in translational ........................ 1

performance between microbes
The concept and the semantics of translational power ................. 4

Translational power measures both the translation rate ................ 6
and the active fraction of ribosomes

Comparisons of translational power ............................................. 9
Selection on translational power ................................................. l3
Tradeoffs affecting translational power ...................................... 15
Ecological strategies of microbes ............................................... 17
Conditions favoring translational power or ................................ 22

translational yield
Chapter 1 References ........................................................................ 25

Chapter 2. Macromolecular composition of bacterial soil isolates .................... 32
representing contrasting ecological strategies

Chapter 2 Abstract ............................................................................ 32
Chapter 2 Introduction ...................................................................... 33
Chapter 2 Materials and Methods ..................................................... 35
Chapter 2 Results and Discussion ..................................................... 60
Chapter 2 References ........................................................................ 77

vi

Chapter 3.

Chapter 4.

Appendix.

TABLE OF CONTENTS, continued

Translational power and the strength of translational ....................... 81
selection vary with ecological strategy in bacteria

Chapter 3 Abstract ............................................................................ 81
Chapter 3 Introduction ...................................................................... 82
Chapter 3 Materials and Methods ..................................................... 86
Chapter 3 Results .............................................................................. 92
Chapter 3 Discussion ...................................................................... 102
Chapter 3 References ...................................................................... 112
Differences in codon bias cannot explain differences .................... 117

in translational power between bacteria pursuing
different ecological strategies

Chapter 4 Abstract .......................................................................... 1 17
Chapter 4 Introduction .................................................................... 118
Chapter 4 Materials and Methods ................................................... 123
Chapter 4 Results ............................................................................ 139
Chapter 4 Discussion ...................................................................... 146
Chapter 4 References ...................................................................... 153
Phylogeny of bacteria used with log likelihood .............................. 159

ratio tests of correlated trait evolution

Appendix References ...................................................................... 164

vii

 

Table 1.1:

Table 2.1:

Table 2.2:

Table 2.3:

Table 4.1:

Table A. 1:

LIST OF TABLES

Comparisons of translational power .................................................... 10
Cell and culture traits of experimental bacteria ................................... 58
Macromolecular content of experimental bacteria .............................. 59
Macromolecular content of E. coli ...................................................... 62
Codon data ......................................................................................... 127
Bacterial taxa analyzed for translational selection ........................... 161

viii

Figure 1.1:
Figure 2.1:
Figure 2.2:
Figure 2.3:
Figure 2.4:
Figure 2.5:
Figure 2.6:
Figure 2.7:

Figure 3.1:

Figure 3.2:

Figure 3.3:

Figure 3.4.

Figure 4.1:

Figure 4.2:

Figure A.1:

LIST OF FIGURES

Contrasting ecological strategies of microbes .................................... 21
Phylogeny and rrn copy number of experimental bacteria ................ 57
Biovolume vs. specific growth rate .................................................... 65
MM/ODm vs. biovolume .................................................................. 68
Photomicrographs of experimental strains ......................................... 69
RNAzDNA ratio vs specific growth rate ............................................ 74
ProteinzDNA ratio vs. specific growth rate ........................................ 75
ProteianNA ratio vs. specific growth rate ........................................ 76
Translational power of rapidly responding and ................................. 95

slowly responding bacteria
Translational output as a function of RNA content ............................ 97

Correlation between rrn copy number and the ................................... 99
strength of translational selection

Relationship between rrn copy number and the codon bias ............. 100
of entire genomes and of highly expressed genes

Translation rate benefit of codon bias in E. coli ............................... 140
Translation rate benefit of codon bias by amino acid ...................... 144
Phylogenetic trees of bacteria .......................................................... 162

ix

KEY TO SYMBOLS AND ABBREVIATIONS

16S rRNA ....... the small subunit ribosomal RNA gene of Bacteria and Archaea

x2 .................... the chi-square distribution of statistics

u ..................... the Specific growth rate of a culture in balanced exponential grth

A562 ................ light absorbance at 562 nm, measure of color development in BCA assay

A-Site .............. the ribosomal site where an amino acyl-tRNA molecule enters the
translational cycle

BCA ............... bicinchoninic acid, key reagent in a colorimetric assay of protein content

BSA ................ bovine serum albumin, a protein standard used for protein assays

CMEIAS ........ Center for Microbial Ecology Image Analysis System, computer software
to facilitate cell measurements from photomicrographs

CMR ............... Comprehensive Microbial Resource, a website maintained by The Institute
for Genomic Research

D ..................... the DNA content (by mass) of a cell or culture

DAPI .............. 4',6-diamidino-2-phenylindole, a DNA-binding fluorescent dye

EDTA ............. ethylenediaminetetraacetic acid, a chelating agent

EF-G ............... translational elongation factor G

EF-Tu ............. translational elongation factor Tu

g ...................... the generation time of a culture in balanced exponential growth

g ...................... the acceleration due to gravity at the Earth’s surface, used as a unit of
centrifiigal acceleration

gp .................... the proteome generation time, an estimate of the time required for
replication of the proteome with different hypothesized degrees of codon
bias

HE .................. highly expressed genes

SDS ................

TCA ................

TE ...................

KEY TO SYMBOLS AND ABBREVIATIONS, continued

the macromolecular content (by mass) of a cell or culture, the sum of the
DNA, RNA and protein content

any one of the 4 possible nucleotides (in a nucleotide sequence)

the effective number of codons, an index of codon bias comparing
observed codon frequencies to uniform synonymous codon use

a modification of Nc, an index of codon bias comparing observed to
expected codon frequencies

the transformation 61—N'c, an index of the strength of translational
selection when observed codon frequencies are based on highly expressed
genes and expected codon frequencies are genome averages

optical density at 420 nm, a measure of culture turbidity used for
determinations of growth rate

photomultiplier tube voltage, a sensitivity adjustment for a fluorescence
scanner

the protein content (by mass) of a cell or culture
the ribosomal site where peptidyl-tRNA is bound during the majority of

the translational cycle, following translocation and prior to the subsequent
peptidyl transfer event

the RNA content (by mass) of a cell or culture
a purine nucleotide, either adenine (A) or guanine (G)
the ribosomal RNA operon

the translation rate benefit of codon bias, estimated as the fractional

decrease in the time required to replicate the proteome given a certain
level of codon bias, in comparison to the time that would be required with
little or no codon bias

sodium dodecyl sulfate, a detergent
trichloroacetic acid, a reagent that precipitates proteins

Tris-EDTA, a buffer used during nucleic acid measurements

xi

KEY TO SYMBOLS AND ABBREVIATIONS, continued

X ..................... an unspecified nucleotide (in a nucleotide sequence)

Y ..................... a pyrimidine nucleotide, either cytosine (C) or, in RNA, uracil (U)

xii

Chapter 1. Introduction: Unexplained variability in
translational performance between microbes

The scientific literature contains a number of explicit comparisons of the concentration of
ribosomes or of RNA between a slowly growing microbe and Escherichia coli 1'4. These
reports note that the concentrations found in their respective strains are not unusual in
absolute terms. The estimates of the number of ribosomes per volume or the mass of
RNA per volume in the slowly growing strains are similar to values determined for E.
coli 5. However, each report emphasizes the unexpectedness of finding similar values in
the slowly growing organism and in E. coli growing at a considerably faster rate. Since
ribosome and RNA concentrations vary with growth rate in E. coli, the slow-growing
organisms have more ribosomes or RNA than expected, if the relationship between

ribosome content and growth rate established for E. coli 5 is extended to other species.

Stated more precisely, these observations violate the expectation that the amount of
protein synthesized per ribosome per time interval is approximately constant across
strains. If the expectation were correct, fewer ribosomes would be needed per mass of
protein in a strain with a longer generation time. This expectation can be decomposed
into two assumptions: that the translation rate is similar between strains, and that the
fraction of active ribosomes is similar between strains. Implicit in these assumptions is
the idea that the quantity of ribosomes is matched to the demand for protein synthesis,
since the presence of superfluous ribosomes at a given level of protein synthesis must
lead to either a decrease in the average rate of translation, or a decrease in the fraction of

ribosomes that are active, or both. All 4 studies that compared the ribosome or RNA

content of a slowly growing microbe to E. coli went on to discuss the validity of these

assumptions in light of their unexpected results; 3 of the reports conclude that slower

translation in the slowly growing bacterial strain is likely 2'4.

Indeed, despite earlier arguments to the contrary 67, it is now generally accepted that the
translation rate varies with growth rate even within E. coli, ranging from 12 to 21 amino
acids polymerized per ribosome per second as the specific growth rate varies from 0.4

tol .7 hr], assuming that the fraction of active ribosomes is constant 5. Two explanations
have been offered for this variation. Koch has suggested that during slow growth, the
ribosome content of E. coli provides a capacity for protein synthesis that is, in fact,
greater than the demand. He argues that maintaining ‘surplus ribosomes’ is beneficial
during slow grth in sub-optimal conditions in a variable environment, because they
permit a more rapid return to faster grth rates when conditions improve 89. Consistent
with this prediction, inactive ribosome dimers have been discovered in E. coli, and the
fraction of ribosomes forming such dimers increases as grth rate declines in E.

coli 10'12. Such dimers could have been mistaken for translationally-active polysomes in
earlier work that appeared to establish a constant fraction of active ribosomes across
growth rates 13. Failure to account for this phenomenon (as in reference 5) would
generate an apparent decline in the translation rate per active ribosome at slower growth

rates, by overestimating the number of active ribosomes.

In contrast, by considering the cost of supplying the ribosomes with substrates as well as

the cost of the ribosomes in a growing cell, Ehrenberg and Kurland have argued that

genuinely slower translation is to be expected at slower growth rates, if organisms have
been selected to maximize their growth rate in a range of growth conditions 14. They find
that the optimal allocation of resources involves an asymptotic approach to saturation of
the ribosome with ternary complexes (i.e. the maximum translation rate) as the organism
approaches its maximum growth rate 14. A number of predictions of the Ehrenberg-
Kurland model are supported by empirical data 15. Although this explanation is quite
distinct from the explanation offered by Koch, the two are not mutually exclusive.
Neither explanation, however, suggests a benefit for inactive ribosomes or slow
translation in organisms that are adapted to stable environments, or that are growing at
their maximum rate. Under these conditions, the selective pressures identified by Koch
and by Ehrenberg and Kurland would favor rapid translation and a high fraction of active
ribosomes. In this context, in organisms growing at or near their own maximal rate but
more slowly than E. coli, the existence of E. coli-like concentrations of ribosomes or

RNA 1‘4 remains unexplained.

To extend these findings, we have searched the scientific literature for data on the protein
and RNA content of microbes during balanced, exponential grth at known rates. To
provide a consistent and rigorous framework for comparison, we will compare the total
rate of protein synthesis of a cell or of a culture, normalized to the biomass invested in
the protein synthesis machinery 15. This type of comparison, made explicitly or
implicitly, is used in each of the studies discussed above that investigate translation in
slowly-growing microbes 1'4. We introduce a new term, ‘translational power’, to refer to

the translational output per mass of the translational apparatus. However, before

presenting the results of our comparison, we must address a semantic issue: Why

introduce a new term for a concept and a quantity that have already been named?

The concept and the semantics of translational power

We use the new term ‘translational power’ to describe the translational output per mass of
the translational apparatus, precisely the same concept (and the same quantitative
parameter, see below) that was originally defined as ‘ribosome efficiency’ 5'7’16. In
recent years, this concept has more commonly been called ‘translational efficiency’ 14,17,
particularly in the context of explaining codon usage bias 15,1849. Although we are
reluctant to depart from established terminology, we do so to avoid an inconsistency with
the meaning of ‘efficiency’ as it is used in many other areas of science and in colloquial
usage. In the physical sciences and in many areas of biology, efficiency refers to a
comparison between the output and the input of a process, particularly the fluxes of
energy and/or mass. For example, the efficiency of a machine is the work performed by
the machine divided by the energy supplied to it 20. Less than perfect efficiency occurs if
not all the energy required to operate the machine is made available for useful work, e.g.,
energy is lost as heat. Similarly, ecologists define ‘trophic transfer efficiency’ as the
fraction of energy contained in the biomass at one trophic level which is transferred to the
biomass of the next trophic level 21. These scientific meanings of ‘efficiency’ are
consistent with the common notion that a process operating with little waste is highly

efficient.

According to these conventions, calculations of efficiency make no direct reference to the
rate at which a process occurs. Instead, physicists and engineers use the term ‘power’ to
refer to the rate of energy consumption or the rate at which work is performed 22. The
semantic distinction between power (or rate) and efficiency is important, because many
real and idealized physical systems can approach maximal efficiency only by occurring at
infinitesimally slow rates; i.e. there is a tradeoff between power and efficiency. Several
attempts to demonstrate that a power-efficiency tradeoff must be an important universal
constraint for biological systems 23:25 have justifiably been criticized for the
inappropriate use of thermodynamic concepts and other flaws 26'28. Despite the failure
of these universal arguments, numerous specific examples of tradeoffs that can be
described in terms of power (ultimately affecting grth rates) and efficiency (of
resource utilization) have been described 29. The organisms involved include

rodents 30'32, frogs 33, snails 34, insects 35, freshwater crustaceans 36'38, trees 39,

plants 40‘42, and phytoplankton 43, based on mechanisms such as foraging behavior,
metabolic rates, resource allocation and resource affinity, as well as other unknown
aspects of cell physiology or biochemistry. A number of comparisons of coexisting
bacterial species have also provided evidence for power-efficiency tradeoffs 44-47, as
have comparisons of engineered mutant strains 48. However, the absence of apparent
tradeoffs in some carefully designed studies of bacteria demonstrates that such tradeoffs
are not inevitable 49'51. Even if power-efficiency tradeoffs occur only in some contexts,
there is value in maintaining a semantic distinction derived from both scientific and
colloquial usage between power (implying a rapid rate) and efficiency (implying low

waste).

However, the term ‘translational efficiency’ blurs this distinction because it refers to a
rate, expressed in units of inverse time. It is a measure of the power of the protein
synthesis subsystem to drive the self-replication of the protein-dominated autocatalytic
system to which it belongs. (It is exactly analogous to the power:mass ratio recognized
as a critical performance characteristic for engines that must move their own mass as one
component of a vehicle.) Furthermore, we will suggest in this dissertation that obtaining
a high translational output per mass invested in the translational apparatus results in a
high frequency of wasteful errors, meaning that a high mass-normalized translational
output trades off with the energetic efficiency of translation. For these reasons, we prefer
the term ‘translational power’ as a description of the translational output per mass of the
translational apparatus that is more consistent with the expected meanings of ‘power’ and

‘efficiency’ based on other areas of science and on colloquial usage.

However, there would be obvious drawbacks in attempting to redefine the established
term ‘translational efficiency’ as our convenient term for referring to the energetic
efficiency of protein synthesis. Hence, in keeping with the meaning of the term ‘yield’ in
microbiology, we define ‘translational yield’ as the mass of protein produced per energy

consumed by the translational apparatus.

Translational power measures both the translation rate

and the active fraction of ribosomes

Translational power measures both the average translation rate and the fraction of active

ribosomes, which we demonstrate as follows, using the approach of chapter 6 of

 

reference 52. The average translation rate of a cell or culture (also called the peptide

chain growth rate 55253) is the rate of amino acid polymerization per active ribosome:

 

. number of amino acids 01 eriz er i im
translation rate: 1) ym edp untt e.

number of active ribosomes

For a culture in balanced, exponential growth, the rate of increase of any culture
component X is dX/dt = ,uX, where ,u is the specific growth rate. Hence, the rate of
protein synthesis (i.e. the translational output) in a culture growing at rate ,u, containing a
mass P of protein, is simply ,uP. Translational output expressed in units of mass can be

converted to a rate expressed in numbers of amino acids by dividing by the average mass

of an amino acid.

number of amino acids polymerized per unit time = #P / (average mass of amino acid)

The number of ribosomes in a culture containing a mass R of RNA can be found by
multiplying R by the fraction of RNA that is ribosomal, and then dividing by the mass of
RNA in a ribosome. However, only a fraction of these ribosomes are active at any given

time. Thus,

R x (ribosomal fraction of RNA)

# of active ribosomes = .
(mass of RNA per ribosome)

 

x (active fraction of ribosomes)

Substituting the two latter equations into the first equation yields:

[JP / (average mass of amino acid)
R x (ribosomal fraction of RNA)

(mass of RNA per ribosome)

translation rate =

 

x (active fraction of ribosomes)

 

After rearranging terms, we have:
. . . . ,uP
(translation rate) x (active fraction of ribosomes): R x C ,

where

_ (mass of RNA per ribosome)/ (average mass of amino acid)

C
(ribosomal fraction of RNA)

 

The quantity ,uP/R represents translational power 5’16, the translational output per
biomass (measured as RNA) invested in the translational apparatus. It is clear that

translational power reflects both the translation rate and the fraction of active ribosomes.

What of the term we have labeled C, implying a constant? The two quantities in the
numerator, the mass of RNA in a ribosome and the average mass of an amino acid, are
indeed constant or nearly constant, both within a strain at different growth rates, and
across strains. However, despite the constant ribosomal fraction of RNA reported in

reference 5, other data indicates that the rRNA fraction decreases from about 85% to

about 75% as grth rate declines in E. coli from 1.7 hr1 to 0.28 hr1 5455, a result
which is expected on theoretical grounds 14,56. This variation is not dramatic; it would
reduce translational power by only 12%, if the translation rate and active fraction of
ribosomes were unchanged. Data are also available fiom 2 of the 4 comparative studies
discussed earlier. The rRNA fraction is reported as 84% for Halobacterium cutirubrum
at specific growth rates of both 0.10 hr'1 and 0.05 hr'l, after the authors made the
deliberately generous assumption that messenger RNA comprises 5% of the total RNA 4.
The rRNA fraction is about 85% for Rickettsia prowazekii at a specific grth rate of
~0.07 hrl, after a correction is made for 2-3% messenger RNA 2. These data do not
suggest that variation in the ribosomal fraction of RNA will obscure the relationship
between translation rate, the active fraction of ribosomes, and translational power, even

for comparisons between strains that grow at very different rates.
Comparisons of translational power

Table 1.1 summarizes comparisons of translational power between E. coli and all other
microbial species for which data could be found in the literature, including the
comparative studies discussed earlier. E. coli is represented by the Bremer and Dennis
data 5, which are typical of the data reported for E. coli in many other studies. Similarly,
comparisons between E. coli and 2 closely related species of enteric bacteria, Salmonella
enterica and Enterobacter aerogenes, are made using only a single representative study
for the latter strains, chosen from among several published reports. For the remaining
species, only a Single published study was available for comparison, except for one

species represented by two studies, both of which were included. For strains not grown

.owma wfiofi no moHoSoom .mm

 

 

 

8 $5». 3o «.5 6w. : Pom v3 ass 258883
E exam 2 I «as 28 <72 3 _ .c 0.2 3.0 @328 BESSIE
E Ramo— moé SAME Oohm cc; 8.28% SRBEBM
mo.mc flaws I .xcom Ewd I mmd <78 Amwdv Ucom Rio I end EEEEEE .3238:me
8 $3 NS <zm am: 9% Ed £88 358%
B :3. 3d <72 Gas 9% 85 ESE... SEES»:
am ._¢\oo: laxmvv Emu. _A I Nwd <72 AENV Oowm Bad I wmd mgwmoewoamé nmoAESthm
mm :9 e3 <72 5.8 can Ed 58 am SUSSEE
N gem Rd mom am: 0.3 Be EREEm Sneeze
v $3 3.6 «~ch Oohm omo EEQEES ESBBBSBE
_ EXLNV EMBAA <72 Aowév Uaom 3.0 33338 BOAESQOEW
hm HER V Emu. ~A <72 €03 Damn ovd 35 ESE mega»: 03—3
M afictv nmhéA mom 8%: Doom mNd Managed? fibQoMfiERM.

9L5 9an

=9 oohoo
ponchomom C833 ohwmuwfimw 13 “MEMBQEOW ~28 £38m Emamwco commas—SO
Reese—want . . 389:8 228% . .
a8 m 550% 33% .
5.5.5 12832295 he 28$anan u—A oEaH

10

8:28:88 08:3 8800.06 E EmENwEo E5 cow 3:850 SE 838w 8:8:88 05 8 8:33:88 850% 329» Mo owgm .

E9388“ 8:268 08% <72: m3 .3 MiamNE. SENEEEBQ 3 333% .

.88 538w :8 .m BEE: DE 8 3358:2088 038 E 883800 63282“ 8m

Ba: 53% How 28 689w 05 S mEBBw :8 .m :2: 8338? E83288: E 80E$>E :owE a wee—«8 8E8~w8 :oEEXEoO .

E8 .m 8 8388088 E88: 3:86 a 888mm 8m8mm8 E8 8.“ 630: 3:08:88: 00 :Omea8oo 25.
$88 526% map—om 3:8 :8 DB no .3688 8888 a :5 TE mo; 00 88 830% a 8 a _.~ .«o 0829 28332.? a wE8:mmm
.m 8:80.38 mo Se: 2: 80¢ cows—:28“: was? :8 MN he.“ 88 530% mo 850:3 a m: 8388088 0882:: .93ng 8388:

Ho: E 00:80.3: REES 05 E 038 55 8585.88 0882:: wEm: :8 Q 8 8m8aw8 E5 .8 855880 05 $056680 8m .

E8 N mo 83:: 3:08:88: 05 mo owfifiéo: a mm @8855 .8m8mm8 82.8388 go .826: 383288.“. .

.v :823 E

8505 «5:6 2: 3 338% £8839“ 182588: E 808:ng Sm 8E5w3 :omEE8oo 3:038 :8 Q :03? an 88 5380 .

.m 8:888 88m :83 Be?
8m: :8 .N .888383 =m com .038 <ZMEE38: 3:86 £78: 638 <ZMHE88: SEER €72 80288880 0882::

3:86 .Om 830:8 mm 3883 :8 .m Em EmanEo 88388 2: 50253 mBEmQQm Egan—EN: on: E H:o:bmo>E 8:85 .

.00 DOGOHOMOH 80.0 want 20 0039 mucuomm fiOflooF—Oo DENHOQEOH .N Gad—GO S CBC—.5 mm

E023 .Ucnm “a @8898 88 530% 88088 2:38:88 2: 0883mm 8 33:2:8 :89 we: 88 838w 388 E023 .3 :98“.— .

.8m8mw8 83888 05 mo 0:880:88 53on BBo< .

ABE: eeaexmzmmvkmvE u 88 853% ommoonm o5 mm 850% ”Emmgwuo 83388

08 How 03232.. 803 83: 885208888 20E? :8 98: cow .9: 8m @8888 2883820 mean: 5.5on $232 BE. .

88:88 “A «fish.

11

at 37°C, we assume that the growth rate, but not the macromolecular content, would be
altered by grth in the same medium at a different temperature 67. The grth rates
reported for these strains were adjusted to the growth rates expected at 37°C using the
linear range of the relationship reported by Farewell and Neidhardt 66. (Although this
relationship was generated with E. coli, the comparison is mathematically identical

whether the temperature correction is applied to E. coli or the comparison strain.)

The comparisons in Table 1.1 are made at the fastest growth rate for which data are
available for each of the organisms compared to E. coli, to reduce the influence of
ribosomes that are either inactive or translating at less than their maximal rates.
However, the E. coli data used in each comparison are taken from a growth rate such that
the investment in the translational apparatus (assessed as ribosome concentration or the
protein:RNA ratio) is similar in the two organisms (see discussion below). A comparison
at similar investment levels reflects the expectation that the selective pressure to
maximize translational output varies with the biomass of the apparatus 14,633. If the
comparisons had always been made to the fastest E. coli growth rate (i.e. where its
translational power is highest), the disparity in translational power would be greater for
most of the comparisons shown. It would appear that the expectation of a constant
translational power is not supported by the published data, particularly for comparisons
between microbes adapted to different ranges of growth rates. While translational power
is higher in E. coli and other fast-growing organisms, it is lower in slow-growing
organisms, ranging from less than 17% to 42% of the value for E. coli. Slowly growing
microbes seem to translate more slowly, or have a larger fraction of inactive ribosomes,

than microbes capable of rapid growth.

12

Selection on translational power

Why should some microbes have low translational power? One answer could be that
translational power doesn’t contribute to fitness for these microbes; indeed, Sharp and his
colleagues have suggested that this is the case for several microbes with unusual patterns
of codon use 69,70 (see also Chapter 3). However, one of the implications of the
Ehrenberg-Kurland model of the growing cell is that all components of the cell can be
considered as growth-limiting, in proportion to their mass 14. Although Ehrenberg and
Kurland obtained this result rather more rigorously via differential calculus, their logic
can be revealed by an evolutionary thought experiment: If, in some specific
environmental context, the metabolic flux through a particular biochemical pathway of a
bacterium is the limiting factor for growth, there is strong selective pressure to increase
the flux through that pathway. Even if the activity of the cell components responsible for
that pathway cannot be improved, the flux through the pathway could be increased by
increasing the abundance of those cell components. Since, by hypothesis, other
metabolic pathways of the cell are not the limiting factors for growth, biomass could be
diverted from those functions without reducing the growth rate. Obviously, such a
process would lead eventually to the growth-rate maximized cell, with biomass allocated
to various cell components so that the flux though all branches of metabolism is

balanced, neither limiting to grth nor in excess of demand.

If the cell components responsible for the limiting pathway were initially a very small
fraction of cell mass, only a slight reallocation of total biomass could increase the flux

through that pathway dramatically. On the other hand, if the limiting pathway already

13

required an investment of the dominant fi'action of cell mass, diverting all the ‘excess’
biomass from all other pathways to the dominant pathway would provide a relatively
small fractional increase in the limiting flux. Similarly, interference with the components
of a minor pathway in the grth rate-maximized cell (e.g., a mutation) can be
accommodated with only a slight growth rate penalty, whereas interference of the same
relative severity in the components of a dominant pathway has a larger effect on the
growth rate. It is in this sense that all pathways in the grth rate-maximized cell

contribute to growth rate limitation in proportion to their mass.

This analysis provides insight into a phrase used above, selection to improve the ‘activity
of the cell components’. In the context of maximizing growth, improved activity must be
interpreted as mass-normalized activity. A cell is poorly served by obtaining a new
enzyme with twice the activity of the old enzyme but triple its mass; the cell would do
better to increase the expression of the old enzyme. Note that the concept of mass-
normalized activity, when applied to the translational apparatus, is precisely translational
power. The Ehrenberg-Kurland model of the cell identifies selection for rapid growth
with selection for high translational power, a particular example of the selection for high

mass-normalized activity, i.e. high power, in all cell components.

Clearly, this model is an idealization unlikely to be matched by actual bacteria, for
several reasons, including spatial and temporal heterogeneity of the environment: the
growth rate-maximized cell is a moving target as environmental conditions change. Cells

will be selected to track the optimal phenotype for their current environment via

14

regulatory changes, but neither the evolutionary nor the physiological adaptation will be
perfect or instantaneous. Even so, the most massive components of the cell will be under
the strongest selective pressure, and the most central components of the cell, those
essential under all environmental conditions, will be under the most continuous selective
pressure. According to both these criteria, the allocation of biomass to the protein
synthesis system is likely to be more constrained by natural selection than allocation
towards other subsystems of the cell. The explanation that translational power may be
low in a particular microbe because protein synthesis makes little contribution to its
fitness is difficult to sustain if a substantial fraction of the cell’s biomass is invested in the
translational apparatus. At least for the slowly growing strains represented in Table 1.1,

measurements of ribosome or RNA content suggest that this is the case.

Tradeoffs affecting translational power

A second potential explanation for low translational power could be the existence of
evolutionary tradeoffs. If high translational power is incompatible with some other trait
that also contributes to fitness, at least under some environmental conditions, the degree
to which organisms display high translational power may depend on the conditions to
which they are adapted. Koch’s explanation of ‘surplus ribosomes’ in E. coli 3 invokes a
tradeoff between high translational power during slow growth and the ability to rapidly
shift up to a faster growth rate. There is evidence for a tradeoff involving not just the
quantity, but the performance of ribosomes as well. Mikkola and Kurland investigated
65 natural isolates of E. coli fiom the ECOR collection 71, and found a wide range of

growth rates during batch culture in either rich or minimal medium 72,73. The fastest

15

grth rates among the natural isolates were comparable to the growth rates of several
laboratory-adapted strains of E. coli in the same media; the slowest grth rate was
nearly 3 times slower. The translation rates in vitro of ribosomes removed from these
strains also varied widely, with the fastest translation rates comparable to the translation
rate of ribosomes extracted from laboratory-adapted strains. The only variable in these
comparisons of translation rate was the source of the ribosomes; all other components of
the in vitro translation system were held constant. There was a linear relationship
between the grth rate of a strain and the translation rate in vitro of the ribosomes taken
from that strain. Seven of the natural isolates that spanned the observed range of growth
rates and translation rates were used to inoculate glucose-limited chemostats.
Remarkably, afier evolving in this environment for only about 280 generations, the

descendents of all 7 strains converged on the values typical of laboratory-adapted strains

for both growth rates in batch culture and translation rates in vitro 73,74.

At the very least, it can be concluded that the selective pressures for rapid grth
experienced in typical laboratory environments are not universally experienced by E. coli
in its native habitat. However, one additional piece of evidence argues for the existence
of an evolutionary tradeoff, with alternative phenotypes apparently favored in different
natural settings, but only one of the phenotypes favored in the lab. The natural isolates
characterized by slower grth and slower translation were better able to survive carbon
starvation than either laboratory strains or fast-growing natural isolates, and selection for
rapid growth and rapid translation led to the loss of this trait. Unfortunately, very little

information is available on the genetic changes that occurred during adaptation of the

16

slowly growing natural E. coli strains to laboratory conditions. Both the rapidity of the
evolutionary change and the trajectory of phenotypic changes over the 280 generations
suggest that only 1-3 mutations are involved 73,74. The linear relationship between the
growth rates of the strains and the translation rates of isolated ribosomes under constant
in vitro conditions, both before and after selection, strongly suggests that the ribosome

itself is involved in the adaptation.

Ecological strategies of microbes

Alternative responses to an evolutionary tradeoff that are favorable under different
circumstances, or that ensure fitness by different mechanisms in the same environment,
can be described as ecological strategies. Low translational power with enhanced
starvation survival, or high translational power with diminished starvation survival, are
ecological strategies available to E. coli. We hypothesize that similar strategic choices
may explain differences in translational power observed over much greater evolutionary
distances, as represented by the comparisons between bacterial species shown in

Table 1.1.

Our treatment of bacterial ecological strategies is informed by the tradition of life history
analysis in ecology, particularly the categorization of traits as r—selected or K-selected
(indicating that the trait enhances intrinsic growth rate or carrying capacity,

respectively) 75‘77. Another source of insight has been the long-standing interest in the
implications of resource availability for microbial physiology, summarized by some as

contrasts between traits found or expected in oligotrophic and copiotrophic

l7

microbes 7879, although such descriptions are not without controversy 80,81. While these
pairs of contrasting terms are related, in that copiotrophic microbes may possess many r-
selected traits and oligotrophic microbes many K-selected traits, these two pairs of
contrasts are certainly not identical 80. However, both of these distinctions capture some
important aspects of the contrasting ecological strategies we believe may be responsible
for differences of translational power between strains. The rapid growth and high
translational power shown by laboratory adapted E. coli are r-selected traits beneficial
during resource abundance. Starvation survival, which we interpret as reflecting better
use of limited endogenous resources, would be a K-selected trait and advantageous in
oligotrophic environments. However, the thrust of this work is not to find correlations
between traits because they are selected under the same ecological conditions, but to
explore the possibility that a tradeoff exists that could explain low translational power in
many diverse strains. Without a compelling reason to assert that either resource
availability or the degree of crowding is the primary determining factor for the ecological
strategies we seek to investigate, we will avoid describing the hypothetical strategies in

these more established, but more general terms.

A more recent and more specific description of alternative ecological strategies involves
differences between microbes in the number of copies of the ribosomal RNA (rrn)
operon per genome. A high rm copy number, because it allows rapid ribosome
synthesis, contributes both to a fast maximal grth rate and to a capacity for rapid
acceleration of the growth rate when growth conditions improve 32. Hence, it is believed

to be an adaptation to niches characterized by episodes of resource abundance. A benefit

18

of low rrn copy number has not been demonstrated unequivocally, but there may be a
benefit related to a reduced burden of unnecessary expression when even basal levels of
rm expression exceed the need for ribosome synthesis during slow growth 33. It has been
demonstrated that a particular E. coli strain with a near-complete deletion of one of the 7
rrn operons experiences a fitness penalty during batch culture that is not apparent during
chemostat grth 34. Hence, low rrn number may be an adaptation to stable niches
characterized by slow grth and low nutrient availability 3:33. In any case, it would be
expected that microbes with lower rrn copy number would be less able to respond rapidly
to a sudden increase in resource availability than microbes with higher rrn copy number.
Klappenbach and his colleagues confirmed this expectation by comparing the rrn copy
number of soil bacterial isolates that first formed visible colonies either within 1-2 days,

or after more than 7 days 35.

We selected strains with known times of initial colony appearance and known numbers of
rrn operons per genome from among the bacteria examined by Klappenbach 85,36 for the
research on translational power reported in this dissertation. To refer succinctly to the
known traits of the organisms, we use the term ‘rapid responders’ for bacteria that formed
colonies early and have high rm copy number, and the term ‘slow responders’ for
bacteria with the opposite traits. These terms apply to the response of the strains to
resource abundance in terms of population grth only; they do not imply different
abilities to respond to the environment in terms of sensory signal transduction or resource
transport. ‘Rapid response’ is a useful description not only of what we know about the

characteristics of a subset of these strains, but also of the strategy we believe these strains

19

represent. It implies both a capacity for rapid acceleration of growth rate and a high
maximal growth rate. Clearly, a rapid growth response can confer a fitness advantage in

some conditions.

‘Slow response’, however, while equally descriptive of another subset of strains, is less
apt as a description of an ecological strategy. We do not imagine that a dilatory growth
response is itself contributing to the fitness of these organisms; rather, it is the
unavoidable consequence of some other, unknown advantageous traits. On the strength
of the E. coli experiments involving starvation survival 74 and the proposed benefit of
low rrn copy number 833536, we surmise that at least one such trait involves maximizing
yield, the conversion of resources into biomass. Additional support for the existence
such a tradeoff will be found in Chapter 3, where we propose mechanisms linking high
translational power (a component of the rapid response strategy) to frequent errors in
translation that would reduce translational yield. However, we have at present no direct
evidence that our slowly responding strains are, in fact, better able to convert resources
into biomass or protein than the rapidly responding strains. Hence, we will continue to
use the term ‘slow responders’ to refer to a particular subset of the organisms in our
collection. The contrasting ecological strategies discussed in this chapter are depicted in

Figure 1.1.

20

Figure 1.1: Contrasting ecological strategies of microbes

early colony formation <:> late colony formation
high rrn copy # <:> low rrn copy #
rapid response <3::> high yield
high translational power <21) high translational yield
rapid translation <:> enhanced starvation survival
copiotrophy <:i> oligotrophy
r-selected traits <::> K-selected traits

Figure 1.1. Examples of contrasting ecological strategies available to microbes. The
pair of characteristics on each line have been suggested to be related by one or more
evolutionary tradeoffs. The characteristics listed in each column are not necessarily
synonymous or equivalent. However, an organism may ofien be described by several of
the terms in a single column, either because of a mechanistic relationship between the
traits or because the traits are favored under similar ecological circumstances. Of course,
none of the contrasts implies merely a dichotomy, but rather a spectrum of possibilities.
As an operational definition, the organisms we refer to as rapid responders or slow
responders in this dissertation were classified as such on the basis of colony appearance
time and rrn gene copy number.

21

Conditions favoring translational power or translational yield

Assuming the existence of a tradeoff between translational power and translational yield,
we will describe conditions in which we expect each strategy to be favored. Dramatic
fluctuations of resource availability occur when bacteria respond to episodes of resource
abundance with exponential grth that continues until the resources are consumed;
these conditions resemble serial batch cultivation of bacteria in the laboratory. We
hypothesize that competition for abundant but ephemeral resources favors high
translational power to permit fast growth. In energy-replete conditions, low translational
yield would be deleterious only insofar as it limits the growth rate; the waste of resources
per se would be irrelevant. The periods of starvation between episodes of resource
abundance may favor high translational yield over high translational power; however, we
anticipate a threshold effect. Unless starvation lasts long enough so that a subset of
organisms experience mortality, or at least some impairment of their subsequent ability to
grow, differences in translational yield may not influence relative fitness. We imagine
that translational yield is only selected to be high enough to ensure survival through the

longest starvation intervals that are typical for a habitat.

If the resource flux in a habitat is reasonably well matched over some period of time with
the rate of bacterial mortality (e. g., from bactivory or viral lysis), conditions may more
closely resemble continuous culture in the laboratory, with continuous growth at a more
or less constant levels of resource availability and bacterial population density. Perhaps
counter-intuitively, selection in such conditions favors the fastest possible growth at the

ambient resource concentration, and hence high translational power over high

22

translational yield 87. This remains true even if the resource concentration is low enough
to constrain the growth rate of a strain far below its maximal value. As long as growth
continues and the resources to support growth are drawn from a common pool, organisms
that grow more quickly are superior competitors regardless of yield. Improvements in
translational yield may improve fitness in these conditions, but only if they have the
effect of increasing growth rates at the ambient resource concentrations. There is no
fitness benefit in conserving resources at the cost of even a slight reduction in growth

rate, if the resources spared from consumption are shared among all competitors.

We predict that conditions favoring high translational yield over high translational power
can occur either when not all organisms are growing, or when resources are not shared.

If the minimal resource levels that can support grth or viability are influenced by the
energy demand for maintenance protein synthesis, higher translational yield would lower
these resource thresholds. If resource levels in a habitat are such that some organisms
can grow while other cannot, or some organisms can remain viable while others cannot,
then selection will favor high translational yield. Selection for improved translational
power would not be absent, but it would occur only among the organisms that are'capable
of growth. Even among growing strains, we expect high translational yield to be favored
if resources are not shared. Imagine a microbial population in a habitat where resource
levels only occasionally rise above the minimal level at which resource uptake occurs,
and at most attain concentrations only slightly above the minimal level. The threshold of
‘only slightly above the minimal level’ is such that all available resources are taken up by

the existing population before cells can grow sufficiently to increase their capacity for

23

uptake. The assimilated resources are endogenous, i.e. not shared. If grth is sustained
primarily by drawing on endogenous resources that are periodically replenished, instead
of by continuous uptake of resources from a common pool, selection will favor increasing
the translational yield. The crucial distinction between these conditions and conditions
favoring translational power is not low resource availability per se, but the fact that an
organism and its progeny have no opportunity to obtain additional resources by growing

more quickly. As long as this condition holds, translational yield will be favored.

The remainder of this dissertation examines the hypothesis that bacteria capable of a
rapid grth response to abundant resources have high translational power, whereas
bacteria with a slow grth response have low translational power. Chapter 2 describes
our choice of bacterial strains to test this hypothesis, the methods developed to measure
bacterial macromolecular content, and the macromolecular data itself, as well as related
data on cell size and cell density. Chapter 3 compares translational power between
rapidly and slowly responding bacteria, using the data of Chapter 2, and applies codon
analysis to test the hypothesis that the strength of selection for translational power varies
along the same ecological axis. Chapter 3 also discusses evidence that low translational
yield is an unavoidable consequence of high translational power, providing a mechanistic
basis for selection against high translational power under some environmental conditions.
Because biased codon use can increase translational power, and because the degree of
codon bias tends to be greater among rapidly responding bacteria, Chapter 4 investigates
the possibility that differences in the degree of codon bias are a sufficient explanation for

the observed differences in translational power among microbes.

24

10.

11.

Chapter 1 References

Shahab, N., F. Flett, S.G. Oliver and RR. Butler, 1996. Growth rate control of
protein and nucleic acid content in Streptomyces coelicolor A3(2) and
Escherichia coli B/r. Microbiology-UK, 142: 1927-1935.

Pang, H.L. and H.H. Winkler, 1994. The concentrations of stable RNA and
ribosomes in Rickettsia prowazekii. Molecular Microbiology, 12(1):115-120.

F egatella, F., J. Lim, S. Kjelleberg and R. Cavicchioli, 1998. Implications of
rRNA operon copy number and ribosome content in the marine oligotrophic

ultramicrobacterium Sphingomonas sp. strain R8225 6. Applied and
Environmental Microbiology, 64(1 1):4433-4438.

Chant, J ., I. Hui, D. Dejongwong, L. Shimmin and PP. Dennis, 1986. The protein
synthesizing machinery of the Archaebacterium Halobacterium cutirubrum -
molecular characterization. Systematic and Applied Microbiology, 7(1):]06-114.

Bremer, H. and RP. Dennis, 1996. Modulation of chemical composition and other
parameters of the cell by growth rate; pp. 1553-1569 in Escherichia coli and
Salmonella: Cellular and Molecular Biology; F.C. Neidhardt, et al., eds. ASM
Press, Washington, D. C.

Maaloe, O. and NO. Kjeldgaard, 1966. Control of Macromolecular Synthesis.
Microbial and Molecular Biology Series, ed. B.D. Davis. W.A. Benjamin, Inc.,
New York.

Maaloe, O., 1979. Regulation of the protein-synthesizing machinery - ribosomes,
tRNA, factors, and so on; pp. 487-542 in Gene Expression; R.F. Goldberger, ed.
Plenum Press, New York.

Koch, AL. and CS. Deppe, 1971. In vivo assay of protein synthesizing capacity
of Escherichia colifiom slowly growing chemostat cultures. Journal of Molecular
Biology, 55:549-562.

Koch, A.L., 1980. Inefliciency of ribosomes functioning in Escherichia coli
growing at moderate rates. Journal of General Microbiology, 116(1): 165-1 71.

Wada, A., K. Igarashi, S. Yoshimura, S. Aimoto and A. Ishihama, 1995.
Ribosome modulation factor - stationary growth phase-specific inhibitor of
ribosome functions from Escherichia coli. Biochemical and Biophysical Research
Communications, 214(2):410-417.

Wada, A., R. Mikkola, C.G. Kurland and A. Ishihama, 2000. Growth phase-
coupled changes of the ribosome profile in natural isolates and laboratory strains
of Escherichia coli. Journal of Bacteriology, 182(10):2893-2899.

25

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

Yamagishi, M., H. Matsushima, A. Wada, M. Sakagami, N. Fujita and A.
Ishihama, 1993. Regulation of the Escherichia coli rmf gene encoding the

ribosome modulation factor - growth phase-dependent and growth rate-dependent
control. EMBO Journal, 12(2):625-630.

Forchhammer, J. and L. Lindahl, 1971. Growth rate of polypeptide chains as a
function of cell growth rate in a mutant of Escherichia coli 15. Journal of
Molecular Biology, 55(3):563-568.

Ehrenberg, M. and CG. Kurland, 1984. Costs of accuracy determined by a
maximal growth rate constraint. Quarterly Reviews of Biophysics, 17(1):45-82.

Kurland, CG, 1987. Strategies for efliciency and accuracy in gene expression.
Trends in Biochemical Sciences, 12(4):126-128.

Kjeldgaard, NO. and CG. Kurland, 1963. The distribution of soluble and
ribosomal RNA as a function of growth rate. Journal of Molecular Biology,
6:341-348.

Kurland, C.G., D. Hughes and M. Ehrenberg, 1996. Limitations of translational
accuracy; pp. 979-1004 in Escherichia coli and Salmonella: Cellular and
Molecular Biology; F.C. Neidhardt, et al., eds. ASM Press, Washington, D. C.

Andersson, S.G.E. and CG. Kurland, 1990. Codon preferences in free-living
microorganisms. Microbiological Reviews, 54(2): 198-210.

Sharp, RM. and W.H. Li, 1987. The Codon Adaptation Index - a measure of
directional synonymous codon usage bias, and its potential applications. Nucleic
Acids Research, 15(3):128l-1295.

Adkins, C.J., 1983. Equilibrium Thermodynamics, 3rd. ed. Cambridge University
Press, Cambridge, UK.

Begon, M., J .L. Harper and CR. Townsend, 1996. Ecology: Individuals,
Populations and Communities, 3rd. ed. Blackwell Science, Oxford.

French, A.P., 1971. Newtonian Mechanics. The MIT Introductory Physics
Series. W. W. Norton & Co., New York.

Odum, HT. and RC. Pinkerton, 1955. T ime’s speed regulator: The optimum

efliciency for maximum power output in physical and biological systems.
American Scientist, 43:331-343.

Smith, CC, 1976. When and how much to reproduce - trade-ofl' between power
and efi‘iciency. American Zoologist, 16(4):763-774.

Silvert, W., 1982. The theory of power and efficiency in ecology. Ecological
Modelling, 1 5(2): 1 59-164.

26

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

Watt, W.B., 1986. Power and efficiency as indexes of fitness in metabolic
organization. American Naturalist, 127(5):629-653.

Mansson, BA. and J .M. McGlade, 1993. Ecology, thermodynamics and H. T.
Odum’s conjectures. Oecologia, 93(4):582-596.

Corning, RA. and SJ. Kline, 1998. Thermodynamics, information and life
revisited. Part I: 'To be or entropy '. Systems Research and Behavioral Science,
15(4):273-295.

Arendt, J .D., 1997. Adaptive intrinsic growth rates: An integration across taxa.
Quarterly Review of Biology, 72(2): 149-177.

Reichman, OJ. and D. Oberstein, 1977. Selection of seed distribution types by
Dipodomys merriami and Perognathus amplus. Ecology, 58(3):636-643.

Brown, J .S., B.P. Kotler and W.A. Mitchell, 1994. Foraging theory, patch use,
and the structure of a Negev desert granivore community. Ecology, 75(8):2286-
2300.

Mueller, P. and J. Diamond, 2001. Metabolic rate and environmental
productivity: Well-provisioned animals evolved to run and idle fast. Proceedings
of the National Academy of Sciences of the United States of America,
98(22):]2550-12554.

Riha, V.F. and K.A. Berven, 1991. An analysis of latitudinal variation in the
larval development of the wood frog (Rana sylvatica). Copeia, (1):209-221.

Schmitt, R.J., 1996. Exploitation competition in mobile grazers: T rade-ofifs in use
of a limited resource. Ecology, 77(2):408-425.

Gotthard, K., S. Nylin and C. Wiklund, 1994. Adaptive variation in growth rate -
life history costs and consequences in the speckled wood butterfly, Pararge
aegeria. Oecologia, 99(3-4):281-289.

Reznick, D., L. Nunney and A. Tessier, 2000. Big houses, big cars, superfleas
and the costs of reproduction. Trends in Ecology & Evolution, 15(10):421-425.

Tessier, A.J., M.A. Leibold and J. Tsao, 2000. A fundamental trade-ofi” in
resource exploitation by Daphnia and consequences to plankton communities.
Ecology, 81(3):826-841.

Tessier, A.J. and P. Woodruff, 2002. Trading ofl the ability to exploit rich versus
poor food quality. Ecology Letters, 5(5):685-692.

Kobe, R.K., S.W. Pacala, J .A. Silander and CD. Canham, 1995. Juvenile tree
survivorship as a component of shade tolerance. Ecological Applications,
5(2):517-532.

27

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50.

51.

Chapin, RS, 1980. The mineral nutrition of wild plants. Annual Review of
Ecology and Systematics, 11:233-260.

McGraw, J .B. and F .S. Chapin, 1989. Competitive ability and adaptation to fertile
and infertile soils in 2 Eriophorum species. Ecology, 70(3):736-749.

Chapin, F.S., K. Autumn and F. Pugnaire, 1993. Evolution of suites of traits in
response to environmental stress. American Naturalist, 142:878-S92.

Sommer, U., 1985. Comparison between steady state and non-steady state
competition - experiments with natural phytoplankton. Limnology and
Oceanography, 30(2):335-346.

Matin, A. and H. Veldkamp, 1978. Physiological basis of selective advantage of a
Spirillum sp. in a carbon-limited environment. Journal of General Microbiology,
105(4):187-197.

J annasch, H.W., 1979. Microbial ecology of aquatic low nutrient habitats; pp.
243-260 in Strategies of Microbial Life in Extreme Environments: report of the
Dahlem Workshop; M. Shilo, ed. Verlag Chemie, Weinhein.

Veldkamp, H., H. Van Gemerden, W. Harder and H.J. Laanbroek, 1984.
Competition among bacteria: an overview; pp. 279-290 in Current Perspectives
in Microbial Ecology; M.J. Klug and CA. Reddy, eds. American Society for
Microbiology, Washington, D. C.

Cavicchioli, R., M. Ostrowski, F. F egatella, A. Goodchild and N. Guixa-
Boixereu, 2003. Life under nutrient limitation in oligotrophic marine
environments: An eco/physiological perspective of Sphingopyxis alaskensis
(formerly Sphingomonas alaskensis). Microbial Ecology, 45(3):203-217.

Helling, RB, 2002. Speed versus efliciency in microbial growth and the role of
parallel pathways. Journal of Bacteriology, 184(4): 1041-1045.

Velicer, OJ. and RE. Lenski, 1999. Evolutionary trade-ofls under conditions of
resource abundance and scarcity: Experiments with bacteria. Ecology,
80(4):l 168-1 179.

Velicer, G.J., T.M. Schmidt and RE. Lenski, 1999. Application of traditional and
phylogenetically based comparative methods to test for a trade-ofl' in bacterial

growth rate at low versus high substrate concentration. Microbial Ecology,
38(3):19l-200.

Vasi, F., M. Travisano and RE. Lenski, 1994. Long term experimental evolution

in Escherichia coli. 2. Changes in life history traits during adaptation to a
seasonal environment. American Naturalist, 144(3):432-456.

28

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

Ingraham, J .L., O. Maaloe and RC. Neidhardt, 1983. Growth of the Bacterial
Cell. Sinauer, Sunderland, Massachusetts.

Neidhardt, F .C., J .L. Ingraham and M. Schaechter, 1990. Physiology of the
Bacterial Cell. Sinauer, Sunderland, Massachusetts.

Dong, H.J., L. Nilsson and CG. Kurland, 1996. Co-variation of tRNA abundance
and codon usage in Escherichia coli at dtfi‘erent growth rates. Journal of
Molecular Biology, 260(5):649-663.

Rosset, R., J. J ulien and R. Monier, 1966. Ribonucleic acid composition of
bacteria as a function of growth rate. Joumal of Molecular Biology, 18:308-320.

Berg, O.G. and CG. Kurland, 1997. Growth rate-optimised tRNA abundance and
codon usage. Journal of Molecular Biology, 270(4):544-550.

Poulsen, L.K., G. Ballard and D.A. Stahl, 1993. Use of ribosomal RNA
fluorescence in situ hybridization for measuring the activity of single cells in

young and established biofilms. Applied and Environmental Microbiology,
59(5):l354-1360.

Parrott, L.M. and J .H. Slater, 1980. The DNA, RNA and protein composition of
the Cyanobacterium Anacystis nidulans grown in light-limited and carbon
dioxide-limited chemostats. Archives of Microbiology, 127(1):53-58.

Riesenberg, D. and F. Bergter, 1979. Dependence of macromolecular composition
and morphology of Streptomyces hygroscopicus on specific growth rate.
Zeitschrifi fur Allgemeine Mikrobiologie, 19(6):415-430.

Mink, R.W. and RB. Hespell, 1981. Survival of Megasphaera elsdenii during
starvation. Current Microbiology, 5(1):51-56.

Boudreaux, DP. and V.R. Srinivasan, 1981. A continuous culture study of growth
of Bacillus cereus T. Journal of General Microbiology, 122(1):129-136.

Mink, R.W. and RB. Hespell, 1981. Long term nutrient starvation of
continuously cultured (glucose-limited) Selenomonas ruminantium. Journal of

Bacteriology, 148(2):541-550.

Mink, R.W., J .A. Patterson and RB. Hespell, 1982. Changes in viability, cell
composition, and enzyme levels during starvation of continuously cultured
(ammonia-limited) Selenomonas ruminantium. Applied and Environmental
Microbiology, 44(4):913-922.

29

64.

65.

66.

67.

68.

69.

70.

71.

72.

73.

74.

75.

Tempest, D.W. and J .W. Dicks, 1967. Inter-relationships between potassium,
magnesium, phosphorus and ribonucleic acid in the growth of Aerobacter
aerogenes in a chemostat; pp. 140-153 in Microbial Physiology and Continuous
Culture, Proceedings of the Third International Symposium; E.O. Powell, C.G.T.
Evans, R.E. Strange and D.W. Tempest, eds. Her Majesty's Stationary Office,
London.

Beresford, T. and S. Condon, 1993. Physiological and genetic regulation of
ribosomal RNA synthesis in Lactococcus. Journal of General Microbiology,
139:2009-2017.

Farewell, A. and RC. Neidhardt, 1998. Eflect of temperature on in vivo protein
synthetic capacity in Escherichia coli. Journal of Bacteriology, 180(17):4704-
4710.

Schaechter, M., O. Maaloe and NO. Kjeldgaard, 1958. Dependency on medium

and temperature of cell size and chemical composition during balanced growth of
Salmonella typhimurium. Journal of General Microbiology, 19(3):592-606.

Mikkola, R. and CG. Kurland, 1988. Media dependence of translational mutant
phenotype. FEMS Microbiology Letters, 56(3):265-269.

Lafay, B., J .C. Atherton and PM. Sharp, 2000. Absence of translationally
selected synonymous codon usage bias in Helicobacter pylori. Microbiology-UK,
146:851-860.

Lafay, B., A.T. Lloyd, M.J. McLean, K.M. Devine, P.M. Sharp and K.H. Wolfe,
1999. Proteome composition and codon usage in spirochaetes: Species-specific
and DNA strand-specific mutational biases. Nucleic Acids Research, 27(7): 1642-
1649.

Ochman, H. and R.K. Selander, 1984. Standard reference strains of Escherichia
coli from natural populations. Journal of Bacteriology, 157(2):690-693.

Mikkola, R. and CG. Kurland, 1991. Is there a unique ribosome phenotype for
naturally-occurring Escherichia coli? Biochimie, 73(7-8):1061-1066.

Kurland, CG. and R. Mikkola, 1993. The impact of nutritional state on the
microevolution of ribosomes; pp. 225-237 in Starvation in Bacteria; S.
Kjelleberg, ed. Plenum Press, New York.

Mikkola, R. and CG. Kurland, 1992. Selection of laboratory wild-type phenotype
from natural isolates of Escherichia coli in chemostats. Molecular Biology and
Evolution, 9(3):394-402.

MacArthur, RH. and E0. Wilson, 1967. The Theory of Island Biogeography.
Monographs in Population Biology. Princeton University Press, Princeton, New
Jersey.

30

76.

77.

78.

79.

80.

81.

82.

83.

84.

85.

86.

87.

Pianka, ER, 1970. r-selection and K—selection. American Naturalist,
104(940):592-597.

Andrews, J .H. and RF. Harris, 1986. r-selection and K-selection and microbial
ecology. Advances in Microbial Ecology, 9:99-147.

Hirsch, P., M. Bernhard, S.S. Cohen, J .C. Ensign, H.W. J annasch, A.L. Koch,
K.C. Marshall, A. Matin, J .S. Poindexter, S.C. Rittenberg, D.C. Smith and H.
Veldkamp, 1979. Life under conditions of low nutrient concentrations group

report; pp. 357-3 72 in Strategies of Microbial Life in Extreme Environments:
report of the Dahlem Workshop; M. Shilo, ed. Verlag Chemie, Weinhein.

Poindexter, J .S., 1981. Oligotrophy - fast and famine existence. Advances in
Microbial Ecology, 5:63-89.

Koch, A.L., 1979. Microbial growth in low concentrations of nutrients; pp. 261-
279 in Strategies of Microbial Life in Extreme Environments: report of the
Dahlem Workshop; M. Shilo, ed. Verlag Chemie, Weinhein.

Morita, R.Y., 1997. Bacteria in Oligotrophic Environments. Chapman & Hall,
New York.

Condon, C., D. Liveris, C. Squires, I. Schwartz and CL. Squires, 1995.
Ribosomal RNA operon multiplicity in Escherichia coli and the physiological
implications of rrn inactivation. Journal of Bacteriology, 177(14):4152-4156.

Stevenson, BS. and T.M. Schmidt, 1998. Growth rate-dependent accumulation of

RNA from plasmid-borne rRNA operons in Escherichia coli. Journal of
Bacteriology, 180(7): 1970-1972.

Stevenson, BS, 2000. Life history implications of ribosomal RNA gene copy
number in Escherichia coli, Ph.D. Dissertation, Microbiology Department,
Michigan State University. East Lansing.

Klappenbach, J .A., 2001. Ecological and evolutionary implications of ribosomal
RNA gene copy number in heterotrophic soil bacteria, Ph.D. Dissertation,
Department of Microbiology and Molecular Genetics, Michigan State University.
East Lansing.

Klappenbach, J .A., J .M. Dunbar and T.M. Schmidt, 2000. rRNA operon copy
number reflects ecological strategies of bacteria. Applied and Environmental
Microbiology, 66(4): 1328-1333.

Hansen, SR. and SP. Hubbell, 1980. Single nutrient microbial competition:

Qualitative agreement between experimental and theoretically forecast outcomes.
Science, 207: 1491-1493.

31

Chapter 2. Macromolecular composition of bacterial soil
isolates representing contrasting ecological strategies

Chapter 2 Abstract

The growth rate, culture density, cell size and macromolecular composition were
assessed in batch culture in two different media for each of 10 bacterial strains. The 8
recent soil isolates and 2 well-characterized strains were chosen to represent either of 2
contrasting ecological strategies. The 'rapid responder' strategy was defined operationally
by the ability to form a visible colony within 2 days during initial isolation and by
possession of 4 or more copies of the ribosomal RNA Operon per genome. 'Slow
responders' required a week or more to form a visible colony during initial isolation and
possessed 1 or 2 ribosomal RNA operons per genome. Each ecological strategy was
represented by members of 4 diverse phylogenetic groups commonly found in soil. Some
trends in the data were consistent with expectations, such as the rank order of DNA, RNA
and protein mass within strains, and a positive correlation of both cell size and the
RNA:DNA ratio with grth rate across strains. However, the variability between
strains is such that trends observed across strains are often of limited value in predicting
the characteristics of individual strains. Overall, the macromolecular composition of
rapidly responding and slowly responding bacteria are similar, although their grth

rates in the same medium are considerably different.

32

Chapter 2 Introduction

Knowledge of the macromolecular composition of bacteria is fundamental for
understanding bacterial physiology and ecology. For example, the investigations of the
Copenhagen School of microbiology into the DNA, RNA and protein content of enteric
bacteria during stable and transient states of growth in a variety of culture conditions 1’2
led to fundamental insights about the control and coordination of bacterial growth 3,4.
These insights provide a system-wide view of the behavior of the self-replicating,
homeostatic metabolic network that is a bacterial cell 5’6. An explanation of this behavior
remains the goal of ongoing efforts to develop highly detailed models of bacterial cell
growth based on genomic, proteomic and metabolomic data 7’8. At the opposite end of
the biological size scale, bacterial macromolecular content influences global cycles of
carbon, nitrogen, phosphorus and other elements 9'11, because macromolecular content is

linked to both the elemental composition and the grth rate of bacteria 12.

An increasing awareness of the scope of microbial diversity reveals that the majority of
previous studies of macromolecular content have focused on a small group of closely
related bacteria that share similar ecological characteristics. Escherichia coli and other
enteric bacteria have received the most attention; relatively few studies have investigated
microbes from the many other phylogenetic lineages that exist. Although some studies in
the latter group have noted contrasts with the patterns observed in E. coli 13‘15, we have
not found any published report comparing the DNA, RNA and protein content of
different bacterial species chosen specifically to represent either evolutionary or

ecological diversity.

33

Because the fundamental roles of the 3 major macromolecules are similar in all known
cellular life, with DNA as an information storage molecule, RNA providing the central
components of the protein synthetic machinery, and protein as the predominant functional
molecule, we expect similarities in the macromolecular composition of microbes
regardless of their phylogenetic affiliation. Nonetheless, comparisons of macromolecular
content across a diverse group of microbes are important for investigating the generality

of the patterns found among the enteric bacteria.

34

Chapter 2 Materials and Methods

Unless specifically mentioned, Sigma was the source for all chemicals.
Strains and media

The soil isolates used in this study were all obtained from long term plating experiments
that followed essentially the same procedure 17'19, designed to permit the isolation of
strains which vary widely in the time required for the formation of a visible colony on
solid media. Briefly, dilutions of a soil suspension were spread onto a dilute nutrient
solution solidified with 1.5% agar; plates were incubated for at least 10 days at room
temperature with adequate humidity to prevent dehydration. Plates were examined daily
for new colonies; colonies were marked uniquely by the day of appearance to allow
subsequent recovery of isolates with known colony appearance times. Strains
S21027/HF 3 and S24542/HSS were a gift from T. Hattori, derived from rice paddy soil in
Japan, using a l/ 100 dilution of Nutrient Broth (Difco) as the nutrient solution 17. The
remaining soil isolates were a gift from J. Klappenbach, derived from agricultural soil at
the Long Term Ecological Research site at the Kellogg Biological Station, Michigan,
USA 18,19. The nutrient solution used for isolation of strains EC2, EC4, ECS and LC9
was a 1/ 100 dilution of Nutrient Broth (Difco); for strains PX3.14 and PX3.15 the
nutrient solution was 5 mM succinate in a basal salts solution. These 8 strains were
chosen from a larger collection of soil isolates obtained in the same fashion, with the goal
of obtaining related pairs of soil bacteria from a number of different phylogenetic groups.
Strains that formed spores or that did not grow with dispersed turbidity in liquid culture

were deliberately avoided to facilitate growth rate measurements.

35

The two members of each related pair of strains were selected to represent contrasting
ecological strategies that we characterize as a rapid growth response or a slow growth
response to resource abundance. The ‘rapid responders’ were isolated from colonies that
were first visible 1-2 days after plating, and contained 4 or more copies of the ribosomal
RNA (rrn) operon per genome USA 13,19. The ‘slow responders’ were isolated from
colonies that were first visible 7 days or more after plating, and contained 1 or 2 rrn
copies per genome 18,19. While the time interval between plating and the formation of
visible colonies decreased for both groups during subsequent cultivation in the lab, the
rapid responders continued to form colonies more quickly than the slow responders on a
variety of solid media. In addition to these 8 soil isolates, we investigated two well-
characterized strains representing the contrasting ecological strategies. E. coli REL607, a
derivative of E. coli B/r, was a gift from R. Lenski 20 and represented a rapidly
responding strain. Sphingopyxis alaskensis RB2256 (originally Sphingomonas), a marine
ultramicrobacterium representing a slowly responding strain, was a gift from R.
Cavicchioli and M. Ostrowski 21,22. This strain was isolated by dilution culture
techniques as described by Schut and colleagues 23:24. For all strains, fi'eezer stocks in
R2BV (see below) with 10% glycerol were prepared and stored at -80°C soon after the
isolation of soil strains or immediately after receiving strains into the lab, to minimize

adaptation to the lab environment.

Each of these 10 strains was grown in batch culture in two media, to make cell

measurements and to harvest biomass during balanced exponential grth at two

36

different growth rates. The first medium for all strains was R2BV, containing per liter:
glucose, 0.5 g; proteose peptone, 0.5 g; yeast extract, 0.5 g; casamino acids, 0.5 g; soluble
starch, 0.5 g; KHZPO4, 0.3 g; sodium pyruvate, 0.3 g; MgSO4, 0.05 g; and 1 ml vitamin
solution. For most strains, the second medium was PVY/ 10, containing per liter:
NazHPO4, 0.71 g; KHZ P04, 0.68 g; (NH4) 2SO4, 0.66 g; MgSO4, 0.12 g; proteose
peptone, 0.5 g; yeast extract, 0.05 g; and 1 ml each vitamin solution and trace element
solution. E. coli, EC2, HS5 and LC9 did not grow adequately in PVY/ 10, so the second
medium for these strains was R2B-GCS, which is identical to R2BV with the omission of
glucose, casamino acids, soluble starch, and vitamins. The vitamin solution contains, per
liter: 200 mg of i-inositol, and 100 mg of each of the following: biotin, choline-Cl, folic
acid, lipoic acid, nicotinamide, pantothenate, para-aminobenzoic acid, pyridoxal HCl,
riboflavin, and thiamine HCl. The trace elements solution contains, per liter:

concentrated (11.6 N) HCl, 6.03 ml; FeSO4-7H20, 2.085 g; ZnSO4, 143.6 mg;
MnC12-4HZO, 89.1 mg; H3BO3 6.2 mg; CoCl-6HZO, 190.3 mg; CuCl4-2H20, 1.7 mg;

NiC12-6HZO, 23.8 mg; NazMoO4°2HZO, 48.4 mg.

Growth and harvest

Cells were grown aerobically at 25°C in 7-10 ml medium in 18 mm diameter tubes
shaking at 200 rpm (E. coli, HF3, EC2, EC4, EC5 and PX3.15), or in 45 ml medium in
baffled nephelometer flasks of 250 ml capacity shaking at 100 rpm (S. alaskensis,
PX3.14, HS5 and LC9). Growth was monitored by turbidity in a Spectronic 20D+
spectrophotometer (Milton Roy) at 420 nm; this short wavelength was chosen to

maximize sensitivity at low cell density. Biomass for macromolecular analysis was

37

obtained by inoculating a pair of tubes from freezer stock; during exponential growth,
one of these cultures provided a 1% or smaller inoculum for 4 additional tubes containing
the same medium. If the exponential grth rates in these tubes were consistent with
each other and with prior experience, one of these tubes provided a 1% or smaller
inoculum for a set of culture vessels (4 tubes or 3 flasks) containing the same medium,
used to harvest bacterial biomass for macromolecular analysis. For strain HF3 growing
in PVY/ 10, one additional transfer was made prior to harvest when the target optical
density for harvest was missed. All culture vessels inoculated at the same time needed to
show essentially the same growth rate, with at least 4 turbidity measurements obtained
during exponential growth, and a final optical density near the target value, in order to be
used for macromolecular analysis. Specific grth rates were calculated as the slope of a

linear regression of the natural logarithm of OD420 with time for each culture vessel,

using the final 4 optical density measurements prior to harvest. All growth rates are

reported as the specific grth rate it = ln(2)/g, where g is the generation time.

The target density for harvest of biomass was half the optical density at which departure
from exponential growth was first detectable during preliminary grth experiments, or

at OD420= 0.2, whichever was lower, so that cultures were at least one generation time

away from a detectable response to changing culture conditions. At harvest, for all
strains except PX3.15, 3-4 replicate aliquots (analytical replicates) containing 2-10 ml of
culture were obtained from each replicate culture vessel (experimental replicates) of a
particular strain-medium combination. For strain PX3.15 in each growth medium, 8

tubes (experimental replicates) containing 10 m1 culture each contributed a single 9.5 ml

38

aliquot for macromolecular analysis. Culture aliquots for determination of DNA, RNA
and protein content were centrifirged at 10,000g and 4°C for 10-30 minutes, medium was
decanted, and cell pellets were stored at —80°C until analysis. Also at the time of harvest,
separate culture aliquots were fixed in formaldehyde for cell enumeration, and

photomicrographs of cells were taken for biovolume determination.
Overview of macromolecular analysis

A single run through the macromolecular analysis protocol involved 1-2 days work with
8 cell pellets; with the exception of strain PX3.15, analysis of pellets from different
culture replicates, media types and strains were dispersed over many days of analysis to
control for any day-to-day variation in measurements. The initial cultures of strain

PX3. 15 in both media provided inadequate biomass for analysis. Larger biomass samples
from subsequent cultures of this strain were analyzed after the analysis of other strains
was completed, with material from replicate cultures in both media analyzed over two

separate runs.

For a single run of the macromolecular analysis protocol, 8 cell pellets were thawed in a
lysis buffer and sonicated to disrupt cells. Sonicated material from each pellet was
subsarnpled for DNA, RNA and protein measurements; each macromolecule was
measured in multiple, independent dilution series of sonicated cell material in a 96-well
plate format. Thus, there was analytical replication at each step of analysis for each
molecule measured fiom a single cell pellet, as well as analytical replication of cell

pellets derived from a single culture vessel (except for strain PX3.15, for which each cell

39

pellet represented a distinct culture vessel). This approach allowed individual outlying

values to be recognized and discarded.

Cell lysis

Sonication was the method chosen for cell lysis (in combination with freeze-thaw
treatment for gram positive strains) to permit rapid, complete lysis without loss of
material or addition of reagents that could interfere with subsequent analysis. To
minimize nuclease and other enzymatic activity, bacterial cell material was held
continuously on ice or in ice water baths after removal from storage at -80°C until shortly
before fluorescence measurements for nucleic acid determination or until the start of
trichloroacetic acid (TCA) extraction for protein determination. Initial fluorescence
values were not affected by an overnight room temperature incubation of the prepared
96-well plates, suggesting that nuclease activity subsequent to warming of the material

did not affect nucleic acid measurements (data not shown).

Cell pellets were generally thawed in 3 ml of ice-cold nuclease-free lysis buffer (10mM
Tris, lmM ethylenediaminetetraacetic acid (EDTA), 2% (v/v) ethanol, 0.05% (v/v) Igepal
CA-630 detergent, 200 mg/l sodium deoxycholate, pH 7.0). However, some strain-
medium combinations with cell pellets of low macromolecular content were thawed in
only 1.5 ml lysis buffer. Cell pellets from gram positive strains EC5 and PX3.15 were
subjected to 5 cycles of freeze-thaw between a dry ice-ethanol bath and a 95°C water
bath prior to the addition of lysis buffer. Cell material and lysis buffer were transferred
to 15 m1 capacity conical bottom glass centrifuge tubes (Kimble) for sonication; the

conical shape maximizes the insertion depth of the sonication probe to reduce foaming,

4O

and glass better conducts heat away from sonicated material to the surrounding ice water
bath. Sonication used a 1/8" tip tapered probe in a Sonifier model 450 (Branson) at half-
maximum power with a 25% duty cycle for 90 seconds. This duration of sonication was
2-3 times the length of time after which no further increase in fluorescent signal was
observed in preliminary experiments; sonication could continue for 8-10 minutes at these
settings with no loss of fluorescent signal (data not shown). Sonicated cell material was
investigated for the efficiency of cell lysis both by plating and by microscopy, which

indicated that neither viable nor intact cells remained after sonication.

Nucleic acid determination

The proprietary fluorescent dyes PicoGreen and RiboGreen (Molecular Probes, 25,25)
were chosen for nucleic acid determination because of their sensitivity, wide dynamic
range, and ease of use in comparison to traditional colorimetric methods such as the
diphenylamine reaction for DNA and the orcinol reaction for RNA. Except as noted
below, measurements followed the procedure suggested by the manufacturer (with the
high range dye concentration for RiboGreen), using 150 pl diluted sample and 150 pl
working dye solution per well in clear-bottom, black 96-well microtiter plates (Costar).
Control experiments with reagents and cell material from all strain-medium combinations
at the same concentrations used for nucleic acid measurements indicated that in the
absence of the fluorescent dyes, fluorescence was negligible. Similarly, we found no
inhibition of the expected fluorescence from known concentrations of DNA and RNA by
cell material of any strain at the concentrations use for nucleic acid determination. A
nonlinear fluorescence response at the highest concentrations of cell material from strain

EC4 during early measurements may have indicated the presence of an inhibitor of

41

RiboGreen fluorescence. Additional dilution of EC4 cell material restored a linear
relationship between the concentration of cell material and RiboGreen fluorescence

response.

Fluorescence was measured from images obtained on a Storm 860 fluorescence scanner
(Molecular Dynamics) using blue laser excitation with 100 micron resolution and 1000
volts PMT; edges of the microtiter plates were removed prior to sample loading so the
flat well bottoms rested directly on the glass optical bed of the scanner. Fluorescent plate
images were analyzed using IrnageQuant 5.0 software (Molecular Dynamics).
Fluorescence per well was calculated by summing pixel intensity within unifonnly sized
circles applied to the image of each well. The circles were sized slightly smaller than the
wells; positions of the circles were adjusted manually as necessary to minimize the
influence of dust or tiny bubbles which were evident in the images of some wells. Use of
a fluorescent plate reader with the ability to stimulate PicoGreen and RiboGreen
fluorescence at their optimal excitation wavelength would improve sensitivity compared
to a fluorescent scanner, but would not provide the ability to recognize and sometimes

compensate for artifacts in the wells.

Immediately after sonication, cell material was vortexed and triplicate aliquots were
diluted 1:10 (or more for some strain-medium combinations) into 1.00 ml TE buffer (10
mM Tris, lmM EDTA) for nucleic acid assays. For strains which required an initial
dilution greater than 1:10 (and hence a smaller aliquot of sonicated cell material added to

1.00 ml T B, see below), a compensatory volume of the lysis buffer was added as well, so

42

the final volume and chemical environment of the triplicate initial dilutions was identical
for all strains and media, equivalent to a 1:10 dilution of the lysis buffer into 1.00 ml TE.
Each of the triplicate initial dilutions fi'om a single cell pellet was assayed in a 3-well
dilution series for each of the two fluorescent dyes. The dilution series comprised no
additional dilution, an additional 2:3 dilution, and an additional 1:3 dilution subsequent to
the initial 1:10 dilution. The diluent for these wells was a 1:10 dilution of lysis buffer in
TE, so that the chemical environment provided by the buffer remained unchanged in all

sample wells; only sonicated cell material was diluted.

Estimates of the DNA concentration in sample wells were derived from PicoGreen
fluorescence, compared to a linear standard curve of PicoGreen fluorescence from a
dilution series of it phage DNA. PicoGreen is advertised as a DNA-specific dye; in a
solution containing an equal mass of single-stranded RNA and double stranded DNA, the
RNA-induced fluorescence is less than 5% of DNA-induced fluorescence when the ratio
of PicoGreen to total nucleic acids is high 25. However, as the ratio of PicoGreen to total
nucleic acids decreases in a solution containing a 1:1 mass ratio of DNA and RNA, RNA-
induced fluorescence can approach 20% of DNA fluorescence 25. Thus, for sonicated
cell material containing as much or more RNA than DNA, the cell material must be
sufficiently dilute to ensure that PicoGreen fluorescence can be interpreted as a DNA
signal. For this study, PicoGreen fluorescence from the dilution series of sonicated cell
material was examined to ensure that different dilutions gave approximately the same
estimate of DNA concentration in the original sample. A trend of higher DNA estimates

with increasing sample concentration in all 3 dilution series from the independent initial

43

dilutions was taken as evidence that RNA-induced fluorescence was making an increased
contribution to total PicoGreen fluorescence in the more concentrated samples. In this
case, DNA estimates were derived only from the fluorescence of more dilute wells, and
future measurements of cell pellets from that strain-medium combination used a higher

initial dilution.

Estimates of RNA concentration in sample wells were derived by a mathematical
subtraction procedure using both RiboGreen and PicoGreen fluorescence data, since
RiboGreen fluorescence is induced by both RNA and DNA 26. The DNA contribution to
the RiboGreen fluorescence of a sample well was estimated from the PicoGreen-derived
estimate of sample DNA content and a linear standard curve of RiboGreen fluorescence
from a A DNA dilution series. This estimated DNA-stimulated RiboGreen fluorescence
was subtracted from the total RiboGreen fluorescence of sample wells; residual
fluorescence was assumed to be due to RNA. The RNA-stimulated RiboGreen
fluorescence was converted to an estimated concentration of RNA using a linear standard

curve of RiboGreen fluorescence from a dilution series of pure E. coli rRNA.

Neither the protocol suggested by the manufacturer for DNase digestion prior to RNA
measurement with RiboGreen, nor a potential alternative of estimating RNA content by
RNase-induced fluorescence loss were followed, both because a preliminary test
indicated that the digestion products of either nuclease inhibited RiboGreen fluorescence
somewhat (data not shown), and because the subtraction protocol we implemented

involved fewer manipulations at the bench. However, estimates of sample RNA

44

concentration from the subtraction procedure are expected to be less precise than
estimates of DNA concentration with our approach, since errors of DNA measurement

affect estimates of RNA concentration as well.

All nucleic acid standard curves used with PicoGreen and RiboGreen were triplicate 5
point, 3-fold dilution series beginning at 1000 ng/ml (rRNA) or 500 11ng (It DNA), plus
5 sonicated reagent blank wells and 5 diluent blank wells. RNA and DNA standard
solutions were prepared at 10x concentrations (10 rig/ml and 5 ug/ml respectively) in
lysis buffer, sonicated, and diluted 1:10 in TE, following the same procedure as for
samples, prior to making 3-fold dilutions to the working standard concentrations and
storing single-use aliquots of standards at -80°C. The sonicated reagent blank was lysis
buffer, sonicated as for samples and diluted 1:10 in TE. Both the PicoGreen and
RiboGreen fluorescence at a given sample dilution were required to be within the range
of fluorescence observed in the nucleic acid standards, otherwise wells at that dilution
were excluded fiom calculations. If the fluorescence was too high from the most
concentrated sample wells, subsequent measurements of cell pellets from that strain-
medium combination used a higher initial dilution of sonicated cell material. If
fluorescence was too low from the least concentrated sample wells, multiple cell pellets
from a single culture vessel of that strain and medium were combined during subsequent
analyses if possible, otherwise single pellets were resuspended in only 1.5 ml lysis buffer

instead of 3.0 ml to attain a higher concentration of cell material.

45

Nucleic acid measurements of individual wells deviating by more than 3 studentized
residuals from the mean of replicate wells were discarded; if 2 of the 3 wells fi'om a
single initial dilution were excluded on this basis, the 3rd well from that initial dilution
was excluded also. The intercept of a linear regression of all remaining fluorescence
values from one cell pellet vs. sample concentration was compared to the mean
fluorescence of blank wells. If the intercept differed by more than 10% from the
measured blank and the omission of the most concentrated sample wells both improved
the fit of the linear regression and brought the intercept within 10% of the measured
blank, it was assumed that the total nucleic acid content of the most concentrated wells
exceeded the linear range of the assay. The nucleic acid estimate derived from lower
concentration wells was retained, and future cell pellets from that strain-medium

combination were subjected to a higher initial dilution.

Protein determination

The bicinchoninic acid (BCA) assay was chosen for protein determination since it is
more sensitive and suffers less from protein-protein variability than other colorimetric
protein assays 27; commercially available fluorescent protein dyes are highly sensitive to
the chemical environment of the assay and not necessarily protein-specific. The
MicroBCA assay kit (Pierce Chemical) was chosen for convenience; procedures followed
the manufacturer’s recommendations for microtiter plate measurements except as noted
below. Hot TCA extraction of protein 28 from sonicated cell material was performed
both to reduce interstrain variability in the chemical environment of the protein assay and

to remove protein from the Tris-based buffer used for cell lysis.

46

Following sonication of all cell pellets during a single run through the macromolecular
analysis protocol, duplicate 1.00 ml aliquots of sonicated cell material from each pellet
were added to 500 ml of 30% TCA (w/v) in locking 1.7 ml capacity microcentrifuge
tubes and immediately incubated at 80°C for 30 minutes. However, for cell pellets
originally diluted into 1.5 ml lysis buffer instead of 3.0 ml, only a single 1.0 ml TCA
extraction was performed. Tubes were then transferred to an ice water bath for 30
minutes, and centrifuged at 10,000g at 4°C for 30 minutes to pellet the precipitated
proteins. TCA was decanted, tubes were spun briefly, and remaining TCA was removed
by aspiration without disturbing the protein pellet. Protein pellets were either analyzed

immediately or frozen at -20°C overnight prior to analysis.

Protein pellets were resuspended by adding 50 ul of alkaline SDS (5% (v/v) sodium
dodecyl sulfate in 0.1 N NaOH) and shaking for 1 hour, followed by addition of 1.00 ml
of saline (0.9% (w/v) NaCl). The protein solution derived from each independent TCA
extraction was analyzed in 6 wells of an untreated, clear, flat-bottomed microtiter plate
(Nunc), 3 wells at the concentration resulting from protein resuspension as described
above, and 3 wells diluted another 1:3. The diluent in this case was alkaline SDS diluted
1:21 into saline, so the chemical environment of the assay was not changed by dilution.
Wells for protein analysis contained 150 pl of protein solution and 150 pl of working
BCA reagent; plates were incubated at room temperature for 90 minutes prior to reading
absorbance at 562 nm on a Biokinetics EL312e plate reader (Bio-Tek Instruments).
Immediately prior to reading the plates, it was necessary to pop any bubbles remaining in

the wells with the tip of a hypodermic needle.

47

Protein standards were dilutions of bovine serum albumin (BSA, Pierce Chemical) in
lysis buffer that had been sonicated and diluted 1:10 into TE as for cell samples.
Multiple 1.00 ml aliquots of sonicated protein prepared in a single batch were added to
500 ml of 30% TCA in looking microcentrifuge tubes and frozen until use. Two such
protein standard tubes were subjected to hot TCA extraction and. protein resolubilization
in parallel with samples; each tube contributed duplicate 5 point 2-fold dilution series of
the protein standard (starting at 1,000 ng/ml or 500 ng/ml depending on the run) for a

total of 4 protein standard dilution series per run of protein analysis.

Sonicated reagent blanks and diluent blanks (8 wells each) were included in each run.
The sonicated reagent blanks were sonicated lysis buffer diluted into TE and subjected to
TCA extraction and resolubilization exactly as for cell material and protein standards.
Diluent blanks were a 1:21 dilution of alkaline SDS into saline, chemically identical to
the material added to samples, protein standards and sonicated reagent blanks subsequent
to TCA extraction. However, color development differed significantly between wells
containing sonicated reagent blanks and diluent blanks (probably due to pH or chemical
effects of residual TCA, data not shown). Thus, we expect the final absorbance of
sample and standard wells to depend not only on protein content, but also on the degree
of dilution subsequent to TCA extraction. Therefore, blank values subtracted from the
raw absorbance of each well were specific to the dilution of the well, interpolated linearly
between the absorbance of sonicated reagent blanks and of diluent blanks. Protein

concentrations in sample wells were estimated from blank-corrected A562 values

48

according to a 2nd order standard curve fitted to the blank-corrected A562 values of

standard wells.

Sample, standard and blank wells were excluded if A562 values differed by more than 3

studentized residuals from the mean of replicate wells. If there were significant
differences between estimates of the original culture protein concentration derived from
the two different assay concentrations, the lower concentration was invariably found to
be close to the detection limit of the assay and those wells were excluded. Subsequent
measurements from that strain-medium combination omitted the 6 wells at the lower

concentration in favor of 2 additional wells at the higher concentration.

Color development in the BCA assay is initially very rapid and then decelerates with
time, but does not reach a true endpoint, which had several practical consequences for our
measurements. First, differences of about 2 minutes in the time of BCA reagent addition
(approximately the time required to add reagent to all wells of the two plates of a single
run) causes significant differences in the color intensity of replicate wells assessed
simultaneously, until the rate of color development has slowed considerably (data not
shown). Practically, this imposes a minimum incubation time on the assay. Second, we
found that incubating the plates at 60°C to accelerate color development (as suggested by
the manufacturer) led to enhanced color development in the outer wells of the plate (data
not shown), presumably because they warmed more quickly than interior wells. We
maintained our reagents and incubations at room temperature, and avoided using the

outermost wells of our plates as sample or standard wells. Instead, we filled the

49

outermost wells with water to provide thermal mass against temperature fluctuations.
Finally, there were subtle differences in the rate of color development between the BSA
protein standard and our samples of total bacterial protein. For all strains, the rate of
color development from bacterial protein was slightly less than that from BSA, so that
longer incubations resulted in lower estimates of sample protein concentration (data not
shown). This result suggested that maintaining consistent incubation times between runs
was essential, and that shorter incubations would be more precise. The incubation time
of 90 minutes was chosen to minimize the effect of different rates of color development
between samples and standards, while permitting adequate color development from
samples of low protein content and reducing the effect of slight differences in the time of

reagent addition to sample wells.

Identification of outliers in macromolecular data

For both nucleic acid and protein measurements, at least 2 cell pellets (analytical
replicates) were analyzed from each of 3-4 separate culture vessels (experimental
replicates) representing each strain-medium combination, except that for strain PX3. 15, a
single cell pellet from 8 separate culture vessels was analyzed for each medium.
Measured macromolecular concentrations from different cultures were normalized to

OD420 = 0.1 to correct for the slight differences in culture density between replicate
vessels at harvest. If the estimated DNA, RNA or protein content per OD420 unit of the
first two analytical replicates from a single culture vessel differed by more than 15%, a
3rd and sometimes a 4th analytical replicate was analyzed as well. Values from an

analytical replicate were discarded if they differed by more than 3 studentized residuals

from the mean of all analytical replicates analyzed for that strain-medium combination;

50

remaining values were averaged by experimental replicate to obtain the estimated
macromolecular content for a particular strain-medium combination. Calculation of the
mean and standard error of compound quantities for each strain-medium combination

was done by calculating the quantity separately for each experimental replicate.

Culture density

Determination of the number of cells per culture volume generally conformed to standard
protocols for counting bacterial cells stained with 4',6-diamidino-2-phenylindole (DAPI)
by epifluorescence microscopy 29,30, with some improvements and modifications as

follows:

Cells were fixed in formaldehyde (2% final concentration) at least overnight; small,
slowly growing cells in particular sometimes appeared dim after only 2-4 hours of
fixation. At least 2 filters were prepared per culture vessel within 48 hours of harvest,

stained filters stored at -20°C remained countable for at least a year.

The glass filter funnel was treated with Rain-X (Sopus Products), a commercial silanizing
reagent to reduce aqueous adhesion. Assembly of the 25 mm diameter glass filter tower
(Millipore) included two Teflon gaskets under the stainless steel fiit and two GF/F filters
(Whatrnan) beneath a 0.45 mm pore size mixed cellulose ester backing filter (Millipore)
to help create a tight seal of the filter funnel against the 0.2 mm pore size Anodisc
working filter (Whatrnan). All filters except the Anodisc (which is bonded to a
polyproylene handling ring) were handled from beneath with a spatula to avoid areas of

concentrated cell deposition on the working filter from the compression or rupture of

51

underlying filters. Such areas were fiequently observed following even gentle handling

of these filters with filter forceps.

20-100 pl of formaldehyde-fixed culture and 5-10 ul of a 0.5 mg/ml working solution of
DAPI were added to 2.5 m1 particle-free water in the fimnel, followed by another 2.5 m1
of water to mix the reagents. After 5-7 minutes staining, vacuum was applied
continuously at a maximum pressure differential of 25 kPa until 2 minutes alter the last
solution had been drawn through the filter sandwich. The Anodisc filter was dried for
several minutes in a foil-covered petri dish on a 60°C dry block heater, then applied to a
~28 mm diameter circle spread from 2 drops of type A (low viscosity, low fluorescence)
immersion oil (Cargille) on a 75 x 38 mm microscope slide. A 50 x 35 mm coverslip
with 1 drop of the same oil was applied to the filter, and the edges of the cover slip sealed

with cosmetic fingernail polish.

The entire surface of all filters was scanned under low power magnification to look for
regions of concentrated cell deposition and areas where liquid apparently passed through
the filter beyond the edge of the bore of the glass filter funnel (as indicated by the
presence of background DAPI fluorescence). Since the latter event was fairly common
(confirmed by the presence of cells in these areas, albeit at lower density) the entire
filterable area of each Anodisc filter was counted as follows: The vernier scales of the
microscope stage were used to estimate the diameter of the filterable area of the Anodisc
filter in both the X and Y direction on >150 filters; the mean value of 18.8 mm was used

to calculate the estimated filter area for all filters. Cells were counted within the ocular

52

grid for 6 randomly chosen, widely dispersed microscope fields between the edge of the
filterable area and the edge of the bore of the filter funnel (diameter 16 mm), and for 17
randomly chosen microscope fields on two perpendicular transects within the bore of the
filter funnel. Since the ratio of 6:17 is within 2% of the ratio of filter area outside and
inside the bore of the funnel, these counts could simply be summed and divided by the
total area covered by the 23 ocular grids to provide an estimate of the cell density on the

entire filter corrected for the two areas of the filter with different cell densities.

Cell counts were performed on an Axioscop 2 (Carl Zeiss) equipped with a 100 Watt Hg
lamp and a 100x Plan NeoFluar objective. Filters were prepared to have an average of
between 20 and 100 cells entirely within or touching two sides of the ocular grid when
viewing the central area of the filter, so that at least 400 cells were counted per filter.
Estimates of cell number per optical density derived from a single filter that differed by
more than 3 studentized residuals from the mean of the 6-10 filters prepared for all
cultures of a particular strain-medium combination were excluded; remaining estimates
were averaged by culture vessel to provide a single estimate of the cell number per

optical density for that strain-medium combination.
Cell size

Cell sizes were determined by image analysis of digital photomicrographs of
immobilized bacteria under phase-contrast illrurrination. Prior to biomass harvest,
agarose coated slides were prepared in a dust-free environment by pipetting 1 ml of a
tempered, molten 1.6% (w/v) solution of quadruple-washed agarose onto cleaned

microscope slides on a horizontal surface. After the agarose was solidified, slides were

53

dried at 60°C and stored without contact to the dried agarose surface. At harvest, 20-26
ml of culture was pipetted onto the dried agarose and covered immediately with a
coverslip. The unfixed, unstained cells were gradually immobilized against the coverslip
by the swelling agarose, with optimum contrast occurring just as the cells became
immobilized. The slides were scanned to find such areas, which were then photographed
in 8-bit grayscale with a Spot 2 cooled CCD camera (Diagnostic Instruments) mounted
on an Axioscop 2 (Carl Zeiss) using a 100x Plan-Neofluar objective under phase contrast
illumination with a green filter. 5-10 such images containing 20-200 mostly non-
overlapping cells were obtained from each culture vessel. For strain PX3. 15, slides
prepared at the time of harvest were flawed; images obtained from earlier cultures in the
same media (4 independent culture replicates per medium) were used for biovolume
determination instead. The highest quality images fiom each strain-medium combination
were processed in Photoshop 6.0 (Adobe) to produce binary images (pixels either black
or white) containing a total of at least 300 nontouching cells; the binary threshold was
chosen interactively so that cell sizes corresponded closely to the apparent cell size on the
original grayscale images. Nonetheless, some image-to-image variation in measured cell
dimensions was observed, reflecting differences in the quality of focus and contrast in the
original digital images. Shape-dependent measurement features available in the
CMELAS software package 31 were used with these binary images to produce biovolume
estimates. Actual dimensions of the imaged cells were calibrated fi'om a digital

photomicrograph of a stage micrometer visualized under identical conditions.

54

Statistical comparisons of data

Potential relationships between two quantities across all strains were examined using
simple linear regression of the mean values of the quantities for each strain. Because data
from the same strain growing in multiple media cannot be considered independent,
comparisons across strains use only data from cultures grown in R2BV, since all strains
were grown in this medium. Differences between the values of a quantity for a single
strain in two media were examined with t tests of values measured fi'om all replicate

cultures of a strain.

Phylogenetic analysis

The phylogenetic affiliations of bacterial strains in our collection were established using
standard protocols for small subunit (16S) ribosomal RNA gene sequence analysis.
Genomic DNA was extracted from pure cultures of the 8 soil isolates and S. alaskensis
using either the UltraClean Soil DNA kit (MoBio Laboratories) or the Bactozol kit
(Molecular Research Center) as directed, and used as a template for amplification of a
near firll length segment of the small subunit (16S) ribosomal RNA gene via the
polymerase chain reaction. Standard bacterial primers (8F paired with either 1492R or
1540R, numbers indicate the starting nucleotide of probe according to E. coli numbering)
were used with the following parameters for thermocycling: 5 minutes at 95°C; 30-35
cycles of 95°C/65°C/72°C held at each temperature for 30 seconds; 5 min at 72°C. (8F:
AGAGTTTGATCCTGGCTCAG, 1492R: GGTTACCTTGTTACGACTT, 1540R:
AAGGAGGTGATCCARCCGCA) Reaction mixtures were purified with either Wizard

PCR Preps (Promega) or Microcon Centrifugal Filters (Millipore) and amplicons were

55

sequenced directly using Big Dye terminator chemistry (Applied Biosystems) on an ABI
377 or an ABI 3700 Gene Sequencer (Applied Biosystems) via sequencing primers that
provided an average of >3x coverage over the length of the amplicon. Assembled
sequences have been deposited in GenBank with accession numbers AY337597-

AY337605.

16S gene sequences from our own sequencing efforts and 16S sequences from fully
sequenced bacterial and archaeal genomes were imported into an Arb database 32 and
aligned automatically to full length aligned sequences obtained from release 8.0 of the
Ribosomal Database Proj ect 33. Initial alignments were optimized manually with
reference to secondary structure information available at the Comparative RNA Website,
www.ma.icmb.utexas.edu 34. We chose a set of high quality, near full length 16S
sequences including our isolates, close relatives of our isolates, and organisms
representing all the major bacterial lineages, along with several diverse archaeal lineages
to serve as outgroups. A mask was created to exclude positions of ambiguous alignment,
leaving 1250 positions for phylogenetic inference using the maximum likelihood
algorithm within Arb. The resulting tree containing 166 bacterial and archaeal sequences
has been ‘pruned’ without altering branching order or branch lengths to obtain the
phylogenic tree displayed in Figure 2.1. The taxonomic affiliations shown in Table 2.1
are derived only from 16S sequence data, not phenotypic characterization, and includes

both the division and family (or genus, where possible) affiliation of each strain.

56

Figure 2.1: Phylogeny and rrn copy number of experimental bacteria

 

 

 

 

_ rm copy #
0.10 Inferred sequence 7 - -
changes per nucleotide _|—l:y9,s,-n,-a pestis 5.7
__ Haemophilus influenzae 6 >-
1 4
Pseudomonas putida 6 :

Duganella zoogloeoides
l I-

Ralstonia solanacaerum

 

 

 

[Rubrivivax gelatinosus
Neisseria meningitidis

 

 

B
Proteobacterla

 

   
  
 

 

 

 

 

 

 

 

 

  
  

 

 

 

 

 

6
4
1
4 _
Agrobacten‘um tumefaciens 4 ‘
Azos irillum Iipoferum -
IBEX!- 2 d
Caulobacter crescentus 2
hln opyxis alaskensis 1
Rickettsia prowazekii 1 - =
4 ,
Pedobacter heparinus - 3 5’3
2 '5 c
. . . a 'U
Sphmgobactenum comltans - m '5
_ ' Porphyromonas gingivalis 4 _
Mycobacten'um chlorophenolicum - -
1 a
'— Mycobacterium tuberculosis 1 2 E
l—__l 4 '5' o
Arthrobacter globifonnis " < .3
L— Streptomyces coelicolor 6 _

 

IE!!!- I slow I

Figure 2.1. Evolutionary relationships among the bacteria examined in this study and
selected reference species were examined by a maximum likelihood algorithm using near
full-length small subunit ribosomal RNA gene sequences. Rapidly responding bacteria
are shown in filled boxes, slowly responding bacteria in open boxes. Sequences from E.
coli and species shown without boxes were obtained from the Ribosomal Database
Project 33; other sequences were obtained in this study. The number of ribosomal RNA
(rrn) operons per genome is shown, where known19r35; taxonomic afiiliations of species
are shown at the right.

57

Table 2.1: Cell and culture traits of experimental bacteria

 

 

Strain Taxonomy: Ref- rm#" 5.3.5511; £533.“... £33.. $33.31;"
7 -l
m «n £35233.
Rapid Responders

E. coli y-Proteobacteria 20 R2BV 0.67 i 0.015 6.4 i 0.4 2.9 i 0.1
Enterobacteriaceae R2B-GCS 0.65 i 0.006 4.9 i 0.2 3.4 d: 0.1

HF3 y-Proteobacteria 17 R2BV 0.710 t 0.008 5.7 :1: 0.5 1.8 t 0.1
Pseudomonas PVY/ 10 0.645 t 0.007 3.2 :1: 0.2 2.9 i 0.3

EC4 B-Proteobacteria l9 R2BV 0.546 at 0.005 5.3 i 0.5 3.6 :1: 0.1
Oxalobacteriaceae R2B-GCS 0.482 t 0.003 5.7 i 0.4 3.4 d: 0.2

EC2 Bacteroidetes l9 R2BV 0.431 :1: 0.002 2.4 :1: 0.1 7.1 i 0.6
Sphingobacteriaceae PVY/ 10 0.236 i 0.002 1.5 d: 0.1 12 :1: 1.8

EC5 Actinobacteriaceae l9 R2BV 0.545 :1: 0.004 2.4 i 0.1 5.6 :t 0.2
Arthrobacter PVY/ 10 0.470 2L. 0.008 2.3 i: 0.2 6.2 i 0.3

Slow Responders

S. alaskensis a-Proteobacteria 23,24 R2BV 0.217 t 0.002 1.9 :1: 0.2 9.1 i 0.4
Sphingomonadaceae PVY/ 10 0.130 t 0.002 0.8 i 0.1 16.5 :t 0.7

PX3.14 or-Proteobacteria l8 R2BV 0.144 :t 0.001 2.8 :h 0.1 3.4 :1: 0.1
Rhodospirillaceae PVY/ 10 0.126 t 0.001 2.9 :t 0.1 2.9 d: 0.1

HS5 B-Proteobacteria l7 R2BV 0.081 i 0.001 3.5 i 0.2 4.8 i 0.3
Comamonadaceae R2B-GCS 0.067 i: 0.001 2.3 :1: 0.1 11.2 :t 0.3
LC9 Bacteroidetes 19 R2BV 0.239 :1: 0.003 1.3 i 0.2 12.1 d: 0.4
Sphingobacteriaceae R2B-GCS 0.118 :b 0.005 0.9 :t 0.1 9.7 i 0.6

PX3.15 Actinobacteriaceae 18 RZBV 0.106 :1: 0.001 1.5 i 0.1 2.1 :t 0.2
Mycobacterium PVY/ 10 0.040 2*: 0.003 1.3 :1: 0.1 2.2 i 0.1

 

a. Taxonomic placement of soil isolates to family or genus level is tentative, based on
’ small subunit ribosomal RNA gene sequence analysis.

b. Number of copies of the ribosomal RNA operon per genome from references 18 and

19.

58

Table 2.2: Macromolecular content of experimental bacteria

 

 

S . C I DNA RNA Protein
(fg cell'l) (fg M) (g cell'l) (fg fl‘l) (fg cell'l) (fg fl-l)
Rapid Responders
E. coli RZBV 31 :t 0.7 4.9 :t 0.4 162 :1: 8 26 :1: 2 428 i 33 69 at 8
RZB-GCS 24 i 0.7 4.9 i: 0.2 123 i 7 25 i l 307 :h 33 63 i 8
HF3 R2BV 54 at 2.2 9.7 :t 0.9 249 d: 6 45 d: 3. 582 :t 35 104 :t 8
PVY/lO 16d:3.0 4.9:t:0.9 59i5 19:2 306i9 97d:2
EC4 R2BV 29 :1: 0.9 5.6 :1: 0.5 54 at l 10 :1: 1 330 :1: 26 64 i 8
R2B-GCS 29i3.2 5.3 $0.9 542t9 10i2 366:1:42 663:1]
EC2 R2BV 133:0.3 5.3i0.l 37i1 15d:1 157i8 65i4
PVY/lO lli0.3 7.3105 18i1 12i1 118i3 81i4
EC5 R2BV 5 :1: 0.5 2.0 :1: 0.3 23 i: 3 10 i l 94 :h 9 39 d: 5
PVY/lO 5120.6 2.1i0.2 293:2 13il 130d:6 58i4
Slow Responders
S. alaskensis R2BV 8 i 0.5 4.3 i 0.4 23 :1: l 12 i l 93 :t 4 49 :t 4
PVY/lO 5i0.3 6.63:0.7 921:0.7 12i2 51i2 68i8
PX3.14 RZBV 1521:].0 5.3:t0.6 38:1 13:1 198:1:7 71:5
PVY/lO l7i1.0 6010.4 52:1:1 18:1:1 254i18 893:7
H85 R2BV 11 21:08 3.2i0.4 16$ 1 4710.4 1571.5 46:1:3
R2B-GCS 12 i 0.5 5.0 :1: 0.3 16 3: 1 6.9 i 0.2 92 i 2 39 i 2
LC9 R2BV 9:1:0.2 7.41: 1.6 222tl 19:4 802t1 66:1:13
R2B-GCS 122t0.7 13.2:1:1.5 20i2 21$] 105:1:1 llld:7
PX3.15 R2BV 172t1.2 11510.8 15:1:1 lOtl 155:1:7 101d:5
PVY/10 173:0.5 13.1:t0.4 162t1 12i1 139i8 1081-7

 

59

Chapter 2 Results and Discussion

Figure 2.1 shows a phylogenetic tree of the inferred evolutionary relationships among
strains in our collection and a selection of familiar bacteria. We were successful in
obtaining both rapidly and slowly responding strains from a number of different
taxonomic groups, so that any differences between bacteria pursuing different ecological
strategies in our analyses will not be explained by evolutionary relatedness among
representatives of one strategy to the exclusion of representatives of the alternative
strategy. The collection contains representatives of each strategy among the
Actinobacteria, the Bacteroidetes and the B-Proteobacteria. However, the soil isolates
derived from long-term plating experiments did not include any y—Proteobacteria that met
the criteria for slow responders, nor any or-Proteobacteria that met the criteria for rapid
responders. Nonetheless, given the topology of the phylogenetic tree, we can still be
certain that the 4 pairs of soil isolates represent 4 evolutionary transitions between the

ecological strategies of rapid and slow response.

Taxonomic information and rrn copy number are shown for all strains in Table 2.1, as
well as the specific grth rate, cell volume, culture density and macromolecular content
for each strain in two growth media. We report the macromolecular content in Table 2.2
both per cell and per biovolume. Normalizing to cell or to biovolume, as opposed to

OD420 unit, facilitates comparison with other published work, although it introduces

additional uncertainty by incorporating one or two additional experimental
measurements, respectively. (Since the average cell volume is multiplied by the culture

density to obtain the denominator for expressing macromolecular measurements per cell

60

volume, these values are influenced by errors in estimating both cell volume and culture
density.) Both in Table 2.2 and in subsequent figures, the uncertainty reported for
compound quantities is the standard error of separate estimates of the quantity for each
experimental replicate, not an estimate of uncertainty based on propagating the estimated

uncertainty of the component measurements.

In general terms, the data conform to expectations, with protein the most abundant and
DNA the least abundant of the 3 macromolecules, for 9 of the 10 strains in both grth
media. For Strain PX3.15, protein is still the most abundant macromolecule, but the mass
of DNA is slightly greater than the mass of RNA in both media tested, although the

difference is not statistically significant.

More specifically, our data for E. coli are consistent with the data of Bremer and

Dennis 36, once a correction is made for differences in grth temperature. It is
generally accepted that the composition of the growth medium determines the
macromolecular composition of bacterial cells, while changing the incubation
temperature influences the grth rate, but not cell size or composition 1. Calculating
from the temperature-dependent growth rate data of Farewell and Neidhardt 37, we
estimate the specific growth rates would be accelerated by a factor of 2.33 by shifting
from our grth temperature of 25°C to 37°C. Table 2.3 shows that with this correction,
our measurements of protein and RNA content per cell are similar to those of Bremer and
Dennis 36; however, our measurements of DNA per cell are 55%-90% higher. We note

that despite the conclusion of reference 1 that growth temperature does not influence

61

Table 2.3: Macromolecular content of E. coli

 

macromolecular content per cell

 

specific
Source of Data growth Protein RNA DNA
rate
(hr-1) (fg) (fg) (fg)
Bremer & Dennis 36
1 .73 448 210 18
1 .39 33 8 13 1 14
this study
1.573 428 162 31
1.50 a 307 123 24

 

a. Growth rate adjusted by a factor of 2.33 from actual grth rates at 25°C
to obtain estimated growth rate at 37°C for comparison to data from
reference 36.

62

the chemical composition of bacterial biomass, the data of reference 1 show 35%-5 0%
higher DNA content at 25°C than 37°C for 3 of the 5 media tested, and essentially the
same DNA content for the 2 remaining media. Given the unknown influence of the
different E. coli strains examined, as well as the differences in the growth temperatures
and the methods used for measuring macromolecular content, we are satisfied with the

agreement between our data and the data of reference 36.

The grth rate of a bacterial strain would be expected to influence the time required for
it to form a visible colony, and early or late colony appearance was one of the criteria for
considering bacteria to be rapid or slow responders. Hence, we are not surprised that
growth rates in liquid media differ for bacteria that pursue different ecological strategies.
The fastest growth rate observed among the slowly responding strains (0.24 hr"1 for
strain LC9 in R2BV) is the same as the slowest growth rate observed among the rapidly
responding strains (0.24 hr'1 for strain EC2 in PVY/ 10). The richer medium, R2BV,
supported faster growth rates than the alternative medium for all strains. However, the
difference between the 2 grth rates is greater than 20% for only 1 of the 5 rapidly
responding strains, whereas it is greater than 20% for 4 of the 5 slowly responding
strains. Because only two growth rates were examined for each strain, and these rates did
not differ greatly in half the strains, we are very cautious in interpreting the failure to
observe growth rate-related trends within a strain, particularly among the fast-growing

strains.

63

Even for the strains that grew at quite different growth rates in the two media, the ability
to distinguish grth rate-related trends is limited by the fact that only two different
growth rates are represented in the data. Because the two growth rates for a single strain
were obtained by varying the composition of the medium, strictly speaking, we cannot
interpret differences in the measurements as a function of growth rate per se. Hence, our
analysis focuses on trends expected across strains, not within strains, although we do at
times use t tests to determine whether the measurements made on a single strain in two
media are statistically distinct. Although subsequent figures will show data derived from
both media for each strain, the data from a single strain grown in two media cannot be
considered independent. Hence, for comparisons across all strains or comparisons
between ecological strategies, we will use only the data from each strain grown in R2BV
medium. However, it could also be argued that if unknown errors have influenced
measurements of a particular strain in R2BV for some quantity, an average of the
measurements in both media would be more likely to be representative of the true value
of that quantity for the strain. For all the comparisons we report between strains, the use
of the average value for the two media instead of the value in R2BV would alter

conclusions in only one case, which is noted below.

Microbiologists have long associated increases in grth rate with increases in cell size

(reference 1, and references therein), although exceptions have been noted 2133. Over all

strains, the regression of cell volume on grth rate is significantly positive (p<0.05,

R2=0.54), determined by the fact that the largest cells occupy a narrow range of rapid

grth rates (Figure 2.2). In contrast, the range of growth rates among small cells is

64

Figure 2.2: Biovolume vs. specific growth rate

 

 

7 _
6 — §
\ I
5 .4
7:: 4 4
Q)
E
.2
O
3 3 ‘ AA + Eco/i
E: + HF3
[I] V §/§ + EC4
2 .. é + EC2
—o— £05
9/0 ; —<>— S.alaskensis
1 d —A— PX3.14
V0 —0— H85
—v— LC9
—-o— PX3.15
I I | I I I I I
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

specific growth rate (hr'1)

Figure 2.2. Biovolume plotted as a function of specific growth rate for 5 rapidly
responding bacteria (solid symbols) and 5 slowly responding bacteria (open symbols).
Considering only data from R2BV medium (the medium providing the faster growth rate
for each strain), the regression of biovolume on specific growth rate is significant
(p<0.05, R2 =0.54). Data points and error bars represent the means and standard errors of
measurements on 3-4 independent replicate cultures per strain medium combination.
Error bars smaller than or of comparable size to plot symbols are not shown.

65

much wider, overlapping at its upper end with the slowest grth rates of the largest cells
(compare strains EC4 and EC5). It has been suggested repeatedly that oligotrophic
bacteria may benefit from the larger surface:volume ratio conferred by small size 39:40;
our data are consistent with this expectation in that slow responders are smaller than rapid
responders (t test, p < 0.05, df = 8). However, the pattern we observe with respect to cell
size may have been influenced by our preference for strains that showed dispersed
growth in liquid culture, to permit accurate growth rate measurements via turbidity.
Bacteria with filamentous growth forms may have been excluded because of a tendency
for flocculent growth, potentially excluding slowly growing strains with large cell

volumes.

Considering size variation within each strain, cell volume is significantly larger in the
medium supporting faster growth for 5 of the 10 strains, including E. coli and HF3,
strains that grew in the two media at rates that differed by less than 20%. Strains PX3.15
and LC9 strains grew in the two media at rates that differed by more than a factor of 2,
but did not differ significantly in cell volume. If we examine the 7 strains with either a
significant difference in cell size or at least a 2-fold difference in growth rates, it is clear
that the relationship between cell volume and growth rate is dependent on the identity of
the strain. Strain PX3.15 shows a (nonsigrrificant) 20% increase in cell size over a 250%
increase in growth rate, whereas HF3 displays an 80% increase in cell size for only a
10% increase in growth rate. Hence, while we do observe a trend across strains of cell
volume increasing with growth rate (with the caveat that filamentous bacteria are not

represented), our data suggest that inferences of grth rate based on observations of cell

66

size should be treated cautiously, unless the observations involve a strain for which the

relationship between grth rate and cell size has already been established.

The relationship between cell volume and the total macromolecular content per OD420
unit (MM/OD, the sum of the DNA, RNA and protein measurements per OD420 unit) is

shown in Figure 2.3. Although units of optical density are generally assumed to be
proportional to dry mass, the proportionality constant may in fact differ between strains
and between grth states, particularly if the cells differ in size and shape 41. The largest
components of a bacterial cell that are not DNA, RNA and protein, at least in E. coli, are
components of the cell envelope 42. Hence, a decline in MM/OD with cell size is
expected in cells of approximately uniform shape and similar cell wall structure, since the
envelope components would be expected to comprise a larger fraction of cell mass and
optical density in smaller cells that have a larger surface:volume ratio. Such a trend is
evident in Figure 2.3. However, it is also apparent that while the relationship between
MM/OD and biovolume is similar for all cultures containing large cells, it is highly
variable among cultures containing small cells. This pattern is not explained by cell size
per se, but may be explained by cell shape and architecture. All the large cells are
regular rods from gram negative lineages, whereas the small cells include a variety of
shapes fiom both gram negative and gram positive lineages (Figure 2.4). The influence
of shape on MM/OD may be both direct, in that the shape of a cell may influence its
light-scattering properties independently of its composition and total mass, and indirect,
since different cell shapes imply different ratios of cell wall material to cytoplasm, and

hence different composition.

67

Figure 2.3: MM/OD420 vs. biovolume

 

 

20 —
V i
16 — \
’3 i
0% vb’V—I El
0 12 _
F0 M + + £00”
1.. + HF3
S, Q l-CH + EC4
3 8 —- + EC2
Q -<>— S.alaskensis
2 —A— PX3.14
5 4 _ O —D— H85
I01 —v— LC9
—O— PX3.15
I I I I I I I
1 2 3 4 5 6 7

cell volume (fl)

Figure 2.3. MM/OD420 (the sum of protein, RNA and DNA per OD420 unit) plotted as a
function of biovolume for 5 rapidly responding bacteria (solid symbols) and 5 slowly
responding bacteria (open symbols). Data points and error bars represent the means and
standard errors of measurements on 3-4 replicate cultures per strain-medium
combination, except strain PX3.15, represented by 7-8 replicates. Variability in
MM/OD420 at smaller cell sizes relects greater variation in cell shape and cell wall
architecture. Error bars smaller than or of comparable size to plot symbols are not
shown.

68

.0082: :8 9 0338 E8 £30880? 3 a 08:: :00 .m 05 E 89 0800 05. 08:90 >m~m
8:80 8 80¢ :08» 83 :03? 08:: n H .33 80008 .3308 830038888 88 808: m0 083 06 8 83.008 >m~d 5 008530
Spa .8080 38890:.” =< 8023 008000888 co “033385: 0:00 “00:88:: 6006:: me £88828 “8.280 08am .v.~ 0.5m:—

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

EMXA 004 mmm 3.me 3880\ch .m.
.. ’72}- w . r I .. x: am
I I I.) I s“ \ 1.. \ f o \ . l./ I,
0 \
n. \ s a/AI. ”I; f / \ mix I / \ 90 a -l . II If
\\\ 0” I! I , x . I ... J
- I o I / l I .v i. x
( ‘ ” ‘ I ’ .l \ / ~‘ .‘I\\ \ \ /\
a It all 01.. / r . d T O 2...: it x
'I‘ .l ' \I 0' ’I’ \ J t\ to /
. v a! I -I O. n- 1 1
I 3’ .‘ I .IHI J , Iv.‘ , I 0 . . x i
« O/\ o w/Ifi \0 \\ I l \ x .\ gs I 4 \\I .
III a I I )o’ \ u
v.0. 0 \l. . Io NI \ I .1. u / \ . s .. .
0.8—E83— 32m
wUm NUm Yum mum N30 .M
. n,
I! I \'. \I \ A .l \I I ~/
J \ \
Il \ I . \
. l. I I a ’l . / w
.I n\ 4 II I
x s I . .1
t .. a /
0.8—E83— 395—

0.58.5 880820.80 .3 manauweuomfieueam "Ya 0.53m

69

Strain PX3. 15 has MM/OD values much lower than those of the other strains. In
contrast, the concentration of macromolecules (MM/cell volume) in PX3.15 is in the
middle of the range observed in other strains (data not shown), suggesting that the low
MM/OD values are a result of high optical density, not low macromolecular content.
PX3.15 is a mycobacterial strain, as determined by full-length 16S rRNA sequence
analysis; both the unique lipid-rich cell envelope characteristic of Mycobacterium and the
thicker peptidoglycan layer of gram positive bacteria could contribute to this effect. The
one other gram positive organism in our study, strain EC5, has the second lowest values
of MM/OD, consistent with there being a larger contribution of the cell wall to optical
density in this strain as well. Since the largest cells in our collection are fairly similar in
shape and share the gram negative architecture of the Proteobacteria, we do not know
whether variability in cell shape and cell wall architecture could influence MM/OD as

dramatically among large cells as among small cells.

Strains EC2, EC5 and H85 display significantly lower MM/OD in the medium which
supports larger cells (R2BV in all cases), a change opposite in sign to both the trend
observed in most other strains and the trend observed across all strains. The change in
biovolume with growth rate is particularly steep in strain HSS (Figure 2.2), which also
shows the greatest decline in MM/OD between the alternative medium and R2BV. These
observations could be related if the nature of the carbon substrates in R2BV induces the
formation or expansion of an extracellular polysaccharide layer that contributes to cell
volume. RZBV contains glucose, casamino acids and soluble starch not found in either

alternative medium.

70

Because of the variability between strains in the extent to which material other than
DNA, RNA and protein contribute to optical density, the mass ratios between
macromolecules may be the best way to investigate trends in macromolecular content
across strains. In E. coli, the mass of all 3 macromolecules increases with grth rate
(leading to larger cells at faster growth rates), but the increase of RNA mass is the
greatest, and the increase of DNA mass is the least for a given increase in growth

rate 1’36. Hence, the RNA:DNA ratio increases with growth rate, and is the
macromolecular ratio most sensitive to grth rate. Because of this sensitivity and the
conserved roles of these molecules, the RNA:DNA ratio has been suggested as a proxy
measurement for the growth rates of bacteria in the environment 43. A 1995 review of
the data available from laboratory-adapted enteric bacteria and several environmental
strains found a positive relationship between the RNA:DNA ratio and growth rate across
taxa and within taxa, but considerable variability was evident in comparisons of the
relationships for individual taxa 44. In particular, the data showed that low RNA:DNA
ratios were obtained over a wide range of slow growth rates, whereas the RNA:DNA

ratio was more sensitive to growth rate during faster growth.

The relationship between the RNA:DNA ratio and growth rate derived from our data is

shown for all strain-medium combinations in Figure 2.5 (p<0.01, R2=O.64, RZBV data).
Our data are similar to those in the literature. Specifically, a wide range of grth rates

generates RNA:DNA values less than 3, and over this range the identity of the strain is a

71

better predictor of the RNA:DNA ratio in a culture than the rate at which the culture is
growing. We have no satisfactory explanation for the unusual RNA:DNA values of
strain EC5. The apparently high value of 6.5 obtained in PVY/ 10 medium at a growth
rate of 0.47 hr‘1 is near the high end of the range of values reported in the literature at
similar growth rates 44, but the steep drop in the ratio with an increase in grth rate is

not consistent with other observations.

The protein:DNA ratio in E. coli is the macromolecular ratio least sensitive to changes in
growth rate, increasing by less than a factor of 2 over a roughly 4-fold increase in growth
rate 36. The plot of the protein:DNA ratio vs. growth rate derived from our data is shown
in Figure 2.6; the slight rising trend is not significant. Once again, both the high
protein:DNA ratio obtained from strain EC5 growing in PVY/ 10 and the steep decline in
the ratio with an increase in growth rate are unusual. The unusual behavior of EC5 in
terms of both the RNA:DNA ratio and the protein:DNA ratio could be explained if the
DNA value obtained from biomass grown in PVY/ 10 medium were low due to some

unrecognized experimental artifact.

The protein:RNA ratio in E. coli declines nearly 2.5-fold over a 4-fold increase in growth
rate 35. This trend is identified with the need for more ribosomes per protein mass in a
fast-growing cell, in order to replicate cell protein more quickly. Figure 2.7 shows the
protein:RNA ratio as a function of growth rate for our data. Across all strains, the
protein:RNA ratio obtained in RZBV medium shows a significant negative trend with

growth rate (p<0.05, R2=O.48). However, if the protein:RNA ratio for each strain is

72

estimated as the average of the values obtained in the two media, the negative trend is not
quite significant at the 95% confidence level (p=0.0575). The product of grth rate and
the protein:RNA ratio is a measure of translational power, as shown in Chapter 1. An
exploration of the differences in translational power between the rapidly responding and

slowly responding bacteria is the subject of Chapter 3.

73

Figure 2.5: RNA:DNA ratio vs. specific growth rate

5- +

° A

4 — |
+ Eco/i
é V + HF3
+ EC4

é/ V + EC2
2 — é__- + ECS
d3 —<>— S.alaskensis

RNA:DNA

 

 

—-A— PX3.14
1 — O\O —D— H85
—v— L09
—0— PX3.15
l l T I l l l l
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

specific growth rate (hr'1)

Figure 2.5. RNA:DNA ratio plotted as a function of specific growth rate for 5 rapidly
responding bacteria (solid symbols) and 5 slowly responding bacteria (open symbols).
Considering only data from RZBV medium (the medium providing the faster growth rate
for each strain), the regression of RNA:DNA on specific growth rate is significant
(p<0.01, R2 =0.64). Data points and error bars represent the means and standard errors of
measurements on 3-4 replicate cultures per strain-medium combination, except strain
PX3.15, represented by 6-7 replicates. Error bars smaller than or of comparable size to
plot symbols are not shown.

74

Figure 2.6: Protein:DNA ratio vs. specific growth rate

 

 

 

 

+ Eco/i
+ HF3
__ —I— EC4
30 _ + EC2
+ EC5
C —<>— S.alaskensis
25 _ —-Ar— PX3.14
‘_ —0— H85
—V— L09
—'0— PX3.15
20 -— .
<
2
g }
c
E 15 — A
e ‘ ,
10 — @70/ A
o a V
5 _
I I I I I I I I
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

specific growth rate (hr'l)

Figure 2.6. Protein:DNA ratio plotted as a fimction of specific growth rate for 5 rapidly
responding bacteria (solid symbols) and 5 slowly responding bacteria (open symbols).
Data points and error bars represent the means and standard errors of measurements on 3-
4 replicate cultures per strain-medium combination, except strain PX3.15, represented by

7-8 replicates. Error bars smaller than or of comparable size to plot symbols are not
shown.

75

Figure 2.7: Protein:RNA ratio vs. specific growth rate

14

T + Eco/i

+ HF3

12 _ + EC4
+ E02
+ E05
—<>— S.alaskensis
—A— PX3.14

§ " —CJ— HS5
—v— L09

3 _ —O— PX3.15

i N;

._ {I——-§

10—

 

ProteianNA

§‘§‘ A

 

 

I I I I I I I I
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

specific growth rate (hr'l)

Figure 2.7. ProteianNA ratio plotted as a fimction of specific growth rate for 5 rapidly
responding bacteria (solid symbols) and 5 slowly responding bacteria (open symbols).
Considering only data from R2BV medium (the medium providing the faster growth rate
for each strain), the regression of protein:RNA on specific growth rate is significant
(p<0.05, R2 =0.48). Data points and error bars represent the means and standard errors of
measurements on 3-4 replicate cultures per strain-medium combination, except strain
PX3.15, represented by 7 replicates in each medium. Error bars smaller than or of
comparable size to plot symbols are not shown.

76

10.

11.

12.

13.

Chapter 2 References

Schaechter, M., O. Maaloe and ND. Kj eldgaard, 1958. Dependency on medium
and temperature of cell size and chemical composition during balanced growth of
Salmonella typhimurium. Journal of General Microbiology, 19(3):592-606.

Kjeldgaard, NO, 0. Maaloe and M. Schaechter, 1958. The transition between
diflerent physiological states during balanced growth of Salmonella typhimurium.
Journal of General Microbiology, 19:607-616.

Maaloe, O. and NO. Kjeldgaard, 1966. Control of Macromolecular Synthesis.
Microbial and Molecular Biology Series, ed. B.D. Davis. W.A. Benjamin, Inc.,
New York.

Cooper, S., 1993. The origins and meaning of the Schaechter-Maaloe—Kjeldgaard
experiments. Journal of General Microbiology, 139:1117-1124.

Ingraham, J .L., O. Maaloe and RC. Neidhardt, 1983. Growth of the Bacterial
Cell. Sinauer, Sunderland, Massachusetts.

Neidhardt, F.C., J .L. Ingraham and M. Schaechter, 1990. Physiology of the
Bacterial Cell. Sinauer, Sunderland, Massachusetts.

Reed, J .L. and B0. Palsson, 2003. Thirteen years of building constraint-based in
silico models of Escherichia coli. Journal of Bacteriology, 185(9):2692-2699.

Selinger, D.W., M.A. Wright and GM. Church, 2003. On the complete
determination of biological systems. Trends in Biotechnology, 21(6):251-254.

Cho, BC. and F. Azam, 1990. Biogeochemical significance of bacterial biomass
in the oceans euphotic zone. Marine Ecology-Progress Series, 63(2-3):253-259.

Ducklow, H.W. and CA. Carlson, 1992. Oceanic bacterial production. Advances
in Microbial Ecology, 12:1 13-181.

Whitman, W.B., D.C. Coleman and W.J. Wiebe, 1998. Prokaryotes: The unseen
majority. Proceedings of the National Academy of Sciences of the United States
of America, 95(12):6578-6583.

Makino, W., J .B. Cotner, R.W. Sterner and J .J . Elser, 2003. Are bacteria more
like plants or animals? Growth rate and resource dependence of bacterial C : N :

P stoichiometry. Functional Ecology, 17(1): 121-130.

Chant, J ., I. Hui, D. Dejongwong, L. Shimmin and RP. Dennis, 1986. The protein
synthesizing machinery of the Archaebacterium Halobacterium cutirubrum -
molecular characterization. Systematic and Applied Microbiology, 7(1): 1 06-1 14.

77

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

Pang, H.L. and H.H. Winkler, 1994. The concentrations of stable RNA and
ribosomes in Rickettsia prowazekii. Molecular Microbiology, 12(1):115-120.

Shahab, N., F. Flett, S.G. Oliver and RR. Butler, 1996. Growth rate control of
protein and nucleic acid content in Streptomyces coelicolor A3 (2) and
Escherichia coli B/r. Microbiology-UK, 142:1927-1935.

Fegatella, F ., J. Lim, S. Kjelleberg and R. Cavicchioli, 1998. Implications of
rRNA operon copy number and ribosome content in the marine oligotrophic

ultramicrobacterium Sphingomonas sp. strain R3225 6. Applied and
Environmental Microbiology, 64(11):4433-4438.

Gorlach, K., R. Shingaki, H. Morisaki and T. Hattori, 1994. Construction of eco-
collection of paddy field soil bacteria for population analysis. Journal of General
and Applied Microbiology, 40(6):509-517.

Klappenbach, J .A., 2001. Ecological and evolutionary implications of ribosomal
RNA gene copy number in heterotrophic soil bacteria, Ph.D. Dissertation,
Department of Microbiology and Molecular Genetics, Michigan State University.
East Lansing.

Klappenbach, J .A., J .M. Dunbar and T.M. Schmidt, 2000. rRNA operon copy
number reflects ecological strategies of bacteria. Applied and Environmental
Microbiology, 66(4): 1328-1333.

Lenski, R.E., M.R. Rose, S.C. Simpson and SC. Tadler, 1991. Long term
experimental evolution in Escherichia coli. 1. Adaptation and divergence during
2, 000 generations. American Naturalist, 138(6): 13 15-1341.

Eguchi, M., T. Nishikawa, K. MacDonald, R. Cavicchioli, J .C. Gottschal and S.
Kjelleberg, 1996. Responses to stress and nutrient availability by the marine

ultramicrobacterium Sphingomonas sp. strain R3225 6. Applied and
Environmental Microbiology, 62(4): 1287-1294.

Ostrowski, M., R. Cavicchioli, M. Blaauw and J .C. Gottschal, 2001. Speczfic
growth rate plays a critical role in hydrogen peroxide resistance of the marine
oligotrophic ultramicrobacterium Sphingomonas alaskensis strain R8225 6.
Applied and Environmental Microbiology, 67(3):1292-1299.

Schut, F., E]. Devries, J .C. Gottschal, B.R. Robertson, W. Harder, R.A. Prins and
BK. Button, 1993. Isolation of typical marine bacteria by dilution culture -

growth, maintenance, and characteristics of isolates under laboratory conditions.
Applied and Environmental Microbiology, 59(7):2150-2160.

Schut, F., J .C. Gottschal and RA. Prins, 1997. Isolation and characterisation of

the marine ultramicrobacterium Sphingomonas sp. strain R3225 6. FEMS
Microbiology Reviews, 20(3-4):363-369.

78

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

Singer, V.L., L.J. Jones, S.T. Yue and RP. Haugland, 1997 . Characterization of
PicoGreen reagent and development of a fluorescence-based solution assay for
double-stranded DNA quantitation. Analytical Biochemistry, 249(2):228-238.

Jones, L.J., S.T. Yue, C.Y. Cheung and V.L. Singer, 1998. RNA quantitation by
fluorescence-based solution assay: RiboGreen reagent characterization.
Analytical Biochemistry, 265(2):368-374.

Smith, P.K., R.I. Krohn, G.T. Hermanson, A.K. Mallia, F.H. Gartner, M.D.
Provenzano, E.K. Fujimoto, N.M. Goeke, B.J. Olson and DC. Klenk, 1985.

Measurement of protein using bicinchoninic acid. Analytical Biochemistry,
150(1):76-85.

Daniels, L., R.S. Hanson and J.A. Phillips, 1994. Chemical analysis; in Methods
for General and Molecular Bacteriology; P. Gerhardt, R.G.E. Murray, W.A.
Wood and NR. Krieg, eds. ASM Press, Washington, DC.

Kepner, R.L. and J .R. Pratt, 1994. Use of fluorochromes for direct enumeration of
total bacteria in environmental samples - past and present. Microbiological
Reviews, 58(4):603-615.

Kirchman, D., J. Sigda, R. Kapuscinski and R. Mitchell, 1982. Statistical analysis
of the direct count method for enumerating bacteria. Applied and Environmental
Microbiology, 44(2):376-382.

Liu, J ., F.B. Dazzo, O. Glagoleva, B. Yu and AK. Jain, 2001. CMEIAS: A
computer-aided system for the image analysis of bacterial morphotypes in
microbial communities. Microbial Ecology, 41(3): 173-194.

Ludwig, W., O. Strunk, R. Westram, L. Richter, H. Meier, Yadhukumar, A.
Buchner, T. Lai, S. Steppi, G. J obb, W. Forster, I. Brettske, S. Gerber, A.W.
Ginhart, 0. Gross, S. Grumann, S. Hermann, R. Jost, A. Konig, T. Liss, R.
Lussmann, M. May, B. Nonhoff, B. Reichel, R. Strehlow, A. Starnatakis, N.
Stuckrnann, A. Vilbig, M. Lenke, T. Ludwig, A. Bode and K.-H. Schleifer, 2004.
ARB: a software environment for sequence data. Nucl. Acids. Res., 32(4):]363-
1371.

Cole, J.R., B. Chai, T.L. Marsh, R.J. Farris, Q. Wang, S.A. Kulam, S. Chandra,
D.M. McGarrell, T.M. Schmidt, G.M. Gan'ity and J .M. Tiedje, 2003. The
Ribosomal Database Project (RDP-II): Previewing a new autoaligner that allows
regular updates and the new prokaryotic taxonomy. Nucleic Acids Research,

31(1):442-443.

Cannone, J ., S. Subramanian, M. Schnare, J. Collett, L. D'Souza, Y. Du, B. Feng,
N. Lin, L. Madabusi, K. Muller, N. Pande, Z. Shang, N. Yu and R. Gutell, 2002.
The Comparative RNA Web (CRVlO Site: An online database of comparative
sequence and structure information for ribosomal, intron, and other RNAs. BMC
Bioinformatics, 3(1):2.

79

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

Klappenbach, J .A., P.R. Saxman, J .R. Cole and T.M. Schmidt, 2001. rrndb: the
Ribosomal RNA Operon Copy Number Database. Nucleic Acids Research,
29(1):181-184.

Bremer, H. and RP. Dennis, 1996. Modulation of chemical composition and other
parameters of the cell by growth rate; pp. 1553-1569 in Escherichia coli and
Salmonella: Cellular and Molecular Biology; F.C. Neidhardt, et al., eds. ASM
Press, Washington, D. C.

Farewell, A. and RC. Neidhardt, 1998. Effect of temperature on in vivo protein
synthetic capacity in Escherichia coli. Journal of Bacteriology, 180(17):4704-
471 0.

Pemthaler, J ., T. Posch, K. Simek, J. Vrba, R. Amann and R. Psenner, 1997.
Contrasting bacterial strategies to coexist with a flagellate predator in an
experimental microbial assemblage. Applied and Environmental Microbiology,

63(2):596-60l.

Hirsch, P., M. Bernhard, S.S. Cohen, J .C. Ensign, H.W. J annasch, A.L. Koch,
K.C. Marshall, A. Matin, J .S. Poindexter, S.C. Rittenberg, D.C. Smith and H.
Veldkamp, 1979. Life under conditions of low nutrient concentrations group

report; pp. 357-3 72 in Strategies of Microbial Life in Extreme Environments:
report of the Dahlem Workshop; M. Shilo, ed. Verlag Chemie, Weinhein.

Button, D.K. and B. Robertson, 2000. Eflect of nutrient kinetics and
cytoarchitecture on bacterioplankter size. Limnology and Oceanography,
45(2):499-505.

Koch, A.L., 1994. Growth measurement; in Methods for General and Molecular
Bacteriology; P. Gerhardt, R.G.E. Murray, W.A. Wood and NR. Krieg, eds.
ASM Press, Washington, DC.

Neidhardt, PC. and HE. Umbarger, 1996. Chemical composition of Escherichia
coli; pp. 13-16 in Escherichia coli and Salmonella Cellular and Molecular
Biology; F.C. Neidhardt, et al., eds. ASM Press, Washington, D. C.

Dortch, Q., T.L. Roberts, J .R. Clayton, Jr. and SJ. Ahmed, 1983. RNA/DNA
ratios and DNA concentrations as indicators of growth rate and biomass in
planktonic marine organisms. Marine Ecology-Progress Series, 13:61-71.

Kemp, RR, 1995. Can we estimate bacterial growth rates from ribosomal RNA
content? ; pp. 279-3 02 in Molecular Ecology of Aquatic Microbes; 1. Joint, ed.
Springer-Verlag, Berlin.

80

Chapter 3. Translational power and the strength of
translational selection vary with ecological strategy in bacteria

Chapter 3 Abstract

All forms of life synthesize protein using a similar process, carried out by similar
molecules, but little is known about how translational performance may differ between
organisms. Phylogenetically diverse bacteria with contrasting ecological characteristics
were used to assess whether variation in translational performance is correlated with
differences in the selective regime experienced by the organism. Bacteria capable of
rapid growth in response to abundant external resources were assumed to pursue a
strategy of exploiting large fluctuations of resource availability. Bacteria incapable of a
rapid growth response were assumed to pursue alternative strategies such as exploiting
conditions of low resource availability or maximizing starvation survival. Two lines of
evidence support a difference in translational performance between rapidly responding
bacteria and slowly responding bacteria. First, translational power, the protein output per
biomass invested in the translational apparatus, was found to differ between rapidly
responding bacteria and slowly responding bacteria. Second, the strength of selection for
translational power experienced by a bacterial species was greater for strains with a
higher number of ribosomal RNA operons per genome, which measures the extent to
which a strain is adapted for a rapid response strategy. These results are consistent with
the existence of an evolutionary trade-off between translational power and translational

yield, the amount of protein synthesized per energy consumed.

81

Chapter 3 Introduction

The activity of a growing bacterial cell is dominated by protein synthesis. The
polymerization of amino acids accounts for 2/3 of total energy expenditure in Escherichia
coli growing on glucose 1, and over half the cell’s dry mass can be devoted to the protein
synthesis system 2. Hence, even small improvements in the operation of the protein
synthesis system — decreasing the energy cost of translation, or increasing the rate of
protein production from the biomass invested in the translational apparatus — may have
significant effects on bacterial fitness. However, despite recent progress in understanding
ribosomal structure 3 and the sequence of events involved in translation 4, interspecies
variation in translational performance has received little experimental attention. This
lack of attention may result from an assumption that since the translational apparatus is

highly conserved, the translational apparatus performs similarly in all organisms.

Since protein synthesis is both essential and ancient, it might be argued that there has
been ample opportunity for natural selection to optimize translational performance.
However, analysis of macromolecular composition data from the scientific literature
indicates that microbes differ in translational power, a measure that reflects both the rate
of translating ribosomes and the fraction of ribosomes that are active (Chapter 1).
Furthermore, the ribosomes of a number of natural E. coli strains show a wide range of
translation rates and missense error rates, and these traits respond rapidly to the selection
for faster grth experienced by the strains in the laboratory environment 5,6. In E. coli,

the evolution of the protein synthesis system is influenced by profound connections

between the rate of translation and the frequency of translational errors 7'13. Hence, if

82

the relative importance of rapid translation and error-free translation depends on the
environment, natural selection may have led to differences in the performance of the

translational apparatus between organisms.

Errors of translation include both inaccurate selection of arninoacyl-tRNAs (missense
errors) and premature translational termination (processivity errors). Processivity errors
are particularly costly, because energy is spent and ribosomes are occupied in the
production of truncated, nonfunctional polypeptides that are subsequently degraded 8. In
contrast, most missense errors result in a protein that retains most of the activity of the
canonical version 3. Nonetheless, translational processivity errors are not rare in
laboratory-adapted strains of E. coli, the only organism for which data are available. An
early experiment demonstrated that the products of processivity errors represent about
25% by mass of all protein synthesized from the IacZ gene 14; other experiments with
lacZ have resulted in similar estimates 10,11. The majority of processivity errors in E.
coli can be attributed to dropoff, the loss of peptidyl-tRNA from the ribosome 8’15. The
average likelihood of a processivity error is about 2.5 X 10'4 per codon; at this rate about
10% of the ribosomes that initiate translation of an average-length gene would fail to
synthesize a full-length protein. Processivity errors increase the energy expended for
protein synthesis. The energy of 4 phosphoanhydride bonds is required to form each
peptide bond — 2 for activating an amino acid for attachment to a tRNA molecule, and 2
for adding an amino acid to the growing polypeptide during translation. While the amino
acids in a truncated polypeptide are recycled via proteolysis, the energy spent to form the

peptide bonds is lost. We define translational yield as the mass of (functional) protein

83

produced per energy consumed. Processivity errors, by wasting energy in the synthesis

of nonfunctional, truncated polypeptides, decrease translational yield.

On the basis of a postulated evolutionary tradeoff between translational power and
translational yield, we predicted that translational power would be higher in bacteria
adapted to exploit large fluctuations of resource availability through rapid growth, and
lower in bacteria adapted to exploit small fluctuations of resource availability, and to
maximize starvation survival. The bacterial soil isolates we have identified as rapid
responders (Chapters 1 & 2) clearly would be expected to have high translational power.
We have fewer reasons to identify the slow responders with high yield strategies, because
these strains have been classified by the absence of a rapid response, rather than by
positive identification of strategy-revealing traits such as low resource thresholds for
grth or viability. Nonetheless, because low rrn copy number limits the capacity of
these strains to make a rapid growth response, we are confident that these strains are not
simply cryptic rapid responders, but instead are pursuing a different ecological strategy.
On the basis of the proposed benefit of low rrn copy number in perpetually resource-
limited habitats and the observed tradeoff between rapid growth and starvation survival in

E. coli, we predicted that the slow responders would have low translational power.

Codon usage analysis was used to develop a second line of evidence supporting a
relationship between translational performance and the ecological strategies of bacteria.
Codon bias refers to the use of some codons significantly more fiequently than

synonymous alternatives; the preferred codons are translated more quickly, and with

84

fewer errors 16:17. Selection for translational power has been invoked to explain a
common pattern in codon bias among the genes of an organism, that the degree of bias is
correlated with the expression level of the gene. Hence, codon usage analysis can be
used to infer the strength of selection for translational power experienced by an organism.
Because a high number of rrn operons per genome is adaptive for a rapid response
strategy 13,19, we predicted that the strength of selection for translational power would be

correlated with rrn copy number.

85

Chapter 3 Materials and Methods

Translational power

Bacterial strains and laboratory techniques used to obtain growth rate and
macromolecular composition data are reported in Chapter 2. These data are the basis of

the comparisons of translational power described in this chapter.

Codon usage analysis

The strength of translational selection was measured by comparing codon usage in a set
of highly expressed (HE) genes to codon usage in the genome as a whole, using 76
bacteria for which the complete genome sequence is available. A complete list of the
genomes analyzed and a phylogenetic tree depicting a hypothesis of their evolutionary
history is presented in the Appendix. The 8 genes comprising the HE set were chosen
because they are known to be highly expressed and to have high codon bias in E. coli 20.
The HE set includes genes for 5 ribosomal proteins rpsA/Sl, rpsB/SZ, rpsI/S9, rplA/Ll
and rle/L13, and 3 elongation factors tuflEF-Tu, tsflEF-Ts and fus/EF -G. We chose not
to include highly expressed E. coli genes in other functional categories (e. g., outer
membrane protein ompA or heat shock protein groEL) in the HE set because we were
more confident that homologs of translation apparatus genes would be identifiable and
would have retained high levels of expression in all organisms. The number of ribosomal
protein and elongation factor genes in the HE set was chosen so that a sufficient number
of codons were included in each subcategory for reliable independent estimations of

codon frequencies. All the results based on the complete HE set are essentially

86

unchanged if the analysis is done using either the ribosomal protein gene subset or the

elongation factor gene subset.

Homologs of the HE set were identified in each bacterial genome via the ‘Name Search’
function of the Comprehensive Microbial Resource (CMR) 21, a website hosted by The
Institute for Genomic Research (www.tigr.org). All homologs of a gene in a single
genome were included in our analysis (i.e., multiple copies of elongation factor genes
were included), as long as protein parameters (length, pI, etc.) were similar between
multiple homologs in a genome. The number of instances of each codon in each gene of
the HE set were downloaded from the CMR website and summed to calculate the
observed relative codon frequencies in the HE set of each genome. Codon data tabulated
for all predicted genes in each genome were downloaded directly fiom the CMR and used
to calculated the expected relative codon fi'equencies for comparison with the HE set..
Procedures to handle unreliable relative codon frequency estimates when amino acids
were observed fewer than 5 times were as suggested by Novembre 22, although the

number of codons included in each dataset was sufficient that such corrections were rare.

The comparison between codon use in the HE set and in the genome as a whole was done
via Novembre’s N'c 22, a generalization of the effective number of codons, Nc, an index
of codon bias proposed by Wright 23. Nc makes an implicit comparison of observed
relative codon frequencies to uniform usage of synonymous codons, reported as an
effective number of codons between 61, the total number of sense codons, and 20, the

minimum number of codons representing all amino acids. A value of 61 indicates no bias

87

(all synonymous alternative codons used at the same fiequency), whereas 20 indicates the

maximum possible bias (only a single codon used for each amino acid).

Instead of making a comparison only to the special case of uniform synonymous codon
use, Novembre’s N'c measures the deviation of the relative codon frequencies observed in
a particular set of genes relative to any explicitly specified expected codon frequency
distribution, e. g., the frequencies with which codons are used in the genome as a whole.
It is based on a x2 comparison of observed and expected frequencies over the codons of
an amino acid. Low x2 values (meaning observed and expected frequencies are similar)
are then transformed to a value equal or close to the total number of codons available for
that amino acid, whereas high 12 values (meaning observed and expected frequencies
differ greatly) are transformed to a value approaching 1. The sum of these values over all
amino acids is N'c for a sequence, which ranges from 61 to 20, as does Wright’s Nc.
However, the scale is not interpreted precisely as an effective number of codons. Instead,
the value of N'c is maximized if the observed sequence makes complete use all the
degrees of coding freedom represented by the expected codon frequencies, even if the
expected frequencies themselves reflect preferential use of only a subset of all sense
codons. If the expected codon frequency distribution is completely unbiased (i.e.,
uniform usage of synonymous codons), N'c reduces to Nc. Neither of these measures is
biased by differences in amino acid composition or gene length, and the increase in
variability of both indices when estimated over short sequences is relatively mild
compared to other indices of codon bias 22,24. Our data are reported using a

transformation of Novembre’s N'c: AN'c = 61 — N'c. We have made this transformation

88

to obtain an index which equals zero when there is no difference in codon use between
the HE set and the genome as a whole, and increases in magnitude as the codon use of the
HE set is increasingly biased relative to codon use over the genome as a whole. Hence,

AN'c is positively correlated with the strength of translational selection.

In addition to calculating AN 'c, which depends on the disparity of codon bias between the
HE set and the genome, we wanted to assess relationship of rrn copy number to the
codon bias of each component separately. We measured the codon bias of the HE set and

the genome separately by calculating Wright’s Nc for each.

The total number of identified tRN A genes in each sequenced genome were obtained

from the CMR website 21, with pseudogenes and selenocysteine tRNA genes excluded.

Comparative analyses via log likelihood ratios

We make comparisons of rrn copy number to 4 different traits across many bacterial
genomes. The four traits are AN 'c for a comparison of the HE set to the genome,
Wright’s No for both the HE set and the genome, and tRNA gene copy number. To test
the significance of an association between rrn copy number and any of these traits, we
used a maximum likelihood log ratio method described by Page] 25 and implemented in
the Continuous software program available at www.ams.rdg.ac.uk/zoology/page1/. The
method calculates the model of evolutionary change of measured traits over a specified
phylogeny that is most likely to have resulted in the observed distribution of trait values

in extant taxa. The likelihood of the best (most likely) evolutionary model representing

89

an experimental hypothesis (e.g., correlation exists between 2 trait values) is compared to
the likelihood of the best evolutionary model representing a null hypothesis (e. g., trait
values are uncorrelated) to calculate the probability (p value) that the null hypothesis

explains the observed distribution of traits as well as the experimental hypothesis.

The explicit phylogenetic information used in the log ratio tests was based on a
phylogeny inferred by a maximum likelihood algorithm 26 from 16S rRNA gene
sequences (see Appendix). To ensure that poorly resolved details of the bacterial
phylogeny did not influence the results, each comparative analysis using the Continuous
software was repeated with 2 different descriptions of bacterial phylogeny: a tree inferred
by maximum likelihood analysis of 16S rRNA sequence data with unmodified internal
branch lengths, and a tree derived from the previous tree in which short internal branches

were collapsed.

Each test for a correlation between two traits using Continuous software followed the
same sequence. First, maximum likelihood estimations of the scaling parameters K, 6,
and A were made with the two traits to be tested included in the model. These scaling
parameters describe how the evolutionary model for the traits fits to the tree topology;
e.g., the extent to which trait evolution is gradual or punctuated, or the extent to which it
has accelerated or decelerated over time (see reference 25 and software documentation
for more details). The scaling parameter values specific to each model were used during
subsequent hypothesis tests. Second, separate models were estimated assuming either

nondirectional (random walk) or directional models of trait evolution, and compared with

90

a log likelihood ratio test; in no case could a nondirectional model be rejected at the 95%
confidence level. Third, the likelihood of the nondirectional model was calculated with
the strength of the correlation between the two traits either constrained to be zero (null
hypothesis), or estimated by maximum likelihood (experimental hypothesis). The
reported significance (p value) for each correlation between two traits represents the log
likelihood ratio these two competing evolutionary models. In other words, it represents
the likelihood that evolution over the specified phylogeny would have generated the
observed trait distribution by chance, if the two traits were unrelated. If the correlation is
significant, the reported coefficient of determination (R2 value) is the square of the

maximum likelihood estimate of the correlation coefficient between the two traits.

91

Chapter 3 Results

Obtaining bacteria with contrasting ecological strategies

Bacteria were tentatively considered to be rapid responders if they were isolated from
colonies first visible 1-2 days after plating a dilute soil suspension on solid media;
isolates from colonies that were first visible after 7 or more days of incubation were
tentatively considered to be slow responders. Several lines of evidence support these
designations. First, the trait of early or late colony appearance is displayed consistently
during subsequent grth of the strains as purified isolates on various media. If delayed
colony appearance during the initial plating experiment were due to variation in the
physiological state of the individual cells when first plated, as opposed to heritable
variation between strains, such consistency would not be maintained. The consistency of
colony appearance time also indicates that late colony formation is not explained by
grth inhibition from co-occurring strains or particular components of the medium.
Second, the identities of the early and late appearing strains are distinct. Although a
single phylogenetic group may contain both rapid and slow responders, the same strain,
as determined by small subunit ribosomal RNA gene sequence, was never isolated from
both early and late appearing colonies. Finally, the number of ribosomal RNA (rrn)
operons per genome, which ranges from 1-15 in bacteria 27, is distinct between the
isolates from early and late appearing colonies. Early colony forming strains from soil
have an average of 5.5 rrn operons per genome, significantly more than the average of
1.4 among the late colony forming strains 18. High rrn copy number is thought to be an

adaptation for conditions of fluctuating resource availability, by permitting both fast

92

grth and rapid acceleration from non-grth to maximal growth rates 18,23. Low rrn

copy number, on the other hand, may be an adaptation for growth in stable conditions of

low nutrient availability 19. Hence, we considered an rrn copy number of 4 or higher as
confirmation of a rapid response strategy, whereas a copy number of 2 or lower

confirmed that a strain pursued an alternative strategy.

To ensure that the comparison of ecological strategies was not confounded by shared
evolutionary history, we sought related pairs of soil isolates, including both a rapid
responder and a slow responder, within phylogenetically diverse taxa. Rapid responder-
slow responder pairs were found among the Actinobacteria, the Bacteroidetes, and the B-
Proteobacteria. A fourth pair included a rapidly responding isolate from the y-
Proteobacteria and a slowly responding isolate from the a-Proteobacteria. In addition to
these 8 recent soil isolates, we tested a well-characterized exemplar of each strategy, E.
coli (y-Proteobacteria) representing rapid response, and the ultraoligotrophic marine
bacterium Sphingopyxis alaskensis RB2256 (on-Proteobacteria, originally Sphingomonas

alaskensis) representing slow response 29:30. The phylogeny and rrn copy number of

these strains are depicted in Figure 2.1.

Comparison of translational power

We compared the translational power of the 10 bacterial strains in our collection using
the metric introduced by Kj eldgaard and Kurland 31 (originally called ‘ribosome

efficiency’, see Chapter 1 for our rationale for the new terminology):

93

translational power = ER}: ,

where ,u is the specific grth rate and P and R are the masses of protein and RNA in a
cell or culture (Chapter 1). Using the macromolecular data reported in Chapter 2, we find
that the ranges of translational power for organisms with contrasting ecological strategies
do not overlap (Figure 3.1). The highest translational power among the slowly
responding bacteria is not as high as the lowest translational power among the rapidly
responding bacteria. Considering data from only the R2BV medium, the average
translational power among the rapidly responding bacteria was 2.2 :1: 0.31 hr'l,
significantly higher than the average translational power among the slowly responding
bacteria of 0.9 :i: 0.07 hr'l, as predicted (mean i se, p < 0.002 by one-tailed t-test, df=8).
The conclusion is unchanged if the translational power of each strain is estimated as the
average value over both media, or if the data fiom both media are included as separate
data points. Our observation that the average translational power of the slowly
responding bacteria in R2BV medium is 41% of the average translational power of the
rapidly responding bacteria in the same medium falls at one end of the range of
comparisons based on data in the literature between various slowly-growing strains and
E. coli, which ranged from <l7% - 42% (Chapter 1). Note, however, that making this
comparison between strains in a consistent environment does not control for different

levels of investment in the translational apparatus between strains.

A more detailed presentation of the data is made in Figure 3.2, which plots translational

output against RNA content for each strain in both media. On this plot, translational

94

Figure 3.1: Translational power of rapidly responding and
slowly responding bacteria

33

translational power pP F1" (hr'1)

 

 

 

 

I WW 7

 

Eco/i HF3 EC4 E02 EC5 IS.ala. sz.14 H85 LC9 PX3.15
I
rapid responders . slow responders

Figure 3.1. Translational power of rapidly responding bacteria (solid bars) and slowly
responding bacteria (open bars) grown in two different media, based on data presented in
Chapter 2. The bar on the left represents data from growth in R2BV medium for each
strain; the bar on the right represents either R2B-GCS (E. coli, EC4, HSS, LC9) or

PVY/ 10 (all remaining strains). Media are described in Chapter 2. The height of each
bar represents the average of measurements on at least 3 independent cultures of a strain-
medium combination; error bars are standard errors of the mean. Based only on data
generated in R2BV, translational power is significantly higher among rapid responders
than slow responders (p<0.002 by one-tailed t test, df=8).

95

power is the ratio of the ordinate to the abscissa, so any straight line passing through the
origin represents a contour of equal translational power. Figure 3.2 shows one such
contour that clearly separates all data points derived from rapidly responding bacteria
from all data points derived from slowly responding bacteria. Making a comparison of

the data in the region where the levels of RNA investment of microbes of both strategies

overlap (boxed region of Figure 3.2, between about 1.25 and 2.75 mg ml'1 0.1 OD420'1

RNA), the average translational output of rapidly responding bacteria is about 3.6-fold
higher than the average translational output of the slowly responding bacteria. This
difference is in the middle of the range of comparisons of translational power between E.

coli and slowly growing microbes presented in Chapter 1.

Correlation of rrn copy number with the strength of translational selection

Translational selection refers to the existence of a translation-related benefit, such as
faster or less error-prone translation, obtained by using certain codons instead of
synonymous alternatives 17. Because the magnitude of any such benefit varies with
expression level, a correlation among the genes in an organism between expression level
and the degree of codon bias is evidence of translational selection. In order to compare
the strength of translational selection between strains, however, it is necessary to
distinguish codon bias that is due to translational selection from codon bias that is due to
mutational bias or other factors that are consistent within a strain but vary between
strains 17. We have made this distinction by measuring the codon bias of a set of highly
expressed genes relative to the codon bias of the genome from which they were derived;

this measure is expressed as the quantity AN'c (Materials and Methods). Using codon

96

Figure 3.2: Translational output as a function of RNA content

 

 

 

 

10—

E
.§ 8—
o
O
o' ,’
L. 6— ’
E
U)
3; I i
“a E V E ,/ +E.coli
.5 4_ 5 g ,’ +I—IF3
9. 5 g//’ +504
8 : V / ’: +EC2
a §+ // i +E05
g 5 / / ’ l-Vrli -<>— S.alaskensis
.3 2_ 5 / / 3 —A—PX3.14
g. g ,/ O ; —G—HSS
,. //[3 ; A19 [3 g —O—PX3.15

8 l l l l l

1 2 3 4 5

RNA content (pg ml'1 0.1004204)

Figure 3.2. Translational output (uP) per optical density (0.1 OD420) as a function of
RNA content per optical density is shown for rapidly responding bacteria (solid symbols)
and slowly responding bacteria (open symbols). In the range of RNA content represented
by both ecological strategies (data points within boxed area), the average translational
output of the rapid responders is 3.6-fold higher than that of the slow responders.
Because translational power is the ratio of the ordinate to the abscissa on this plot, a
straight line through the origin represents a contour connecting points of equal
translational power. One such contour has been drawn that separates all data derived
from rapid responders (higher translational power, above and left of contour) from all
data derived from slow responders (lower translational power, below and right of
contour). Data points connected by a line represent the same strain grown in two
different media. Data points and error bars are the means and standard errors of
measurements on at least 3 independent replicate cultures per strain-medium
combination. Error bars smaller than or of comparable size to plot symbols are not
shown.

97

data from 76 completely sequenced bacterial genomes 21, we have found that AN'c is
correlated with rrn copy number (Figure 3.3a), implying that stronger translational

selection is associated with a rapid growth response strategy, as predicted.

A likelihood ratio method 25 using explicit phylogenetic information (see Appendix)
demonstrates that the association of strong translational selection and high rrn copy is
significant throughout the phylogeny, explaining about a quarter of the evolutionary
variation in these traits across bacteria (R2=O.29, p<10'6). For this likelihood ratio test as
well as the test of tRN A gene number reported below, the conclusions are not altered by
the inclusion of poorly resolved details in the bacterial phylogeny (see Appendix). A
traditional correlation analysis (which would show a stronger correlation between the
traits) is inappropriate for demonstrating a biological relationship between translational
selection and rrn copy number, since it assumes the statistical independence of data

derived from species with various durations of shared evolutionary history 25.

Since our measure of the strength of translational selection depends on both the codon
bias of the genome as a whole and the codon bias of the HE set, we investigated whether
either of these component traits were correlated with rrn copy number, using Wright’s Nc
as our measure of codon bias. Considered separately, the codon bias of neither the highly
expressed genes nor the entire genome is correlated with rrn copy number, as shown in
Figure 3.4. This result indicates that rrn copy number is related specifically to the
strength of translational selection, not to trends in codon bias that may be influenced by

factors other than translational selection.

98

Figure 3.3: Correlation between rm copy number and
the strength of translational selection

 

 

 

 

30-
a
0 8 g 0 O O
20_ O O 0
§ 8 0 8 3
<1 9 o
10— 0
8 e 8
e O 9 ° ° 0
g g p<10'6,R2= 0.29
O ' l l l l l l
120-
b
90 g 0 O O
O
.5 . 8 .
t». g 8 g 8
§ § 0
9.5.
§
p<10"‘, R2 = 0.46
O l l l l l I
2 4 6 8 10 12

rrn copy number

Figure 3.3. Correlation between rrn copy number and the strength of translational
selection, measured either as AN‘c (panel a, see Materials and Methods for explanation)
or as tRNA gene copy number (panel b). Each data point represents a sequenced genome
(n=76); list of taxa and phylogenetic information is provided in the Appendix. Strength
and significance of correlations between rrn copy number and measures of translational
selection across taxa are assessed by log likelihood ratio tests (see Materials and Methods
for explanation).

99

Figure 3.4: Relationship between rm copy number and the codon bias
of entire genomes and of highly expressed genes

 

 

 

 

60 -
a 8 O O O

’6‘ <> <> 0 o O
z 50 8 g 0
7; _ 0 g o 0 8
(U
'5 g g 8 8
8 O o O O
'o 40 - o
8 8 0
8
r: 30 —
8

20 I I I I I I

60 Q

h
0 O

’5 50 -
g 8 8 O
.8 g 8 0
.a
8 g g o <> <> 0 O
o 9 O 9 8 8
Lu 0

20 I I I I I I

2 4 6 8 10 12

rrn copy number

Figure 3.4. Relationships between rrn copy number and the degree of codon bias
either in all predicted genes of a genome (panel a) or in a set of 8 highly expressed (HE)
genes (panel b, see Materials and Methods for a list of genes). Codon bias is measured as
Wright's Nc23. Each data point represents a sequenced genome (n=76); list of taxa and
phylogenetic information is provided in the Appendix. The correlation between rrn copy
number and codon bias across genomes is significant in neither set of genes, as assessed
by log likelihood ratio tests (see Materials and Methods for details).

100

An independent measure of the strength of translational selection is the number of tRNA
genes per genome. All organisms use ‘wobble’ pairing in the 3rd codon position to
translate the 61 sense codons with a reduced number of unique anticodons, typically
around 40. However, each anticodon can be represented in the genome by 1 or more
tRNA genes. Since higher gene dosage is associated with the abundant tRNAs that
translate preferred codons, the total number of tRNA genes reflects the overall strength of
translational selection 3233. Consistent with a previous study using a smaller set of
microorganisms 33, we found that tRNA gene copy number in 76 sequenced bacterial
genomes correlates with rm copy number (R2=O.46, p<10'11 by likelihood ratio test),
supporting our inference of an association between strong translational selection and the

rapid response strategy (Figure 3.3b).

101

Chapter 3 Discussion

The data of Figures 3.1 and 3.2 demonstrate that variation in translation power among
bacteria is consistent with ecological factors, but not phylogeny. Specifically, rapidly
responding bacteria have high translational power, whereas slowly responding bacteria
have low translational power. Protein synthesis is commonly assumed to be
fundamentally equivalent in all bacteria. Indeed, we would agree that all organisms
accomplish translation by the same sequential reactions, using homologous components
with conserved functions. However, these data do not show equivalent performance, but
rather systematic differences between rapidly responding bacteria and slowly responding
bacteria either in the translation rate, or in the fraction of active ribosome, or both. Our
results are consistent with comparisons made from macromolecular data available in the

literature (Table 1.1).

Codon usage analysis provides supporting evidence that translational performance differs
between bacteria, and that this variation is influenced by ecological factors. Preferred
codons tend to be translated by abundant tRNA species, or to form 3 canonical base pairs
with their cognate tRNA, instead of relying on ‘wobble’ pairing in the third codon
position. This allows preferred codons to be translated more quickly and with fewer
errors 16,34,35, enhancing translational power. Hence, selection for translational power is
one of the two major factors contributing to codon bias. However, to use codon bias as
an indicator of the strength of selection for translational power, the effects of translational
selection must be distinguished from those of the other major factor affecting codon use,

which is mutational bias. Mutational bias establishes a characteristic, organism-specific

102

nucleotide composition in the chromosome 36; it is generally supposed to be selectively
neutral. It is also fairly uniform throughout a genome; spatial heterogeneity of mutational
bias in the genome has been identified as the major source of variability in codon use
within an organism only when the mutational bias is not extreme and translational
selection is weak or absent 37'39. This has allowed us to use genome-average codon use
to control for differences in mutational bias and other genome-wide processes between
organisms; we can measure the strength of translational selection consistently across

organisms by the deviation of codon use in the HE set relative to this control.

The complexity of correcting for mutational bias in a codon-based measure of
translational selection can be circumvented by using a different measure of the strength
of translational selection, the total number of tRNA genes per genome. Because the
benefit of codon bias depends in large part on having abundant tRNA molecules cognate
to the preferred codons 16, and because the abundance of a particular tRNA species is
strongly influenced by gene dosage 40,41, the total number of tRNA genes in an organism
is a measure of the benefit it obtains from codon bias. As shown in Figure 3.3, both
measures of the strength of translational selection are consistent, indicating that the
strength of selection for translational power is strongest in bacteria with high rrn copy
number. Our results relating both codon bias patterns and tRNA gene copy number to
rrn copy number are similar to those of an earlier study based on 18 microbial

genomes 33. Both data from the scientific literature (Table 1.1) and our own results

(Figure 3.1) indicate that translational power is high in organisms with high rrn copy

103

number, consistent with the predicted ecological benefit of this trait for rapid growth and

rapid acceleration of growth 13,19.

Codon bias is not merely a passive indicator of translational selection; codon bias itself
directly influences translational power 16. It is reasonable to ask whether the stronger
selective pressure for translational power identified in the rapidly responding bacteria
needs any additional mechanism, beyond codon bias, in order to exert its effect. In other
words, we are asking if the variation in codon bias between rapid and slow responders is
a sufficient explanation for the observed differences in translational power. It is not a
trivial task to address this topic. A precise estimate of the extent to which codon bias
accelerates translation in an organism would require knowledge of the translation rate for
each codon as a function of the concentration of each of its cognate tRNA species, as
well as knowledge of both codon expression frequency and individual tRNA
concentrations in vivo as a function of growth rate. Such data is not yet available for E.
coli, much less any other organism. However, the available data are sufficient for a
rough estimate of the magnitude of the translation rate benefit of codon bias in E. coli;
this estimate is the subject of Chapter 4. Our conclusion is that translation is at most 60%
faster in E. coli than it would be in the absence of codon bias. While substantial, this
effect is much smaller than the variation in translational power inferred from
macromolecular data in the literature (Table 1.1) or from our own measurements (Figures
3.1 and 3.2). Hence, differences in translational performance between microbes adapted
for different ecological strategies must reflect differences in the translational apparatus

itself.

104

A tradeoff mediated by processivity errors

One mechanism that could account for varying performance of the translational apparatus
between organisms is an evolutionary tradeoff, so that organisms with high translational
power would necessarily be worse in some other fitness-related trait. Organisms would
be selected to have low translational power relative to others, if they occupied a niche
where the alternative trait made a larger relative contribution to their fitness. In Chapter
1, we noted evidence for such a tradeoff involving starvation survival, obtained during
the adaptation of natural E. coli isolates to laboratory conditions. The introduction of this
chapter reviews evidence that processivity error rates, specifically dropoff rates, are
surprisingly high in laboratory adapted E. coli. Accordingly, we propose the existence of
a tradeoff between high translational power and high translational yield, mediated by the

frequency of processivity errors.

Although a complex tradeoff in which missense accuracy affects both the translation rate
and the dropoff frequency has been proposed on theoretical grounds 9’42, we know of no
direct experimental evidence indicating that rapid translation must be associated with
frequent dropoff errors. However, evidence for the existence of some sort of mechanistic
connection between the missense error rate, the processivity error rate, and the translation
rate is strong. Ribosome mutations derived from laboratory-adapted strains of E. coli that
alter the frequency of missense errors in either direction consistently decrease the
translation rate and increase the frequency of processivity errors, most of which are
dropoff events 3:10. In vitro measurements of the rates of peptidyl-tRNA dissociation

from two different sites on the ribosome suggest the involvement of two classes of

105

dropoff events: missense-restrictive mutations may facilitate dropoff from the A-site,
whereas missense-enhancing mutations may facilitate dropoff from the P-site 12. Hence,
the laboratory ‘wild-type’ ribosome of E. coli appears to combine optimality according to
two criteria simultaneously: rapid translation, and a low frequency of processivity errors,
obtained by balancing two different classes of dropoff events. However, the frequency of
processivity errors for these rapidly translating ribosome remains high enough that only
about 90% of ribosomes successfully translate the length of an average gene. If the
selective pressure for rapid translation were relaxed, further reductions in the rate of
processivity errors may be possible. The discovery that selection for rapid translation is
linked to a decline in the ability to survive starvation among natural E. coli strains 5’6 is
consistent with such a tradeoff between translation rate and the frequency of dropoff
errors. If starvation survival requires energy derived from endogenous resources, some
of which is used for protein synthesis, enhancing translational yield by reducing the
frequency of processivity errors could promote survival by conserving endogenous

I'CSOLIICCS.

There may be a second, independent link between processivity errors and translational
power, mediated by protein factors involved in normal translational termination. The
tradeoff would involve not translation rate, but the active fraction of ribosomes, the
second component of translational power. When a stop codon occupies the ribosomal A-
site, either release factor 1 (RF 1) or release factor 2 (RF 2) enters the A-site to facilitate

cleavage of the nascent protein from the P-site tRNA; release factor 3 (RF 3) then

displaces RFl or RFZ from the ribosome 43,44. Ribosome release factor (RRF) replaces

106

RF3 in the A-site, and with the help of EF-G, it releases the deacylated tRNA from the P-
site and allows the ribosome either to dissociate from or slide along the mRN A 45,46.
However, RF3, RRF and EF-G can also act with a sense codon in the A-site and peptidyl-
tRNA in the P-site, generating a dropoff error 47. Nonetheless, at an individual sense
codon, elongation is far more likely than dropoff, which may be explained, at least in
part, by the fact that amino acyl-tRNA selection is much faster than release factor
activity. The ribosome typically requires ~20—50 ms to find a cognate ternary complex
during elongation 48,49, whereas protein release by RF 1/2 during termination occupies
~1-2 s 50 and the release of the ribosome appears to be even slower 51,52. However, the
inherent competition between elongation and factor-catalyzed dr0poff at sense codons
may explain why E. coli has not evolved to reduce the amount of time ribosomes spend
sequestered at the end of messenger RNA. Decreasing the fiaction of inactive ribosomes
in the cell by accelerating post-termination ribosomal release (whether by regulatory
changes to increase the concentration of RF3 and/or RRF, or by structural changes that
accelerate the activity of these factors) may also increase the frequency of dropoff errors
during elongation. A similar argument can be made that the slow action of RFl and RF2
in releasing the nascent polypeptide is related to avoidance of false stops, when these
factors act inappropriately during translational elongation. However, misrecognition of

sense codons by RFl and RF2 occurs less frequently than dropoff errors 50.

An alternative explanation

We are not the first to note a link between patterns of codon bias and ecological factors.

Sharp and his colleagues have suggested that differences between strains in the strength

107

of translational selection (recognized as variation in patterns of codon bias) reflect
differences in bacterial life history 17, suggesting that the importance of translational
power varies with ecological strategy. More recently, this group has specifically
identified a gradient across species in the strength of translational selection on codon use
with a gradient in the rate of exponential growth or importance of exponential growth for
the fitness of the organism 1753. Helicobacter pylori provides the most extreme contrast
to the paradigm of translational selection represented by E. coli; reportedly it shows no
evidence of translational selection in its codon use, despite the fact that its mutational
bias is not strong enough to obscure translational selection if it exists 54. Lafay et al.
argue that translational power offers no benefit in the niche inhabited by H. pylori,
suggesting that it does not experience competitive exponential growth 54. This appears to
be consistent with the harsh environment of the gastric mucosa, where no other microbial
species are known to persist 54. In effect, these authors have conceived of a spectrum of
ecological strategies very similar to the spectrum we describe between rapidly responding
and slowly responding bacteria, and suggested that the relative positions of strains on this
spectrum could be inferred by assessing the strength of translational selection. In this
light, the correlations shown in Figures 3.3 can be seen as quantitative support for their
expectation, confirming the utility of rrn copy number as an index of ecological strategy

along this axis.

The explanation offered by Sharp and colleagues appears to be simpler than an

explanation involving tradeoffs. It depends solely on variation in the strength of positive

selection for translational power, rather than invoking as-yet untested mechanisms that

108

link translational power to other traits. We would certainly concur that the benefit of
translational power varies between strains according to their ecological strategy, and that
such variation contributes to patterns in codon use and translational power. However, we
consider it unlikely that a benefit of translational power is completely absent in any
organism. In H. pylori, for example, Lafay et al. acknowledge that the population
structure of this organism reflects high rates of interstrain recombination, with well
documented coinfection by multiple strains in a single individual 54. These results make
it more difficult to accept the claim 54 that episodes of intraspecific competition are

completely absent.

As explained in Chapter 1, even conditions where grth is severely constrained by
resource availability do not necessarily eliminate selection for higher translational power.
Furthermore, the Ehrenberg-Kurland model of the cell 7 (see also Chapter 1) suggests
that the contribution of a cell component to the fitness of an organism can be measured
by its mass. If protein synthesis has little influence on the fitness of an organism, why
would it devote a large amount of biomass to the translational apparatus? A large
biomass investment in the protein synthesis system suggests that the rate of protein
synthesis has limited the growth rate of the strain, at least in some of the conditions it has
experienced over its evolutionary history. We have no information on the protein or
RNA content of H. pylori; however, we know that some microbes with low translational
power are making substantial biomass investments in ribosomes or in RNA (Table 1.1,
Figure 2.7). The phenomenon of a large investment in the protein synthesis machinery

along with low translational power becomes easier to understand if high translational

109

power has not only beneficial effects (which may or may not be realized depending on

the ecological context), but deleterious effects as well.

One small, puzzling detail remains. If, as we suggest, reduced but positive selection for
high translational power is opposed in some conditions by stronger selective pressure to
avoid processivity errors, strains adapted to these conditions might be expected to have a
translational apparatus with a low inherent frequency of processivity errors, but also to
have biased codon use to increase translational power. If the translational apparatus itself
is constrained to adopt a higher yield, lower power phenotype, selection for higher
translational power will be focused on codon use and tRNA abundance. Why do we not
find stronger evidence of translational selection among the slowly responding, low rm
copy number strains? It may be that selection to avoid processivity errors directly
influences the degree of codon bias as well the phenotype of the translational apparatus.
In order to supply a larger concentration of tRNAs cognate to preferred codons, either the
total tRNA concentration must increase (hence diminishing the translation rate benefit of
codon bias by imposing an increased biomass cost), or the concentration of tRNAs
cognate to nonpreferred codons must decrease. However, the expression of codons
translated by rare tRNAs increases the frequency of processivity errors (at least in E.
coli), an effect that can be reversed by increasing the supply of the rare tRNAs 34. The
selective pressure to restrict processivity errors in slowly responding bacteria may
constrain the disparity in abundance of different tRNA species, which in turn reduces the
potential for a co-evolutionary response in codon frequency and tRNA abundance

distributions that is the hallmark of translational selection.

110

Conclusion

Translation is an ancient process, predating the last common ancestor of all known life.

It is both essential and expensive, representing a significant fraction of all biological
activity, particularly for unicellular organisms. Most improvements to the translational
apparatus that do not involve compromising one beneficial feature for another may well
have become universal through more than 3.5 billion years of evolution by natural
selection. Thus, engineers, ecologists, and evolutionary biologists might not be surprised
at the suggestion that optimizing this complex, well-integrated biological apparatus for
one aspect of performance involves sacrifices according to another criterion. We believe
that evidence for such a tradeoff is found in correlations between the ecological strategy
of a bacterial species and both the power of its translational machinery and the strength of
selection influencing its codon use. Certainly, the specific mechanism we have proposed
may be confirmed or refuted as our knowledge of translation becomes more complete.
Nonetheless, these results demonstrate that the performance characteristics of the
translational apparatus are not invariant, but adapt in response to selection. Comparative
analysis of the translational apparatus among bacteria with different ecological strategies
may improve our understanding of translation, and recognizing that translational
performance is adaptive will contribute to our understanding of the diversity and

distribution of microbial life on Earth.

111

10.

11.

12.

13.

Chapter 3 References

Russell, J .B. and GM. Cook, 1995. Energetics of bacterial growth - balance of
anabolic and catabolic reactions. Microbiological Reviews, 59(1):48-62.

Ingraham, J .L., O. Maalee and RC. Neidhardt, 1983. Growth of the Bacterial
Cell. Sinauer, Sunderland, Massachusetts.

Yusupov, M.M., G.Z. Yusupova, A. Baucom, K. Lieberman, T.N. Earnest, J .H.D.
Cate and HF. Noller, 2001. Crystal structure of the ribosome at 5.5 angstrom
resolution. Science, 292(5518):883-896.

Rarnakrishnan, V., 2002. Ribosome structure and the mechanism of translation.
Cell, 108(4):557-572.

Kurland, CG. and R. Mikkola, 1993. The impact of nutritional state on the
microevolution of ribosomes; pp. 225-237 in Starvation in Bacteria; S.
Kjelleberg, ed. Plenum Press, New York.

Mikkola, R. and CG. Kurland, 1992. Selection of laboratory wild-type phenotype
from natural isolates of Escherichia coli in chemostats. Molecular Biology and
Evolution, 9(3):394-402.

Ehrenberg, M. and C .G. Kurland, 1984. Costs of accuracy determined by a
maximal growth rate constraint. Quarterly Reviews of Biophysics, 17(1):45-82.

Kurland, C.G., D. Hughes and M. Ehrenberg, 1996. Limitations of translational
accuracy; pp. 979-1004 in Escherichia coli and Salmonella: Cellular and
Molecular Biology; F.C. Neidhardt, et al., eds. ASM Press, Washington, D. C.

Kurland, CG. and M. Ehrenberg, 1985. Constraints on the accuracy of messenger
RNA movement. Quarterly Reviews of Biophysics, l8(4):423-450.

Dong, H.J. and CG. Kurland, 1995. Ribosome mutants with altered accuracy
translate with reduced processivity. Journal of Molecular Biology, 248(3):551-
561.

J orgensen, F. and CG. Kurland, 1990. Processivity errors of gene expression in
Escherichia coli. Journal of Molecular Biology, 215(4):511-521.

Karimi, R. and M. Ehrenberg, 1996. Dissociation rates of peptidyl-tRNA from the
P-site of E. coli ribosomes. EMBO Journal, 15(5): 1 149-1 154.

Karimi, R. and M. Ehrenberg, 1994. Dissociation rate of cognate peptidyl-

transfer RNA from the A-site of hyper-accurate and error-prone ribosomes.
European Journal of Biochemistry, 226(2):355-360.

112

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

Manley, J .L., 1978. Synthesis and degradation of termination and premature

termination fragments of ,B-Galactosidase in vitro and in vivo. Journal of
Molecular Biology, 125:407-432.

Menninger, J.R., 1976. Peptidyl transfer RNA dissociates during protein synthesis
from ribosomes of Escherichia coli. Journal of Biological Chemistry, 251 :3392-
3398.

Andersson, S.G.E. and CG. Kurland, 1990. Codon preferences in free-living
microorganisms. Microbiological Reviews, 54(2):198-210.

Sharp, P.M., M. Stenico, J .F . Peden and A.T. Lloyd, 1993. Codon usage -
mutational bias, translational selection, or both? Biochemical Society
Transactions, 21(4):835-841.

Klappenbach, J .A., J .M. Dunbar and T.M. Schmidt, 2000. rRNA operon copy
number reflects ecological strategies of bacteria. Applied and Environmental
Microbiology, 66(4):]328-1333.

Stevenson, BS. and T.M. Schmidt, 1998. Growth rate-dependent accumulation of

RNA from plasmid-borne rRNA operons in Escherichia coli. Journal of
Bacteriology, 180(7):1970-1972.

Eyre-Walker, A., 1996. Synonymous codon bias is related to gene length in
Escherichia coli: Selection for translational accuracy? Molecular Biology and
Evolution, l3(6):864-872.

Peterson, J .D., L.A. Umayam, T. Dickinson, E.K. Hickey and O. White, 2001.
The Comprehensive Microbial Resource. Nucleic Acids Research, 29(1): 123-125.

Novembre, J .A., 2002. Accounting for background nucleotide composition when
measuring codon usage bias. Molecular Biology and Evolution, 19(8):1390-1394.

Wright, F., 1990. The eflective number of codons used in a gene. Gene, 87(1):23-
29.

Comeron, J .M. and M. Aguade, 1998. An evaluation of measures of synonymous
codon usage bias. Journal of Molecular Evolution, 47(3):268-274.

Page], M., 1999. Inferring the historical patterns of biological evolution. Nature,
401(6756):877-884.

Felsenstein, J ., 1981. Evolutionary trees from DNA sequences - a maximum
likelihood approach. Journal of Molecular Evolution, 17(6):368-376.

Klappenbach, J .A., P.R. Saxman, J .R. Cole and T.M. Schmidt, 2001. rrndb: the
Ribosomal RNA Operon Copy Number Database. Nucleic Acids Research,

’ 29(1):]81-184.

113

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

Condon, C., D. Liveris, C. Squires, I. Schwartz and CL. Squires, 1995.
Ribosomal RNA operon multiplicity in Escherichia coli and the physiological
implications of rrn inactivation. Journal of Bacteriology, 177(14):4152-4156.

Cavicchioli, R., M. Ostrowski, F. Fegatella, A. Goodchild and N. Guixa-
Boixereu, 2003. Life under nutrient limitation in oligotrophic marine
environments: An eco/physiological perspective of Sphingopyxis alaskensis
(formerly Sphingomonas alaskensis). Microbial Ecology, 45(3):203-217.

Schut, F., J .C. Gottschal and RA. Prins, 1997. Isolation and characterisation of
the marine ultramicrobacterium Sphingomonas sp. strain R3225 6. FEMS
Microbiology Reviews, 20(3-4):363-369.

Kjeldgaard, NO. and CG. Kurland, 1963. The distribution of soluble and
ribosomal RNA as a function of growth rate. Journal of Molecular Biology,
6:341—348.

Ikemura, T., 1985. Codon usage and transfer RNA content in unicellular and
multicellular organisms. Molecular Biology and Evolution, 2(1): 13-34.

Kanaya, S., Y. Yarnada, Y. Kudo and T. Ikemura, 1999. Studies of codon usage
and tRNA genes of 18 unicellular organisms and quantification of Bacillus
subtilis tRNAs: Gene expression level and species-specific diversity of codon
usage based on multivariate analysis. Gene, 238(1):143-155.

Sorensen, H.P., H.U. Sperling-Petersen and K.K. Mortensen, 2003. Production of
recombinant thermostable proteins expressed in Escherichia coli: Completion of
protein synthesis is the bottleneck Journal of Chromatography B-Analytical
Technologies in the Biomedical and Life Sciences, 786(1-2):207-214.

Akashi, H. and A. Eyre-Walker, 1998. Translational selection and molecular
evolution. Current Opinion in Genetics & Development, 8(6):688-693.

Sueoka, N., 1988. Directional mutation pressure and neutral molecular evolution.
Proceedings of the National Academy of Sciences of the United States of
America, 85(8):2653-2657.

McInemey, J .O., 1998. Replicational and transcriptional selection on codon
usage in Borrelia burgdorferi. Proceedings of the National Academy of Sciences
of the United States of America, 95(18): 10698-10703.

Lafay, B., A.T. Lloyd, M.J. McLean, K.M. Devine, P.M. Sharp and K.H. Wolfe,
1999. Proteome composition and codon usage in spirochaetes: Species-specific
and DNA strand-specific mutational biases. Nucleic Acids Research, 27(7):1642-
1649.

114

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

Romero, H., A. Zavala and H. Musto, 2000. Codon usage in Chlamydia
trachomatis is the result of strand—specific mutational biases and a complex
pattern of selective forces. Nucleic Acids Research, 28(10):2084-2090.

Ikemura, T., 1981. Correlation between the abundance of Escherichia coli
transfer RNAs and the occurrence of the respective codons in its protein genes - a
proposal for a synonymous codon choice that is optimal for the Escherichia coli
translational system. Journal of Molecular Biology, 151(3):389-409.

Dong, H.J., L. Nilsson and CG. Kurland, 1996. Co-variation of tRNA abundance
and codon usage in Escherichia coli at difi’erent growth rates. Journal of
Molecular Biology, 260(5):649-663.

Kurland, CG. and M. Ehrenberg, 1987. Growth-optimizing accuracy of gene
expression. Annual Review of Biophysics and Biophysical Chemistry, 16:291-
3 17.

Freistroffer, D.V., M.Y. Pavlov, J. MacDougall, R.H. Buckingham and M.
Ehrenberg, 1997. Release factor RF 3 in E. coli accelerates the dissociation of

release factors RF 1 and RF 2 from the ribosome in a GT P-dependent manner.
EMBO Journal, 16(13):4126-4133.

Kisselev, L., M. Ehrenberg and L. Frolova, 2003. Termination of translation:
interplay of mRNA, rRNAs and release factors? EMBO Journal, 22(2):175-182.

J anosi, L., S. Mottagui-Tabar, L.A. Isaksson, Y. Sekine, E. Ohtsubo, S. Zhang, S.
Goon, S. Nelken, M. Shuda and A. Kaji, 1998. Evidence for in vivo ribosome

recycling, the fourth step in protein biosynthesis. EMBO Journal, 17(4):1141-
1 15 1 .

Kaji, A., M.C. Kiel, G. Hirokawa, A. Muto, Y. Inokkuchi and H. Kaji, 2001. The
fourth step of protein synthesis: Disassembly of the posttermination complex is
catalyzed by elongation factor G and ribosome recycling factor, a near-perfect
mimic of tRNA; pp. 515-529 in The Ribosome; B. Stillman, ed. Cold Spring
Harbor Laboratory Press.

Heurgue-Hamard, V., R. Karimi, L. Mora, J. MacDougall, C. Leboeuf, G.
Grentzmann, M. Ehrenberg and RH. Buckingham, 1998. Ribosome release factor

RF 4 and termination factor RF 3 are involved in dissociation of peptidyl-tRNA
from the ribosome. EMBO Journal, 17(3):808-816.

Bilgin, N., F. Claesens, H. Pahverk and M. Ehrenberg, 1992. Kinetic properties of
Escherichia coli ribosomes with altered forms of SI 2. Journal of Molecular
Biology, 224(4): 101 1-1027.

Bilgin, N., L.A. Kirsebom, M. Ehrenberg and CG. Kurland, 1988. Mutations in
ribosomal proteins L 7/L12 perturb EF-G and EF - T u functions. Biochimie,
70(5):611-618.

115

50.

51.

52.

53.

54.

Freistroffer, D.V., M. Kwiatkowski, R.H. Buckingham and M. Ehrenberg, 2000.
The accuracy of codon recognition by polypeptide release factors. Proceedings of
the National Academy of Sciences of the United States of America, 97(5):2046-
205 1.

Pavlov, M.Y., D.V. Freistroffer, V. HeurgueHamard, R.H. Buckingham and M.
Ehrenberg, 11997. Release factor RF 3 abolishes competition between release
factor RF 1 and ribosome recycling factor (RRF ) for a ribosome binding site.
Journal of Molecular Biology, 273(2):389-401.

Pavlov, M.Y., D.V. Freistroffer, J. MacDougall, R.H. Buckingham and M.
Ehrenberg, 1997. Fast recycling of Escherichia coli ribosomes requires both
ribosome recycling factor (RRF) and release factor RF 3. EMBO Journal,
16(13):4134-4141.

Grocock, R.J. and RM. Sharp, 2002. Synonymous codon usage in Pseudomonas
aeruginosa PA01. Gene, 289(1-2):131-139.

Lafay, B., J .C . Atherton and RM. Sharp, 2000. Absence of translationally

selected synonymous codon usage bias in Helicobacter pylori. Microbiology-UK,
146:851-860.

116

Chapter 4. Differences in codon bias cannot explain
differences in translational power between
bacteria pursuing different ecological strategies

Chapter 4 Abstract

Translational power is a comparison of translational output to the biomass invested in
translational machinery. Translational power is higher among bacteria adapted to exploit
large fluctuations in resource availability via rapid growth, and lower among bacteria
adapted to exploit small fluctuations in resource availability (Chapters 1 and 3). The
selective benefit of high translational power among Escherichia coli and other microbes
capable of rapid growth is believed to explain a correlation between the degree of codon
usage bias and gene expression level in such organisms. Conversely, the absence of such
a benefit has been inferred from the absence of a correlation between codon bias and
gene expression level among at least some slowly growing microbes. To investigate
whether differences in the degree of codon usage bias could explain differences in
translational power between bacteria pursuing these contrasting strategies, we estimated
the extent to which codon bias could accelerate translation in E. coli. We conclude that
while codon bias offers a substantial benefit in terms of faster translation, it is insufficient

to explain the observed differences in translational power.

117

Chapter 4 Introduction

Translational power is the translational output of a cell or culture normalized to the
amount of biomass invested in the translational apparatus. The concept was first
introduced to facilitate quantitative comparisons of translational performance between
different growth rates within a single bacterial strain 1. (Translational power is a new
term we are introducing to replace the terms ‘ribosome efficiency’ and ‘translational
efficiency’ that have previously been used for this concept; the rationale for this semantic
distinction is provided in Chapter 1.) The initial belief that translational power is nearly
constant across a wide range of growth rates, based both on empirical data and ecological
arguments 23, has gradually given way to the current understanding that translational
power increases with growth rate, at least in E. coli 4'7. The question of whether
translational power varies between microbial species has been addressed only rarely, in
four studies that have made comparisons between E. coli and one other strain 8‘1 1. Each
of these studies found that translational power was higher in E. coli than in a slowly
growing strain; reanalysis of these studies as well as other data available in the literature
suggests that the translational power of bacteria incapable of rapid growth ranges fiom
less than 17% to 42% of the translational power of E. coli (Chapter 1). Furthermore, an
experiment comparing bacteria of contrasting ecological strategies showed higher
translational power among strains adapted for rapid grth in response to resource
abundance, and lower translational power among strains incapable of such a response
(Chapter 3). The quantitative differences were similar in magnitude to those found in

data from the literature.

118

One factor capable of affecting translational power is biased use of synonymous
alternative codons. In the standard translational code, 18 of the 20 amino acids are
encoded by more than a single codon, but in many microorganisms, synonymous codons
are not equally abundant 12. The pattern first found in E. coli and Bacillus subtilis turns
out to be common: the majority of genes within an organism show a preference for the
same subset of codons, but the degree of bias towards the preferred subset is correlated
with the expression level of the gene 13,14. For some time, the consensus has been that
such a pattern reflects selection for translational power 15,16. Codon bias enhances
translational power because preferred codons are translated more rapidly than
synonymous alternatives, either because the cognate tRNA species are relatively
abundant, or because the codon-tRNA interaction is relatively efficient, or both 15,15.
Additional translation-related selective pressures, such as reducing the frequency and
severity of translational errors, may have influenced the choice of preferred codons in

some instances 17, but have not been demonstrated generally.

Since the benefit of using a preferred codon in a particular gene depends on how often the
gene is translated, selection for preferred codons increases with expression level,
accounting for the correlation noted above between the degree of codon bias and gene
expression level. Although a number of examples have been documented where
particular rare codons play a regulatory role 13'20, most non-preferred codons are not
under strong selection 21,22, and a general role for non-preferred codons as down-

regulators of gene expression has been rejected 16,23. Rather, non-preferred codons

become more abundant in genes expressed at lower levels because the benefit obtained

119

from preferred codons is reduced, allowing a higher proportion of random synonymous

mutations to become fixed by genetic drift 24.

In contrast to codon bias due to translational selection, codon bias that is consistent in
both magnitude and direction in genes that vary widely in expression level is explained
most easily by mutational bias acting on DNA 15. Generally, mutational bias has been
associated with deviations from 50% G+C content that are uniform throughout the
genome, but a regional variation in G+C content has been identified in Mycoplasma
genitalium 25 and strand-specific mutational bias has been identified in a number of
organisms 25. However, spatial heterogeneity of mutational bias in the genome has been
identified as the major source of variability in codon use within an organism only when
translational selection is weak or absent 27'29. While the effects of both translational
selection and mutation bias are evident in some microbial genomes with moderately
biased G+C content 3031, several organisms with extremely high or low G+C content
have been reported to show very little 32 or no 33'” evidence of translational selection.
Theoretical calculations indicate that if the strength of mutational bias exceeds a certain
critical threshold, any pre-existing codon preferences that conflict with the mutational
bias will be reversed 36. In this case, codon use is almost entirely determined by the
mutational bias, which influences genes equally regardless of expression level. Note that
while a comparison between codon use and expression level within such a genome would
not provide evidence of translational selection, this does not imply that deviations from
the average codon usage would be selectively neutral in such an organism, or that the

fitness penalty of such deviations would be independent of gene expression level. Some

120

organisms with moderate G+C content have also been reported to lack correlations
between codon usage and gene expression level, particularly the spirochete T reponema
pallidum 28 and the proteobacteria Helicobacter pylori 37. The absence of evidence for
translational selection is much more striking in these organisms, since they lack a strong
mutational bias that could obscure such evidence. It has been suggested that for such
organisms, competitiveness during exponential growth confers little or no fitness
benefit 15,37, consistent with their slow grth rate and other characteristics of their

ecological niche.

If variation in the strength of selection for translational power leads to differences in the
degree of codon bias between microbes (superimposed on any differences in codon bias
that can be attributed to variation in mutational bias), we wondered whether differences
in codon bias could in turn explain the observed differences in translational power
between microbes. To address this issue, we frame the following question: To what
extent does codon bias accelerate the translation rate of E. coli, in comparison to a
hypothetical organism with the same proteome composition and the same investment in
the translational apparatus, but which lacks bias in codon use? For convenience, we will
refer to this hypothetical E. coli-like organism with uniform use of synonymous
alternative codons as ‘Uni’. By ‘same proteome composition’, we mean that over a cell
generation, the number of times each amino acid is used in translation is the same in Uni
and in E. coli, although the frequencies of the codons directing that translation will differ
for the 18 amino acids specified by multiple codons. By ‘same investment in the

translational apparatus’, we mean that the total biomass of the translational apparatus is

121

the same in Uni as in E. coli, although ideally the allocation of that biomass among the
translational components would be optimized for the unbiased codon use postulated for
Uni. However, in order to apply empirical codon-specific translation rate data, we will
impose a more stringent requirement on Uni, that the abundance of each individual
component of the translational apparatus will be unchanged in comparison to E. coli.
Due to this restriction and for other reasons, we make no claim to be able to answer our
question precisely. However, our approximations are sufficient to conclude that
differences in codon bias alone are unlikely to account for differences in translational
power of the magnitude inferred to exist from macromolecular analysis of slowly

growing and rapidly growing microbes.

122

Chapter 4 Materials and Methods

Calculation of the translation rate benefit of codon bias

Consider a cell in which a total of C ,- codons of type i are translated during a single cell
generation, so that the sum over all sense codons C = ZCl- is the total number of codons

translated during a cell generation. (Hereafter we refer to the translational output over a

cell generation as the proteome.) If we define ci = C/C as the proportion of all codons of
type i in the proteome and r, as the average translation rate of codons of type i, the total

time required for replication of the proteome (i.e., the proteome generation time) will be

 

gp -.- Zi/r‘ = C/RZCI/ri Equation 1

where R is the average number of ribosomes active in translation over the cell cycle.

Codon bias in favor of rapidly translated codons will reduce gp in comparison to uniform

codon use. If a mutation changes the fitness of an organism from w to w ’, the benefit of
the mutation is typically described as s, where w ’/w = 1 + 3. By analogy, and considering

gp to be inversely related to fitness, we can express the translation rate benefit of codon

bias as

g p (uniform codon use)

bias _

 

g p (biased codon use)

123

The protein content is the same in Uni as in E. coli by hypothesis, and with the restrictive
condition that the abundance of each individual component of the translational apparatus
is unchanged in Uni, ribosome content will be the same also. Hence, the C/R term

cancels from gp in both the numerator and denominator of Sbias’ leading to

c r. ni
sbm(E. coli) =——Z '/ ' (U ) 1 Equation2

ECU/’1- (E. coli) _

 

Since amino acid frequencies are identical in E. coli and Uni, the disparity in translation
rates between synonymous codons largely determines the magnitude of the translation

rate benefit of codon bias.

We will use the same codon—specific translation rates (the ri’s) for both Uni and E. coli,

again invoking the restrictive stipulation that the abundance of each individual tRNA
species is unchanged. If rate constants for the interaction of each codon with each of its
cognate tRNA species were known, we could calculate the optimal tRNA abundance
distribution for the codon frequencies of Uni, and infer the resulting codon-specific
translation rates 3839. However, in vivo codon-specific translation rate data are available
only as codon averages, including translation from all tRNA species cognate to each
codon. Hence, rate constants specific to each codon-cognate tRNA pair cannot be
calculated from the available data for the codons translated by multiple tRNA species,
and we cannot calculate an optimal tRN A abundance distribution for Uni. Insofar as the

codon-specific translation rates measured in E. coli reflect an allocation of tRNA

124

abundance that would be sub-optimal for Uni (as we argue below), using the same codon-
specific translation rates in both cases leads to an overestimate of the benefit of codon

bias, a conservative error for our purposes.

To investigate the influence of comparing E. coli to a biologically realistic standard,
instead of imposing strictly uniform use of synonymous codons on Uni, we also apply
Equation 2 while allowing Uni to display as much codon bias as might be found in an
actual organism with limited codon bias. We take T. pallidum as our example of a low-
bias microbe, since it is a slowly-growing bacterium with little mutational bias (52.7%
G+C), which reportedly lacks evidence of translational selection as well 28. T. pallidum
has the second-most uniform codon use over all predicted genes in the genome (assessed
as Wright’s effective number of codons 40) for 108 bacterial and archaeal species for

which complete genome sequences were available in June, 2003 (data not shown).

Data Sources

For the codon frequencies used in synthesizing the proteome of E. coli, we rely on the
data of Dong et al. at 2.5 dbl hr‘1 41, compiled from public gene sequence databases and
protein abundance data derived from 2D gel electrophoresis studies 42,43. The absolute
frequencies shown in Table 4.1 have been recalculated from 41 with initiation and stop
(including selenocysteine) codons removed. To compare the effect of codon bias in E.
coli to the effect of a small, but biologically plausible, amount of codon bias, we
generated a low bias set of codon frequencies (Table 4.1) that retain the same amino acid

frequencies as E. coli, as well as the same rank order of codon frequency within

125

Table 4.1 footnotes

. Proteome codon frequencies from reference 41 for E. coli growing at 2.5 dbl hrl,
modified slightly as described in Materials and Methods.

. Low bias codon fiequencies representing the degree of codon bias present in the
genome of T. pallidum, generated as described in Materials and Methods.

. Summed abundance of all tRNA species cognate to the listed codon, expressed as a
percentage of total tRNA, based on tRNA abundance data of references 41 and 44 and
cognate specificity of reference 45, modified slightly as described in Materials and
Methods. Values for all codons sum to >100%, a result of the partially overlapping
codon specificity of many tRNA species.

. Empirically determined relative rates of ternary complex selection at the listed codon
from reference 46, expressed relative to the rate of a uniform competing frameshift
event. Rate for codons CGC and CGA modified as described in Materials and
Methods. Estimates of Sbias were made using the rates as listed or with a correction

for the duration of translocation, as described in Materials and Methods.

. Predicted relative translation rates based on the empirical rates of column 6 and
scenarios as described in Materials and Methods. Estimates of Sbias were made using

empirical rates of column 6 for YNN codons in preference to the predicted rates
shown in parentheses; predicted rates are shown for comparison only.

. Predicted relative translation rates from theory of reference 39, modified slightly as
described in Materials and Methods, using the codon frequency data of column 3 and
the cognate tRNA abundance data of column 5.

126

Table 4.1: Codon data

 

 

Codon Frequency tRN A Empirical Predicted Rel. Translation Rates
Codon AA E. colia low biasb Abund.c RCI- Trans.
(x 10-3) (x 10_3) (%) Ratesd Sc. 2c Sc. 36 Sc. 46 Se. 5f

UUU Phe 8.0 9.6 1.5 8.5 (7.6) (6.3) (3.2) 3.7
UUC Phe 23.4 21.8 1.5 12.0 (12.0) (10.8) (10.8) 6.4
UUA Leu 2.8 6.6 2.7 4.3 (5.1) (5.2) (1.9) 7.1
UUG Leu 4.3 17.1 3.8 8.7 (6.0) (6.5) (6.5) 8.9
UCU Ser 16.5 10.7 3.4 11.6 (10.3) (9.5) (9.5) 7.1
UCC Ser 11.8 7.2 1.2 14.7 (9.0) (8.0) (7.3) 6.0
UCA Ser 2.0 5.4 2.1 7.0 (4.6) (3.3) (4.6) 2.5
UCG Ser 2.5 7.1 2.6 9.0 (5.0) (3.7) (0.4) 2.8
UAU Tyr 6.8 11.3 2.7 4.3 (7.2) (6.1) (2.8) 7.3
UAC Tyr 16.6 12.1 2.7 8.4 (10.4) (9.5) (9.5) 11.5
UGU Cys 2.8 3.1 2.1 4.0 (5.2) (5.4) (1.9) 7.6
UGC Cys 3 .8 3.6 2.1 7.0 (5.8) (6.3) (6.3) 8.9
UGG Trp 7.1 7.1 1.5 5.0 (7.3) (7.3) (7.3) 6.4
CUU Leu 3.9 15.2 2.6 8.4 (5.8) (3.9) (9.8) 6.3
CUC Leu 4.1 16.0 1.7 11.0 (5.9) (4.1) (12.0) 6.5
CUA Leu 0.8 4.4 0.9 0.6 (3.6) (0.6) (0.6) 2.9
CUG Leu 61.2 17.9 7.3 14.4 (18.6) (15.7) (15.7) 24.9
CCU Pro 4.4 9.0 1.8 8.4 (6.1) (4.5) (7.3) 2.7
CCC Pro 1.1 6.2 1.1 9.6 (3.9) (2.3) (0.5) 1.3
CCA Pro 5.2 12.1 0.8 1.6 (6.5) (4.9) (9.0) 2.9
CCG Pro 29.0 12.4 1.5 2.5 (13.2) (11.6) (11.6) 6.8
CAU His 6.8 9.0 1.2 4.0 (7.2) (6.3) (2.7) 3.5
CAC His 14.3 12.1 1.2 8.0 (9.7) (9.1) (9.1) 5.1
CAA Gln 7.1 10.7 1.2 5.6 (7.3) (5.8) (3.4) 3.3
CAG Gln 27.5 23.9 2.3 10.0 (12.9) (11.4) (11.4) 6.4
CGU Arg 44.2 21.4 7.5 14.0 (16.0) (13.7) (13.7) 31.3
CGC Arg 20.8 19.1 7.5 11.5 (11.4) (9.4) (10.5) 21.4
CGA Arg 0.7 11.3 7.5 3.0 (3.5) (1.7) (0.6) 3.9
CGG Arg 0.6 5.1 0.6 0.8 (3.4) (1.6) (0.6) 2.6

 

a-f. See facing page for footnotes.

127

Table 4.1: continued

 

 

Codon Frequency tRN A Empirical Predicted Rel. Translation Rates
Codon AA E. colia low biasb Abund.° Rel. Trans.
(x 103) (x 10-3) (%) Ratesd Sc. 26 Se. 3e Sc. 46 Sc. 5*”
AUU Ile 15.9 21.5 6.8 10.2 8.2 4.1 17.1
AUC Ile 44.2 28.0 6.8 16.0 13.7 13.7 28.6
AUA Ile 0.5 11.2 0.3 3.3 1.5 0.6 1.4
AUG Met 21.8 21.8 1.4 11.7 10.5 10.5 5.7
ACU Thr 20.8 15.0 3.8 11.4 9.9 8.7 7.6
ACC Thr 26.9 18.7 1.8 12.8 11.3 11.3 8.7
ACA Thr 2.6 10.0 2.0 5.1 3.5 0.5 2.7
ACG Thr 4.2 10.8 2.9 6.0 4.5 7.1 3.4
AAU Asn 5.7 16.7 2.1 6.7 5.1 3.5 3.9
AAC Asn 29.4 18.4 2.1 13.3 11.7 11.7 8.9
AAA Lys 55.4 39.6 3.1 17.8 15.0 15.0 12.8
AAG Lys 17.4 33.2 1.2 10.6 8.4 4.5 2.9
AGU Ser 2.2 6.8 1.7 4.8 3.8 2.3 3.4
AGC Set 9.4 7.2 1.7 8.2 7.9 7.9 7.0
AGA Arg 0.6 5.4 1.1 3.4 4.9 0.6 3.7
AGG Arg 0.0 4.5 1.7 2.6 1.1 0.6 0.8
GUU Val 43.5 38.8 7.9 15.9 13.6 13.6 23.3
GUC Val 7.7 13.2 2.0 7.5 5.7 0.6 9.8
GUA Val 22.5 20.1 6.0 11.8 9.8 10.5 16.8
GUG Val 15.1 16.8 6.0 9.9 8.0 8.5 13.7
GCU Ala 39.8 38.6 7.2 15.3 13.2 13.2 19.4
GCC Ala 11.9 15.7 1.1 9.0 7.2 0.5 10.6
GCA Ala 25.1 30.6 6.1 12.4 10.4 10.1 15.4
GCG Ala 24.3 16.2 6.1 12.2 10.3 8.2 15.1
GAU Asp 19.4 22.3 4.4 11.1 9.3 3.7 14.0
GAC Asp 34.0 31.1 4.4 14.2 12.4 12.4 18.5
GAA Glu 58.3 39.8 8.5 18.2 15.3 15.3 35.7
GAG Glu 17.1 35.6 3.4 10.5 8.3 4.5 7.7
GGU Gly 45.9 23.7 9.3 16.3 13.9 13.9 21.0
GGC Gly 34.4 23.0 7.3 14.3 12.1 10.7 18.2
GGA Gly 1.3 17.8 1.9 4.1 2.3 0.6 3.5
\ GGG Gly 2.4 19.5 3.2 4.9 3.2 8.7 4.8

 

a-f. See page 126 for footnotes.

128

synonymous groups of codons. The low bias frequencies were generated from relative
codon frequencies over all predicted genes in the complete genome sequence of T.
pallidum 47 (obtained from the website of the National Center for Biological Information,
www.ncbi.nhn.nih.gov). By relative codon frequencies, we mean the absolute frequency
of a codon divided by absolute frequency of the amino acid it encodes. The set of T.
pallidum relative codon frequencies for a particular amino acid were multiplied by the
absolute frequency of that amino acid in the E. coli proteome; the resulting set of absolute
codon fi'equency values were assigned to the codons of that amino acid in the low-bias
set so as to retain the same rank order of codon frequency among synonyms as exists in
the E. coli proteome. For example, the total number of isoleucine codons and the identity
of the 15‘, 2nd and 3rd most common isoleucine codons is the same in the low bias set as
in the E. coli proteome. However, the relative frequencies of the 15‘, 2nd and 3rd most
common isoleucine codons in the low bias set are the same as the relative frequencies of

the 15‘, 2nd and 3rd most common isoleucine codons in the T. pallidum genome.

To represent codon-specific translation rates, we use the relative rate data (the quantity
RtRNA/Rshifi) of Curran and Yarus 46 for the 29 sense codons beginning with U or C
(YNN codons, Y = pyrimidine). Although incomplete, this is by far the largest data set
available for in vivo translational kinetics. The original publication transposed values
reported for two arginine codons, CGC and CGA 48; we have corrected this error. We
also revised the rate measured for CGA downward, to account for interference fiom the
bulky wobble position inosine-adenine base pair in the P site that results from translation

of a CGA codon. Such interference is strongly suggested to slow selection of a ternary

129

complex at the codon subsequent to CGA 48; such an effect would not have been
measured with the experimental system of reference 46, but is appropriate to include as a
codon-specific effect of CGA on translation rate. In the absence of more precise data, we
reduced the rate measured for CGA by a factor of 3, the factor by which CGA reduces
read-through of a following stop codon by a suppressor tRNA in comparison to CGC 48.
This adjustment to the CGA rate brings these results into rough agreement with those of
Sorensen and Pedersen 49, who used an experimental approach that would have detected
a consistent effect of CGA on the translation rate of the subsequent codon, attributing it
to slow translation of CGA itself. The relative rates of reference 46, modified as

described above, are listed in Table 4.1.

The relative rates reported by Curran and Yarus 45 do not reflect the entire translational
cycle, but rather the time required for selecting a cognate ternary complex at an empty,
codon-programmed ribosomal A site, which is believed to occupy the majority of the
elongational cycle 5051. Although peptide bond formation may be very rapid, the time
required for the EF—G-catalyzed translocation of the ribosome to the subsequent codon
(and the associated movement of P- and A-site tRNAs) may not be much shorter than the
time needed for EF-Tu-catalyzed ternary complex selection 50. Hence, in addition to
calculations made using ternary complex selection to represent an entire cycle of
translational elongation (assuming, in effect, that the duration of translocation is
negligible), we also made calculations after modifying the reported rates by adding an
invariant ‘translocation time’ to the variable ‘ternary complex selection time’ for all

codons. The duration of translocation per codon was set at 40% of the average time

130

required to select a ternary complex containing tRNAPhe at a UUU codon, consistent with
the only quantitative measure of translocation rate that has been made in conditions
approximating those in vivo 50. Results from both sets of calculations are presented for
each scenario (described below) that is based on these ternary complex selection rates.
For convenience, hereafter we will refer to the relative rates of reference 46 as translation
rates, rather than using the more accurate but cumbersome expression ‘ternary complex

selection rates’.

To calculate the total abundance of cognate tRNA for each codon, we assign cognate
specificity largely according to Bjork 45, and use the tRNA abundance data from Dong et
al. (at 2.5 dbl hrl) 41. We differ from Bjork only in assuming that the leucine and
glycine tRNAs with uridine in the anticodon wobble position (for which nucleotide
modifications have not been characterized) will read codons ending in U, A and G,
instead of A and G only. This would be the case if the wobble position U is modified to
cmOSU, as is done for each of the other 6 amino acids encoded by a full box of the
translational code (i.e., amino acids for which the four XXN codons are synonyms).
Following Bjork, we assume that 40% of the tRNAs for glutamate, glutamine and lysine
with uridine in the anticodon wobble position are modified to mnm5Ser and thus read
codons ending in A or G; the balance of these tRN A species are assumed to have
mnm5S7-U in the wobble position and read A-ending codons only 45. The abundance of
two pairs of isoaccepting tRNA species (Glnl + Gln2 and Ilel + Ile2) were reported as
sumed values by Dong et al. 41, since these individual species were not separated under

the experimental conditions applied. We have resolved the summed values to the

131

abundance of individual species using the ratios of the individual abundance values as
reported by Ikemura 44. We show cognate tRNA abundance data in Table 4.1 as a
percentage of total tRNA, omitting initiator and selenocysteine tRNAs; the sum of all
values is greater than 100%, reflecting the partially overlapping specificity of many

tRNA species.

Scenarios for applying incomplete empirical data

We address the incompleteness of codon-specific translation rate data in several ways. In
Scenario 1, we assume that the effects of biased use of YNN codons on translation rate
can be used to represent the effects of bias over all codons, without assigning particular
translation rates to the unmeasured codons. However, since the YNN codons are ahnost
half of all sense codons but only account for about a third of all expression (Table 4.1),
they must be less highly expressed, on average, than the RNN codons (R = purine).
Consequently, selection for translational power may have been weaker among YNN
codons than RNN codons. Scenarios 2-4 address this potential deficiency by applying
various strategies of assigning translation rates to the unmeasured codons that are
consistent with observed patterns, but that could allow the effect of codon bias on
translation rate to be greater among RNN codon than YNN codons. Scenario 5 abandons
empirical codon-specific translation rate measurements completely, assigning translation
rates to all codons on the basis of the proteome codon frequency and cognate tRNA
abundance of E. coli, assuming optimality (i.e., maximal translation rate) according to
theory developed by Solomovici et al. 39. For the scenarios based on the empirical rates
of reference 46 (i.e., Scenarios 1-4), we calculate the translation rate benefit of codon bias

in two ways, making different assumptions about the duration of translocation. The two

132

assumptions are either that translocation is essentially instantaneous in comparison to the
duration of ternary complex selection, or that translocation requires a short amount of
time that is the same for all codon-cognate tRNA pairs. For all scenarios, we calculate
the translation rate benefit of codon bias in E. coli in comparison to both uniform
synonymous codon usage and a biologically plausible set of low-bias codon fi'equencies,
derived as described above. The comparisons using the low bias codon frequencies

assume a short, codon-independent duration of translocation.

Scenario 1: The 29 YNN codons encode 10 amino acids, 9 of which have multiple
codons. For 7 of these 9 amino acids, the most common synonym is the codon with the
fastest translation rate. One of the remaining amino acids is serine, for which the two
fastest-translated codons are the two most abundant, although in reverse order, with
relatively small differences between the two in both rate and abundance. Only proline
appears to be anomalous; the 2 most abundant codons encode over 90% of all proline
residues in the proteome 41, but support ternary complex selection about 3.5-fold more
slowly than the 2 least abundant codons 46. It has been suggested 46 that this anomaly
could be adaptive; if proline, because of its unique structure, is found preferentially
between protein domains 52 where slow translation may be important to permit
cotranslational folding 5354. If proline is the only amino acid for which such contrarian
selection pressure is more important than selection for translational power, including
proline codons in a sample intended to represent all codons will lead to an underestimate

of Sbias- Hence, in Scenario 1 we apply Equation 2 over YNN codons, with the

calculated translation time for non-proline YNN codons weighted by a factor of 3.2,

133

which scales the expression level of these codons to the expression level of all non-

proline codons. In other words, we assume the effects of codon bias on translation rate
among the 25 non-proline YNN sense codons are representative of the effects of codon
bias among all 57 non-proline sense codons, whereas the translation rates measured for

proline codons are applied only to themselves.

Scenario 2: Curran and Yarus noted that among highly expressed genes, there is a
significant tendency for rapidly-translated codons to be used frequently, although the
relationship appears to be nonlinear 45. We observe the same pattern comparing their
relative rate data to the proteome codon frequency data of Dong et al. 41 at the highest
growth rate. For non-proline YNN codons, the best fit (R2 = 0.56) of a quadratic
relationship passing through the origin between the codon frequency and translation rate

data of Table 4.1 is cl- = 0.205 ri — 0.522 r,2. We use this equation to predict translation

rates from codon frequency for all RNN codons, as shown in Table 4.1. Since our
objective is to obtain a reasonable estimate the codon—specific translation rate for codons
which have not been measured, not to defend a particular model of the relationship
between codon frequency and translation rate, we make no attempt to justify a quadratic
fit in comparison to other possible functional relationships. The predicted rates for RNN
codons and the measured rates for YNN codons (Table 4.1) are used with Equation 2 to

estimate the translation rate benefit of codon bias under Scenario 2.

Scenario 3: The preceding scenario applied to the YNN codons tends to predict

translation rates among synonymous alternatives that are not as disparate as those

134

actually observed. Furthermore, the fit of a functional relationship between codon
frequency and translation rate among YNN codons is better when only preferred codons
are considered, instead of all codons. Hence, we fit a quadratic relationship passing

through the origin to data from 10 preferred non-proline YNN codons, obtaining
c, = 0.352 rl- -— 1.611 r,2 (R2 = 0.81). Among the 10 preferred codons, we include UGG,

the sole tr'yptophan codon, and UUG, the preferred leucine codon within the UUR split
box although not the preferred leucine codon overall. We then applied this equation to
predict translation rates from codon frequencies for 12 preferred RNN codons, including
AUG, the sole methionine codon, and AGG and AGC, the preferred arginine and serine
codons within their respective split boxes, although not the preferred codons overall. For
non-preferred RNN codons, translation rate is predicted by multiplying the predicted rate
for the preferred synonym (within the full or split box) by the ratio of the square roots of

the codon frequencies for the non-preferred and preferred codons:

V cnonpref

= predicted rpm] —— Equation 3
cpref

predicted r"

onpref

This relationship was chosen both because a dependence on the square root of codon
frequency has been suggested repeatedly in theoretical investigations of optimal
translation rates 38,39,55'57, and because for all non-preferred RNN codons, this
relationship leads to a greater disparity of predicted translation rates compared to the
preferred synonym than the regression of Scenario 2. (It also predicts a greater

translation rate disparity than is observed for the majority of non-preferred YNN codons.)

135

When both the quadratic regression for preferred codons and Equation 3 for non-
preferred codons are applied to predict the translation rate of non-proline YNN codons,
the correlation of predicted with measured translation rates is comparable to that attained
with Scenario 2 (R2 = 0.57). The predicted rates for RNN codons and the measured rates
for YNN codons (Table 4.1) are used with Equation 2 to estimate the translation rate

benefit of codon bias under Scenario 3.

Scenario 4: This scenario is generated in three steps, with the goal of generating an
estimate of the translation rate benefit of codon bias that is consistent with the most
extreme empirical observations. First, three rare RNN codons (AGG and AGA for

arginine and AUA for isoleucine, all with c ,- < 0.1%) are assigned the slowest relative
translation rate observed among YNN codons (r i = 0.6 for the rare leucine codon CUA).

Second, the translation rates for preferred RNN codons within full or split boxes (except
AGG) are estimated according to the regression equation described for Scenario 3.
Finally, the translation rates for non-preferred codons (except AGA and AUA) are
predicted from the preferred synonym using the ratios of the most disparate translation
rates observed empirically among synonymous alternatives, treating split boxes and full
boxes of the translational code separately. The most extreme ratio observed among
translation rates in a split box is 3.375, for glutamate codons in the study of Sorensen and
Pedersen 49. The most extreme ratios observed for translation rates of codons in a full
box is 1:1.3:1.6:24 for the CUN leucine codons in the study of Curran and Yarus 46.
(Exploring other rate values 1 S x S y S 24 in ratios of the form l:x:y:24 failed to find any

that greatly increased the estimated benefit beyond that using the leucine ratios, data not

136

shown.) Although this scenario is based on extreme observations, applying these 3 rules
to the non-proline YNN codons leads to a correlation of predicted and measured
translation rates (R2 = 0.67) somewhat better than that obtained under Scenarios 2 and 3.
The predicted rates for RNN codons and the measured rates for YNN codons (Table 4.1)
are used with Equation 2 to estimate the translation rate benefit of codon bias under

Scenario 4.

Scenario 5: In contrast to the preceding scenarios that extend codon—specific translation
rate measurements of a subset of codons to make an estimate of the translation rate
benefit of codon bias over all codons, Scenario 5 incorporates a theoretical prediction of
optimal translation rates based only on empirical codon fiequency and cognate tRNA
abundance data. Solomovici et al. 39 assume that selection on synonymous codon
frequencies reflects intrinsic differences in rate constants for a cognate tRNA interacting
with preferred and non-preferred codons, while the total tRNA abundance and amino acid
composition are fixed. They demonstrate that the fastest overall translation rate is
obtained when the square roots of synonymous codon frequencies are proportional to the
rate constants for cognate tRN A interacting with the codons. They assume further that
the rate constants for the interaction of all non-degenerate or preferred codons with their
preferred cognate tRN A are identical, so the translation rate for these codons is
proportional to cognate tRNA abundance. We modified the approach of reference 39 to
allow for greater degeneracy in translation 45 (see also the earlier comment regarding the
codon specificity of leucine and glycine tRNAs), and applied it using the codon

frequency and tRNA abundance data of Dong et al. 41, modified as shown in Table 4.1.

137

The predicted relative translation rates for YNN codons (i.e., the recalculated quantities

dij and dimJ- of reference 39 for codons with single or multiple cognate tRNAs,

respectively) are not in good agreement with observed relative rates of Curran and
Yarus 45 (R7- = 0.30). However, the empirical codon frequencies of Dong et al. 41 are
correlated more closely with predicted relative rates of Scenario 5 (R2 = 0.70) than with
the empirical relative rates of Curran and Yarus 46 (R2 = 0.31). A good correlation
between the predicted translation rates and the empirical codon frequencies is expected,
since the latter were used in generating the former. However, the poor correlation
between predicted and empirical translation rates could reflect the inadequacies in 1) the
assumptions of Solomovici et al. 39, 2) the rate measurements of Curran and Yarus 45,
and/or 3) the codon and tRNA data of Dong et al. 41. Alternatively, the phenotype of E.
coli may not be perfectly optimized for maximal translation rates 33, either because of
genetic drift or because of conflicting selection pressures. Nonetheless, the disparity
between the relative rates of synonymous preferred and non-preferred codons for most
amino acids are greater with the predicted rates of Scenario 5 than with the observed
rates. Hence, Scenario 5 will overestimate the translation rate benefit of codon bias
compared to a strict application of the empirical codon-specific translation rates. (None
of our scenarios are, in fact, strict applications of the empirical rates. Scenarios 1-4
extend the empirical rates in ways that will increase the estimated benefit of codon bias.)
The predicted translation rates for all codons (Table 4.1) are used with Equation 2 to

estimate the translation rate benefit of codon bias under Scenario 5.

138

Chapter 4 Results

As expected, the translation rate benefit of codon bias increases monotonically with
growth rate, when calculated by any of the scenarios described in Materials and Methods,
using proteome codon frequencies and tRNA abundance data from the range of growth
rates reported in reference 41 (data not shown). This increase reflects simply the
increasing bias in both proteome codon usage and relative tRNA abundance with
increasing grth rate. Since we are interested in the maximum effect of codon bias, we
report results from only the highest growth rate for which data are available, 2.5

doublings per hour.

We define sbias, the translation rate benefit of codon bias in E. coli, as the fractional

increase in the time required to replicate the E. coli proteome if the actual codon bias in
E. coli were replaced with uniform use of synonymous codons (Equation 2). Estimates of

Sbias in E. coli according to several scenarios (described in Materials and Methods) are

presented in Figure 4.1. All scenarios are based on the data shown in Table 4.1, but use
different assumptions for extending incomplete codon-specific relative translation rate
data to estimate the benefit of bias over all codons (Scenarios 1-4), or use a theoretical
prediction of optimal codon-specific translation rates based on reasonably complete
codon frequency and cognate tRNA abundance data (Scenario 5). Since the empirical
rate data used in Scenarios 1-4 do not account for translocation, for these scenarios two
estimates are shown that make different assumptions regarding the relative duration of

translocation and ternary complex selection. The estimated values Obeias range from

0.6 — 1.4 if translocation time is neglected, or from 0.4 — 1.1 with the more realistic

139

Figure 4.1: Translation rate benefit of codon bias in E. coli

Benefit of codon bias (shias)

 

Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5

Figure 4.1. The estimated translation rate benefit of codon bias in E. coli, using 5
difl’erent scenarios to obtain an complete set of relative codon—specific translation rates
from incomplete empirical data (see Materials and Methods for explanation). White bars:
duration of translocation assumed to be negligible in comparison to the duration of
ternary complex selection. Cross-hatched bars: duration of translocation assumed to be
constant for all codons and short in comparison to the duration of ternary complex
selection. Both white and cross-hatched bars: benefit of codon bias in E. coli estimated
in comparison to strictly uniform codon use. Black bars: duration of translocation
assumed to be constant for all codons and short in comparison to the duration of ternary
complex selection; benefit of codon bias in E. coli estimated in comparison to degree of
codon bias found in an actual low-bias organism. The white, cross-hatched and black
bars represent a series of increasingly realistic estimates.

140

assumption that translocation requires a short amount of time. In a further attempt to

introduce biological reality to our calculations, we also estimated Sbias for E. coli, not in

comparison to strictly uniform synonymous codon use, but in comparison to the limited
degree of codon bias that might be found in an actual low-bias organism. In this case, the
estimated benefits are lower still, ranging from 0.2 — 0.6, again assuming a short duration

of translocation The estimates of Sbias from the more theoretically-based Scenario 5

(using codon frequency and tRNA abundance data) fall into the middle of the range of

Sbias estimates from the more empirically-based Scenarios 1-4 (using codon frequency

and codon—specific translation rate data). Hence, we are confident that our conclusions
are not overly sensitive to specific details of the empirical translation rate measurements

or of the methods used to extend the empirical measurements to unmeasured codons.

Our definition of Sbias can be applied over any subset of codons, in particular, it can be

applied to the codons of each amino acid separately. While all amino acids with multiple

codons except proline contribute positively to Sbias in all scenarios, the magnitude of that

contribution is highly variable between amino acids (Figure 4.2). Codon bias accelerates
the translation of most amino acids only slightly in E. coli, because most non-preferred
codons are not particularly rare in the E. coli proteome, compared to the preferred
synonym. For example, among the 9 amino acids encoded by 2 codons, there is an
average 2.9-fold difference in the frequency of preferred and non-preferred codons;
asparagine shows the greatest difference with GAC being 5.2-fold more abundant than
GAU. Even if the disparity in translation rates is unrealistically large, the ratio of

preferred to non-preferred codons in E. coli constrains the maximum possible value of

141

Sbias' For asparagine, even if the preferred codon were translated instantaneously (i.e.,

infinitely faster than the non-preferred codon), the difference between 50% usage of the
slowly translated codon in Uni compared to 16% usage in E. coli corresponds to only

about a 3-fold acceleration of translation (Sbias 2 2) for this amino acid. With more

realistic disparities in the translation rates of the preferred and non-preferred codons, the

largest estimate of Sbias for asparagine in any of our scenarios is less than 0.2, i.e., codon

bias leads to a 20% increase in translation rate over all asparagine codons (Figure 4.2).

The amino acids making the largest contribution to Sbias are leucine, isoleucine, and

arginine (Figure 4.2). Although these amino acids are not rare, they possess between
them the six rarest codons in E. coli, each encoding less than 0.1% of the proteome. The
disparities in frequency between the most and least abundant synonym for leucine,
isoleucine and arginine are 74—fold, 83-fold, and 1460-fold, respectively. (The ratio for
arginine reflects the extreme rarity of AGG, which is 17-fold less abundant than the
second rarest E. coli codon, AUA encoding isoleucine.) Since the translation rates
measured or assumed for these rare codons are quite slow, their increased abundance in
Uni accounts for the much of the additional time required for replicating the Uni
proteome. If these six codons were as rare in Uni as they are in E. coli, while all other
synonymous codons were used without bias, the translation rate benefit estimated under
Scenario 4 (the scenario producing the largest benefit estimates) would be reduced by
almost half (data not shown). The influence of these 6 codons is such that the estimate of

Sbias is quite sensitive to the translation rates assigned to them, in contrast to the relative

insensitivity of Sbias to the exact translation rates assigned to most codons.

142

Figure 4.2: Translation rate benefit of codon bias by amino acid

Figure 4.2. The translation rate benefit of codon bias in E. coli, estimated separately for
each amino acid, is shown by the ordinate; the frequency of the amino acid in the E. coli
proteome is shown by the abscissa. Each amino acid is represented by its one-letter
abbreviation. Panels a-e represent Scenarios 1-5 respectively, described in Materials and
Methods. The estimate of Sbias is made in comparison to uniform use of synonymous
codons, and assumes that the duration of translocation is negligible (corresponding to the
first set of bars in Figure 4.1). Only a few amino acids, those encoded by one or more
rare codons, contribute disproportionately to the total translation rate benefit of codon
bias.

143

s bias by ammo acrd

SDI-as by amino acrd

Sb,-as by amino acrd

Figure 4.2: Translation rate benefit of codon bias by amino acid

 

 

 

 

 

 

4-

a R
3‘ L
2—4
1-
o_ W l-K flap
'1 I I I I I I
4— 9 R
3‘ L
24
14 I G

T

o— w 161‘ Pups D KE V A
'1 I I I I I I
5-

E
4- R
3_ L
2_ l

G
1.—
T
o— W W FdPS D KE V A
'1 I I I I I I
20 40 60 80 100

Codon frequencies (x 103)

144

120

s bias by ammo acrd

s bias by ammo acrd

 

ID.

Figure 4.2: continued

 

 

IGD

P
T
NKFuSD KE

V

A

 

l l l

l

20 40 60 80

Codon frequencies (x 10

145

-3)

I
100

120

Chapter 4 Discussion

We want to know whether reduced codon bias could account for the lower translational
power measured in at least some slowly growing bacteria, compared to E. coli. We
approach this issue by its converse, calculating the acceleration of the translation rate due
to codon bias in E. coli. If we take our estimates at face value, we would conclude that

even during rapid growth when the proteome is most biased, Sbias is unlikely to be as

large as 1, corresponding to a doubling of the average translation rate. However, there
are two reasons to think that the benefit of codon bias for E. coli, in comparison to most

actual slow-growing organisms, is even less than this.

The first reason is that we have prevented our hypothetical Uni from adapting to the
codon frequencies we have assigned to it, by keeping the abundance of each component
of the translational apparatus fixed. The data do not suggest that maximizing
translational power has been the only selective pressure influencing codon use in

E. coli 38,46. If it had been, the codon with the highest rate constant for ternary complex
selection among synonymous alternatives would always be the preferred codon, since it
would permit faster translation with a lower investment in cognate tRNA. Of 10 amino
acids with multiple codons for which codon-specific translation rate measurements

exist 46,58, leucine, serine and proline are not consistent with this prediction. On the
other hand, it seems clear that selection for rapid translation has exerted some, and
perhaps the major influence on the evolution of codon frequencies and tRN A abundance
in E. coli. The codon with the highest rate constant is the preferred codon for 7 of the 10

amino acids for which data are available. Other considerations (possibly including error

146

avoidance 17, interactions between adjacent tRNA anticodons 59, or factors unrelated to
translation 60) may have been more influential than the inherent characteristics of the
codon-anticodon interactions for determining the preferred codons encoding leucine,
serine and proline. However, the importance of rapid translation remains evident in that
E. coli still translates the preferred codons quickly for 2 of these 3 amino acids, albeit
with a larger investment in tRNA than would be necessary if the interaction between the

preferred codon and its cognate tRNA occurred more readily.

At a larger scale, the correlation across all codons between frequency and cognate tRN A
abundance 41,44 is best explained as a response to selection for rapid translation, as is the
pattern of increased bias towards rapidly translated codons with increased gene
expression 46. Without asserting that the distribution of tRNA abundance in E. coli
necessarily produces the fastest possible translation rate for the E. coli codon frequency
distribution, it is clear that selection for translational power has been a significant factor
in the coevolution of codon frequencies and cognate tRNA abundances in E. coli. Thus,
it is very unlikely that the distribution of tRNA abundance values in E. coli, and the
resulting codon-specific translation rates, produces the fastest possible translation rate
when matched with different codon frequencies in Uni. For this reason, our estimates
confound the translation rate benefit of codon bias in E. coli with the penalty of a

suboptimal allocation of translational resources in Uni.

The second reason that the benefit of codon bias in E. coli is unlikely to be as large as our

estimates, when the comparison is made to actual slow-growing organisms, is that actual

147

microbes are not completely devoid of codon bias. Assessing Sbias in E. coli in

comparison to a biologically plausible standard for low codon bias, instead of the
implausible standard of no codon bias whatsoever, reduces the estimated benefit in E.
coli by about half (Figure 4.1). The low bias codon frequencies derived from T. pallidum
can accelerate translation about half as much as the more highly biased E. coli codon
frequencies because only a few codons in E. coli are translated much more slowly than
the median rate. Moderate avoidance of these few codons can provide a considerable
acceleration of the average translation rate without generating a dramatic bias in overall

codon use.

Our estimate of biologically plausible low-bias codon frequencies is deliberately
conservative, underestimating the degree of bias expected in most slowly growing
microbes, for two reasons. First, we used the genome codon frequencies of T. pallidum,
as if all predicted genes in the genome were expressed equally. Although analysis of
codon use at the level of individual genes failed to uncover evidence of translational
selection in T. pallidum 28, calculating codon frequencies over a set of putative high
expression genes shows that codon use in such genes is, indeed, more biased than codon
use in the genome as a whole. This conclusion is based on a comparison of Wright’s
effective number of codons (Ne) 40 for codon frequencies summed over ribosomal
proteins and translation elongation factors (Nc = 52.7) or for codon frequencies summed
over all predicted genes in the genome (Nc = 55.2). The failure to observe this low level
of codon bias in the earlier analysis 23 can probably be attributed to high variability in

estimates of codon frequencies derived from small samples of codons in each gene.

148

Thus, we believe that even for T. pallidum, proteome codon frequencies will be more
biased than the low-bias codon frequencies derived from the T. pallidum genome that
were used to generate the estimates shown in Figure 4.1. Second, the choice of T.
pallidum to represent slow-growing strains is conservative, because its codon bias is
essentially free of the influence of mutational bias, with a genome G+C content of 52.7%.
In contrast, many slow-growing microbes have more extensive codon bias that can be
attributed mostly or entirely to biased nucleotide composition (e.g., R. prowazekii 35,
Helicobacter pylori 37, Borrelia burgdorferi 27,28, Buchnera aphidicola 61, Mycoplasma
genitalium 25, and Chlamydia species 29,52). If codon bias derived from biased
nucleotide composition, like codon bias derived from translational selection, permits
more rapid translation, the use of low bias codon frequencies derived from T. pallidum

will underestimate the translation rate of many slow growing strains.

The advantage of codon bias depends to a large extent on matching the preferred codons
with abundant cognate tRNAs. Even if codon use is determined by mutational bias in the
DNA replication and repair systems 63, not by selection acting simultaneously on
individual codons and their cognate tRNAs via translation-associated effects, selection
for translational power can influence the relative abundance of tRNA species. Relatively
few mutations may suffice to influence the relative tRNA abundance in an organism, in
comparison to the number of mutations required to influence proteome codon
frequencies. (Consider that ~40 mutations could allow a single mutation in the regulatory
region of many or even all tRNA genes, depending on the organism, but could alter the

identity of less than 0.5% of the >9,000 codons in genes encoding ribosomal proteins and

149

translational elongation factors.) Hence, the mutation-selection balance argument
invoked to explain diminished codon bias in genes expressed at low levels in many
strains 1534 also suggests that the tRNA abundance distribution can be influenced by
translational selection that may be too weak to create a dramatic effect on codon usage.
In fact, if codon use is highly biased in the same direction in all genes (as expected if the
source of codon bias is mutational bias), instead of being biased only in highly expressed
genes, it increases the selective pressure for adaptation of the tRNA pool. Hence, it
would be very surprising if the anticodon sequences and the tRNA abundance distribution
in organisms with high or low G+C content did not reflect their biased use of codons.
This prediction is confirmed by the only two studies of tRNA abundance in microbes
with extreme G+C content, involving Mycoplasma capricolum (25% G+C) 54 and
Micrococcus luteus (74% G+C) 65. M. capricolum, but not M. luteus, can be considered
a constitutively slow-growing strain. As expected, cognate tRNA abundance in both
organisms is correlated with codon frequency, both across all codons and within
synonymous codon families. For M. capricolum, this is accomplished largely without the
tRNA gene dosage effects that are important for E. coli 41 and B. subtilis 66, since 28 of
the 29 M. capricolum tRNA genes are present in only a single copy 64. These examples
indicate that selection for translational power is operative even for organisms in which
the codon bias is determined by mutational bias instead of translational selection, and
even if the organisms are slow growers. Because codon bias from any source can be

exploited to obtain higher translational power, the estimates of sbm for E. coli compared

to codon frequencies derived from T. pallidum will overstate the benefit expected in

comparison to most other slowly growing microbes.

150

In summary, we believe the translation rate benefit of codon bias in E. coli is likely to be
less than 0.6 when the comparison is made to an actual slow-growing organism that
shows limited codon bias, such as T. pallidum, and substantially less than 0.6 in
comparison to a slow-growing organism with more extensive codon bias, regardless of its
source. We do not mean to suggest that an increase of up to 60% in the overall
translation rate is unimportant. Clearly, the aggregate benefit of codon bias must be
substantial, considering that many thousands of preferred codons are stably maintained in
the E. coli genome, despite the randomizing influence of mutation acting at each

individual codon.

On the other hand, the influence of codon bias on the average translation rate is far
smaller than the differences in translational power observed between microbes pursuing
different ecological strategies. In order to explain the difference in translational power
between rapidly and slowly responding bacteria growing in R2BV medium (Chapter 3),

Sbias would have to be ~1.4, and in order to explain the comparison of translational power
made at similar levels of RNA investment, Sbias would have to be ~2.6. The latter value

also corresponds to about the middle of the range of comparisons of translational power
between E. coli and slowly growing microbes presented in Chapter 1. Our estimate of

the maximum value of Sbias in E. coli is less than half of the lowest of these estimates of

translational power differences. Hence, the parsimonious explanation that known
differences in the degree of codon bias in the proteome account for observed differences

in translational power between microbes of contrasting ecological strategies is not

151

plausible. Instead, the macromolecular data that demonstrate differences in translational
power must be interpreted as evidence for some more fundamental difference in
translational performance. The hypothesis proposed in Chapter 3, that there is an
evolutionary tradeoff between translational power and translational yield mediated by

processivity errors, is one example of such a fundamental difference.

 

152

10.

11.

Chapter 4 References

Kjeldgaard, NO. and C .G. Kurland, 1963. The distribution of soluble and
ribosomal RNA as a function of growth rate. Journal of Molecular Biology,
6:341-348.

Maalee, O., 1979. Regulation of the protein-synthesizing machinery - ribosomes,
tRNA, factors, and so on; pp. 487-542 in Gene Expression; R.F. Goldberger, ed.
Plenum Press, New York.

Ingraham, J .L., O. Maaloe and RC. Neidhardt, 1983. Growth of the Bacterial
Cell. Sinauer, Sunderland, Massachusetts.

Ehrenberg, M. and CG. Kurland, 1984. Costs of accuracy determined by a
maximal growth rate constraint. Quarterly Reviews of Biophysics, 17(1):45-82.

Koch, AL. and CS. Deppe, 1971. In vivo assay of protein synthesizing capacity
of Escherichia coli from slowly growing chemostat cultures. Journal of Molecular
Biology, 55:549-562.

Bremer, H. and PP. Dennis, 1996. Modulation of chemical composition and other
parameters of the cell by growth rate; pp. 1553-1569 in Escherichia coli and
Salmonella: Cellular and Molecular Biologi; F.C. Neidhardt, et al., eds. ASM
Press, Washington, D. C.

Wada, A., K. Igarashi, S. Yoshimura, S. Aimoto and A. Ishihama, 1995.
Ribosome modulation factor - stationary growth phase-specific inhibitor of
ribosome functions from Escherichia coli. Biochemical and Biophysical Research
Communications, 214(2):410-417.

Chant, J ., I. Hui, D. Dejongwong, L. Shimmin and PP. Dennis, 1986. The protein
synthesizing machinery of the Archaebacterium Halobacterium cutirubrum -
molecular characterization. Systematic and Applied Microbiology, 7(1):106-114.

Pang, H.L. and H.H. Winkler, 1994. The concentrations of stable RNA and
ribosomes in Rickettsia prowazekii. Molecular Microbiology, 12(1):115-120.

Fegatella, F., J. Lim, S. Kjelleberg and R. Cavicchioli, 1998. Implications of
rRNA operon copy number and ribosome content in the marine oligotrophic

ultramicrobacterium Sphingomonas sp. strain R3225 6. Applied and
Environmental Microbiology, 64(11):4433-4438.

Shahab, N., F. Flett, S.G. Oliver and PR. Butler, 1996. Growth rate control of
protein and nucleic acid content in Streptomyces coelicolor A3 (2) and
Escherichia coli B/r. Microbiology-UK, 142:1927-1935.

153

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

Sharp, RM. and W.H. Li, 1986. An evolutionary perspective on synonymous
codon usage in unicellular organisms. Journal of Molecular Evolution, 24(1-
2):28-38.

Sharp, P.M., E. Cowe, D.G. Higgins, D.C. Shields, K.H. Wolfe and F. Wright,
1988. Codon usage patterns in Escherichia coli, Bacillus subtilis, Saccharomyces
cerevisiae, Schizosaccharomyces pombe, Drosophila melanogaster and Homo

sapiens - a review of the considerable within species diversity. Nucleic Acids
Research, 16(17):8207-821 1.

Akashi, H. and A. Eyre-Walker, 1998. Translational selection and molecular
evolution. Current Opinion in Genetics & Development, 8(6):688-693.

Sharp, P.M., M. Stenico, J .F. Peden and A.T. Lloyd, 1993. Codon usage -
mutational bias, translational selection, or both? Biochemical Society
Transactions, 21(4):835-841.

Andersson, S.G.E. and CG. Kurland, 1990. Codon preferences in free-living
microorganisms. Microbiological Reviews, 54(2): 1 98-2 1 0.

Parker, J ., T. Johnston, P. Borgia, G. Holtz, E. Remaut and W. Fiers, 1983. Codon
usage and mistranslation - in vivo basal level misreading of the MSZ coat protein
message. Journal of Biological Chemistry, 258(16):10007-10012.

Yanofsky, C., 1987. Operon-specific control by transcription attenuation. Trends
in Genetics, 3(12):356-360.

Takano, E., M. Tao, F. Long, M.J. Bibb, L. Wang, W. Li, M.J. Buttner, Z.X.
Deng and K.F. Chater, 2003. A rare leucine codon in adpA is implicated in the

morphological defect of bldA mutants of Streptomyces coelicolor. Molecular
Microbiology, 50(2):475-486.

Dobrindt, U. and J. Hacker, 2001. Regulation of the tRNA (5)(Leu)-encoding gene
IeuX that is associated with a pathogenicity island in the uropathogenic
Escherichia coli strain 536. Molecular Genetics and Genomics, 265(5):895-904.

Sharp, RM. and W.H. Li, 1987. The rate of synonymous substitution in

Enterobacterial genes is inversely related to codon usage bias. Molecular
Biology and Evolution, 4(3):222-230.

Sharp, RM. and W.H. Li, 1986. Codon usage in regulatory genes in Escherichia
coli does not reflect selection for rare codons. Nucleic Acids Research,
14(19):7737-7749.

Kurland, CG, 1993. Major codon preference - theme and variations.
Biochemical Society Transactions, 21(4):841-846.

154

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

Buhner, M., 1991. The Selection-Mutation-Drift theory of synonymous codon
usage. Genetics, 129:897-907.

Kerr, A.R.W., J .F. Peden and RM. Sharp, 1997. Systematic base composition
variation around the genome of Mycoplasma genitalium, but not Mycoplasma
pneumoniae. Molecular Microbiology, 25(6):1177-1 179.

Frank, AC. and J .R. Lobry, 1999. Asymmetric substitution patterns: A review of
possible underlying mutational or selective mechanisms. Gene, 238(1):65-77.

McInemey, J .O., 1998. Replicational and transcriptional selection on codon
usage in Borrelia burgdorferi. Proceedings of the National Academy of Sciences
of the United States of America, 95(18):10698-10703.

Lafay, B., A.T. Lloyd, M.J. McLean, K.M. Devine, P.M. Sharp and K.H. Wolfe,
1999. Proteome composition and codon usage in spirochaetes: Species-specific
and DNA strand-specific mutational biases. Nucleic Acids Research, 27(7):1642-
1649.

Romero, H., A. Zavala and H. Musto, 2000. Codon usage in Chlamydia
trachomatis is the result of strand-specific mutational biases and a complex
pattern of selective forces. Nucleic Acids Research, 28(10):2084-2090.

Andersson, S.G.E. and RM. Sharp, 1996. Codon usage in the Mycobacterium
tuberculosis complex. Microbiology-UK, 142:915-925.

Shields, DC. and RM. Sharp, 1987. Synonymous codon usage in Bacillus subtilis
reflects both translational selection and mutational biases. Nucleic Acids
Research, 15(19):8023-8040.

Wright, F. and M.J. Bibb, 1992. Codon usage in the G+C-rich Streptomyces
genome. Gene, 113(1):55-65.

Ohama, T., A. Muto and S. Osawa, 1990. Role of GC-biased mutation pressure

on synonymous codon choice in Micrococcus luteus, a bacterium with a high
genomic GC content. Nucleic Acids Research, 18(6):1565-1569.

Ohkubo, S., A. Muto, Y. Kawauchi, F. Yamao and S. Osawa, 1987. The
ribosomal protein gene cluster of Mycoplasma capricolum. Molecular & General
Genetics, 210(2):314-322.

Andersson, S.G.E. and RM. Sharp, 1996. Codon usage and base composition in
Rickettsia prowazekii. Journal of Molecular Evolution, 42(5):525-536.

Shields, DC, 1990. Switches in species-specific codon preferences - the influence
of mutation biases. Journal of Molecular Evolution, 31(2):71-80.

155

37.

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

Lafay, B., J .C. Atherton and RM. Sharp, 2000. Absence of translationally
selected synonymous codon usage bias in Helicobacter pylori. Microbiology-UK,
146:851-860.

Berg, O.G. and CG. Kurland, 1997. Growth rate-optimised tRNA abundance and
codon usage. Journal of Molecular Biology, 270(4):544-550.

Solomovici, J ., T. Lesnik and C. Reiss, 1997. Does Escherichia coli optimize the

economics of the translation process? Journal of Theoretical Biology, 185(4):511-
521.

Wright, F., 1990. The effective number of codons used in a gene. Gene, 87(1):23-
29.

Dong, H.J., L. Nilsson and CG. Kurland, 1996. Co-variation of tRNA abundance
and codon usage in Escherichia coli at difi’erent growth rates. Journal of
Molecular Biology, 260(5):649-663.

Vanbogelen, R.A., P. Sankar, R.L. Clark, J .A. Bogan and F .C. Neidhardt, 1992.
The gene-protein database of Escherichia coli - edition 5. Electrophoresis,
13(12):1014-1054.

Pedersen, 8., PL. Bloch, S. Reeh and RC. Neidhardt, 1978. Patterns of protein
synthesis in Escherichia coli - catalog of amount of 1 40 individual proteins at
difl'erent growth rates. Cell, 14(1):179-190.

Ikemura, T., 1981. Correlation between the abundance of Escherichia coli
transfer RNAs and the occurrence of the respective codons in its protein genes - a
proposal for a synonymous codon choice that is optimal for the Escherichia coli
translational system. Journal of Molecular Biology, 151(3):389-409.

Bjork, GR, 1996. Stable RNA modification; pp. 861-886 in Escherichia coli and
Salmonella: Cellular and Molecular Biology; F .C. Neidhardt, et al., eds. ASM
Press, Washington, D. C.

Curran, J .F. and M. Yarus, 1989. Rates of aminoacyl-tRNA selection at 29 sense
codons in vivo. Journal of Molecular Biology, 209(1):65-77.

Fraser, C.M., SJ. Norris, G.M. Weinstock, 0. White, G.G. Sutton, R. Dodson, M.
Gwinn, E.K. Hickey, R. Clayton, K.A. Ketchum, E. Sodergren, J .M. Hardham,
M.P. McLeod, S. Salzberg, J. Peterson, H. Khalak, D. Richardson, J .K. Howell,
M. Chidambararn, T. Utterback, L. McDonald, P. Artiach, C. Bowman, M.D.
Cotton, C. Fujii, S. Garland, B. Hatch, K. Horst, K. Roberts, M. Sandusky, J.
Weidman, H.O. Smith and J .C. Venter, 1998. Complete genome sequence of

T reponema pallidum, the syphilis spirochete. Science, 281(5375):375-388.

Curran, J .F., 1995. Decoding with the A-I wobble pair is inefficient. Nucleic Acids
Research, 23(4):683-688.

156

49.

50.

51.

52.

53.

54.

55.

56.

57.

58.

59.

60.

Sorensen, M.A. and S. Pedersen, 1991. Absolute in vivo translation rates of
individual codons in Escherichia coli - the 2 glutamic acid codons GAA and GAG

are translated with a threefold diflerence in rate. Journal of Molecular Biology,
222(2):265-280.

Bilgin, N., L.A. Kirsebom, M. Ehrenberg and C.G. Kurland, 1988. Mutations in
ribosomal proteins L 7/L12 perturb EF-G and EF - T u functions. Biochimie,
70(5):611-618.

Schillingbartetzko, S., A. Bartetzko and K.H. Nierhaus, 1992. Kinetic and
thermodynamic parameters for transfer RNA binding to the ribosome and for the
translocation reaction. Journal of Biological Chemistry, 267(7):4703-4712.

Thanaraj, TA. and P. Argos, 1996. Ribosome-mediated translational pause and
protein domain organization. Protein Science, 5(8):1594-1612.

Cortazzo, P., C. Cervenansky, M. Marin, C. Reiss, R. Ehrlich and A. Deana,
2002. Silent mutations affect in viva protein folding in Escherichia coli.
Biochemical and Biophysical Research Communications, 293(1):537-541.

Komar, A.A., T. Lesnik and C. Reiss, 1999. Synonymous codon substitutions
aflect ribosome trafi‘ic and protein folding during in vitro translation. F EBS
Letters, 462(3):387-391.

von Heinje, G., 1979. The concentration dependence of the error frequencies and
some related quantities in protein synthesis. Journal of Theoretical Biology,
78:1 13-120.

Garel, J .P., 1974. Functional adaptation of tRNA population. Journal of
Theoretical Biology, 43:21 1-225.

Xia, X., 1998. How Optimized Is the Translational Machinery in Escherichia coli,
Salmonella typhimurium and Saccharomyces cerevisiae? Genetics, 149237-44.

Sorensen, M.A., C.G. Kurland and S. Pedersen, 1989. Codon usage determines
translation rate in Escherichia coli. Journal of Molecular Biology, 207(2):365-
377.

Smith, D. and M. Yarus, 1989. Transfer RNA-transfer RNA interactions within
cellular ribosomes. Proceedings of the National Academy of Sciences of the
United States of America, 86(12):4397-4401.

Antezana, M.A. and M. Kreitrnan, 1999. The nonrandom location of synonymous
codons suggests that reading frame-independent forces have patterned codon
preferences. Journal of Molecular Evolution, 49(1):36-43.

157

61.

62.

63.

65.

66.

Wemegreen, J .J . and NA. Moran, 1999. Evidence for genetic drift in

endosymbionts (Buchnera): Analyses of protein-coding genes. Molecular Biology
and Evolution, 16(1):83-97.

Dalevi, D.A., N. Eriksen, K. Eriksson and S.G.E. Andersson, 2002. Measuring
genome divergence in bacteria: A case study using Chlamydian data. Journal of
Molecular Evolution, 55(1):24-36.

Sueoka, N., 1988. Directional mutation pressure and neutral molecular evolution.
Proceedings of the National Academy of Sciences of the United States of
America, 85(8):2653-2657.

Yamao, F ., Y. Andachi, A. Muto, T. Ikemura and S. Osawa, 1991. Levels of
transfer RNAs in bacterial cells as affected by amino acid usage in proteins.
Nucleic Acids Research, 19(22):6119—6122.

Kano, A., Y. Andachi, T. Ohama and S. Osawa, 1991. Novel anticodon
composition of transfer RNAs in Micrococcus luteus, a bacterium with a high

genomic G+C content - correlation with codon usage. Journal of Molecular
Biology, 221 (2):387-401.

Kanaya, S., Y. Yamada, Y. Kudo and T. Ikemura, 1999. Studies of codon usage
and tRNA genes of18 unicellular organisms and quantification of Bacillus
subtilis tRNAs: Gene expression level and species-specific diversity of codon
usage based on multivariate analysis. Gene, 238(1):143-155 .

158

Appendix. Phylogeny of bacteria used with log likelihood
ratio tests of correlated trait evolution

Full length small subunit ribosomal RNA (16S) gene sequences were obtained either
from the Comprehensive Microbial Resource 1 of The Institute for Genomic Research
(www.tigr.org) or from GenBank (www.ncbi.nlm.nih. gov) for 78 completely sequenced
bacterial genomes (Table A.1); the operon labeled ‘A’ or ‘1’ was chosen from genomes
containing multiple operons. Sequences were imported into Arb software 2 and aligned
automatically to related full length sequences obtained from release 8.0 of the Ribosomal
Database Project 3; initial alignments were optimized manually with reference to
secondary structure information available at the Comparative RNA Website,
www.ma.icmb.utexas.edu 4. Additional near firll length sequences from RDP release 8.0
were included in the phylogenetic analysis to ensure that all deeply branching bacterial
lineages were represented by at least 3 sequences 5. A mask was created to exclude
positions of ambiguous alignment, leaving 1250 positions for phylogenetic inference
using the maximum likelihood algorithm within Arb. The resulting tree containing 166
bacterial and archaeal taxa has been ‘pruned’ without altering branching order or branch
lengths to obtain the phylogenic hypothesis displayed in Figure A.1a, which is rooted

with archaeal sequences that are not shown on the tree.

To ensure that poorly resolved details of the bacterial phylogeny did not influence results
of the comparative analyses, a second tree was derived in which short internal branches

corresponding to 8 or fewer inferred nucleotide changes (in 1250 positions) were

159

collapsed, depicted in Figure A. 1b. The choice of 8 changes as the threshold for collapse
was based on a comparison with the assessment in reference 5 of the reliability of
maximum likelihood rRNA-based phylogenetic inference for reconstructing ancient
events in bacterial evolution. Collapsed branches were assigned the minimal nonzero
length since the Continuous software program used for our analysis requires a bifurcating
tree; these collapsed lengths represented less than 0.01% of a single nucleotide change
between sequences. For both descriptions of bacterial phylogeny, branches of zero length
at terminal nodes (i.e., separating strains with identical 16S gene sequences at the 1250
positions analyzed) were changed to a length representing half a nucleotide change

between sequences.

Of the 78 bacterial genome sequences depicted on the phylogenetic trees of Figure A. 1,
codon data was not available for Mycoplasma pulmonis and Ureaplasma urealyticum,
and tRNA gene annotation was not available for Streptococcus pneumoniae R6 and
Pseudomonas putida. Hence, 76 taxa were analyzed for the strength of translational
selection either using AN 'c, based on codon use (Figure 3.3a), or using tRNA gene copy

number (Figure 3.3b).

160

Table A.1: Bacterial taxa analyzed for translational selection

Agrobacterium tumefaciens C58 Cereon
Agrobacterium tumefaciens C58 Uwash
Aquifex aeolicus VF5

Bacillus halodurans 0-125

Bacillus subtilus 168

Borrelia burgdorferi B31

Brucella melitensis 16M

Brucella suis 1330

Buchnera aphidicola Sg

Buchnera APS

Campylobacterjejuni NCTC1 1 168
Caulobacter crescentus 0B15
Chlamydia muridarum strain Nigg
Chlamydia pneumoniae AR39
Chlamydia pneumoniae CWL029
Chlamydia pneumoniae J138
Chlamydia trachomatis serovar D
Chlorobium tepidum TLS

Clostridium acetobutylicum ATCC 824
Clostridium perfringens 13
Corynebacterium glutamicum AT0013032
Deinococcus radiodurans R1
Enterococcus faecalis V583
Escherichia coli K12-MG1655
Escherichia coli O157:H7 EDL933
Escherichia coli O157:H7 VT2-Sakai
Fusobacterium nucleatum ATCC 25586
Haemophilus influenzae KW20
Helicobacter pylori 26695

Helicobacter pylori J99

Lactococcus lactis subsp. Lactis |L1403
Listeria innocua CLIP 11262

Listeria monocytogenes EGD-e
Mesorhizobium Ioti MAF303099
Mycobacterium leprae TN
Mycobacterium tuberculosis 0001551
Mycobacterium tuberculosis H37 Rv
Mycoplasma genitalium G37
Mycoplasma pneumoniae M129

Mycoplasma pulmonis UAB CTIP
Neisseria meningitidis M058

Neisseria meningitidis serogroup A Z2491
Nostoc sp. PCC 7120

Oceanobacillus iheyensis HTE831
Pasture/Ia multocida PM70
Porphyromonoas gingivalis W83
Pseudomonas aeruginosa PAO1
Pseudomonas putida KT 2440

Ralstonia solanacearum GMI1000
Rickettsia conon'i Malish 7

Rickettsia prowazekii Mad rid E
Salmonella enterica serovar Typhi CT18
Salmonella typhimurium LT2 SGSC1412
Shewanella oneidensis MR-1
Sinorhizobium meliloti 1021
Staphylococcus aureus COL
Staphylococcus aureus Mu50
Staphylococcus aureus MW2
Staphylococcus aureus N315
Streptococcus agalactiae 2603V/ R
Streptococcus pneumoniae R6
Streptococcus pneumoniae TIGR4
Streptococcus pyogenes MGA8315
Streptococcus pyogenes MGAS8232
Streptococcus pyogenes SF350 serotype M1
Streptomyces coelicolor A3 (2)
Synechocystis sp. P006803
Thermoanaerobacter tengcongensis MB4(T)
Thermosynechococcus elongatus BP1
Thermotoga maritime MSBB

Treponema pallidum Nicholas
Ureaplasma urealyticum serovar 3

Vibrio cholerae El Tor N16961
Xanthomonas axonopodis pv citri 306
Xanthomonas campestris pv.campestris
Xylella fastidiosa 9a50

Yersinia pestis 0092

Yersinia pestis KIM

Table A.1. 78 bacterial genome sequences analyzed for the strength of translational
selection, according to two measures. AN 'c could not be assessed for Mycoplasma
pulmonis and Ureaplasma urealyticum; tRNA copy number could not be assessed for
Streptococcus pneumoniae R6 and Pseudomonas putida. (Chapter 3).

161

Figure A.1: Phylogenetic trees of bacteria

Mycoplasma genitalium
a I E M coplasma pneumoniae
° 4 l—— reaplasma urealyficum
‘ Mycoplasma pulmonis
Streptococcus pneumoniae R6
0.10 Streptococcus pneumoniae TIGR4
Streptococcus agalactiae
tre tococcus o enes 315
§tre tococcus Biogenes $993
Streptococcus pyogenes MGA88232
Lactococcus lactis
Enterococcus faecalis
Staphylococcus aureus Mu50
Staphylococcus aureus COL
Staphylococcus aureusMW2
Sta hylococcus aureus N315
Bacillus subti is
Oceanobacillus iheyensis
Listeria innocua
Listeria mon ogenes
Bacillus halodurans 125
:Clostridium pertringens
Clostridium acetobutylicum
Therrnoanaerobacter ten con ens
Nostoc sp. PC 712
Synechocystis sp P00 6803
Thermosynechococcus elon atus
Mycobacterium tuberculos s H37FIV
Mycobacterium tuberculosis 0001551
Mycobacterium leprae
Corynebacterium glutamicum
Streptomyces coelicolor
Treponema pallidum
L—————Borrelia burgdorleri
Chlamydophila pneumoniae A839
Chlamydophila pneumoniae .1138
Chlamydophila pneumoniae CWL029
LLChIamydia muridarum
Lfl Chlamydia trachomatis
Chlorobium te idum
— Escherichia coli 01 7:H7 Sakai
Escherichia coli K12
Escherichia coli O157:H7 EDL933
Salmonella enterica sv Typhi
Salmonella himurium
Yersinia pas 's 0092
Yersinia pestis KIM
Buchnera aphidicola Sg
Buchnera sp. APS
Haemophilus influenzae

P I ’
Vibrio choleraeasteure Ia multocida

Shewanella oneidensis
Pseudomonas aeruginosa
Pseudomonas putida
Ralstonia solanacearum
l INeisseria meningitidis 22491
Neisseria menin itidis M058
Xanthomonas axono is pv citri .
annthomonas campestris pv.campestris
X ella fastidiosa
rucella melitensis

Brucella suis

Mesorhizobium Ioti

Sinorhizobium meliloti
Agrobacterium tumefaciens 058 Cereon
Agrobacterium tumefaciens 058 UWash
Caulobacter crescentus
Rickettsia conorii
Rickettsia prowazekii
Helicobacter pylori J99
Helicobacter pylori 26695

Campylobacter jejuni
Fusobacterium nucleatum
Porphyromonas gingivalls
Deinococcus radiodurans
Thermotoga maritime

Aquifex aeolicus

 

 

 

 

 

   
  
 

 

 

  
  
 
  
 
    
 

 

 

 

 

 

 
 
 
  
 
 
  
 
  
   
 
   

 

 

  
 
 
 
  

 

 

 

 

 

 

 

162

Figure A.1: continued

Mycoplasma genitalium
| C M coplasma pneumoniae
l 1—— reaplasma urealyticum
_ ‘ Mycoplasma pulmonis
Streptococcus pneumoniae R6
0 10 Streptococcus pneumoniae TIGR4
' Streptococcus agalactiae
tre tococcus o enes A 315
tre tococcus Biogenes SENS
Streptococcus pyogenes MGA88232
Lactococcus lactis
Enterococcus faecalis
Staphylococcus aureus Mu50
Staphylococcus aureus COL
Staphylococcus aureusMW2
Sta hylococcus aureus N315
Bacillus subti is
—- Oceanobacillus iheyensis

Listeria innocua
__L—‘l Listeria monoc enes
Bacillus halodurans 12
T :Clostridium perfringens
Clostridium acetobutylicum
Thermoanaerobacter ten n ens

co
T Nostoc sp. P00 71 0
___E Synechocystis sp PCC 6803
Thermosynechococcus elongatus
Mycobacterium tuberculosrs H37FiV
Mycobacterium tuberculosis 0001551
Mycobacterium leprae
Corynebacterium glutamicum
Streptomyces coelicolor
Treponema pallidum
—l:Borrelia burgdorteri
Chlamydophila pneumoniae AR39
Chlamydophila pneumoniae J138
Chlamydophila pneumoniae CWL029
Chlamydia muridarum
__ Chlamydia trachomatis
Chlorobium tepidum
”P Escherichia coli 0157:H7 Sakai
Escherichia coli K12
Escherichia coli O157:H7 EDL933
Salmonella enterica sv Typhi
Salmonella typhimurium
Yersinia pes is 0092
Yersinia pestis KIM
Buchnera aphidicola Sg
Buchnera sp. APS
Haemophilus influenzae
, Pasteurella multocida
Vlbno cholerae
Shewanella oneidensis
Pseudomonas aeruginosa
Pseudomonas putida
Ralstonia solanacearum

I lNeisseria meningitidis 22491
Neisseria menin itidis M058
Xanthomonas axono is pv citri
CXanthomonas campestris pv.campestris
Xylella lastidiosa
E Brucella melitensis
Brucella suis
Mesorhizobium Ioti
Sinorhizobium meliloti

Agrobacterium tumefaciens 058 Cereon
Agrobacterium tumetaciens 058 UWash

—— Caulobacter crescentus
Rickettsia conorii
L Rickettsia prowazekii
Helicobacter pylori J99
| l-HeIicobacter pylori 26695
1———0ampylobacter jejuni
F usobacterium nucleatum

Porphyromonas gingivalis
Deinococcus radiodurans
Thermotoga maritime
Aquifex aeolicus

163

 

 

 

 

 

 

 

 

 

 
 
 
 
 
  
   
  
   
    

 

 

 

 

 

 

 

 

 

 

Appendix References

Peterson, J .D., L.A. Umayam, T. Dickinson, E.K. Hickey and 0. White, 2001.
The Comprehensive Microbial Resource. Nucleic Acids Research, 29( 1):123-125.

Ludwig, W., O. Strunk, R. Westram, L. Richter, H. Meier, Yadhukumar, A.
Buchner, T. Lai, S. Steppi, G. Jobb, W. Forster, I. Brettske, S. Gerber, A.W.
Ginhart, 0. Gross, S. Grumann, S. Hermann, R. J ost, A. Konig, T. Liss, R.
Lussmann, M. May, B. Nonhoff, B. Reichel, R. Strehlow, A. Stamatakis, N.
Stuckmann, A. Vilbig, M. Lenke, T. Ludwig, A. Bode and K.-H. Schleifer, 2004.
ARB: a software environment for sequence data. Nucl. Acids. Res., 32(4):1363-
1371.

Cole, J .R., B. Chai, T.L. Marsh, R.J. Farris, Q. Wang, S.A. Kulam, S. Chandra,
D.M. McGarrell, T.M. Schmidt, G.M. Garrity and J .M. Tiedje, 2003. The
Ribosomal Database Project (RDP-II): Previewing a new autoaligner that allows

regular updates and the new prokaryotic taxonomy. Nucleic Acids Research,
31(1):442-443.

Cannone, J ., S. Subramanian, M. Schnare, J. Collett, L. D'Souza, Y. Du, B. F eng,
N. Lin, L. Madabusi, K. Muller, N. Pande, Z. Shang, N. Yu and R. Gutell, 2002.
The Comparative RNA Web (CR W) Site: An online database of comparative
sequence and structure information for ribosomal, intron, and other RNAs. BMC
Bioinformatics, 3(1):2.

Ludwig, O. and HP. Klenk, 2001. Overview: A phylogenetic backbone and
taxonomic framework for prokaryotic systematics; pp. 49-65 in Bergey’s Manual
of Systematic Bacteriology; D.R. Boone and R.W. Castenholz, eds. Bergey's
Manual Trust, East Lansing, Michigan.

164

   

IIII‘IIIIIIIIIIIIII