.u

»
a:

Jan ‘
a w“;

£4

(:4: .|

hair

an. ran

:31 A .
x;

‘. g.u}mm..)
.ppacfi‘m‘w
@ Sufism. 3,}
. mummy...
{:2

i“.

‘»
$5514

.5.

,,r

.3133

I...

.. 22.512
v! ,-

2.MI..LOR.HH~ .2

itHQI-n

I

 

 

 

 

 

2‘ -.....__,.,, ff..-“
\ I LIBRARY
1006’ Michigan State
University

 

 

This is to certify that the
thesis entitled

RELEASE FROM ANT AGONISTIC PLEIOTROPY AND
COEVOLUTION FOLLOWING GENE DUPLICATION IN
FUNGAL MITOCHONDRIAL HEAT SHOCK PROTEINS

presented by

Krista Gudrais Reitenga

has been accepted towards fulfillment
of the requirements for the

 

 

 

 

Master of degree in Microbiology & Molecular
Science Genetics
//
L/§::m / fl
ji__;Major Piofgsor’ 3 Signature
5/ 61/ 1
Date

MSU is an Affirmative Action/Equal Opportunity Employer

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SIOB KzlProj/Acc8Pres/CIRC/DateDuohdd

~--‘ “3....—

 

 

RELEASE FROM ANT AGONISTIC PLEIOTROPY AND COEVOLUTION
FOLLOWING GENE DUPLICATION IN FUNGAL MIT OCHONDRIAL HEAT
SHOCK PROTEINS

By
Krista Gudrais Reitenga

A THESIS

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

MASTER OF SCIENCE
Microbiology & Molecular Genetics

2009

F0

that g3
multpl
Mai

F65 cl

ABSTRACT

RELEASE FROM ANTAGONISTIC PLEIOTROPY AND COEVOLUTION
FOLLOWING GENE DUPLICATION IN FUNGAL MIT OCHONDRIAL HEAT
SHOCK PROTEINS

By
Krista Gudrais Reitenga

SSCl is a gene that encodes a multifunctional mitochondrial heat shock protein
that gave rise to SSQl by gene duplication in a subset of yeasts. In contrast to the
multiple chaperone functions carried out by most heat shock proteins, Ssqlp is
specialized in Fe/S cluster assembly. Ssclp and Ssqlp both participate in the formation of
Fe/S clusters and require interaction with Jaclp. Biochemical experiments and genetic
manipulation of Saccharomyces cerevisiae have provided evidence that Ssqlp and J aclp
may have coevolved to optimize a specialized interaction. Together, these factors present
a unique opportunity to understand how natural selection shapes the functional
coevolution of gene duplicates. We hypothesized that the divergence of SSCl and SSQl
resulted in the coevolution of the JACl-SSQl pair. Here, we report that, in the presence
of a rapidly evolving SSQl, the average rate of J AC1 evolution has decreased. Our study
also supports a burst of adaptive evolution in SSQl immediately following its inception.
Additionally, both SSCl and SSQl exhibit elevated rates of evolution when co—
occurring. When taken together, the signatures of ancestral and present-day selection
point to a release from antagonistic pleiotropy that facilitated coevolution between J AC1
and SSQl. This study offers detailed evidence that the duplication of multifunctional

genes allows for the coevolution of interacting proteins to optimize a paired function.

LIST (
LIST (

SECII
NRC

little
filter

.V‘n
D
d P?

FD
z.
rd
,‘Qr,cf?."_"—4€I7V’;r

TABLE OF CONTENTS

 

LIST OF TABLES ............................................................................ v
LIST OF FIGURES ........................................................................... vi
SECTION I. BACKGROUND
INTRODUCTION ............................................................................ 1
Hsp70s: Characteristic Features .................................................... 2
Hsp70 Phylogenetic Distribution and Gene Family Evolution ................ 3
Yeast Mitochondrial Hsp7OS ........................................................ 7
88C]: A Multifunctional Mitochondrial Hsp7O ................................. 9
Iron-Sulfur Cluster Assembly ...................................................... 10
SSQl: The Mitochondrial Hsp70 Iron-Sulfur Cluster Specialist .............. 13
J-protein Co—chaperones ............................................................ 15
J AC1: The Mitochondrial J -protein Iron—Sulfur Cluster Specialist ............ 17
Patterns and Mechanisms of Gene Evolution Following Duplication. . . . . l9
Detecting Signatures of Selection by Evolutionary Rate Comparisons ...... 25
Correlated Evolution vs. Co-Adaptation ................................ 27
Site-Specific Models ........................................................ 29
Branch-Specific Models .................................................... 31
Branch-Site Models ......................................................... 32
C lade Models ................................................................ 33
Tools for Evolutionary Rate Analysis .................................... 35
Methodological Limitations ................................................ 37
SECTION H. EXPERIMENTAL STUDY
HYPOTHESES AND PREDICTIONS .................................................... 40
METHODS ..................................................................................... 47
Fungal Taxa and Gene Sequence Alignments .................................... 47
Cladogram Construction for PAML Input Trees ................................. 49
Data Partitioning ............................................................ 49
Maximum Parsimony ....................................................... 51
Maximum Likelihood ....................................................... 53
Bayesian Inference .......................................................... 54
Constructing Composite Input Tree Topologies ........................ 56
thsp70 Clade Model Rate Comparisons ........................................ 58
Site-Specific Rate Tests ............................................................. 6O
Branch-Site Test ...................................................................... 61
RESULTS ...................................................................................... 64
85C] evolution accelerated in the presence of SSQl ........................... 64
SSQl has evolved at a faster rate than SSCl ..................................... 65
JACl evolution has decelerated in the presence of SSQl ...................... 68
SSQl has evolved under positive selection ....................................... 71
iii

DlSCl

APPEf

REE}

DISCUSSION .................................................................................. 77

Future Directions ..................................................................... 87
APPENDICES

Appendix A: Fungal Mitochondrial Heat Shock Protein Coding

Region DNA Sequence Sources ................................................... 90

Appendix B: Fungal Mitochondrial Heat Shock Protein Multiple

Sequence Alignments ......... - ....................................................... 94

Appendix C: Fungal Mitochondrial Heat Shock Protein
Phylogenetic Gene Tree Input Topologies for codem'l

Evolutionary Rate Analysis ......................................................... 122

Appendix D: Evolutionary Rate Test Specifications Used in

Control Files Used to Run codem'l of PAML .................................. 141

Appendix E: Likelihood Ratio Tests of codeml Evolutionary

Rate Analyses ........................................................................ 143
REFERENCES ................................................................................ 153

 

labia
its: C

Table
VF ‘2!

Table i
Model

Table l
llcdel

LIST OF TABLES

Table 1: Evolutionary Rate ((1)) Estimation Under the Branch-Site Model ............ 62
Table A1: SSCl Sequence Sources ......................................................... 91
Table A2: SSQl Sequence Sources ......................................................... 92
Table A3: I AC1 Sequence Sources ......................................................... 93

Table El: Likelihood Ratio Test Comparison of SSCl Clade Model _
Test Outputs ................................................................................... 144

Table E2: Likelihood Ratio Test Comparison of 88C] and SSQl Clade
Model Test Outputs ......................................................................... 146

Table E3: Likelihood Ratio Test Comparison of JACl Site-Specific
Model Test Outputs ........................................................................ 149 ‘

Table E4: Likelihood Ratio Test Comparison of SSQl Branch-Site
Model Test Outputs ......................................................................... 151

hgur

reigns

duplz.

Figlin
distril

Figurt
coast!
Figure
{Lplll l

Figure
and :21!

LIST OF FIGURES

Figure 1: A simplified Cladogram representing the evolutionary
relationships among selected fungi in relation to thsp70 gene
duplication events ............................................................................. 8

Figure 2: Summary of mitochondrial heat shock protein (thsp)
distribution among fungal clades ............................................................ 41

Figure 3: Average within-clade pair-wise sequence divergence of J AC1 ............. 48

Figure 4: Summary of data partitions and phylogenetic tree
construction for evolutionary rate analysis ................................................. 58

Figure 5: a priori defined lineages used for clade and branch-site model

input trees ...................................................................................... 63
Figure 6: Comparison of SSC 1 codon evolution from taxa encoding SSQl

and taxa lacking SSQl ........................................................................ 65
Figure 7: Comparison of SSCl and SSQl codon evolution ............................. 67
Figure 8: Site-specific to estimations for J AC1 from clades encoding SSQl ......... 70
Figure 9: Site-specific a) estimations for J AC1 from clades lacking SSQl ........... 71

Figure 10: Comparison of ancestral SSQl codon evolution to SSQl and
88C] evolution within all other lineages ................................................... 74

Figure 11: Comparison of posterior probabilities of placement of sites
into a divergent rate class by the branch-site model, among input tree
topologies ....................................................................................... 75

Figure 12: Amino acid sequence of Ssql encoded by Saccharomyces
cerevisiae YJM789 showing sites inferred to exhibit relaxed selective
constraint and ancestral positive selection ................................................. 76

Figure Bl: SSCl amino acid multiple sequence alignment ......................... 95

Figure B2: SSCl and SSQl combined amino acid multiple sequence
alignment ..................................................................................... 106

Figure B3: Saccharomyces clade J AC1 amino acid multiple sequence
alignment ..................................................................................... 118

vi

 

Figure B4: Candida clade JACl amino acid multiple sequence

alignment ..................................................................................... 119
Figure B5: F usarium clade JACl amino acid multiple sequence

alignment ...................................................................................... 120
Figure B6: Aspergillus clade J AC1 amino acid multiple sequence

alignment ....................................................................................... 121
Figure Cl: SSCl Bayesian Inference Tree 1 .............................................. 123
Figure C2: SSCl Bayesian Inference Tree 2 .............................................. 123
Figure C3: SSCl Bayesian Inference Tree 3 .............................................. 124
Figure C4: SSCl Bayesian Inference Tree 4 .............................................. 124 '
Figure C5: SSCl Maximum Likelihood Tree 1 .......................................... 125
Figure C6: SSCl Maximum Likelihood Tree 2 ........................................... 125
Figure C7: SSCl Maximum Likelihood Tree 3 ........................................... 126
Figure C8: SSCl Maximum Parsimony Tree 1 ........................................... 126
Figure C9: SSCl Maximum Parsimony Tree 2 ........................................... 127
Figure C10: SSCl Maximum Parsimony Tree 3 .......................................... 127
Figure C11: SSCl Maximum Parsimony Tree 4 .......................................... 128
Figure C12: SSCl and SSQ1 Bayesian Inference Tree 1 ................................ 129
Figure C13: SSC] and SSQ1 Bayesian Inference Tree 2 ................................ 130
Figure C14: SSCl and SSQ1 Bayesian Inference Tree 3 ................................ 131
Figure C15: SSC] and SSQ1 Bayesian Inference Tree 4 ................................ 132
Figure C16: SSC] and SSQ1 Bayesian Inference Tree 5 ................................ 133
Figure C17: SSCl and SSQ1 Bayesian Inference Tree 6 ................................ 134
Figure C18: SSCl and SSQ1 Maximum Likelihood Tree ............................... 135
Figure C19: SSC] and SSQ1 Maximum Parsimony Tree 1 ............................. 136

vii

Figure C20: SSCl and SSQ1 Maximum Parsimony Tree 2 ............................. 137

Figure C21: JACl Saccharomyces Bayesian Inference/Maximum

Parsimony Tree ................................................................................ 138
Figure C22: J AC1 Saccharomyces Maximum Likelihood Tree ........................ 138
Figure C23: J AC1 Candida Bayesian Inference Tree .................................... 138
Figure C24: JACl Candida Maximum Likelihood Tree .................. A ............... 138
Figure C25: JACl Candida Maximum Parsimony Tree ................................. - 139
Figure C26: J AC1 F usarium Bayesian Inference/Maximum Likelihood/

Maximum Parsimony Tree ................................................................... 139
Figure C27: JACl Aspergillus Bayesian Inference Tree ................................. 139
Figure C28: J AC1 Aspergillus Maximum Likelihood Tree ............................. 139
Figure C29: J AC1 Aspergillus Maximum Parsimony Tree .............................. 140

viii

 

E

[\TR(

gulps
races it

ligand

SECTION I. BACKGROUND

INTRODUCTION

Coevolution has long been appreciated as a mechanism that operates between
groups of organisms with the potential to create ecological mutualisms and initiate arms
races for adaptation. However, coevolution is a pervasive phenomenon which extends F
beyond macroscopic interactions such as among flowers and their pollinators or hosts and

their parasites. Phenotypes that determine ecological fitness are the result of complex

 

biochemical pathways. Coevolution, therefore, also takes place among the molecules E
within organisms, and at times, may even be responsible for species-level
interdependencies and competitive strategies. Through molecular coevolution, proteins
can exert a selective influence over interacting partners or components of a biochemical
pathway to favor molecular cooperation or antagonism. Proteins may become specialist
or generalist as a result. Therefore, the evolutionary success of organisms hinges upon the
fitness advantages conferred by molecular components. Additionally, molecular
coevolution may influence genetic interactions, which can, among other things, lead to
congenital diseases and contribute to the process of speciation. Coevolution thus merits .
careful study to facilitate our understanding of many fundamental aspects of biology.
Heat shock proteins (Hsps) constitute a group of proteins that are of central
importance to nearly all organisms. A great deal of data has been amassed concerning the
biochemical and genetic properties of Hsps and has led to detailed understanding of the
many known functions of these proteins. While highly conserved and slowly evolving,

one class of Hsps exhibits dynamic variation in their gene copy number. In one

BSSC

deci

inter

CXCF

interesting case, gene amplification has led to the specialization of an Hsp in Fe/S cluster
assembly, an essential pathway for which biochemical mechanisms are only now being
deciphered. Furthermore, the multiple functions carried out by Hsps necessitate
interaction with a wide variety of protein partners and creates ample potential for Hsps to
exert a reciprocal influence on other constituents of networks. Combined, the
characteristics of Hsps present a unique opportunity to study how changes in gene copy
number affect coevolution of interacting partners within an essential biochemical

pathway. i

 

Hsp70s: Characteristic Features

So named for their discovery (Ritossa 1962) as a group of proteins that exhibited
increased abundance in cells following heat stress, heat shock proteins of the 70
kiloDalton (kDa) class (Hsp703) represent a multi-gene family of protein chaperones with
a nearly ubiquitous distribution within the tree of life. Homologs have been found
throughout the Bacteria and Eukarya, as well as some representatives in Archaea (the
absence of Hsp70s in Archaea has been reported by (Gribaldo et a1. 1999)). Hsp7OS are
known to participate in an array of indispensable functions associated with the folding, .
transport, and degradation of a wide variety of polypeptides. Hsp7OS may perform house-
keeping functions constitutively under many physiological conditions or exhibit
transcriptional up-regulation in response to environmental stresses in order to protect the
integrity of polypeptide components of the cell (Boorstein et al. 1994). Since their first
identification in heat stressed drosophila cells in the 1970’s (T issieres et a1. 1974; Bukau

and Horwich 1998) many other stimuli have been demonstrated to trigger increased

SIT

In)
an
Egg.

mu“.

\1(

synthesis of Hsps, including exposure to ethanol, anoxic conditions, heavy metal ions,
and ultraviolet light (Lindquist and Craig 1988). The Hsp703 have an extremely slow rate
of evolution and share a common t1i~domain protein structure across all three domains of
life. The canonical form comprises a 44 kDa amino-terminal ATPase domain, an 18 kDa
peptide binding domain (Wang et a1. 1993), and a 10 kDa carboxy-terminal domain of ’
variable amino acid composition. Hydrolysis of adenosine-5'-triphosphate (ATP)
regulates the induction of a conformational change within the Hsp703’ substrate binding
pocket and consequent binding and release of hydrophobic regions of the substrate
polypeptide (Bukau and Horwich 1998). The functions of the proteins comprising the
Hsp70 family are so well conserved that, when expressed by a mammalian cell, an Hsp70

protein from a fruit fly is able to perform heat stress protection (Pelham 1984).

Hsp70 Phylogenetic Distribution and Gene Family Evolution

Though many Hsp7O homologs have retained equivalent functional abilities
across divergent organismal taxa, the number of Hsp70 genes encoded within a genome
shows plasticity, a dynamic rife with evolutionary and ecological potential. Comparative
sequence analyses have revealed that the eukaryotic Hsp70 genes, all encoded within the
nuclear genome, constitute four phylogenetically distinct clades. The clades are
characterized by common intracellular localization of the protein products to either the
mitochondria, endoplasmic reticulum, plastids, or cytoplasm (Boorstein et a1. 1994).
Nearly all eukaryotes contain at least three Hsp70 gene copies; the budding yeast
Saccharomyces cerevisiae possesses 9 cytosolic (cyt), 3 mitochondrial (mt), and 2

endoplasmic reticulum (er) isoforms of Hsp70. However, the number of paralogs

cncod
Hspil'.
specie
the Eli

encodi

genes 1

CK‘Clll'it'

genes. I
Wage
Nil “1'

m.“

and Fe:

encoded by different eukaryotic genomes can vary widely, as exemplified by the 10
Hsp70 genes found in the nematode Caenorhabditis elegans and 19 in the closely related
species C. briggsae (Nikolaidis and Nei 2004). An early gene duplication event prior to
the radiation of eukaryotic species gave rise to the cytHsp7OS and erHsp7OS. The genes
encoding Hsp7OS of the mitochondria and plastids are likely of bacterial origin. After
establishment of the bacterial endosymbionts that are hypothesized to have gaven rise to
the mitochondria and plastids in an ancestral eukaryote, lateral transfer of the Hsp70
genes from the organellar genomes to the nuclear chromosome is thought to have
occurred (Muhlenhoff and Lill 2000).

Gene duplication is known to play an important role in the amplification of Hsp70
genes. Duplication is likely facilitated by inverted and tandem cytHsp7O gene pair
arrangements common to the genomes of the Caenorhabdid nematodes (N ikolaidis and '
Nei 2004), mosquito (Benedict et a1. 1993), rat (Walter et a1. 1994), fruit fly (Bettencourt
and Feder 2002), fugu (Lim and Brenner 1999), and human (T avaria et a1. 1996). A
biological cost-benefit balance may play a role in governing the cytHsp70 copy number
within genomes. Cells sustain a cost of deleterious effects on growth, imposing an upper
limit on the optimum Hsp7O expression level due to a cost of replicating additional
Hsp70 genes, energy required for additional translation, or a toxic effect associated with
Hsp70 expression above a certain threshold. Conversely, an increase in Hsp70 expression
may offer the benefit of an enhanced ability to survive environmental stresses. Evidence
of a correlation between Hsp70 expression level and degree of thermotolerance has been
well documented in Drosophila (Feder et a1. 1996). Thermotolerance and survival in the

face of other environmental stressors by Hsp70 buffering therefore constitute ecologically

relevant

gene cor
frequent
iBCllEl'lC
concurs
spreadin
rimmed
Within 31
dlltlgen
mutation
inurlf}
lititltlio
k a ton
m51mm
Hép‘Os f

Nei 2011

relevant phenotypes on which natural selection may act.

Examination of two cytHsp7O paralog clusters from Drosophila revealed that
gene conversion between and among groups of physically clustered genes is likely to be a
frequent event which contributes to the homogenization of Hsp70 copies within a group
(Bettencourt and Feder 2002). Gene conversion maintains sequence similarity, while
concurrently enabling a subgroup of cytHsp703 to diverge in a concerted manner by
spreading new mutations among copies. Gene conversion among cytHsp70 has also been
reported in the nematodes (Nikolaidis and Nei 2004) and has been suspected to occur
within angiosperm plants (Renner and Waters 2007). Alternatively, the lack of
divergence among a group of Hsp705 may be due to slow evolutionary rates. The bias of
mutations exhibited among paralogs toward synonymous changes implies the large role
of purifying selection. In conjunction with gene homogenization, the spread of
deleterious changes among Hsp70 paralogs is disfavored (Bettencourt and Feder 2002).
As a consequence, Hsp70 sequences of proteins localized to the same cellular
compartment from distantly related organisms tend to share greater similarity than
Hsp708 from different cellular compartments within the same organism. (N ikolaidis and
Nei 2004).

Unlike the mechanisms of convergent evolution that characterize many
cytHsp7OS, the mt- and erHsp7OS show evidence of divergent evolution. Diversifying
selection is a mechanism which drives divergent evolution and is facilitated by the
process of independent gene duplication and loss events among lineages (Ota and Nei
1994). The birth and death of paralogs is a feature of the mt- and erHsp70s. While many

divergent eukaryotes, including Drosophila, nematodes, and the marine diatom

‘i rabid:

Waters

plastids
to the H
ilk plat
cyanobe
encode '
nm3h
{lambs
Show ll
finals 1
minty;

Wile:

klflligl
“mtg;
Willa
53 come
Sign? Us.

The-.1.

Le;-

Thalassiosira pseudonana encode a single thsp70, the thsp7OS have undergone
duplication in other eukaryotic lineages, with Saccharomyces cerevisiae possessing 3,
Arabidopsis thaliana with 2, and Plasmodium falciparum genomes with l (Renner and
Waters 2007).

Congruent with the hypothesis for the origin of eukaryotic mitochondria and
plastids from ancient bacterial endosymbionts, thsp70 genes display greatest similarity
to the Hsp70 bacterial homologues from representatives of the a-Proteobacteria, whereas

the plastid Hsp70 genes most closely resemble the heat shock protein genes of

 

cyanobacteria (Boorstein et a1. 1994). Within the Bacteria, some organisms may also
encode multiple Hsp705 (referred to as dnaK or heat shock cognate, hsc, in the bacteria),
with 3 homologs in the Escherichia coli genome (Itoh et a1. 1999) and the
cyanobacterium Synechococcus (W ard-Rainey et al. 1997). Bacterial Hsp7OS have been
shown to display paralog-specific localization patterns. In the case of Synechococcus,
dnaK3 localizes specifically to the cytosolic thylakoid membrane of an oxygen-producing
photosynthetic system (Nimura et a1. 1996), analogous to the plastid-specific organellar
localization observed in some eukaryotes.

In contrast to the ever-present status of Hsp70 in Eukarya and Bacteria, the
detection of gene homologs within Archaea has been patchy, with presence reported in
some taxa (Macario et al. 1991; Gupta and Singh 1992, 1994), but absence of
recognizable homologs in others (Lange et a1. 1997). These observations have given rise
to controversy surrounding the origin of the archaeal Hsp70 and challenge the reliability
of the use of Hsp70 as a phylogenetic marker with respect to the three domains of life.

The alternative hypotheses of lateral acquisition in a subset of lineages (Philippe et a1.

Yea

Pr,—
r.l

1999) and differential gene loss (Gupta 1999) have also been proposed.

Yeast Mitochondrial Hsp705

The plasticity of gene copy number within the Hsp70 gene family has produced a
particularly interesting outcome within the yeast thsp70s. Gene duplication has given
rise to a functionally specialized protein that can be readily studied in the experimentally
tractable model eukaryote, Saccharomyces cerevisiae. S. cerevisiae encodes three
thsp7OS: Ssclp, the most abundant Hsp70 that functions within the organelle, plus
Ssqlp and EcmlOp, two constitutively present forms of rarer abundance. Included in all
three yeast thsp70 sequences is a leader sequence that targets the protein products for.
import into the mitochondria, where they function in the matrix (Craig 1989). In an event
independent of the whole genome duplication estimated to have occurred about 150
million years ago in yeast (Langkjaer et a1. 2003), SSQ1 arose from SSCl by gene
duplication prior to the most recent common ancestor of S. cerevisiae and Candida
albicans (see Figure 1). Additionally, the paralog SSQ1 has been identified in all
descendent fungal taxa studied (Schilke et a1. 2006). The duplication of SSCl is in
agreement with the observation that slowly evolving genes in S. cerevisiae tend to
duplicate, with subsequent retention of paralogs, more frequently than fast evolving
genes (Davis and Petrov 2004). Later, ECMlO, a third yeast thsp70, was generated
during the whole genome duplication believed to have occurred in the most recent
progenitor of the clade that includes S.cerevisiae and S. castellii (Kellis et a1. 2004) (see
Figure 1). While ECMlO now shares 82% amino acid sequence identity with SSCl of S.

cerevisiae (Baumann et a1. 2000), SSQ1 has undergone greater divergence, particularly

 

effit'
com

81011

first

 

within the substrate-binding domain, sharing an overall amino acid identity of only 52%
with SSCl (Schilke et a1. 2006). Each yeast thsp7O is located on a separate nuclear
chromosome, a feature which has the potential to result in disparate mutation rates and
efficiencies of natural selection which act on the three thsp7O genes. The genomic
context within which the thsp7O genes reside can therefore exert an influence on
evolutionary rates of these genes independent of their respective protein structure and

function (Pal et a1. 2006).

 

Schizosocchoromyces pombe

l———Aspergillus niduians
I L———Neosartoryafischeri

I .——Neurospora crassa

 

 

 

—Fusarium verticiliioides

 

 

‘r’orrowio Iipolytico

r—Deboryomyces hansenii
L—Candida olbicans

Soccharomyces cos tel/ii

 

 

andida giabrata

Sacchoromyces cerevisiae

Figure 1: A simplified Cladogram representing the evolutionary relationships among
selected fungi in relation to thsp70 gene duplication events. This Cladogram is a
modified version of that constructed by Fitzpatrick et a1. (2006) using maximum
likelihood to infer the organismal relationships among fungi based on a concatenated
alignment of 153 universally distributed fungal genes. All branches shown were
supported with a bootstrap value of 100. The gray star indicates the lineage within which
a thsp7O gene duplication gave rise to SSQ1. ‘WGD’ indicates the lineage within
which a whole genome duplication took place, giving rise to ECMlO.

moi

HUT

rial

Pf?-

‘a ‘,I

iii-Lu

SSC]: A Multifunctional Mitochondrial Hsp70

Ssclp is a constitutively expressed, essential protein that functions as the major
molecular chaperone within the matrix of the yeast mitochondrion and interacts with a
myriad of different peptides. The constitutive chaperone tasks of Ssclp involve peptide
chain folding, unfolding and translocation necessary for mitochondrial biogenesis. About
10% of the Hsp70 protein present in the mitochondria acts as a component of the pre-
protein translocase of the inner membrane (TIM) complex. As a TIM constituent, Ssclp
cyclically binds and releases polypeptides to assist the pumping of nuclear-encoded
peptide chains across the inner membrane of the mitochondrion. Subsequently, Ssclp
facilitates folding of the chains into their native conformation as they emerge into the
matrix (Neupert 1997). Because many proteins translated in the cytosol become folded
prior to their import across the mitochondrial membranes, protein unfolding into linear
peptide chains appropriate for translocation through the TIM complex is also critical, and
is yet another function performed by Ssclp via interaction with a substrate peptide’s N-
terrninal pre-protein signal sequence (Lim et al. 2001). Ssclp can also be found
associated with mitochondrial ribosomes to fold proteins into their native conformation
as newly synthesized peptides emerge during translation.

Under conditions of heat stress, Ssclp protects the cell from the toxic effects of A
protein denaturation and aggregation within mitochondria. For instance, Ssclp is
responsible for maintaining Varlp, a subunit of mitochondrial ribosomes, in a soluble
form to prevent aggregation or misfolding prior to ribosome assembly, a danger met with
' increased potential during heat shock (Herrmann et al. 1994). Further, Ssclp plays a role

in the synthesis of mitochondrial DNA as a partner in the Hsp70—Hsp78 mitochondrial

 

bicha
a pan
[harm

mch

cluster
creatior
linked 1

one dep

ImIl-Su

J

{‘1-

CT

Lille
.ml‘i‘sxg
Film
Elfin.
z“Slam
£65361:
mine in

55331] it

 

bichaperone system. In yeast, this system is critical to the maintenance and restoration of
a particularly thermosensetive enzyme, Miplp, the mtDNA polymerase, during severe
thermal stress. The Hsp70-Hsp78 bichaperone is known to localize within protein-
mtDNA complexes known as nucleoids, where the bichaperone may act to quickly refold
Miplp within the nucleoid scaffold leading to protection and reactivation of the mtDN A
polymerase. Reactivation of Miplp is more efficient than importing newly synthesized
Miplp into the mitochondrion (Germaniuk et al. 2002).

In addition to these classical roles as a chaperone, Ssclp is involved in Fe/S

 

cluster biosynthesis, a function that was encoded by the ancestral mtI-Isp70 prior to the
creation of SSQ1 (Schilke et a1. 2006). The process of Fe/S cluster assembly is tightly
linked to the mitochondria in eukaryotes and has been appreciated only in recent years as

one dependent on an enzyme-mediated biochemical pathway (Zheng et al. 1993).

Iron-Sulfur Cluster Assembly

From a broad perspective, it is no understatement to characterize Fe/S clusters as
ubiquitous chemical structures that enable biochemical reactions essential to the
processes that drive Earth’s ecology, since these units make photosynthesis, cellular
respiration, and nitrogen fixation possible. Serving as inorganic cofactors for a variety of
proteins, Fe/S clusters participate in substrate binding and dictate many catalytic
mechanisms via oxidation and reduction within enzymes. Fe/S cluster proteins are thus
necessary for the citric acid cycle, haem biosynthesis, DNA repair, protein synthesis, and
purine metabolism (Rouault and Tong 2005). Additionally, Fe/S clusters have been

shown to sense oxidative stress and intracellular concentrations of iron to mediate cellular

10

16S;

I":

Slm

suit.

 

responses, sometimes as Fe/S cluster-containing transcription factors (Kiley and Beinert
2003).

Though many details remain to be clarified, the general mechanism for the
synthesis of Fe/S cluster assembly involves an initial step of structurally coordinating
sulfur and iron into a cluster on a scaffold protein and the subsequent transfer of the
metallocluster to a substrate apoprotein. While enzymatic abstraction from cysteine
residues is known to supply the sulfur for Fe/S cluster biogenesis, the source of iron has
yet to be elucidated (Lill and Muhlenhoff 2008). Several roles have been proposed for
Hsp70 chaperones in the context of Fe/S cluster assembly, though none have been proven
experimentally. Hypothesized Hsp70 functions in Fe/S cluster biogenesis include
assisting the transfer of assembled Fe/S clusters from the scaffold protein to the recipient
apoprotein, or binding to Fe/S assembly proteins and/or substrate apoproteins to prevent
inappropriate oxidation of cysteine residues that serve as ligands to coordinate the Fe/S
structure (Muhlenhoff and Lill 2000). One certainty that emerges regarding the process of
Fe/S cluster assembly is that this multi-step pathway is rife with ample potential for
SSCl, SSQ1, and their co—chaperone, .1 AC 1, to interact with many protein players.

As a testament to Fe/S cluster essentiality, three different pathways have arisen
throughout the tree of life dedicated to Fe/S cluster biogenesis: the nitrogen fixation
(NIF), iron-sulfur cluster (ISC), and sulfur utilization factor (SUF) pathways. The N IF '
pathway consists of a set of genes highly conserved in azototrophic bacteria and is
devoted to the formation of Fe/S clusters exclusively for the maturation of the
nitrogenase enzyme. The more general ISC pathway genes interact to assemble Fe/S

prosthetic groups onto a variety of apoproteins (Zheng et al. 1998). This second system is

11

 

uti,

it be r
al. If!)

the Sl‘

first

 

utilized by a much broader distribution of organisms and shows strong conservation
throughout the Bacteria, particularly within the a-proteobacteria, the mitochondrial
ancestor of which is hypothesized to have bestowed an intact ISC biosynthesis system to
the Eukarya with subsequent preservation from yeast to humans (Lill and Muhlenhoff V
2008). While the Archaea encode many proteins which rely on Fe/S clusters for their
functions, this domain of life lacks homologs of both NIF and ISC assembly systems.
Instead, these microbes encode genes homologous to some of the genes of the third Fe/S
cluster assembly pathway, SUF. The SUF operon encodes a redundant pathway
discovered in Escherichia coli when a small degree of Fe/S enzyme activity was retained
following deletion of the bacteria’s ISC operon (Takahashi and Tokumoto 2002). Later,
in contrast to the housekeeping function of the ISC pathway, the SUF pathway was found
to be required by E. coli under conditions of Fe starvation and oxidative stress (Outten et
al. 2004). SUF homologs have also been identified within plastid genomes. Additionally,
the SUF system may have served as the origin for the scaffold protein of the ISC cluster
assembly pathway in some bacteria (Takahashi and Tokumoto 2002).

Though the NIF, ISC, and SUF pathways function independently, similarities
among the systems abound, which have facilitated the identification of the functional
components that perform analogous tasks within yeast mitochondria for Fe/S biogenesis.
All together, at least 15 proteins have been implicated as Fe/S cluster assembly proteins
that cooperate in the mitochondrial matrix (Lill and Muhlenhoff 2008). Though many of
the proteins that require Fe/S clusters function within mitochondria, some cytosolic
proteins also contain Fe/S clusters and are believed to receive Fe/S clusters exported from

the mitochondria, since Fe/S cluster biogenesis has not been demonstrated to occur in the

12

5501

ales;
biochl
SSQI
mind
come
fittest
Conic;

cflzlnj

 

cytosol. The reducing chemical conditions and lower partial pressure of 02 within the
mitochondrial matrix relative to the cytosol may have favored the establishment of the

Fe/S cluster biogenesis pathway within this organelle (Muhlenhoff and Lill 2000).

8801: The Mitochondrial Hsp70 Iron-Sulfur Cluster Specialist

SSQ1 has become specialized in the assembly of Fe/S clusters, but at the price of
a loss in the multifunctionality displayed by its paralog SSCl. Recent genetic and in vitro
biochemical experiments offer support of the functional specialization of SSQ1. When
SSQ1 was deleted from the S. cerevisiae genome, mutants accrued iron within the
mitochondrial matrix with a concurrent reduction in Fe/S cluster-containing enzyme
concentrations and protection against oxidative agents (Voisine et al. 2000). To further
investigate this phenotype, authors of another study used an assay to observe the
. conversion of ferredoxin, a mitochondrial protein that requires an Fe/S cluster for
enzymatic function, from its apo-forrn to its holo-form within isolated mitochondria.
Within mitochondria extracted from an S. cerevisiae SSQ1 deletion strain, the majority of
ferredoxin failed to mature into a holoenzyme. Ferredoxin that did achieve the
holoenzyme state was found to have reduced kinetic character (Lutz et al. 2001). The
interaction of Ssql with known components of the Fe/S cluster assembly pathway has
also been tested, and investigators observed efficient binding of a purified protein binding
domain fragment of Ssqlp to a peptide fragment of the scaffold protein involved in Fe/S
cluster formation (Schilke et al. 2006). These results suggest that Ssqlp is important for
Fe/S biogenesis and is able to physically interact with a key component of the pathway.

Consistent with the hypothesis of specialization and the concomitant loss of

13

ances
bin-iii
llllHSj
follou
mcmb
bind 1(
ct al. 3
require
fading
I'Sc'm'iit

These n

3301 d.
mam v
PIOitClic

“PINK

 

ancestral thsp70 function after duplication, Ssqlp was found to have very weak
binding specificity to peptides known to be bound by Ssclp of both pre- and post
thsp70 duplication (Andrew et al. 2006). Additionally, Ssqlp was not detected
following co-irnmunoprecipitation with Tiin44p, a component of the inner mitochondrial
membrane peptide translocase, was attempted. This demonstrated that Ssqlp does not
bind to Tim44p, in contrast to Ssclp, which acts as a subunit of the TIM complex (Lutz
et al. 2001). Ssql has lost the ability to bind the same variety of peptide substrates that
require Hsp70s for general protein translocation and folding. In accordance with this
finding, the inability of SSQ1 over-expression to complement an SSCl null mutation
(Schilke et al. 1996) is consistent with the loss of general chaperone function by SSQ1.
These results support the conclusion that Ssqlp is no longer a generalist thsp70.

SSQ1 is dispensable for yeast survival due to some functional overlap of 38C].
SSQ1 deletion mutants have been observed to accrue iron within the mitochondrial
matrix with a concurrent reduction in Fe/S cluster-containing enzyme concentrations and
protection against oxidative agents, phenotypes that can be partially rescued by the over-
expression of SSCl (V oisine et al. 2000). Furthermore, because the mechanism by which
SSCl and SSQ1 participate in Fe/S cluster biogenesis seems to require interaction with
the same conserved motif of Isu, the Fe/S cluster biogenesis scaffold protein (Schilke et
al. 2006), SSCl is likely to assist Fe/S cluster formation in fungi lacking SSQ1. Because
of this overlap, Ssclp and Ssqlp compete for nucleotide exchange factor Mgelp, which
allows ADP and Pi to be released from the thsp7OS and is present in limiting amounts.
The greater abundance of Ssclp in the mitochondrial matrix, compared to Ssqlp, may

limit the amount of Mgel p that can interact with Ssqlp to be recycled to its active form.

14

As a n

“€811“

the H5]
yeah.
obligrt
protein
al. 300
have in
and one
bite be
mciudir.

neitifur

A.—
6,,

slim

3‘9.‘ 1'
‘thsi

J'Pmlei

t.
‘Qii'l‘ a}

liq-232d

1501.:

s19"

I‘Lbel

.J
r—9

.3
51. £9»

rural

 

 

As a result, the reduced proportion of activated Ssqlp may only be sufficient to carry out
a restricted task load compared to Ssclp (Schmidt et al. 2001).

The role of Ssqlp in Fe/S cluster formation is analogous to the specialized task of
the Hsp70 HscA in bacteria and it appears that, after arising independently in a subset of
yeasts, SSQ1 has undergone functional evolution. In the process, Ssqlp has acquired an
obligatory protein interaction with the yeast orthologs of the cluster assembly scaffold
protein and the co-chaperone proteins with which the bacterial HscA interacts (Schilke et
al. 2006). A specialized Hsp70 committed to Fe/S cluster biogenesis therefore appears to
have independently arisen twice throughout the course of evolution- once in the bacteria
and once in the yeast. The initial discoveries of SSQ1 in E. coli and S. cerevisiae seem to
have been serendipitous; SSQ1 homologs remain undetected in many eukaryotes,
including humans (Schilke et al. 2006). Given that most eukaryotes utilize a
multifunctional thsp7O in the Fe/S cluster biogenesis pathway, the advantage of
dedicating a separate thsp70 to assist exclusively in this process in yeasts remains to be

established.

J-protein Co-chaperones

J—domain protein co-chaperones belong to the 40 kDa heat shock protein (Hsp40)
family and engage in an obligate, physical interaction with Hsp70s J-proteins are
required to stimulate the activity and mediate the function of Hsp705; thus, J-protein
isoforms are active in all cellular compartments containing Hsp70s. While the J -proteins
represent a disparate group of proteins with little gene sequence conservation or protein

structural organization among members, J -proteins do all share a defining feature called

15

the 14
contai
actii'it
proteir

Hsp?)

intent:
b} Still
the abs
differet
tenets}
domain
133p. Vi
Y itxct
Sahi an-
E'Oi YEP]
imger d
”Win
3.1“]

“Um

"SWIG;

 

the J-domain, named for its sequence similarity to the E. coli DnaJ protein. All J-domains
contain a histadine-proline-aspartic acid motif essential for stimulation of the ATPase
activity of the Hsp70 partner (Cheetham and Caplan 1998). Both general and specialist J -
proteins exist in yeast, with several unique J-proteins that function to assist general
Hsp70 functions or specialized Hsp70 roles, depending on the specific J—protein/Hsp70
interaction. A distinction between generalist and specialist J -proteins was demonstrated
by Sahi and Craig (2007) in S. cerevisiae when the deleterious growth effect caused by
the absence of J -protein Ydjlp was rescued by expressing J-domain fragments of several
different J -protein co-chaperones, indicating that Ydj 1p is a generalist J -protein. Such
general J-proteins may thus work to indiscriminately stimulate the ATPase functional
domain common to all Hsp70s. When specialist J -proteins ch23p, Sislp, Jjj 1p, and
Jjj3p were deleted from the yeast genome, however, the deleterious phenotype could not
be rescued by expression of any other gene. Thus, in contrast to the generalist Ydj 1p,
Sahi and Craig (2007) showed that the J —domain fragment of specialist Jjj3p alone could
not replace the function of full-length specialist J-proteins. For Jjj3p, an additional zinc
finger domain was shown to be required for the J -protein’s specialized role as a
component in the diphtharnide biosynthesis pathway. Some specialist J-proteins form an
exclusive thsp7O partnership to perform a single function, as in the case of a
chaperone-co-chaperone pair, Sszlp and Zuolp, which associates with translating
ribosomes to fold newly synthesized peptides. J-protein Zuolp interacts solely with the
Hsp70 Sszl p, and Sszlp does not pair with any other J —protein, despite the co-occurrence
of several other types of J-proteins.

In some cases, J -proteins have been shown to bind substrate peptides themselves,

16

indej

azin

or re.
sub~t

coil

.___4

andr

speei

IAC'

getter

twist

liken

independent of the formation of a complex with an Hsp70. For the E. coli DnaJ homolog,
a zinc finger-like region and the carboxy-terminal region are required for ligand binding
function (Han and Christen 2003). Some J -proteins may deliver substrates to the Hsp70
or recruit the Hsp70 to a peptide when they exhibit a ligand binding function. Similar
substrate polypeptide binding features in specialized yeast J-proteins may also act to
localize the J-protein to a particular site within the cell, thereby sequestering a J -protein
and rendering it unavailable to function in place of other J-proteins, thus conferring

specificity (Sahi and Craig 2007).

JACl: The Mitochondrial J -protein Iron-Sulfur Cluster Specialist

J AC1 is an essential gene and encodes one of 22 J-proteins in the S. cerevisiae
genome. The J acl p protein contains an N-terminal mitochondrial signal sequence and is
imported into the mitochondrial matrix where it serves as a specialized co—chaperone to
assist in Fe/S cluster generation (Voisine et a1. 2001). Its task is to bind the Fe/S cluster
assembly scaffold protein Isup for delivery to a thsp70 and stabilize the Isulp-Hsp70
interaction (Andrew et al. 2006). J aclp serves as the only known J-protein capable of
interaction with thsp70 Ssqlp and together, the J-protein/Hsp70 pair has become
specialized in the yeast Fe/S cluster assembly pathway. However, because J AC 1 and
88C] orthologs have been preserved together from bacteria to humans as components of
the ISC Fe/S cluster formation pathway, they retain the ability to cooperate in yeast,
explaining why the effects of deleting SSQ1 from the S. cerevisiae genome may be
compensated for by the over-expression of J AC1 (Andrew et al. 2006).

Several pieces of evidence demonstrate that J aclp and Ssqlp are a functional pair

17

5r

pit

Il'lE

IOI

C01

1?:

rL_

.5}

 

specialized in Fe/S cluster biogenesis and are consistent with the coevolution of the two
proteins. When the mitochondria of mutant J AC1 S. cerevisiae strains are isolated and
manipulated to contain normal Fe concentrations, a decrease in the activity of Fe/S
cluster enzymes was reported. Further, the J AC1 mutation created in this study was found
to display a negative genetic interaction with deletion of SSQ1, as these double mutants
could not be recovered (Voisine et al. 2001). Additionally, Both J aclp and Ssqlp have
been demonstrated by Andrew and colleagues (2006) to bind the C-terminal domain of
Isup. This has revealed that J ac 1p and Ssqlp interact with a common component of the
Fe/S cluster biogenesis pathway. Importantly, although J aclp also has the potential to
pair with the more abundant thsp7O SSCl, J aclp displays a greater degree of in vitro
stimulation of Ssqlp ATPase activity compared with the efficiency of the Jaclp — Ssclp
interaction of both pre- and post thsp7O duplication yeasts (Schilke et al. 2006).
Recently, new insight into genetic basis of differences that have evolved at the
J AC1 locus and are responsible for the increased efficiency of Ssqlp ATPase stimulation
have been elucidated, and involve shortening of the J-domain (Marszalek, unpublished).
JACl from S. cerevisiae was engineered to include a section of the J-domain from the
pre-duplication yeast Y. lipolytica. The elongated J -domain more closely resembled J AC1
sequences from yeasts encoding Ssclp, but lacking Ssqlp. The ability of the chimeric
protein to stimulate Ssclp in S. cerevisiae, relative to native J aclp, was increased.
Therefore, the portion of the J-domain lost in yeasts encoding Ssqlp may be important
for interaction with thsp705 and the increased affinity of Jaclp for Ssqlp compared to
Ssclp may have been due to this J -domain modification. The functional specialization of

the Jaclp - Ssqlp pair emerged through the sequence of events in evolutionary history

18

Pine:

fineag
genom-
fittest
it get
multifi
novel:
hast
It’lfil‘im
di) erg;-

Ihese d

i .,

r611

 

that followed the duplication of an ancient, multifunctional thsp70. Conversely, the
divergence of paralogs SSC] and SSQ1 may have shaped the evolution of J AC1 and

molded this J -protein into an Fe/S cluster assembly specialist as well.

Patterns and Mechanisms of Gene Evolution Following Duplication

Because SSQ1 and SSC] originated from a gene duplication event in a yeast
lineage, it is important to understand how the presence of paralogous genes within a
genome can affect evolutionary divergence. Gene duplication plays a prominent role in
molecular evolution as a mechanism of spawning the genetic material needed to generate
the genomic variation responsible for biological diversity. When the ancestral,
multifunctional mtI-Isp70 gene duplicated, a potential was created for the development of
novel adaptation unattainable in the single gene copy state. However, the evolutionary
fate of gene duplicates depends on two distinct types of mechanisms: 1) one of initial
retention within a population and 2) one of several alternative modes of paralog
divergence. While many models exist to describe the modes of gene duplicate evolution,
those described here have emerged to the forefront of research studies (Hurles 2004).

Neofunctionalization and nonfunctionalization are two models of gene duplicate
evolution first put forth by Ohno (1970) to describe the resolution of functionally
redundant paralogs. Common to both models is the assumption that a gene duplication
event has no effect on organismal fitness because immediately after duplication, the
paralogs are equivalent, with each gene copy capable of fulfilling all functions of the
ancestral gene equally well. The gene copies are expected to be interchangeable while

both paralogs retain high sequence identity, rendering the new duplicate gene immune to

19

forces 1
muunc
lossof-
gene to
bOIh le
copy su

exhibiti

duplical
all illilC'
of deleti
or both.
fate of g
longer f-

linen;

Dip-ii (t g

forces of selective constraint. Therefore, the duplicate gene is free to accumulate
mutations that would have been forbidden in the ancestral single copy state because any
loss-of-function that the duplicate gene copy sustains would be rescued by the redundant
gene copy. Under this premise, the neofunctionalization and nonfunctionalization models
both predict an asymmetry in the evolutionary rates between paralogous genes, with one
copy subject to purifying selection to retain ancestral functions and the other copy
exhibiting accelerated substitution due to relaxed constraint.

Nonfunctionalization occurs when the period of relaxed constraint on the
duplicated gene copy results in the accumulation of deleterious mutations that degenerate
all functions of the ancestral gene, without the creation of new functions. Accumulation
of deleterious mutations may occur within the protein-coding region, regulatory region,
or both, and eventually leads to pseudogenization. This is likely to be the most common
fate of gene duplicates (Li 1980). Once a gene has sustained a null mutation and is no
longer functional, it is selectively eliminated from the genome and leads to the permanent
preservation of the non-mutated paralogs.

N eofunctionalization describes a scenario in which, during the period of initial
relaxed selection on the duplicate gene, mutations are acquired in the coding or
regulatory sequence that lead to a novel function of the encoded protein. These mutations
are thought to be rare, relative to nonfunctionalization. Positive selection to optimize the
novel function of the neofunctionalized paralog is then followed by reassertion of
selective constraint to preserve the new function. Assuming that neofunctionalization
results in the loss of an ancestral gene function, this process too, can lead to non-mutated

paralog retention.

20

tests 4
genes.
proces
iaeilitz
subset
dislillt‘
inplice
period

ICCOl'lSl

 

Under a third model of paralog resolution, known as subfunctionalization, the
tasks of a multifunctional ancestral gene become partitioned between the two duplicate
genes. Duplication-Degeneration-Complementation (DDC) (Force et al. 1999) is one
process by which subfunctionalization is thought to occur, where degenerative mutations
facilitate the preservation of both paralogs that have become dedicated to complimentary
subsets of modular ancestral functions. DDC assumes that the ancestral gene expresses
distinct functions ascribed to independent, modular regions of the gene. Following
duplication, both paralogs acquire complementary loss-of-function mutations during a
period of relaxed selection such that the expression of both paralogs is necessary to
reconstitute the repertoire of functions encoded by the ancestral gene.

Escape from adaptive conflict is yet another alternative model of paralog
divergence. The premise of this model is that, if an ancestral gene gains an additional
novel utility that is in adaptive conflict with the first function, the creation of a duplicate
gene could confer an immediate fitness advantage by breaking the ancestral gene free of
antagonistic pleiotropy. Through divergent selection, each paralog would have the
opportunity to individually specialize in at least one of the ancestral functions to a greater
degree than was possible in the ancestral gene. Assuming the ancestral gene was
constrained by competing phenotypes conferred by a single gene, the functional
partitioning between duplicates or the rise of a novel function after duplication could
proceed in a non-neutral manner, driven by an adaptive advantage.

Strong evidence for gene duplicate evolution by escape from adaptive conflict has
been shown in the regulatory divergence of paralogs of the S. cerevisiae galactose

utilization pathway (Hittinger and Carroll 2007). GALl and GAL3 arose as duplicates of

21

a hifu
induc
bifum
prnmt
trans;
in mm
replae.
found

these c
GAD

lilietg:
Promo:

manna

Clflas

 

a bifunctional ancestral gene and encode the galactokinase enzyme Gallp and a co-
inducer Gal3p, respectively. While they once shared a common promoter in the
bifunctional ancestral gene, near complete subfunctionalization of the upstream
promoters between the descendent paralogs has resulted in stringent control of GALl
transcriptional regulation, contrasting with a more modest GAL3 transcriptional response
to induction. The authors swapped the promoter sequences of GALl and GAL3, and also
replaced native paralog promoters with that of a bifunctional GALl/GAL3 promoter
found in another yeast species, and subsequently evaluated the fitness consequences of
these changes. The results from these experiments revealed that switching GALl and
GAL3 promoters was detrimental, indicating that each promoter had undergone
divergence to optimize the expression of GALl and GAL3 individually. While the
promoter of the bifunctional gene performed well in driving the expression of GAL3 and
maintaining yeast fitness, regulation of GALl by the bifunctional promoter reduced basal
expression and decreased yeast fitness. The spacing of transcriptional activator binding
sites was then altered within the promoter sequence of the bifunctional gene to mimic the
binding site arrangement of the GAL1 promoter. The manipulated bifunctional gene
promoter increased the expression control of the galactokinase function in response to the
presence of galactose. Adaptive conflict was therefore proposed to have compromised the
expression optimization of galactokinase in the bifunctional ancestral gene with a single
promoter. Only after duplication and promoter divergence was the expression of the
galactokinase function brought under tighter regulation.

Recently, an additional example of biochemical evidence for gene evolution via

escape from adaptive conflict was presented in a study focused on a set of genes involved

22

 

in at
and 1
then
conic
insert

Mil

 

in a pigment biosynthetic pathway in the morning glory, Ipomeoea purpurea (Des Marais
and Rausher 2008). The dihydroflavonol-4-reductase (DFR) gene, responsible for the
chemical reduction of flavonoid precursors of anthocyanin, has given rise to three gene
copies, DFR-A, DFR-B, and DFR-C, through two gene duplication events. Biochemical
assays for enzymatic reduction of five substrates (three commonly reduced by DFR and
two rarely reduced by DFR) by the DFR copies encoded by both pre- and post-
duplication species were conducted. Severe reductions in the capacity of post-duplication
DFR-A and DFR-C to act on any of the five substrates tested, and an increase in post-
duplication DFR-B to reduce all substrates when compared to the activity of pre-
duplication DFR enzymes were demonstrated. The authors therefore concluded that the
function of the ancestral gene was improved by the DFR-B copy. The release of adaptive
constraint, imposed by antagonistic pleiotropy on the ancestral DFR, following the
creation of duplicate genes, was consistent with comparative DNA sequence-based
evidence of adaptive molecular evolution.

While the above models of paralog evolution offer gene optimization through
specialization or the acquisition of novel roles as long-term fitness advantages of gene
duplication, short-term benefits must exist to govern the retention of paralogs
immediately after gene duplication. The presumed selective neutrality of gene duplication
fundamental to Ohno’s models fails to offer a short-term benefit of duplication events.
The interim retention of duplicate genes on the path to neo- or subfunctionalization
through DDC or escape from adaptive conflict also requires a fitness advantage.
Kondrashov et al. (2002) have suggested that gene duplication itself may be a mechanism

of adaptation by hypothesizing that survival in the face of environmental stresses may

23

 

mind:
throng
n13} tl
Sher.
infiner
ripen
tl'Ollt'
the an.
result:
1998 1.
in yet

erplnr

mandate an increase in protein and/or RNA dosage that can be immediately achieved
through an increase in gene copy. An environmentally determined optimum copy number
may thus exist for each gene under a given set of conditions (Kondrashov et al. 2002).
Several studies performed with yeast suggest that environmental conditions may
influence gene copy number. For example, when a population of S. cerevisiae was
experimentally propagated for 450 generations in glucose-limited media, the population
evolved the ability to reproduce at a higher cell yield per unit of glucose compared with
the ancestral strain (via a glucose transport system with enhanced glucose affinity),
resulting from multiple tandem duplications of two hexose transport genes (Brown et al.
I 998). The amplification of genes within the Hsp70 gene family may similarly be driven
in yeasts as a mechanism to tolerate variations in heat, pH, ethanol, etc., to facilitate the
exploration of new environments.

The role that selection plays in detemrining the evolutionary fates of gene
duplicates, from initial retention in the genome to degeneration, neofunctionalization, or
subfil netionalization, or other intermediate states of paralog divergence, distinguish the
different patterns of gene evolution following duplication described above. Though the
Previously discussed modes may not be mutually exclusive and no one model of paralog
eVOllltion may serve as a general mechanism applicable to all gene duplication events, the
ability to characterize the direction and strength of past and present selective forces

aeting on paralogs are proving to be keys to the elucidation of molecular evolutionary

outcomes.

24

Detectin

ultimatel
the rent
ll'llhlll a
lit the 0
changes
those the
and tram
are those
mutitinr
acids to
31? 91851
hillmcd

Wise

Detecting Signatures of Selection by Evolutionary Rate Comparisons
At the level of DNA, nucleotide mutations arise in a stochastic manner and
ultimately rely on either the forces of natural selection acting on the fitness conferred by
the resulting phenotype, or random genetic drift within the population to achieve fixation
within a population. Within protein-coding DNA, signatures of selection can be identified
via the comparison of the proportion of nonsynonymous to synonymous nucleotide
changes that have occurred though time. Nonsynonymous nucleotide substitutions are
those that result in the substitution of an amino acid, via a change in both the DNA codon
and translated peptide sequence. Synonymous nucleotide substitutions on the other hand,
are those that do not alter the amino acid of the corresponding protein. Synonymous
mu tations exist due to the degeneracy of the genetic code, which allows some amino
acids to be specified by several unique nucleotide triplet sequences. Synonymous changes
are presumed to be invisible to selection acting on protein phenotypes and are therefore
assumed to represent the locus-specific background level of mutations fixed by neutral '
processes such as population bottleneck events or mutational hitchhiking.

In the context of sequence evolution, the proportion of nucleotide differences per
site between two genes that result in nonsynonymous codon changes represents the
norIS)Inonymous substitution rate, dN, while the synonymous substitution rate is given as
(18- Funher, expressing these two rates as the ratio at) = dN/ds can be interpreted as a gauge
of the direction and strength of selection. An co value of less than 1 indicates that the
number of mutations resulting in amino acid changes that reach fixation is more restricted

than the basal mutation level. Therefore, w < l is indicative of negative or purifying

Seleection, which reduces the rate of fixation of deleterious mutations. When a) is greater

25

enti:

SW.

Thea]
{hf}
pres.-

that

may:
5‘ A“ I
UJJUI

Clint

than 1, it may be inferred that positive selection is responsible for the greater-than-
expected fixation of amino acid changes. An co = 1, due to equal rates of nonsynonymous
and synonymous nucleotide changes, points to neutral evolution of codons, with amino
acid substitutions neither being selected for, nor against. Therefore, the greater the
deviation of (0 is from 1, the greater the influence of selection. While evidence of
negative selection may identify DNA sequence coding for regions of proteins that require
strict structural conservation, uncovering signatures of positive selection is of particular
importance in the search for evidence of adaptive evolution.
Calculating an average a) for an entire protein-coding gene through pair-wise
sequence comparison detects evidence of positive or negative selection throughout an
en tire gene, evidence found in only a very small proportion of gene sequences. For
ex ample, in a large-scale study conducted with 3,595 groups of homologs, comprising
24, 832 unique sequences, only 17 gene groups (or 0.45% of the total groups) emerged as
candidates of positive selection (Endo et al. 1996). Such estimations of the prevalence
and scope of the role of positive selection, however, may be misleading. Since gene-wide
mean a) values mask site specific heterogeneity with which natural selection may act,
they may not provide an accurate representation of the strength and direction of selective
Pressures experienced by a gene. Strict interpretations of gene-wide average 0) values
may overlook the high to with which a few sites of a gene are evolving, veiled by the low
(0 values which characterize the evolution of the majority of sites. To bolster this line of
reaSOning, Yang and Swanson (2002) used several models to estimate the number of
codons subjected to positive selection in two gene sequence alignment sets: 192 human

c1388 I MHC glycoprotein alleles, and abalone sperm lysin genes from 25 different

26

spec
nail}

abet

10 ill”.-

pm.

Corr

El?»

ml”:
‘

5L

11c; ,

a}

x
kl

species. For both groups of genes, when a) was averaged across all sites of the sequences,
neither the class I MHC glycoproteins, nor the sperm lysin genes yielded an a) value
above 1. On the other hand, when a model permitting 0) to vary from codon to codon was
applied, a number of sites emerged as likely targets of positive selection with a)
significantly greater than 1. This result was upheld regardless of whether sites of each
gene alignment were partitioned into two groups with different evolutionary rates a priori
(based on functional information of the encoded structures), or whether sites of different
evolutionary rate classes were assumed to be distributed randomly across the sequence.
The results of this study highlight the need to account for evolutionary rate heterogeneity
to uncover patterns of selection. Several models have been developed to identify

differential rates of molecular evolution that can result from selective pressures unique to

particular sites and organismal lineages.

Correlated Evolution vs. Co-Adaptation

Differential selective pressures that act on individual sites may be particularly
relevant to the detection of coevolution of interacting protein partners. Hakes et al. (2007)
have suggested that a distinction must be made between correlated evolution and co-
a~df'-lli>i::—:1tion among protein sites to more specifically describe coevolution. Correlated
eVOIUtion is the concurrent change among interface residues of interacting proteins that
may not necessarily be directly influenced by selective forces due to the protein-protein
interaction itself. Co-adaptation, however, is driven by selection to maintain functional
and Structural integrity of the protein pair to preserve cooperative abilities and results in

the c()mpensatory change among interacting protein partners. The compensatory mutation

27

ma

sen

touj

spec

com

lllZL’l

COlT

seq;

Can
rest
“'h‘;
Valid
1'10

ldfif

may be fixed in response to an amino acid substitution in a region of either protein, which
serves as the point of contact with the other partner.

Of the proteins investigated by Hakes et al. (2007), an average of only 13% of the
total protein sequence was found to correspond to exposed residues directly involved in
specific binding activity at the interface of an interacting protein. Patches of proteins that
comprise only a minority of residues may experience selective pressure exerted by

interacting partners for inter-protein compensatory change. Therefore, correlated
increases in the evolutionary rate of whole protein sequences of protein-protein
interaction partners do not constitute conclusive evidence for co-adaptation. Instead,
correlated coevolution among physically interacting proteins detected by whole gene
seq uence evolutionary rate analysis may point to other targets of selection unrelated to
the interaction of residues at binding surfaces. For instance, gene expression is known to
heavily influence the rate of gene evolution (reviewed in Pal et al. 2006). Because
cooperative proteins often depend on specific stochiometric ratios of active partners
Witlji m the cell for efficient interaction, selection for changes in expression of one protein
Can lead to selective pressure for a corresponding expression change in the other. The
resul taut expression levels may then be the cause for evolutionary rate changes across the
Who le protein sequence in both partners, detected as correlated evolution without an
underlying adaptation of optimizing inter-protein residue binding. Therefore,
evolutionary rate models that account for site-to-site differences are more likely to
idel’ltify compensatory mutations resulting from co—adaptation.
In addition to acting in a targeted manner within a protein coding gene, adaptive

coevOlution has been shown to occur in episodic patterns of bursts (Messier and Stewart

28

19%. 1
over th:
ratios n
inform:
history.
ma) im
selectio

differen

Silt-Spg

codon-t
ll): cod
nucleon
rtWM
flUcleotl
“equal
be diffs
simplgf)
Willem
filth as
the P085

‘1

l“; .3“

E

(I'D

1997). By effectively averaging any strong, but transient periods of positive selection
over the phylogenetic history of two sequences, pair-wise calculation of whole-gene 0)
ratios may miss evidence for divergent adaptive evolution when lineage-specific
information is not taken into account. Events at particular time points in a phylogenetic
history, such as environmental changes impacting ecological niches or gene duplication,
may impose divergent selective pressures on two protein-coding sequences. Divergent

selection is implicated as a cause for (0 values of a gene to differ among clades, reflecting

different selective pressures influencing different branches of a phylogeny.

S ire-Specific Models
Site-specific models define the codon as the unit of evolution and employ a
codon-based substitution model to describe site-specific variations in evolutionary rate.
The codon substitution model utilizes all of the information encoded within DNA at the
nucleotide level, but improves upon the nucleotide substitution model in its
representation of molecular evolution by recognizing the amino acids that are encoded as
nucleotide triplets. Importantly, considering the amino acid sequence that will result from
a Sequence of nucleotide codons allows synonymous and nonsynonymous mutations to
be differentiated (Goldman and Yang 1994). In employing codon models, several
Sin'1l.3lifying assumptions must be made. First, the codon model assumes that the DNA
Seqllences under study are protein-coding and does not consider untranslated sequences
such as introns. Second, codons which signal translational termination are not included in
the Possible codons allowed to result from substitution, since these stop codons most

often generate a truncated protein and are generally not tolerated within organisms

29

(Nielsc
are m
mutant

(Goldn

each cc
predeli
models
datum
tstimat
Calegor

llllffpre

(Nielsen and Yang 1998). Lastly, only one of the three nucleotide positions of a codon
are assumed to undergo substitution per mutation event (for example, an AGG to CGA
mutation would require more than one step under the codon model of evolution)
(Goldman and Yang 1994).
The site-specific model of codon substitution assigns a probability with which
each codon of a multiple sequence alignment is expected to fall within a particular
predefined number of evolutionary rate categories. By conducting this test using nested
models of increasing rate categories, the optimum number of rate categories can be
determined by statistical tests. The evolutionary rate for each of the rate categories is
estimated from the data. Each codon can be assigned to a particular rate class and
categorized as evolving under positive or negative selection, and at what magnitude, by
interpreting the sign and value of (1).

Maximum likelihood estimation of site-specific rates of evolution can be
conducted using fixed-site models or random-sites models. Fixed-site models utilize
Structural and functional information about a protein of interest to identify specific amino
acids predicted to be under equal selective pressures a priori, while random-site models
do not make any prior assumptions about the evolutionary rate of any particular site.
when Yang and Swanson (2002) analyzed the site-specific rates of evolution of MHC
c:IaSS I and sperm lysin genes, the proportion of codons belonging to each evolutionary
rate Class and the values of to that were estimated exhibited a high degree of consistency
among both fixed- and random-sites models for both gene data sets. The authors

demonstrated that partitioning codons into rate classes prior to a) estimation is not

necessary; the random-sites model was just as powerful. The residues were classified into

30

evolutionary
functional or

pressures.

Brunch-Spat

Serc
heterogencit
pmneter. u
phylogenetit
multiple line
lineage will
llOmOgeneoL
numbfil’ 0f 65
Mel comp
Wide (1) film
Mfll~specj
inlemaj bran
ml Il0l eXls
Solution Un
Mel-w
historical en

evolutionary rate groups corresponding to the groups of residues constituting evolving

functional or structural regions of the proteins believed to be under unique selective

pressures.

Branch-Specific Models

Several models have also been developed to examine lineage-specific a) value
heterogeneity among genes. The simplest lineage-specific model includes only one a)
parameter, which assumes the same gene-wide average to for each branch of a
phylogenetic tree. The number of different on values represented by a gene across
multiple lineages may be increased to test whether a gene along one a priori identified
lineage (with (0;) is evolving with an overall rate that is significantly different from a
homogeneous rate (we) characterizing that gene from all other branches of the tree. The
number of estimated branch-specific to parameters may be increased until maximum
model complexity is reached with the “free-ratio” model, in which an independent gene-
wide a) value is estimated for each branch of the tree. An important distinction of the
branc h-specific model compared to the site-specific model is the ability to examine
internal branches of a phylogenetic tree. Because known DNA sequence representatives
may not exist for internal branches, the ability to detect evidence of ancestral sequence
eVOII—ltion under positive selection makes this model powerful. A test conducted using a
branCh-specific model allows one to correlate a phylogenetic branch with known
hiSt()l‘ical events, such as ecological shifts or a gene duplication, to hypothesize the

Soul‘ce of increased selection.

Yang (1998) used a branch-specific test to demonstrate that a lysozyme gene,

31

present in t
had a highe
the other at
Monkey pr
along the b
that the lyst

the ph}loge

Brunch-Sin

The
combined It
more specil
particular p
for Clumplc
55mg llll'eg'

A ‘h
may hate 8‘
Wired a]
TOrcgmunc
mk‘hes, u
abng m6 \‘p
bl ”‘6 null t

when [he nu

present in the ancestral primate leading to the divergence of the Hominoid species group,
had a higher overall nonsynonymous to synonymous substitution rate ratio compared to
the other ancestral and present-day Colobine, Cercopithecine, Hominoid, and New World
Monkey primates examined. Furthermore, the average to of the lysozyme gene inferred
along the branch leading to the Hominoids was found to be greater than one, indicating
that the lysozyme gene was likely under a divergent positive selection during this time in

the phylogenetic history of primates, rejecting a strictly neutral mechanism of evolution.

Branch-Site Models
The principles of site— and branch- specific estimation of to have also been
combined to design models that are used to test gene evolution hypotheses with even
more specificity. These methods allow one to gain evidence for hypotheses concerning
particular points in evolutionary history. An instance in which ancestral gene duplication,
for example, may have given rise to changes in the selective pressures acting on a gene
bei n g investigated, could be identified.

A ‘branch-site’ test allows one to test for the presence of individual codons that
may have evolved under positive selection along specified branches. The branch
SPeCifred a priori as that hypothesized to be under positive selection, is denoted the
“foreground” branch and is compared to all other branches of the tree, the “background”
bral'l<:hes, with respect to site-specific to distribution. The detection of positive selection
along the specified branch relies on the rejection of neutral evolutionary rates predicted

by the null hypothesis of a fixed on = l for the gene on the foreground branch. Therefore,

when the null hypothesis is rejected, codons are identified along the foreground branch

32

r7
...__<
A~

U11

Ih:

~l.
Ltd

Nil

that exhibit a) both greater than that of the background branch sequences and greater than
1. To increase the rigor with which false positives arise, the subset of positively selected
foreground sites are divided among two categories: 1) a class where the co of background
sites is free to vary from 0 < (o > 1 and 2) a class where the (D of background sites is fixed
at 1. This technique provides a more accurate estimation of the (o of background
branches, to which the foreground sites are compared for evidence of positive selection

(Zhang et al. 2005).

Clade Models

In addition to the branch-site model, the clade model allows evolutionary rate
comparisons to be made simultaneously among codons within a gene and among
branches of a phylogeny. The clade model rate test combines patterns of substitution rate
heterogeneity across a gene sequence and lineage-dependent rate disparities. However,
clade models differ from branch—site models in two important respects: 1) the sequences
under analysis must represent at least two clades, defined as a group that includes all of
the taxa descended from a common ancestor, a situation described as monophyly, and 2)
clade models do not require an (n > 1 to detect a significant difference in evolutionary
rates between foreground and background branches. Statistical comparisons of nested
clade models can show evolutionary rate accelerations or decelerations that represent a
potential increase or relaxation of selective constraint, respectively.

A clade chosen a priori is compared with all other clades on the tree with respect
to its site-specific a) distribution. These two clades are often called the “foreground” and

“background” clades, respectively. Ultimately, individual codons that are evolving at a

33

different rate in one lineage compared to equivalent codons from another lineage are
identified. Individual amino acids may therefore be examined as candidates responsible
for functional divergence within a protein.

The clade model was first used to test for divergence in selective pressure
between the 8 and y globin genes, paralogs which encode subunits of the hemoglobin
oxygen binding protein products in placental mammals (Bielawski and Yang 2003).
Following the gene duplication that created the e and y globins, selection is thought to be
responsible for the divergence in observed expression patterns, leading to delayed, post-
embryonic y globin expression in the simian primate lineage. In contrast, a globin
expression has maintained ancestral gene expression patterns and remains confined to the
embryonic life stage of all placental mammals. Under application of the clade model,
approximately 16% of the codons common to the e and y globins were found to be
evolving under divergent selective pressures, with e globin codons in this rate category
evolving under very strong purifying selection (a) = 0.008) and orthologous y globin
codons in the divergent rate category evolving under weak purifying selection (a) = 0.79).
The twelve codons that comprised the class of divergently evolving sites among the a and
y globin clades were subsequently mapped onto three-dimensional globin protein
structures to verify that the majority of the encoded residues are part of major structural
and functional features of the hemoglobin holoenzyme, one such region being that
responsible for oxygen affinity. The authors concluded that, while the majority of globin
sites evolve at similar rates when the e and y globin clades are compared and display
substantial selective constraint, the twelve codons of the divergent (1) category are

residues likely to have been important for the expression-niche expansion of y globin to

34

the fetal developmental stage following gene duplication.

Tools for Evolutionary Rate Analysis

One popular tool that has been developed to model the heterogeneous nature of
molecular evolutionary rates is the package of computer programs known collectively as
PAML, or Phylogenetic Analysis by Maximum Likelihood. Among other functions,
PAML implements maximum likelihood statistical methods in the context of a phylogeny
to estimate synonymous and nonsynonymous substitution rates. The estimates can then
be used to test hypotheses of site- and lineage-specific a) variation given a sequence
alignment and phylogenetic tree topology. Included within PAML is codeml , a program
that can perform the site-specific, branch-specific, branch-site, and clade model tests. The
user inputs 3 multiple sequence alignment file, a tree topology which describes a
hypothesis of evolutionary relationships among the input sequences, and a control file
which specifies the model with either initial or fixed parameter values.

A strength of PAML is the ability to optimize parameters that define trends
unique to individual data sets of protein-coding sequences through the numerical
maximization of the log likelihood value. The likelihood score is indicative of the
probability of observing a set of data given a particular model of evolution and
phylogenetic tree. Parameters used to describe patterns of sequence change upon which
the model and tree are dependent are optimized simultaneously within the likelihood
score calculation. Optimized parameters include the transition/transversion rate ratio (1c),
and total genetic distance among sequences used to infer branch lengths (t), and

nonsynonymous to synonymous substitution rate ratio (0)). Equilibrium codon

35

fret

CV3

Obs:
an i
It‘ll.
1m

CVO

out;

like

frequencies exert an influence on the optimization of K, t, and w, and are therefore
evaluated by PAML analytically from the sequence alignment.

Recognizing the possibility that multiple local maxima may occur within the
distribution of likelihood values (Suzuki and Nei 2001), it is important to allow PAML to
optimize parameter values using several different initial parameter input values to ensure
that the likelihood space is sufficiently explored. The use of different codon frequency
calculation methods is also encouraged to ensure that the parameters are optimized
robustly and result in the greatest likelihood score. Ignoring codon bias has been
observed to impose an even greater influence on 0) estimations than K, since codon bias is
an influential source of unequal substitution rates among codons (Bielawski and Yang
2004b). Additionally, replicate PAML tests should be performed using alternative input
tree topologies, if multiple tree topologies exhibit strong statistical support. Because the
“true” phylogeny of a set of sequences cannot be known, it is important to show that
evolutionary rate analysis results are not dependent on any one tree topology and that test
outputs are in agreement with a common conclusion (Bielawski and Yang 2004a).

Outputs obtained from multiple runs can subsequently be compared by their log
likelihood scores in a likelihood ratio test, which evaluates the differences between a pair
of nested models with different parameters. In this “goodness-of—fit” test, the simpler
model represents the null hypothesis. To perform a likelihood ratio test, twice the log
likelihood difference between the competing models, defined as the log likelihood test
statistic, is first calculated. The log likelihood test statistic is assumed to approximately
follow a x2 distribution. Therefore, the x2 distribution is used to determine an expected

value of the log likelihood test statistic, using the number of additional parameters

36

incorporated into the more complex model relative to the simpler model, as the
appropriate degrees of freedom. The null hypothesis is accepted if the log likelihood test

statistic falls within the expected distribution (Bielawski and Yang 2004a).

Methodological Limitations

Interpreting the role of selection on a gene through (D estimations of protein-
coding regions has the potential to be misleading. For one, the calculation of d5 ignores
the cases where a nucleotide substitution that fails to change the encoded amino acid may
confer a fitness difference. The value of ds may therefore be erroneously assumed to be a
rate of neutral mutation. For example, biased abundances of iso-accepting tRN As
containing different anticodons, within the cellular pool of tRNAs, may result in
differential translational efficiency of sequences containing different nucleotide triplets
for the same amino acid. Synonymous substitutions may also violate the assumption of
neutrality when a nucleotide is shared between genes, as in the case of genetic material of
many viruses (Diamond et al. 1989) for which the mutation is nonsynonymous for an
overlapping reading frame. Moreover, nucleotide changes may affect the stability of
DNA or RNA molecules if the substitution results in disruption of secondary structure
through elimination of a crucial hydrogen bond. Hammerhead ribozymes, for instance,
rely on stem-loop features for recognition, binding, and subsequent cleavage of substrates
(T uschl and Eckstein 1993).

In addition, the alignment of DNA and amino acid sequences is implied to be
error—free, such that each nucleotide within a ‘column’ corresponds to the same codon

position of all other genes. However, the “true” alignment of a group of sequences is

37

unknown, and even computer programs using sophisticated algorithms to align sequences
can only make an inference of sequence relationships. Assessment of simulated DNA
sequence data alignments has shown that the reliability of computer— generated alignments
for correctly recognizing homologous sites decreases when the length of sequences that
contain insertions and deletions is increased (Nuin et al. 2006). Similarly, the estimation
of site or lineage-specific to values relies on the topology of a cladogram which serves as
a description of the ancestral origins and relationships among the sequences in question.
However, cladograms represent hypothesized phylogenetic relationships; the true
phylogenetic history of a set of gene sequences can never be known with certainty.

Furthermore, factors other than positive selection can cause an 0) >1. For instance,
the severe reduction in population size caused by a population bottleneck can decrease
the effectiveness of purifying selection, allowing deleterious mutations that would
otherwise be eliminated, to rise to fixation and oppose selection via drift. In some cases,
the random nature of mutation may result in the absence of synonymous substitutions.
Thus, a codon may show a) >1 simply due to the stochastic nature of mutation.
Likelihood tests of evolutionary rate heterogeneity do not yet allow such alternative
explanations for w >1 to be statistically considered (reviewed in Hughes, 2007).

Finally, and perhaps most importantly, recovering molecular signatures indicative
of the direction and intensity of selection are not adequate to make conclusions about the
phenotypes and subsequent fitness effects of observed mutations. Instead, evolutionary
rate analysis should be used as a springboard for the formulation of hypotheses that may
directly (i.e. biochemically, at the molecular level) investigate the fitness costs and

benefits to organisms conferred by the products of genes evolving at elevated or

38

dec:

ecoi

decelerated rates. Ultimately, the goal of this line of research should be to seek the

ecological origins for evolutionary forces that lead to adaptation.

39

SECTION II: EXPERINIENTAL STUDY

HYPOTHESES AND PREDICTIONS

We sought to investigate the evolutionary patterns of the mitochondrial heat
shock proteins involved in Fe/S cluster biogenesis: the paralogous genes SSC] and
SSQ1, plus their interacting J-protein partner, J AC1. Motivation for this study comes
from the observation that SSQ1 represents an example of a heat shock protein that has
become specialized in a particular sub-function of its ancestral gene, and interestingly,
one unusual to chaperones. In this study, we analyzed sequences of monophyletic fungal
groups of comparable within-clade relatedness, two of which diverged from a common
ancestor prior to the gene duplication that created SSQ1 (Aspergillus and F usarium), and
two of which diverged after the duplication event (Saccharomyces and Candida) (see
Figure 2). The presence of J AC1 within each of these clades has given us the opportunity
to investigate how the duplication of the ancestral thsp70 has influenced the
evolutionary paths of SSCl, SSQ1, and J AC1, via extensive comparative analyses of the

rate of gene sequence evolution.

40

SSC1 SSQ1 JACl

'I'III'I'

   
 

Candida

III'd-"I"

+-'l-

M
<+—+

Figure 2: Summary of mitochondrial heat shock protein (thsp) distribution among
fungal clades. The monophyletic fungal groups examined were the Saccharomyces,
Candida, F usarium, and Aspergillus clades. All four clades encode the gene for the
multi-functional thsp70 SSC 1, as well as the interacting mitochondrial J -protein co-
chaperone encoded by JAC] present in all four clades. The “+” and “-” symbols represent
the presence or absence of a gene within a clade, respectively. In the lineage indicated by
the star, prior to the divergence of the Saccharomyces and Candida clades, the ancestral
thsp70 underwent a gene duplication, which gave rise to the gene for the specialized
thsp70 SSQ1 carried in Saccharomyces and Candida taxa.

 

Our objective was to elucidate molecular patterns of selection to test the following

hypotheses:

Hypothesis 1: Selective constraint has been relaxed in 88C] in the presence of its
paralog, SSQ1.

Ho: The rate of SSC] evolution in clades encoding the paralogs 8801 is equal to the
rate of 88C] evolution in clades that lack SSQl

Inability to reject the null would suggest that SSC] and SSQ1 paralogs are
equivalent and therefore, functionally redundant. This outcome seems unlikely, given that
evidence indicates SSCl and SSQ1 cannot replace one another and rules out the

functional equivalence of the encoded proteins.

41

H1: The rate of 88C] evolution in clades encoding the paralog SSQ1 is not equal to
the rate of 88C] evolution in clades that lack 8801

Evidence to support this hypothesis would be consistent with a relaxation of
selective constraint following duplication of the ancestral thsp70 gene if SSCl is
evolving at a faster rate than SSQ1. Our a priori prediction is that the rate of SSC 1
evolution will be elevated in clades encoding SSQ1, versus clades that lack SSQ1.
Biochemical evidence for the increased affinity displayed by J aclp for Ssqlp suggests
that Ssqlp may be capable of fulfilling the role of Ssclp in Fe/S cluster biogenesis. Thus,
SSQ1 may have the ability to compensate any loss-of-function mutations affecting SSC 1
at sites important for Fe/S cluster biogenesis. SSQ1 would negate the need for SSCl to
maintain sites used for the Fe/S cluster assembly pathway, allowing a greater proportion
of mutations to be fixed at these sites in SSCl.

Alternatively, the rate of 88C] evolution in clades encoding SSQ1 could be
decreased relative to the rate of SSC 1 evolution in clades that lack SSQ1. Evidence for
this result would be consistent with an increase in selective constraint on SSCl when co-
occurring with SSQ1. It is difficult to identify possible sources of increased constraint on
SSCl, given that evidence does not exist to suggest that SSCl has attained a novel

function or adaptive peak since the thsp70 duplication event.

Hypothesis 2: 8801 is under less selective constraint than SSCl because 8801 has
fewer encoded functions to maintain.

Ho: 8801 and 88C] evolve at equal rates
Inability to reject the null would be consistent with the conclusion that SSQ1 and

SSC] are not under current divergent selective pressures. One possible explanation for

42

this result could be that, while functional divergence of Ssqlp and Ssclp occurred in an
ancestral lineage, current selective pressures in extant taxa are now acting with the same
direction and magnitude on each paralog. However, the functionally divergent paralogs
interact with different groups of substrates and therefore have different sources of
possible coevolutionary influence. Thus, it seems unlikely that SSCl and SSQ1 would be
evolving at equal rates.

H1: 8801 and SSCl evolve at unequal rates

Evidence to support the alternative hypothesis would be consistent with the

functional specialization of Ssqlp in Fe/S cluster biogenesis if our a priori prediction,
that SSQ1 is evolving at an elevated rate compared to SSCl, is observed. According to
biochemical experiments, J aclp stimulates the ATPase activity of Ssqlp to a greater
extent than Ssclp of both pre- and post— thsp70 duplication species. Therefore,
mutations must have been fixed in SSQ1 since the time of the duplication event to afford
functional distinction from SSCl. Additionally, Ssqlp has a diminished functional
repertoire. Therefore, SSQ1 must have undergone fixation of mutations that result in loss
of function. On the other hand, Ssclp has not been shown to have gained any novel
functions subsequent to the creation of SSQ1. Because Fe/S cluster biogenesis constitutes
only one of many roles encoded by SSCl, any loss of performance in Fe/S cluster
assembly sustained by SSC] is predicted to have occurred with a small number of
mutations. The number of mutations likely to have occurred in SSQ1, to degenerate the
many lost roles in protein folding and translocation, would be comparably large.
Therefore, a greater number of mutations are likely to have occurred in SSQ1 than SSCl

since the time of duplication.

43

Alternatively, SSQ1 could be evolving at a decreased rate compared to 88C]. An
increase in selective constraint on SSQ1 would be a possible explanation for this result.
However, because fewer functions have been ascribed to Ssql, relative to Ssclp, this
outcome would support the need to further investigate the functions of Ssql to identify

additional sources of constraint that could be acting on SSQ1 compared to SSC 1.

Hypothesis 3: The rate of JACl evolution is positively correlated with the rate of
SSQ1 evolution because JACl and 8801 are coevolving.

Ho: The rate of JAC] evolution in clades encoding 8801 is equal to the rate of JAC]
evolution in clades that lack SSQl

Inability to reject the null would be consistent with the absence of an influence by
SSQ1 on the direction and magnitude of selection acting on J AC1. This outcome does not
seem likely, given that J aclp has been demonstrated to result in different magnitudes of
ATPase stimulation for Ssqlp and Ssc 1 p.Therefore, the selective pressures exerted by
Ssqlp and Ssclp on Jaclp are probably not equivalent. Alternatively, the increased
efficiency of the J aclp - Ssql interaction could be due to changes at very few sites in
J AC1, or entirely independent of J AC1 evolution, resulting from the specialization of
Ssql alone.

H1: The rate of JACl evolution in clades encoding SSQ1 is not equal to the rate of
JAC] evolution in clades that lack 8801

Evidence to support this hypothesis would be consistent with the coevolution of
J AC1 with a duplicate thsp70 under increased or decreased selective constraint relative
to the ancestral pre-duplicate thsp70. Our a priori prediction is that J AC1 evolves at an
increased rate in the presence of SSQ1, compared to J AC1 from clades that lack SSQ1.

Because Jaclp must stimulate the ATPase activity of an thsp70, evolution of J AC1

44

would be necessary to accommodate any changes in a thsp70 that might hinder the
ability of J aclp to physically interact with the thsp70. Molecular coevolution of J AC1
with SSQ1 could account for the specialized interaction that has given rise to the ability
of Jaclp to stimulate Ssqlp to a greater extent than Ssclp. Furthermore, because SSQ1 is
a duplicate gene, it is expected to be under relaxed selective constraint compared to
thsp70s in the single gene state. Therefore, if SSQ1 is evolving at a faster rate, the rate
of JACl evolution would be expected to accelerate when co-occurring with SSQ1 to
maintain the ability to physically interact. This assumes that the faster rate of SSQ1
evolution is due to changes at sites critical to interaction with J AC1. Regardless of
whether SSQ1 — JACl coevolution was instigated by initial changes in SSQ1 or J AC1,
the exertion of reciprocal selective pressures could result in correlated rate acceleration of
SSQ1 and JACl. Thus, J AC1 would be observed to evolve faster in clades encoding
SSQ1 compared to clades lacking SSQ1. Conversely, if the evolutionary rate of SSQ1 is
observed to be slower than that of SSC], we predict the rate of J AC 1 evolution will be
decelerated in the presence of SSQ1.

A negative correlation between J AC1 and SSQ1 evolution would be indicative of
antagonistic coevolution. A coevolutionary relationship of this nature could result if
either JACl or SSQ1 constrain the evolution of the other, such as if the proteins had
reached an adaptive peak in their interaction. Alternatively, another factor (perhaps an
unidentified component of the Fe/S cluster biosynthesis pathway) could increase selective
constraint on J AC] or SSQ1, while releasing constraint on the other.

Hypothesis 4: 8801 has undergone adaptive evolution to optimize the Ssql - Jaclp
interaction important for Fe/S cluster biogenesis.

Ho: 8801 has not evolved under positive selection

45

confer
relarat
the di\

11 5.1168

fl'olut
functit
111 the

33km

Inability to reject the null would be consistent with the fixation of mutations that
confer the functional differences between Ssqlp and Ssclp to have occurred by a
relaxation of selection and/or genetic drift. A possible evolutionary history to account for
the divergence of Ssqlp without positive selection would include relaxation of constraint
at sites required for protein folding, translocation, and stress responses. A relaxation of
selective constraint at those sites would allow deleterious mutations to accumulate and
degenerate the encoded functions. The increased ATPase activity in the presence of J ac l p
could have arisen in SSQ1 due to the random fixation of beneficial mutations with weak
fitness effects. Alternatively, the increased ATPase activity could be due to evolution
within JACl alone.
H1: 8801 has evolved under positive selection

Evidence to support this hypothesis would be consistent with a period of adaptive
evolution in the history of SSQ1. The premise for this proposal is that SSQ1 has become
functionally specialized since its divergence from SSCl. SSQ1 shows increased activity
in the presence of J AC1, an improvement of ancestral function important for Fe/S cluster
assembly. To improve upon the ancestral function, SSQ1 must have acquired mutations
beneficial to the Jac lp-Ssqlp interaction, potentially at sites critical to physical contact
between the two proteins. If increased efficiency of Ssqlp ATPase stimulation by J aclp
results in an increase in adaptive fitness, perhaps by improving Fe/S cluster biogenesis,

then positive selection could drive the new SSQ1 allele to fixation.

46

METHODS

Fungal Taxa and Gene Sequence Alignments

Gene sequences were retrieved from seven to eight fungal species, from each of
four monophyletic clades. The taxa from two of these clades, the Saccharomyces and
Candida groups, encode the duplicate Hsp70, SSQ1, while the Aspergillus and F usarium
clades lack SSQ1. SSC], SSQ1, and J AC1 coding region sequences (exons only) were
taken from Saccharomyces cerevisiae RM1 11, Saccharomyces cerevisiae YJM789,
Saccharomyces paradoxus, Saccharomyces mikatae, Saccharomyces bayanus,
Saccharomyces castellii, and Candida glabrata genomes, which comprise the
Saccharomyces clade, and from Candida lusitaniae, Candida guilliermondii,
Debaryomyces hansenii, Candida parapsilosis, Candida tropicalis, Candida dubliniensis,
and Candida albicans genomes, which comprise the Candida clade. SSCl and J AC1
sequences from the Aspergillus nidulans, Aspergillus niger, Aspergillus terreus,
Aspergillus oryzae, Aspergillusflavus, Aspergillus clavatus, Aspergillusfirmigatus, and
Neosartorya fischeri genomes comprise the Aspergillus clade, while sequences from the
Podospora anserina, Trichoderma reesei, F usarium solani, F usarium graminearum,
F usarium verticilliodes, and F usarium oxysporum genomes comprise the F usarium
clade. The complete genome of each of the above fungal species has been sequenced and
has been made available through sequence databases curated by the Saccharomyces
Genome Database, the BROAD Institute, the Joint Genome Institute, the Wellcome Trust
Sanger Institute, and Génoscope. Nucleotide and protein sequence BLAST searches were

performed to identify orthologs, with reciprocal best BLAST hits used to confirm

47

orthology and reject paralogy of SSCl and SSQ1 sequences. The sources and genome
coordinates of all sequences are presented in Appendix A.

Because J AC1 is a fast evolving gene and differs by more than 80% at the
nucleotide level between fungal clades, J AC1 is too divergent to confidently generate a
multiple alignment of J AC1 sequences from all four fungal clades. Therefore, it was
necessary to carry out I AC1 sequence alignments, subsequent gene tree construction, and
rate analyses, separately for each of the four fungal clades. However, a similar average
I AC1 sequence divergence and number of taxa for each clade facilitates comparison of
JAC] sequences among the clades (see Figure 3). Within each clade, any two JAC]

nucleotide sequences differ by approximately 25% to 35%.

Average Within-Clade Pairwise Sequence Divergence

 

0.5 7

 

r: Clade: encoding 8801
. Clade: lacking $801

 

Uncorrected p-distances

 

 

 

 

 

 

Saccharomyces Candida Aspergillus Fusarium

Figure 3: Average within-clade pair-wise sequence divergence of J AC1. The average

I AC1 divergence, calculated as the uncorrected p-distance, between any two J AC1
sequences encoded by taxa belonging to the same clade is shown. P-distances are
expressed as the percent sequence dissimilarity. Error bars represent standard deviations.

48

Multiple alignments of translated amino acid sequences were performed using
CLUSTAL W (Thompson et al. 1994), with default gap penalties and subsequent manual
trimming to remove gaps. The J AC1 alignment of each clade contains the following
number of amino acids: Saccharomyces: 177, Candida: 167, F usarium: 189, and
Aspergillus: 185. In contrast, the more conserved nature of the thsp70 genes permitted
alignment of sequences from all taxa. The SSCl alignment includes 603 amino acid sites
from all four fungal clades, and the combined SSCl and SSQ1 alignment includes 580
amino acid sites from SSCl of all four fungal clades and SSQ1 from the Saccharomyces

and Candida clades. All amino acid alignments are shown in Appendix B.

Cladogram Construction for PAML Input Trees

Data Partitioning

Figure 4 depicts a graphical summary of gene tree construction. Analysis of
separate partitions of data with independent evolutionary models has been demonstrated
to fit heterogeneous data better when compared to un-partitioned data. Further, data
partitioning may also yield support for alternative tree topologies (DeBry 1999). Analysis
of partitioned sequence data is a technique used to accommodate evolutionary
heterogeneity within subsets of the sequences. The first and second nucleotide positions
of codons within protein coding regions are expected to evolve at a slower rate than third
positions, due to the fact that most substitutions at third positions are synonymous. At
first positions, however, most substitutions are nonsynonymous, and all substitutions are

nonsynonymous at second positions. Therefore, selective constraint is expected to be

49

weakest for third positions, of intermediate strength at first positions, and strongest for
second positions. Thus, the fastest rate of change is expected to take place at third
positions and result in greater ability to resolve phylogenetic relationships among closely
related or slowly evolving sequences. In such cases where few sequence changes are
expected to have accumulated among taxa, first and second positions may not contain
sufficient variation to resolve evolutionary histories. First and second positions are often
more useful in resolving deep branches of a phylogenetic tree, where the sequences in
present-day taxa may be very divergent. Given greater sequence divergence, the chance
increases for third positions to become saturated with homoplasies, at which point these
nucleotides no longer provide a reliable signal to distinguish basal relationships.

Tree construction of translated amino acid sequences is another method to achieve
robust branch resolution, given evolutionary rate variation within genes. A model used to
describe patterns of amino acid substitution may be more appropriate than models that
use DNA units of evolution, and is useful to complement results of cladogram
construction using nucleotide substitution models. While information held within DNA is
lost when sequences are examined at the amino acid level due to the degenerate nature of
the genetic code, modeling amino acid substitutions releases analyses from biases in
nucleotide base composition and mutation more prevalent in nucleotide sequences. For
example, unlike peptides, nucleotide evolution is often influenced by structural
constraints that favor a particular nucleotide sequence for hairpin or loop regulatory
features that result when the DNA is transcribed into RNA. Selection for codon bias also
falls into the category of nucleotide compositional bias. Moreover, far more character

types make up peptide sequences compared to DNA sequences (there are more types of

50

amino acids than nucleotide bases), therefore making amino acids less prone to mutations
that revert a site back to its ancestral state. Additionally, because amino acid substitutions
often require more than one nucleotide substitution, the rate of amino acid evolution is
slower than that of nucleotides. Together, the reduced homoplasy and slower
evolutionary rate observed at the amino acid level confers the advantage of better
phylogenetic resolution of distantly related taxa or fast-evolving genes than might be
possible by nucleotides.

Therefore, different subsets of the sequence data were considered here
individually. Unrooted gene trees were constructed using the following partitions: first,
second, third, first with second, and all nucleotide positions of codons, as well as amino

acids.

Maximum Parsimony

Constructing the phylogenetic tree topologies to be used in the estimation of
evolutionary rates is a critical initial step that can be accomplished using several different
methods of inference. Ideally, given a set of properly aligned sequences, the inferred
phylogeny would be identical, regardless of the method used to construct the tree, if the
“true” evolutionary history is to be accurately represented. In practice, however, each
method of phylogenetic tree construction possesses unique strengths and pitfalls, and
therefore, can influence the outcome of phylogenetic analyses. For this reason, it is
prudent use more than one method in parallel, and, given a sequence data set, to examine
alternative trees in subsequent analyses when possible. Maximum parsimony (MP),

maximum likelihood (ML), and the Bayesian inference (BI) methods were used in this

51

study.

The MP method of phylogenetic inference is a character-based method that seeks
to recover tree topologies that minimize the number of evolutionary transitions necessary
to explain the distribution of characters among taxa (Hennig 1966). A tree search
algorithm is used to evaluate tree topologies according to the minimum number of steps
required. The occurrence of convergent evolution, parallel evolution, or character
reversals to the ancestral state, may cause two sequences to appear more closely related
than they actually are. These are sources of homoplasy; the opportunity for their
occurrence increases with the time since divergence from a common ancestor and are
assumed to be minimized in the most parsimonious tree. However, MP tree construction
has a tendency to erroneously group highly divergent sequences together, particularly
when the sequences are distantly related or have undergone very rapid evolution. This
problem is known as long-branch attraction (Felsenstein 1978).

MP trees were constructed in PAUP* v 4.0b10 (Swofford 2000), with a heuristic
search using the tree-bisection-reconnection (T BR) branch-swapping algorithm and equal
weighting for all characters. The TBR method of tree searching starts with an initial tree
topology, breaks the tree into two sub-trees, and then reconnects the halves at all possible
nodes. Here, the initial topology was generated by the random, stepwise addition of
sequences and heuristic tree search proceeded by random addition sequence replications.
One hundred bootstrap replacement replicates were performed to determine statistical

support for branches of each topology.

52

Maximum Likelihood

Another commonly used method for inferring phylogenies is the MLmethod
introduced by Felsenstein (Felsenstein 1981). ML tree construction can use many
different models of sequence substitution in conjunction with the powerful statistical
inference of optimizing a likelihood function. This allows the ML method to more
efficiently distinguish homoplasy from synapomorphy, an advantage that provides greater
accuracy of phylogenetic inference of very divergent taxa or sequences with very
different rates of evolution, compared to the MP method. The ML method examines all
possible pathways of sequence change possible for a given data set in order to identify
the hypothesis most likely for the data. Within the likelihood calculation used to evaluate
hypotheses, the tree topology, branch lengths, and evolutionary model components are
simultaneously optimized. When these parameters have been optimized to maximize the
likelihood, the best evolutionary model and tree have been found (according to ML). This
is analogous to reaching a peak in a multi-dimensional parameter landscape. Parameter
values at the peak reached in the parameter space are estimated from the data, and
therefore do not need to be specified a priori by the investigator before examining the
data (Holder and Lewis 2003). However, ML method calculations are computationally
intensive and may propose an incorrect evolutionary relationship if an inappropriate
substitution model is chosen (Huelsenbeck and Crandall, 1997).

Maximum likelihood trees were inferred using PhyML v2.4.4 (Guindon and
Gascuel 2003) by applying the general time reversible (GTR) model of nucleotide
substitution. The GTR model estimates an independent frequency with which each

nucleotide base is observed within a set of sequences and an independent substitution rate

53

for each pair of nucleotide substitutions. Additionally, each substitution type is assumed
to be equally reversible to allow, for instance, G —> T and T —> G to occur at equal rates.
Furthermore, parameters such as the proportion of invariant sites and the gamma shape
parameter, used to describe the distribution of substitution rates among sites, account for
site-to-site evolutionary patterns. All parameters for the model were estimated from the
data, with four discrete categories in the gamma rate distribution. The amino acid model
of substitution indicated as the best model for protein evolution by ProtTest v1.4
(Abascal et al. 2005), according to a likelihood ratio test, was specified for each multiple
amino acid sequence alignment as follows: Saccharomyces J AC1: WAG, Candida J AC1:
RtREV, Fusarium JACl: WAG, Aspergillus JACl: J'I'I', SSCl: RtREV, SSCl and SSQ1
combined: RtREV.

A neighbor-joining tree was generated in PhyML to serve as the starting tree in
the tree search. A hill-climbing algorithm was then used to optimize the maximum
likelihood. One hundred bootstrap replicates were performed to determine statistical
robustness of trees and yielded a bootstrap consensus tree used to assess clade support for

both MP and ML methods.

Bayesian Inference

The BI method of phylogenetic reconstruction resembles the ML method in that
the BI method can incorporate many different molecular evolutionary models in the
search for the best tree. However, the BI method samples from the posterior probability
distribution to identify the most probable phylogenetic tree, given a data set. This

requires an investigator to assign prior probabilities for all parameters, i.e. predictions

54

made before examination of the data, a potential source of bias that some consider a
disadvantage of the BI method (Felsenstein 2003). In phylogenetic analysis, prior
probabilities are usually given an uninforrnative or “flat” distribution, to regard all
possible trees as equal hypotheses until the data are examined.

While the BI method evaluates the likelihood of a hypothesis to calculate the
posterior probability, parameters are not optimized as in ML. Instead, the Markov Chain
Monte Carlo (MCMC) algorithm is used to estimate the probability distribution of a
hypothesis. This algorithm constructs chains to move from one location to the next within
a multi-dimensional space of hypotheses, periodically sampling the posterior probability,
and moving toward successively greater probability densities. Each “link” within the
chain, or location within the tree space, is termed a “generation.” The goal is to reach an
equilibrium posterior probability distribution, at which time a move to a new location
within the tree space does not yield a greater posterior probability. Separate chains
running in parallel converge at similar posterior probability values. To avoid becoming
stuck in local regions of high posterior probability density and allow more efficient
exploration of the hypothesis landscape, the tree and evolutionary model evaluated at a
location by one chain may be periodically swapped between other parallel chains
(reviewed in Holder and Lewis, 2003). Heated chains are freer to traverse peaks and
valleys in the landscape, and are thus useful when posterior probabilities are swapped
among parallel chains. The cold chain is more restricted in its movement and is the chain
from which sampled posterior probabilities are used as the output for a run. Because the
initial locations of the chains (termed the “bum-in”) in the tree space is often far from the

greatest posterior probability density, a proportion of the first generations are discarded

55

from the final evaluation of posterior probability distributions.

Bayesian inference trees were constructed in MrBayes v3.1.2 (Ronquist and
Huelsenbeck 2003) using the same nucleotide substitution model as described for ML
tree construction of nucleotide sequences and mixed model optimization for amino acids.
The default assumption of flat prior probability densities was implemented for all
parameters. Two parallel Markov chain Monte Carlo processes were initiated, consisting
of three hot chains and one cold chain. The chains were run for 1,000,000 generations
each, with a sampling frequency of once per 100 generations. The initial 2,500 trees were
discarded as the burn-in. Chain parameter and tree convergence within one run and
between parallel runs was assessed by likelihood scores. When the likelihood scores of
the cold chains were no longer increasing and showed fluctuation within a narrow range,
the chain was assumed to have reached stationarity within the parameter space. In
addition, plots of generation versus the log posterior probability were also generated for
each run to visually detect stationarity via absence of increasing or decreasing posterior

probability value trends.

Constructing Composite Input Tree Topologies

For each method of tree construction, the most parsimonious or most likely (as
appropriate to the tree method) trees were visually examined for branch resolution on
bootstrap consensus trees. Note that in instances where more than one tree topology was
returned as the most parsimonious tree by the MP method, computation of the bootstrap
consensus negated the need to examine multiple MP trees for each data partition.

Bootstraps of Z 90% or posterior probability values of _>_ 0.9 were considered sufficiently

56

well supported. In cases where branch resolution could not be achieved using one data
partition, but could with another, branches were manually inserted to produce the best
resolved, composite tree. In instances where evolutionary relationships among sequences
could not be resolved with high statistical support by any combination of sequence
partitions, tree branches were collapsed into polytomies. The number of unique
composite tree topologies obtained by each phylogenetic inference method for each gene
alignment are as follows: Saccharomyces J AC1: 2, Candida J AC 1: 3, F usarium J AC 1: 1,
Aspergillus J AC1: 3, SSCl: 11, SSC] and SSQ1 combined: 9. All composite tree

topologies used as input trees for codeml are displayed in Appendix C.

57

1‘t 2”

/‘——rfi

1" 2'

1‘2

Figur
1316 E
three
Ukel
were

that

play}.

530:1
arm
SSC

M

El‘ol

Maximum Parsimony (PAUP*)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Composite
Partitioned Sequence Data ]Branch sulDPOrt inputtree
assessment -
,1 u {l u u U
1d 20!! 3M 1“ 20d 2nd ammo
acids
Maxrmum Likelihood (PhyML) Composite
Partitioned Sequence Data I BfaflCh SUPP!”t inputtree
assessment
topologies
V v v V V
1st 2nd 3"I 1st 2Ml 1st 2"“ 3" mm
acids
Bayesran Inference (MrBayes) Composite
Partitioned Sequence Data g ]Branch support in uttree
]assessment p
topologies
1n 2M3” 1“ 2nd 2,“, amino
acids
toAnalysis
(codeml)

Figure 4: Summary of data partitions and phylogenetic tree construction for evolutionary
rate analysis. J AC1, SSCl, and combined SSC] and SSQ1 gene trees were inferred using
three different methods of tree construction: Maximum Parsimony, Maximum
Likelihood, and Bayesian Inference. Five different partitions of each sequence alignment
were used individually for tree construction: all three nucleotide bases of a triplet codon,
the 1St and 2“d nucleotide positions only, each nucleotide position individually, and the
amino acid residues. Each tree resulting from each data partition analyzed by the three
phylogenetic methods was assessed for branch support to generate all possible unique,
strongly supported composite topologies for to analysis using the cod eml program of
the PAML package.

thsp70 Clade Model Rate Comparisons

To investigate the potential role selective constraint has played in the evolution of
SSCl since the inception of SSQ1, as stated in Hypothesis 1, a clade model test was
performed to examine the influence of the presence of SSQ1 on the rate of SSCl

evolution. The rate of SSCl codon evolution from taxa possessing SSQ1 (Saccharomyces

and Candida clades) was compared to the rate of SSCl codon evolution from taxa

58

lacking SSQ1 (Aspergillus and F usarium clades) via application of the clade model rate
test. The foreground clade (SSC 1 from taxa co-occurring with SSQ1) was distinguished
from the background clade (SSCl from taxa lacking SSQ1) at the node representing the
most recent common ancestor of the Aspergillus and F usarium sub-clades for each of the
eleven input trees (see Figure 5A). Due to a program glitch that we found in the
application of model=3 in PAML v. 4.0 (Bielawksi 2008), clade model analyses were
carried out in version 3.0 of PAML (Yang 1997).

Tests were conducted under three different clade models that varied in the number
of pre-defined rate categories. The null model had one rate category, while the two
alternative models were specified to have either two or three rate categories. likelihood
ratio tests were then performed to determine the most appropriate rate test model.
Likelihood ratio tests comparing models are shown in Appendix E. For all clade models,
initial 0) and K values of 0.5, 1.0, and 1.5 were tested. In addition, two different methods
of codon frequency adjustment were applied in all codem'l tests: 1) codon frequency
model F3x4, where 15‘, 2nd, and 3rd base frequencies from the data were used to estimate
codon frequencies, and 2) a table of codon frequencies observed within the data. Because
varying initial a) and K values and codon frequency models does not alter the number of
parameters used by the model, the effect of altering these settings cannot be determined
by a likelihood ratio test. However, clade tests conducted with initial values set to 0.5 and
1.0 tended to give slightly higher likelihood scores than with initial values set to 1.5.
Initial values of 0.5 and 1.0 gave very similar, and often identical, likelihood scores. Use
of the observed codon frequency table always resulted in the highest likelihood scores of

all codem'l tests. The output that yielded the highest likelihood score is reported in the

59

RESULTS.

To investigate the possibility that SSQ1 is under weaker selective constraint than
SSCl, as stated in Hypothesis 2, clade model rate analyses were also performed as
described above to compare the rate of SSQ1 sequence evolution to that of SSC 1.
Cladograms constructed for all taxa in a single tree, with gene sequences from both
thsp7OS, were divided into a foreground clade of SSQ1 and background clade of SSC 1

(see Figure 5B).

Site-Specific Rate Tests

To investigate the potential role that the thsp70 gene duplication played in
altering the rate of J AC1 evolution in the presence of SSQ1, as outlined by Hypothesis 3,
J AC1 site-specific rate tests were conducted. J AC1 sequences were separately evaluated
for each of four fungal clades using a site-specific model of gene evolution applied in the
codem'l program of version 4.0 of PAML (Yang 2007). The site-specific rate model is
used to estimate to values for a pre-defined number of rate categories and, subsequently,
each codon is assigned to the most likely category. We used this test to look for evidence
of increased or decreased selective constraint acting on individual amino acids in
sequences derived from clades in which J AC1 co-occurs with both Hsp70 paralogs, SSC 1
and SSQ1, compared to J AC1 sequences obtained from clades possessing only SSCl.

Each J AC1 cladogram was subjected to rate analysis using models consisting of
either three or ten possible rate categories. High and low initial values (1.3 and 0.3) for w
and K were tested, and it was found that in all cases the analyses reached convergence

under both starting values for both parameters. Codon frequency models were varied as

60

described above for the clade model rate analyses above. The number of 0) categories
which best modeled site-specific rates of evolution for each J AC1 clade was determined
by likelihood ratio tests (see Appendix E). The use of ten rate categories was found to
confer a significantly greater likelihood of predicting the data for the Aspergillus clade
when either the BI or ML input trees were used. The results obtained from the simpler

model, using three rate categories, was superior in all other cases.

Branch-Site Test

To investigate the possibility that positive selection played an historical role in the
adaptation of SSQ1 to Fe/S cluster biogenesis specialization, as stated in Hypothesis 4,
we conducted a branch-site test to analyze the codon-specific selection pressures of the
ancestral SSQ1. The ancestral SSQ1, which existed immediately after the thsp70 gene
duplication, was defined as the foreground branch (see Figure 5C). We expected sites
along the foreground branch to show evidence of positive selection. The model placed
each codon into one of four 0.) rate categories, with restrictions placed on a) values as
shown in Table l. Codons were placed into two classes for which a) was constant among
ancestral and descendent sites, and two classes for which 0) was variable between
ancestral and descendent SSQ1 sites. The value of a) was estimated to be 0 < (0 < 1 for
common rate class 1, while the proportion of sites with 03:10 shared among ancestral
and descendent sites was estimated for common rate class 2. To test the alternative model
of evolution under selection, the estimated a) of ancestral SSQ1 sites was free to vary
with (o > 1, while holding descendent SSQ1 sites at 0 < a) < 1 for divergent rate class 1.

The background a) for divergent rate class 2 was held at 1.0. Posterior probabilities for

61

site classes were calculated by the Bayes empirical Bayes (BEB) method (Zhang and
Yang 2005). The same eleven SSCl and SSQ1 combined gene trees used for clade model
analyses were input into the branch-site test, with a 3X4 codon frequency model and the
parameters K and to estimated from the data. The results of these tests were compared by
likelihood ratio test to the null model under which all sites of the ancestral SSQ1 branch

evolving at a divergent rate were modeled with a fixed a) = 1.

 

 

 

Table 1: Evolutionary Rate (to) Estimation Under the Branch-Site Model
“0 Hr
Evolutionary Rate Class Descendent lineages Ancestral SSQl Ancestral SSQl
(background) (foreground) (foreground)
Common rate class 1 0 < w >1 0 <0» <1 0 <w <1
Common rate class 2 w = 1 w = 1 w = 1
Divergent rate class 1 O <w <1 w = 1 w > 1
Divergent rate class 2 w = 1 w = 1 w > 1

 

62

   

'IIIIII. IIIIOIOIO.
COO-IOIIIIOIICOIII'

 

l

mesa

 

l;

mesa

 

 

     

Figure 5: a priori defined lineages used for clade and branch-site model input trees. The
phylogenetic relationships among SSCl fungal sequences, shown in dark gray, and SSQ1
fungal sequences, in light gray, are depicted in these simplified schematic trees. Dotted
boxes are used to encompass foreground clades in trees A and B. The clade model was
used to test for divergent selection pressures among SSC] of pre- and post- thsp70
duplication clades (A), and among SSQ1 and 88C] (B). The branch-site test was used to
look for evidence of positive selection along the highlighted ancestral SSQ1 branch (C).
A starred thick gray line is used to indicate the foreground lineage in tree C, representing
the ancestral SSQ1 sequence present following the mitochondrial heat shock protein 70
(thsp70) gene duplication event and prior to the divergence of the Saccharomyces and
Candida SSQ1 clades

63

RESULTS

SSCl evolution accelerated in the presence of $801

The clade model test was conducted to examine whether altered selective
constraint affected the evolutionary rate of SSCl in the presence of SSQ1 (Hypothesis
1). Rates of codon evolution were compared between SSCl DNA sequences derived
from fungal clades that differed with respect to the presence of the fungal paralog, SSQ1
(Candida and Saccharomyces vs. Aspergillus and F usarium, harbor the presence and
absence of SSQ, respectively). The purpose of this test was to identify the proportion of
SSCl codons evolving at different rates between those SSCl sequences that co-occur
with SSQ1 (foreground clade) and those evolving in the absence of SSQ1 (background
clade), and to determine the w of those sites evolving at differential rates. The results
presented were obtained using the SSCl MLTree 2, the tree that gave the highest
likelihood score when used as the input tree. Similar results were attained with all tree
topologies tested, and are thus independent of tree topology. More than half (61.6%) of
sites in all SSCl genes exhibited an a) of 0.001 (common rate class 1), and just under a
third (28.8%) of sites showed an a) value of 0.038 (common rate class 2), regardless of
the presence or absence of the duplicate gene (Figure 6). However, about 9.6% of SSC]
codons differ in their rate of evolution, depending on the presence or absence of SSQ1
(Figure 6). The faster evolving codons, belonging to clades lacking SSQ1, show an m of
0.107. In contrast, these same SSCl codons evolved more than twice as fast, with (0 =

0.284, in taxa possessing SSQ1 (Figure 6).

64

 

 

Average in for SSC1 codons in

divergent rate class
__ _ 0.234 ,_,_

 

 

 

Ciades lackingSSQl Ciedos onde 5501

Distribution of SSC1 codons among
three evolutionary rate classes

r l “”4“” ; .
' i- . 28.896 "“911!
L'u- 0.038
t " .
61.6% Cicommon rate class 1
to = 0-001 a common rate class 2

I divergent rate class

Figure 6: Comparison of SSC1 codon evolution from taxa encoding SSQ1 and taxa
lacking SSQ1. The pic graph depicts the distribution of SSC1 codon evolutionary rates.
Common rate classes are comprised of codons common to SSC1 from all taxa that evolve
at the same rate. Codons of the divergent rate class are those common to SSC1 from all
taxa that show two different rates of evolution, corresponding to the co—occurrence or
absence of SSQ1. The largest proportion (61.6%) of SSC1 codons belong to common rate
class 1, with an a) = 0.001. The second largest proportion (28.8%) of SSC1 codons
belong to common rate class 2, with an a) = 0.038. The smallest proportion (9.6%) of
SSC1 codons were placed into the divergent rate class. The bar graph depicts the
difference in evolutionary rates between SSC1 from clades lacking SSQ1 and clades
encoding SSQ1. The codons of the divergent rate class of clades encoding SSQ1 evolve
with an a) = 0.107, while the codons of the divergent rate class of clades lacking SSQ1
evolve with an (0 = 0.284.

8801 has evolved at a faster rate than SSC1
Additionally, the clade model test was used to examine the rate of SSQlevolution
relative to SSC, in order to determine whether or not there is evidence for an increase or

decrease in selective constraint acting on SSQ1 (Hypothesis 2). By designating the

65

monophyletic group formed by all SSQ1 sequences as the foreground clade and the
monophyletic group comprised of all SSC1 sequences as the background clade, the clade
model test was used to determine the magnitude and direction of selection acting on a
proportion of codons evolving at different rates between SSC1 and SSQ1. The results
presented were obtained using the SSC1 and SSQ1 combined BI Tree 4, the input tree
which yielded the most likely clade model outputs. SSQ1 sequences were found to
contain a subset of sites evolving faster than those of SSC1 (Figure 7). Most of the sites
conserved between SSC1 and SSQ1 are evolving at equal (slow or intermediate) relative
rates, with about 43.5% having an a) = 0.002 and about 40.8% having an a) =
0.031(Figure 7). Approximately 15.6% of codons estimated to have a differential rate
ratio of about 0.209 in SSQ1 and about 0.077 in SSC1, which is nearly three times as fast
in SSQ1 than in SSC1 (Figure 7). A total of 82 codons comprise the 15.6% of SSC1 and
SSQ1 in the divergent rate class. The encoded amino acids are highlighted within the

Ssqlp amino acid sequence in Figure 12.

66

 

Average 00 for SSC1 and $50.1
codons In divergent rate class

0.25 T — -~
I 0.203

 

 

 

 

Distribution of SSC1 and SSQ1 codons
among three evolutionary rate classes

1 .
'."r . _,‘
-“gJ-‘h‘ , r;
_ . , o
, . . 'v

gd T’s-

” " ‘1‘“ K, '
1..
‘33” D cOmmon rate class 1
u t 0.002 B Common rate class 2

I divergent rate class

Figure 7: Comparison of SSC1 and SSQ1 codon evolution. The pie graph depicts the
distribution of SSC1 and SSQ1 codon evolutionary rates. Common rate classes are
comprised of codons common to SSC1 and SSQ1 that evolve at the same rate in all taxa.
Codons of the divergent rate class are those common to SSC1 and SSQ1 from all taxa
that show two different rates of evolution unique to each paralog. The largest proportion
(43.3%) of SSC1 and SSQ1 codons belong to common rate class 1, with an 00 = 0.002. A
nearly equal proportion (41.6%) of SSC1 and SSQ1 codons belong to common rate class
2, with an (0 = 0.031. The smallest proportion (15.1%) of SSC1 and SSQ1 codons were
placed into the divergent rate class. The bar graph depicts the difference in evolutionary
rates between SSC1 and SSC1. The SSC1 codons of the divergent rate class evolve with
an 0) = 0.208, while the SSQ1 codons of the divergent rate class evolve with an 0) =
0.082.

For both thsp70 comparative analyses, the clade model that grouped all codons
into one of three rate categories was significantly more likely to predict the data, as
indicated by likelihood ratio test, than when only two rate categories were used.
Likelihood ratio test results are presented in Appendix E. The null model, with all codons

constrained to have evolved at equal rates, was also rejected in every instance by

67

likelihood ratio tests. Statistical validation of the use of the clade model with three 00
categories held among all tree topologies examined (11 input trees for SSC1 and eight
input trees for SSC1 and SSQ1 combined). Results of the clade model tests indicated that
SSC1 evolved at an elevated rate when co-occurring with the duplicate gene, while SSQ1

evolved faster than SSC1.

JAC] evolution has decelerated in the presence of SSQ1

A site-specific model was used to examine the direction and strength of selection
that acted on individual codons of J AC1 among the Candida, Saccharomyces,
Aspergillus, and F usarium fungal clades. Our purpose was to assess possible trends in
J AC1 evolution from clades possessing duplicate Hsp703 compared to clades lacking the
duplicate thsp70 (Hypothesis 3).

Figures 8 and 9 show the distribution of codon evolutionary rates across the J AC1
sequences. The alternative tree topologies tested closely agree in the magnitude and
location of elevated codon rates for the Saccharomyces and Candida clades. In the case
of the Aspergillus clade, examination of alternative, strongly supported tree topologies
resulted in some variation in the magnitude, but not location, of elevated codon rates.
Only one JACl tree topology was used in the analysis of F usarium clade sequences
because the topologies generated by each phylogenetic inference method were identical.
In the clades containing the duplicate gene, SSQ1, J AC1 shows similar a) values across
the gene sequence, rarely rising above 0.1 (Figure 8). In contrast, when the sequences
from fungi lacking SSQ1 are examined, the average 00 of J AC1 is greater (Figure 9). The

average 0) across the J AC 1 sequence and corresponding standard errors from each clade

68

were as follows: Saccharomyces: 0.0546 1 0.0028, Candida: 0.0348 :1: 0.0024,
Aspergillus: 0.0711 :1: 0.0061, and F usarium: 0.0812 :1: 0.0056. The variance of (0 values
estimated for J AC1 from the clades lacking SSQ1 was also greater than from the clades
co-occurring with SSQ1 (Saccharomyces : 0.0014, Candida: 0.0010, Aspergillus: 0.0070,
and F usarium: 0.0059). Additionally, none of the J AC1 site-specific analyses produced 0)
estimates of 0, excluding the possibility of the absence of nonsynonymous mutations at a
particular site across the sequences of a clade. Thus, the results of our codon-specific rate
analysis of J AC1 from four fungal clades has opposed our prediction; the rate of
evolution of J AC1, the J-protein co-chaperone specialized in Fe/S cluster assembly,

slowed down following the duplication of the thsp70.

69

A Saccharomyces Clade iac1

 

 

 

 

0.4 , —— __L ____.
l
0.3 - e e e -
00 0.2 i _ __
l —~—MP/Bl tree
01 l m "V M Mi ' _ l i l' "I M" —MLtree
o . r r ‘6 r m T r a T i F m . l
”seassasaaa§§§§§§§§g
codon
B Candida Clade iac1
0.4 7 - ~ 7 ,, , . — *x—flw~-m—H-- ,1 i- A e
0.3 i- -— — - - , ,, we r n
w 0'2 if,“ HM“ H MD I U _.~-,.-_. '_ Vii—~8—MPtree

i

0.1 l -_-_,_ —.. _...._—._.—ML tree

    

_...... Bl tree

codon

Figure 8: Site-specific 00 estimations for J AC 1 from clades encoding SSQ1. Evolutionary
rates (0)) for J AC1 codons from the Saccharomyces and Candida clades are shown as a
function of codon position within the gene sequence. Codon numbers represent column
positions within trimmed nucleotide sequence alignments. Results from each input tree
topology are represented: (A) Saccharomyces clade, MP/BI tree shown in dark gray, ML
tree shown in black, (B) Candida clade, MP tree shown in dark gray, ML tree shown in

black and BI tree shown in light gray.

70

 

      

Fusarium Clade JAC1
0.47,”: , , . ~77 , ,, r
0.3 f~ ,-,_,_,,._., ' W _
0.1 ’ ’7 " l v ' 7’ l ' l 'H t, ' ’ Bltl'ee
o . , . . . . A . - , r . .

H 0 0'1 00 i\ to In 3 m N H O 0" W l\ {D in Q m N H
"""“"""‘“ “wmggfififlfifltfihfl
codon

B .
Aspergillus Clade iac1
0.4
0.3 l, l l
i i
w 0'2 l1 . l l ‘ '1 —MPtree
0.1 * 3" , j M . l “r"- —Bltree
0 . . l. i I . I l 1 MLtree
"sssssrsssaseaaasaesa
codon

Figure 9: Site-specific w estimations for J AC1 from clades lacking SSQ1. Evolutionary
rates (0)) for J AC1 codons from the F usarium and Aspergillus clades are shown as a
function of codon position within the gene sequence. Codon numbers represent column
positions within trimmed nucleotide sequence alignments. Results from each input tree
topology are represented: (A) F usarium clade, MP/ML/BI tree, (B) Aspergillus clade, MP
tree shown in black, BI tree in light gray, and ML tree shown in dark gray.

SSQ1 has evolved under positive selection

Because .1 AC1 is evolving slowly in the presence of SSQ1, we suspected that

JACl and SSQ1 have reached an optimum coevolutionary state among the extant taxa.

This suggests that the potential for adaptive coevolution may have occurred between

SSQ1 and JACl (Hypothesis 4). Ideally, we would test for evidence of positive selection

71

along the ancestral branch of J AC1 corresponding to the lineage in which SSQ1 arose.
However, such a J AC1 branch-site test would require a single phylogenetic tree that
incorporated sequences from all fungal clades, in order to reconstruct ancestral states at
critical points in evolutionary history. Due to our inability to generate the needed multiple
sequence alignment, the required tree could not be inferred. However, such tests are
possible with SSQ1. Therefore, we conducted a branch-site test to detect evidence of
positive selection affecting sites along the tree branch giving rise to SSQ1.

Sites with constant evolutionary rates in both the ancestral SSQ1 branch (inferred
sequence of the foreground branch) and all other sequences (background branches) were
grouped into two categories (Figure 10). A proportion of codons (81.3%) were estimated
to have evolutionary rates of (0 = 0.034, representing common rate class 1, and 4.4%
exhibited an to: 1.000, representing common rate class 2. Hence, these rate categories
were constant regardless of whether the sequence was that of the ancestral SSQ1 gene or
a background gene (Figure 10). For 11.8% of codons, 0) was estimated at 1.994 within
the ancestral SSQ1 and 0.034 for all other genes, designated divergent rate class 1 (Figure
10). A very small fraction of sites (0.6%) were placed into divergent rate class 2, which
evolved at a rate of 00 = 1.994 in the ancestral gene, while these same codons evolved at
0) = 1.000 in derived sequences. This suggests that 12.4% of ancestral SSQ1 codons,
representing codons from both divergent rate classes 1 and 2, were subjected to positive
selection immediately following SSC1 gene duplication.

Though the posterior probabilities associated with the placement of each codon
into a given rate category varied according to tree topology, five out of the nine tree

topologies agreed on four candidate sites for the initial fixation of adaptive mutations

72

following the birth of SSQ1. These four codons, corresponding to amino acids His”,

Lysm, Glum, and Leu346 of the raw SSQ1 sequence from S. cerevisiae YJM789, were
given a posterior probability of Z 0.90 of having an (0 of approximately 2 by at least 5 of
the tree topologies tested (shown in Figures 11 and 12). Several other residues were given
a high probability of having undergone positive selection in ancestral SSQ1 by some tree
topologies (see Figure 11). The results obtained using tree topology B15, however,
identified a different set of residues with high probabilities of belonging to a rate category
with 00 > 1 and did not support evolution under positive selection for the residues shown
in Figure 11. The source of this anomaly is unclear, given that the topology of the B15
tree does not show any large deviations from the other topologies used. All likelihood
ratio tests allowed for the rejection of the null model of neutral evolution, validating the
branch-site test model incorporating sites evolving under positive selection, as a
statistically significantly better fit to model early SSQ1 evolution.

The branch-site test was thus able to detect evidence of positive selection within
the ancestral SSQ1 lineage immediately following gene duplication, and thereby rejects
evolution by neutrality. Together, the two variable rate categories suggest that adaptive
evolution in SSQ1 decelerated in descendent gene sequences after a burst of stronger

selection immediately following the inception of SSQ1.

73

 

Average 00 for codons in
divergent rate class 2

2.5 7
2.0 -l—~

m 1.5 7—— ~
1.0 y
0.5 1. ~~ .
0.0 + - 7

 

 

 

 

   

 

 

Distribution of ancestral and
descendent codons among four
evolutionary rate classes

 

 

 

 

 

 

 

 

 

 

0.6%
Average 00 for codons in
divergent rate class 1
2: 71.994 " ' if \
., —— :r------——— —— a ”33,136,,
1.0 « _-_ A ,. " ' a common rate class 1
0'5 4 0.034 N i 1 common rate class 2
0.0 J AI other i 4.4% I divergent rate class 1
“"009” “l = 1:000 I divergent rate class 2

 

 

 

Figure 10: Comparison of ancestral SSQ1 codon evolution to SSQ1 and SSC1 evolution
within all other lineages. The pie graph depicts the distribution of SSQ1 codon
evolutionary rates. Common rate classes are comprised of codons that evolve at a
constant rate. Codons of the divergent rate classes are those common to ancestral and
present-day descendent lineages that show two different rates of evolution, for each
divergent rate class, unique to the ancestral and descendent lineages. The largest
proportion (83.1%) of codons belong to common rate class 1, with an 00 = 0.002. A much
smaller proportion (4.4%) of codons belong to common rate class 2, with an 0) = 1.000. A
proportion of 11.8% of codons were placed into divergent rate class 1. The bottom bar
graph depicts the difference in evolutionary rates between ancestral SSQ1 codons and
those of descendent sequences in divergent rate class 1. The ancestral SSQ1 codons of
divergent rate class 1 evolve with an 0) = 1.994, while SSC1 and SSQ1 codons from all
other lineages of the tree of divergent rate class 1 evolve with an 0) = 0.034. The top bar
graph depicts the difference in evolutionary rates between ancestral SSQ1 and descendent
codons of divergent rate class 2. The ancestral SSQ1 codons of divergent rate class 2
evolve with an 0) = 1.994, while the present-day descendent SSQ1 codons of divergent
rate class 1 evolve with an 00 = 1.000.

74

SSQI Amino Acid Residues

 

 

 

 

 

 

 

 

 

 

 

Lys Asn Hrs Lys Glu Leu Arg Tyr

Alignment Sequence 171 219 256 258 279 287 327 579

Raw Sequence 224 274 315 317 338 346 386 649

I 8'1 ** ** *** *4!!! *** *Iluk *** #31!
n 81 2 ** ** *** *** ** ** ** **

p 8' 3 ** *t *1“! *** ##4! #134! ** ***

I] 8'4 *4"! *** *** *** *** *** ** ***

t

BIS - - - - - - - -

T Bl 6 ** ** *** *** *** *** ** **

r ML ** *** *** *** *** *** ** ***
e MP 1 _ at: an: an: at: an: an all
e MP2 _ _ *4!!! *4! t 4! *** *

 

 

 

 

 

 

 

 

 

 

 

Figure 11: Comparison of posterior probabilities of placement of sites into a divergent
rate class by the branch-site model, among input tree topologies. All residues assigned to
divergent rate category 1 or 2, with to > 1 and a posterior probability of at least 0.9 in at
least one of the tested tree topologies is shown. Posterior probabilities for placement in
divergent rate class 2 of 0.70-0.79 (*), 0.80-0.89 (**), and 0.90-0.99 (***) are shown,
with residues given a posterior probability of less than 0.70 indicated by (-). The His,
Lys, Glu, and Leu residues shaded in gray are those residues of ancestral SSQ1 believed
to have evolved under positive selection, given that at least five of the nine tree
topologies tested resulted in those residues with a posterior probability of 0.90-0.99 of
evolving with (0 > 1.

75

Ssq 1p Saccharomyces cerevisiae YJM789

IOOVIVVVV'VIVVII'VV|YVI'l!"""""!"l"'1|V'QV'VV'QII'V'I'V’VI'1P'|I

10 20 30 40 50 60 70
VIGIDLGTTNflAukglkDKgAgIIBNREGRTTPSIVAPLVGMAAKRQNAINSENTFFATKRLIGRAFNDEB

V'7'99'9l'VV'I'UIVl9...IY"!""Il""|""l'VV'IPI'II'V'VIV'VVIYQVV'V

BO 90 100 110 120 130 140
‘ VQRDHAVNPYKIVKCGQIYLSTSGLIQSPSQIASILIKYLKQTSEEYLGVNLAVITVPAYFNDSQRQATK
O O 0 0‘0. 0 O O O

I!’l!'!""""'9!|vvvvIvvrv'vv'vlvvvv'7111|vvvv'VVVVI'VVVIICIV'QVPOI9

150 160 170 180 190 200 210
DAGKLAGLNVLRVINEPTAAAL3FGIDD§§N§PIAVYDLGGGTPDISILDIBDGVFEVRATNGDTHLGGE
9'9'0'!!|V9"|QP"|VVI1|vvov'vv1vlvvvv'vvvv' v v'vvvvlvvvv'919'|9!' 1

220 230 240 250 60 270 0

DPDNVIVNYIIDTPIEITREEITKNRETKQRLKDVSBRAKIDISHVRKTPIELPFUYKSRHLRVPMTEEE
O O O. .0. .0 I O O

""'¢"IVV'VITVVV'V'09'99'9l'9'9|""|""""'|"'VI'P"|""|'V'Vl'

290 300 310 320 330 340 350
LDNMTLSLINRTIPPVKQALKDADIBPEDIDBVILVGGHTRNPKIRSVVKDLFGRSPNSSVNPDBTVALG
I O O I O O I O. O.

"'|"9'|?'QI|"'V|"99I99""i"|"'?|""""'l""'VP'VIV'V'IVV'II'

360 370 380 390 400 410 420
AAIQGGILSGBIKNVLLLDVTPLTLGIETPGGAPSPngkflTTVPVKKTEIPSTGUDGQTGVDIKVFQGB

'V'I'999'9'VC'I'V'I',"|0",'99'9'9'9'I'VVVI'VIIIVQI'I'V'VI'I',|!9,'I!

430 440 450 460 470 480 490
RGLVfiNNKLIGDLKLTGITPLPKGIPQIYVTPDIDADGIINV§AABK38%FQQSITSGLSE§EI:§PIBB

I'QIOOUOI'OOO|DOIQI999!'0'?!I'9'!|QIOI|9999|9099|9909|9909'Q'IVIVQQ!|9

500 510 520 530 540 550 560
ANANRAODNLIRQRLELISKADIMISDTENLFKRYEKLISSKEYSNIVEDIKALROAIKNFKSIDVNGIK
oee oooo 00 on oooooooe ooooooee
Q'Q'OO'O'QPVQ'IOQVI
570 580
KATDALQGRALKLPQSATK
O 000 0

Figure 12: Amino acid sequence of Ssql encoded by Saccharomyces cerevisiae YJ M789
showing sites inferred to exhibit relaxed selective constraint and ancestral positive
selection. The 82 amino acids indicated with dots are the amino acids identified as
belonging to the divergent rate class in the SSC1 and SSQ1 clade model test (see Figure
7), which evolve at an accelerated rate compared to SSC1. The four residues indicated by
the arrows correspond to the sites identified via the branch-site model test as those
estimated to have evolved under positive selection in the ancestral SSQ1, immediately
following the mitochondrial heat shock protein 70 gene duplication. The Ssqlp sequence
shown is from the trimmed SSC1 and SSQ1 combined sequence alignment.

76

DISCUSSION

Molecular coevolution among interacting proteins can confer fitness
consequences to crucial enzymatic pathways and can be initiated by the ubiquitous
genetic phenomenon of gene duplication. The findings presented here in the evolutionary
rate analyses of the fungal thsp70 paralogs SSC1 and SSQ1, and the interacting J -
protein co—chaperone ] AC 1, together with previous observations concerning the functions
of the encoded proteins, bear evidence consistent with release from antagonistic
pleiotropy following a gene duplication event. Subsequent subfunctionalization has
facilitated the coevolution of SSQ1 and J AC1 to optimize a J -protein co-chaperone-
thsp70 chaperone interaction dedicated to activity in the Fe/S cluster biogenesis
pathway in yeast.

The thsp70 paralogs investigated here show a history of selection similar to that
inferred for morning glory dihydroflavonol-4-reductase (DFR) duplicate genes. PAML
rate analyses of the anthocyanin biosynthesis pathway DFR genes were consistent with
paralog divergence via escape from adaptive conflict (Des Marais and Rausher 2008).
Evidence from clade and branch—specific 00 value estimates of codon rate evolution for

each of the three DFR copies indicated an ancestral single-copy DFR that was subjected
to purifying selection, followed by a relaxation of selective constraint after gene
duplication. Evidence for positive selection within the lineage immediately following the
second duplication was observed. Positive selection early in the history of the paralogs of
the most recent DFR duplication potentially enabled a burst of adaptive mutation fixation

within these paralogs. Combined with biochemical evidence of optimization from an

77

ancestral sub-function of one of the DFR paralogs, and the loss of the ability to perform
other ancestral functions in paralogs, the authors concluded that antagonistic pleiotropy
enforced selective constraint to prevent full optimization of all ancestral DFR functions in
single-copy form. In analogy to the DFR study, one of the fungal thsp70s, SSQ1, was
found to have undergone positive selection in its ancestral sequence shortly following the
gene duplication event from which it was created. Like DFR-B, SSQ1 became
specialized in a role performed by the pre-duplication gene, and may have even evolved
to outperform its paralog, SSC1, in terms of increased affinity for Jaclp and greater
ATPase activation.

SSQ1 shows biochemical evidence of ATPase activity improvement in response
to J AC1 stimulation, with the potential to improve Fe/S cluster biogenesis efficiency, an
ancestral pre-duplication function. Concornitantly, SSQ1 can no longer perform the
ancestral thsp70 functions of protein folding and translocation functions, nor provides
protection to cellular integrity from environmental stresses. The functional evolution of
SSQ1 thus fits the criteria for a case of subfunctionalization. Furthermore, I AC1
evolution resulting in the loss of J ~domain residues important for Ssclp ATPase
activation has occurred in yeasts encoding SSQ1. Therefore, an alteration of the J -domain
of Jaclp may have been necessary for improved affinity to Ssqlp and may have been
evolutionarily favored only in the presence of a thsp70 specialized in Fe/S cluster
biogenesis. Coevolution of J AC1 with SSQ1 would have thus been a consequence of
thsp70 paralog evolution following escape from adaptive conflict.

However, evidence is lacking to meet the more stringent criteria of SSC1 and

SSQ1 evolution by escape from adaptive conflict. There is no direct proof of a novel

78

function arising in the pre-duplication thsp70 that reduced the ability of the ancestral
protein to perform any of its other tasks. This would require the biochemical
characterization of the protein translated from an ancestral gene reconstruction.
Additionally, future investigation of EcmlOp functions, and the selective forces acting on
this third yeast mtI-Isp70 duplicate, could bolster the case for adaptive conflict in the pre-
duplication thsp70 if ECM10 has also optimized an ancestral SSC1 function. Finally, it
remains to be determined if a more efficient J aclp—Ssclp interaction optimizes the Fe/S
cluster assembly pathway to increase yeast fitness.

The functional specialization of SSQ1 also resembles the subfunctionalization for
optimization of GALl and GAL3 functions, after release from antagonistic pleiotropy, by
gene duplication (Hittinger and Carroll 2007). While promoter divergence resulted in the
evolved phenotypes of differential control over GALl and GAL3 transcription,
regulatory evolution of the thsp705 was not examined in this study. However, previous
observations of decreased SSQ1 expression compared to SSC1, within S. cerevisiae
mitochondria, suggests that SSQ1 and SSC1 have also undergone regulatory divergence.

Another possibility is that the specialized function of SSQ1 hinges on a mutation
analogous to a GALl Ser-Ala di-peptide identified to be sufficient for galactokinase
activity when added to the active site of GAL3, the co-inducer of the galactose uptake
pathway. Because deletion of the di-peptide from GALl, and a pre-duplication
GALl/GAL3 bifunctional protein, did not improve the co-inducer function of the
encoded proteins, the Ser-Ala mutation of GAL] could not be ruled a source of adaptive
conflict. The effect of the Ser-Ala mutation on galactokinase function was dependent on

the background of residues present at other sites within GALl. It is possible that

79

mutations have similarly arisen in SSQ1 that now contribute to functional specialization,
but were fixed as compensatory mutations secondary to mutations fixed as a direct result
of release from antagonistic pleiotropy.

It is reasonable to hypothesize that opportunity for functional specialization of
proteins like pigment biosynthesis enzymes, galactose pathway components, or
thsp708, may extend to molecules that participate within a common biological
pathway, by coevolution. The release of SSQ1 from antagonistic pleiotropy has
influenced the evolution of J AC1, the J -protein partner also specialized in this pathway.
J AC1 coevolution with the thsp70 paralogs has allowed its interaction with SSQ1 to

become more efficient, while decreasing its efficiency of ATPase stimulation in SSC1.

Support for Hypothesis 1: Selective constraint has been relaxed in SSC1 in the
presence of its paralog, SSQl.

An equal rate of SSC1 evolution, in the presence versus absence of SSQ1, was rejected.
SSC1 evolved faster in the presence of its paralog, SSQ1.

The result that SSC1 evolved faster when co-occurring with SSQ1 is consistent
with the conclusions of Scannell and Wolfe (2008), who found that recent paralogs tend
to evolve at an increased rate compared to singleton genes. Here, we suggest that the
functional specialization of SSQ1 has relieved SSC1 of the Fe/S cluster biogenesis task,
thereby relaxing selective constraint acting on SSC1 for this particular function. The'
availability of the SSQ1 :J AC1 specialized pair could have rendered the SSC1 :J AC1
cooperation less important, thus allowing a greater proportion of nonsynonymous codon
changes to be tolerated in SSC1, particularly if those sites encode residues that contribute

to interaction with J AC 1 , or other unidentified aspects of Fe/S cluster biogenesis.

8O

In the absence of SSQ1, however, antagonistic pleiotropy would continue to
impose evolutionary constraint on SSC1, because SSC1 would be required to perform
Fe/S cluster formation, in addition to protein import and folding. While evidence does not
yet exist to suggest that SSC1 has improved any other pre-duplication thsp70 function
in the presence of SSQ1, it could be that escape from adaptive conflict may allow SSC1
to perform a chaperone task, such as peptide translocation across the inner mitochondrial
membrane, with greater efficiency if optimization is permitted in the presence of
paralogs. This seems plausible if the relaxation of selective constraint on SSC1 among
clades that harbor duplicate genes persists for tens of millions of years (Scannell and
Wolfe 2008). An extended period of relaxed constraint may have the potential to fix

many mutations via drift, and as a composite, could result in an altered phenotype.

Support for Hypothesis 2: SSQl is under less selective constraint than SSC1 because
SSQl has fewer encoded functions to maintain.

Evolution of SSQ1 and SSC1 at equal rates was rejected. SSQ1 evolved faster than
SSC1.

When the average rate of codon evolution was compared between SSQ1 and
SSC1, we were able to conclude that SSQ1 evolved faster than SSC1. This rate
asymmetry is consistent with other published analyses of evolutionary rate asymmetry in
paralogs (Conant and Wagner 2003; Zhang et al. 2003). An examination of gene
duplicates created by a whole genome duplication in yeast revealed that genes with the
most dramatic evolutionary rate increase, immediately following duplication, remained
the “faster” evolving gene of the two paralogs. Therefore, it is likely that SSQ1 will

continue to evolve with a greater (0 than SSC1. While evidence for sites under positive

81

selection (an to greater than 1) in extant taxa was not identified, the faster rate of SSQ1
evolution compared to SSC1 is interpreted as a result of relaxed constraint, depressed
expression level, or both.

The increased rate of evolution for SSQ1 could be due to relaxation of selective
constraint that is independent of gene expression in order to allow for specialization on a
single function. Relaxed constraint on SSQ1, compared to the ancestral single-copy
thsp70, likely initially resulted from the ability of SSC1 and SSQ1 to reciprocally
compliment one another, and subsequently also provide robustness against deleterious
mutations. For example, if a mutation in SSC1 resulted in diminished function as an Fe/S
cluster biogenesis chaperone, SSQ1 would have been able to restore this function. We
propose that SSQ1 would then have been free to optimize efficiency for its role in Fe/S
cluster assembly in the presence of SSC1, which could functionally replace SSQ1 for any
of the many sub-functions that may have been compromised during Fe/S cluster assembly
optimization. As a result, disproportionately many sites in the protein may now be under
relaxed selection and thus evolve at a faster rate compared to the multifunctional SSC1.

Alternatively, gene expression divergence of SSQ1 and SSC1 alone is a viable
explanation for the faster rate of SSQ1 evolution. This line of reasoning is supported by
the Drummond et a1. (2005) study, which concluded that gene expression level is the
single greatest determinant of protein evolution, explaining more than half of the
variation in nucleotide substitution rates of genes in S. cerevisiae. Though gene length,
dispensability, and recombination have also been suggested as factors aiding to predict
evolutionary rates of genes, these variables seem to play a minor role in the determination

of evolutionary rates. In addition, expression levels have been shown to exert control

82

over these factors, often confounding efforts to link these factors as direct causes.
Drummond et al. (2005) revealed that genes with a lower level of expression tend to
evolve faster and offer an explanation for this observation independent of selection on
protein function. It is known that errors during mRN A translation lead to the
accumulation of mis-folded and toxic protein products that impose fitness costs to a cell
by disrupting metabolic processes (Bucciantini et al. 2002). It was therefore proposed that
selection acts to increase the translational accuracy of a sequence, (i.e. using the most
abundant tRNA anticodons for amino acids), and to increase the robustness of a sequence
to translational errors. Favoring amino acid sequences that fold into functional proteins,
regardless of the generation of missense errors, increases translational robustness
(Drummond et al. 2005).

A subsequent study (Drummond and Wilke 2008) identified protein misfolding
costs as the underlying selective pressure responsible for the co-variation in evolutionary
rates, codon preference, and gene expression within and between genes, observed for
model organisms ranging in complexity from E. coli to humans. The authors revealed
that translational accuracy, translational robustness, the synthesis of full—length peptides,
and the tendency to fold properly, all correlate positively with gene expression level. The
cost of protein misfolding thus provides a reason for the selective constraint that gives
rise to a greater proportion of optimal codons observed at conserved sites within a protein
in genes that are most highly expressed.

When a gene is expressed at a higher level, as with SSC1, translation occurs more
frequently, increasing the number of opportunities for detrimental errors, so that accuracy

and robustness become more influential to the cell’s overall fitness. Therefore, by

83

selecting against protein sequences with toxic characteristics (such as a propensity for
aggregation) when translated incorrectly, the same evolutionary forces may indirectly
select for a protein structure with enhanced thermodynamic stability. Together, selection
which results in an increase in translational accuracy and robustness may have the effect
of lowering both the rate of synonymous and nonsynonymous mutation fixation,
imposing a form of evolutionary constraint at the sequence level. Higher expression level
may therefore bring about increased evolutionary constraint on SSC1, while the relatively
decreased level of expression of SSQ1 may result in relaxation of constraint. Divergence
in the expression level of paralogous genes could occur as a consequence of accelerated
promoter or regulatory region evolution by adaptive or neutral evolution. This has been
suggested to be a common phenomenon in eukaryotes (note that evidence of regulatory
sequence evolution would go undetected in protein-coding 0) analyses) (Zhang, 2003).
Alternatively, divergence in paralog expression levels can result from other sequence
changes that contribute to mRNA stability or chromatin structure differences between the
gene duplicates (Li et al. 2005).

Indeed, the approximately 1000-fold lower concentration of Ssql protein present
in the mitochondria of S. cerevisiae (Voisine et al. 2000) is accompanied by a decreased
codon usage bias and an increased overall rate of nucleotide substitution, indicative of
relaxed selective constraint. While the codon adaptation index for SSC1 is reported to be
0.521, the codon adaptation index of SSQ1 is much lower, at 0.148 (SGD project, Sept.
2008) and is indicative of less selective constraint acting on third position nucleotides of
SSQ1 codons. Less constraint on these nucleotides could allow SSQ1 to tolerate more

synonymous substitutions than SSC1. Therefore, the elevated 00 of SSQ1 is impressive in

84

the face of an elevated d5, as was observed in a gene-wide average of site-specific ds
values estimated across tree branches and compared to SSC1 d5 averages (data not

shown).

Lack of support for Hypothesis 3: The rate of J AC1 evolution is positively
correlated with the rate of SSQl evolution because JAC] and SSQl are coevolving.

An equal rate of JAC] evolution in clades encoding SSQ1 compared to the rate of
JAC] evolution in clades that lack SSQ1 is rejected. However, JAC] evolution
decelerated after thsp70 gene duplication.

Here we have examined the influence of a gene duplication event on the selective
forces driving the molecular evolution of protein partners specialized in Fe/S cluster
assembly. We have demonstrated that J AC1 evolves faster in the absence of SSQ1. Our
proposed explanation is that selective constraint is acting on J AC1 to preserve an
optimized, physical interaction with SSQ1, which resulted from the coevolution of JACl
and the duplicate, specialized thsp70. While J AC1 now evolves slowly in the presence
of SSQ1, it is conceivable that the rate of evolution of J AC1 was initially accelerated
after the thsp70 gene duplication that gave rise to SSQ1. Subsequently, J AC1 may
have quickly reached an adaptive peak, together with SSQ1, in its ability to facilitate Fe/S
cluster assembly. Or, J AC 1 was brought under constraint by some other influence. The
rapid rate of J AC1 evolution, however, precludes the testing of this hypothesis, as carried
out for SSQ1, since we could not reconstruct J AC1 ancestral states. We speculate that,
subsequent to initial rate acceleration during a co-adaptive arms race to fix
complimentary changes in the sites that physically interact between J aclp and Ssqlp,

Jaclp evolution slowed to maintain efficient cooperation with Ssqlp. An alternative

explanation for the faster rate of J AC 1 evolution in the Aspergillus and F usarium clades

85

could be a smaller effective population size of representative species compared to the
Candida and Saccharomyces clades, which would in turn result in a reduced efficiency of
purifying selection.

Though expression data of J AC1 in the fungal species from which sequences were
analyzed is unavailable, it is possible that the expression of JACl has been increased in
the organisms possessing SSQ1 to balance molecular stoichiometry. Indeed, a higher
average codon bias, consistent with higher levels of gene expression (Wang et al. 2005),
was observed for J AC1 from clades encoding SSQ1. J AC1 CAI value means and
standard errors calculated for each clade were as follows: Saccharomyces: 0.273 :t 0.013,
Candida: 0.249 1 0.010, F usarium: 0.195 t 0.013, and Aspergillus: 0.183 :t 0.010. Thus,
the third nucleotide positions of J AC 1 codons from clades encoding SSQ1 are likely to be
under stronger selective constraint than third nucleotide positions within JAC] from
clades lacking SSQ1. Increased constraint on third position nucleotides, as well as an
overall increase in constraint to preserve translational robustness when gene expression is

elevated, may be depressing w in J AC1 from Saccharomyces and Candida taxa.

Support for Hypothesis 4: 8801 has undergone adaptive evolution to optimize the
Ssql - Jaclp interaction important for Fe/S cluster biogenesis.

Evolution of SSQ1 in the absence of positive selection is rejected. SSQ1 evolved under
positive selection in the lineage immediately following its inception.

The antagonistic pleiotropy that characterized the ancestral thsp70 prior to gene
duplication may have been broken by positive selection in ancestral SSQ1, immediately
following its gene duplication. Positive selection may have enabled a burst of adaptive
evolution to optimize the J aclp-Ssqlp partnership important for Fe/S cluster biogenesis

and promoted rapid subfunctionalization of SSQ1, thus relaxing constraint on SSC1 at

86

sites necessary for interaction with J aclp. The retention of SSQ1 within the genome
following the gene duplication event may be attributable to this subfunctionalization,
possibly having involved an adaptive sweep at ancestrally positively selected sites His3'5,
Lys3 '7, Glum, and Leu346 within the ATPase domain. Given that past studies have shown
how significant adaptive shifts can be instigated by very few amino acid substitutions
(Golding and Dean 1998), rapid thsp70 SSQ1 evolution may have been responsible for
its coevolution with J AC1 to specialize the J -protein-thsp70 pair. Alternatively, the
signature of an initial burst of selection detected in the ancestral sequence of SSQ1 may
not have been accompanied by functional adaptation at all sites, but instead reflect the
fixation of compensatory substitutions to rescue a decrease in fitness arising from

deleterious mutations within the gene or even elsewhere within the genome (Pal et a1.

2006).

Future Directions

Possible future lines of research include conducting protein structural and
functional analyses, via experimental genetics and biochemistry, in order to elucidate the
role of particular SSC1, SSQ1, and J AC1 sites in Fe/S cluster biogenesis. Site—directed
mutagenesis and reconstruction of inferred ancestral gene sequences, followed by
biochemical characterization of ‘resurrected’ ancestral proteins, is a technique that has
been successfully used in the past to gain insight into the fates of paralogous genes
following gene duplication (Zhang and Rosenberg 2002). Additional experiments could
include mutating SSC1 sites to those corresponding to SSQ1 sites that were identified to

have undergone positive selection immediately after the gene duplication. It would be

87

interesting to determine if those sites from SSQlimprove the efficiency of Ssclp ATPase
activity in the presence of J aclp, and if so, whether those sites are necessary for direct
contact with Jaclp, the nucleotide exchange factor protein, or the nucleotide. SSC1
engineered to encode an ATPase domain that more closely resembles that of SSQ1 might
also be predicted to have decreased chaperone and stress mediation functions. Such a
result would directly demonstrate the tradeoff between optimization of J aclp-mediated
ATPase activity and loss of performance in other functions within the thsp70. The
source of antagonistic pleiotropy in the ancestral thsp70 would thus be pinpointed
within the ATPase domain. Conversely, manipulation of sites in SSQ1, where
homologous positions in SSC1 are under relaxed selection, are predicted to be involved
in Fe/S cluster biogenesis, as these were the sites predicted to be released from selection
by subfunctionalization. On the other hand, independent manipulation of the sequences
encoding the substrate binding, ATPase, and variable domains of SSQ1, to contain those
sites that are under strict selective constraint in SSC1, should be performed. Such
manipulation might lead investigators to attribute the increased Ssqlp ATPase activity to
a domain other than the ATPase domain. Identifying sites in J AC1 that have evolved at a
f«'tlSl: rate in the F usarium and Aspergillus clades, but have evolved at a slower rate in the

Candida and Saccharomyces clades, might also be informative in guiding similar site-

SPGCific mutation construction of J AC 1.

The role of regulatory sequence evolution should also be explored in the future,

per haps by evaluating the effect of exchanging the promoters of the paralogous
rntI‘ISp703. One expectation might be that replacing the SSC1 promoter with that

regulating the transcription of SSQ1 will decrease the expression level of SSC1 within

88

the m
[18.1151

resul‘

more
stud

and

chill
the
AT]
pro
clus
phe
Van
H01
clue

3V0

the mitochondrial matrix. Alteration in the number, orientation, and/or sequence of
transcription factor binding sites after mtI-lsp70 gene duplication might be expected to
result in such regulatory differences. Another outcome of thsp70 promoter swapping
might be that, when under the control of the SSC1 promoter, SSQ1 is increased in its
degree of expression. However, the extent to which active Ssqlp is produced may still be
less than Ssclp levels, given that SSQ1 has a lower codon bias and therefore might be
more prone to translational errors that result in truncated or misfolded proteins. Such
studies would be important to verify that the expression level difference between SSC1
and SSQ1 is due to cis-regulatory evolution and is not an effect of other forms of
regulation, such as feed-back inhibition.

The ultimate goal should be to elucidate details of how the thsp70 paralogs
differ in their interaction with J AC1 and how these changes confer fitness differences via
the execution of Fe/S cluster biogenesis. Thus, the direct impact that the increased Ssqlp
ATPase stimulation by J ac 1p confers upon the level of active Fe/S-containing proteins
produced in vivo must be established. Further, the fitness advantage of an optimized Fe/S
cluster biosynthesis pathway must be demonstrated by the observation of an adaptive
phenotype. This will not be a trivial undertaking, as the advantage of a phenotype often
varies under different growth conditions and the presence of ecological competitors.
However, as with any molecular process, if we are to advance our understanding of Fe/S
cluster biosynthesis, we must study the pathway components in the context of

evolutionary and ecological dynamics.

89

Fun.

Appendix A

Fungal Mitochondrial Heat Shock Protein Coding Region DNA
Sequence Sources

90

lrl. .l‘. | . F

I I .. .... l. l a. . lib l u v v I at ill-fillili \ lit! I:I~.tr\...whnr_pd
_ IOWCOucurthOUGDNWW .IEW.~.UU(H .G.EC.U..OOU OEOCOU 09.30% COCODUOAM OEOCQUN CORR

 

 

IDOL—40m 0020330” 90%“ up.‘ 0~OBL

58 §§E 2a: .568 a 3 a 852 28528 8a n. :33 _> .5338 use: 80822.8: E82 39.2 2: ..
0:9.» 05009:on 20:8 .-. 5 .+.

2.8350.

uses... 0583 2.2. u 6..

223:. 96% n 90%

838.5 oEocoe $388. u now

mommnvo 32 8.3008 xcwncoo . . . .
«.0893 Ex . n . .5.an «.8808 832

mosafltfios E:
EEooSEEo. E33:
E38 E: :38

 

 

 

 

 

 

 

 

 

 

 

|_|

N Eco m— 20

 

 

 

 

 

 

 

838 (EZ 300.—
Ndmmmo mum? 300.
@883 —ooooO< 300.
mevvomuwh< 300.

 

 

mag flak: u . .
ego «segue.
$81.2 macho. was .11 .

 

 

 

 

 

 

 

 

 

 

 

 

3.0338 32 830008 xcamcoo
830mg? maoo. m EEO . maovmmmmhm—mmm 00w
owowoo <40< 300: I mm .250 . «52.9335 com
o. Fowz< mace. now 4:84. FEB + vomnéwvm . v<0m0m 85:9: me: .113“
89 2:0 9.50. N area + mmvnomwé Enem— lomunfluo Shrub o<OIm 380w 3853 929.60
l L F — SSMMTIO .250 it; 5 Tmnmwom— 2359... coucom EFF 058:0; 800 m6§£3=n 8.3.30
mdwto 05.0 v.38.— N 45233. .+ 0359.0980— é

 

 

         
   

 

9280 6‘50 «:00. on— .Eoccuasm _. Swen—.583
.R8000<Imn_ n=00.

p.033 0301 0:00.

089950 $50 D<Omm
anon—Sun «RENO .:<Om
omen—Sac SEED o<Omm

   

  

  

 

 

 

 

 

 

 

 

 

 

323m cocoa—Sm 5mm u..< 03a...

91

 

 

C‘r‘L-ucu 1‘“ 1- 1|

 

38.» 258528 283 ... ..o .+.
2:59... 23:00 EB... n .9.

age. 90mm n 90mm

3333 2550 $5383 u now

 

33:5 88.—

 

 

 

   

188 23% .835
88 acacia 8.28

 

 

_
328 050 88:

 

 

888 oz 588% €850
$08835. c9388 Eamcoo

 

 

 

 

 

 

 

mo—ooopoam: 3.3008 scamcow 3 32m $6328 .3003
38288.85... .Icoilua sang—28025.30 8338.3? .2550 :93...

828m 8..ng 5mm "2 oz.»

92

mwogaom aucmzcwm “Ova. ..MQ 2cm...

“...-i

:58 Sarawak Eu: 3363 a: E 2 3:82 zSEE8 8% 2 c>m2v _> 8.52.0; 95w: 8080352 «$82 26:3 2:. ..
ucwi oESoEeco 833 3 .+.

08883.

83.2. ancoo E2. u .01

83:2. 905 u 90mm

3338 0:880 gluon“ u now

 

 

 

 

 

 

 

 

     

ooiboSuz 980.
Nmtvomfiu? 260.

m woooomwooaoo< 300.
2.3.: So 32 c9338 x5980—

 

 

 

 

mté Ea $6.

38868? cou$oou xcamcoc
035° So< 260.

8w 6F c2w88u xcamcoo

35 a :‘6 300. N 4:00..” . an .N EEO + mowsnéoowm 338.60 3:58 .:<Om_ 3.80m 3&3 «3ng

 

 

  

 

_ 508.95 .28 .+ Save—8m .8 .3: 2.8.3; 88 mafia» £28

 

 

98° 605.0 2.30. N 4523:» .. 38783 raglan. 83:8 :29...
0.830 0<n6 260. . . . $3500 83ch :28

   

 

 

 

awoono<o<xmo goo.
fimmowowamvm v.80.

3338 33:3 901m
33800 3.8an .Q<Omm

   

 

 

 

 

 

 

vsmooo SIO<< c9338 xcamcomu
monooo Fo>m<< c9333 x5956
«9808; c2338 xcumcoa
$88885 8.8.5.“ €380

8058...... 8:2... ice—=21

 

 

 

8:2; 33:28 a... E2388
2 E2: 3%.289; Epafioum

 

 

cox-h

3958 8:258 FOE. "2 03a...

93

Apmndix B

Fungal Mitochondrial Heat Shock Protein Multiple Sequence
Alignments

Multiple alignments of amino acid sequences translated from protein-coding regions of
mitochondrial heat shock proteins (thsps) were performed using CLUSTAL W

(Thompson et al. 1994) with default gap penalties, and subsequent manual trimming to
remove gaps. Alignment columns highlighted in black denote sites sharing 100% identity
among all taxa. Taxon name abbreviations used are listed in the table below:

 

Taxon

Taxon

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Abbreviation Fungal Species Abbreviation Fungal Species
Sce r_Y Saccharomyces cerevisiae RM11 Fgra Fusarium graminearum
Sce r_R Saccharomyces cerevisiae YJ M789 Fver Fusarium verticilliodes

Spar Saccharomyces paradoxus Fsol Fusan'um solani
Smik Saccharomyces mikatae Ncra Neurospora crassa
Sbay Saccharomyces bayanus Tree Trichoderma reesei
Scas Saccharomyces caste/Iii Pans Podospora anserina
Cgla Candida glabrata Nfis NeosarIOIya fischen‘
Calb Candida albicans Anid Aspergillus nidulans
Ctro Candida tropicalis Ater Aspergillus terreus
Cpar Candida parapsi/osis Acla Aspergillus clavatus
Cgui Candida guiI/iennondii Afla Aspergillus flavus
Cdub Candida dub/iniensis Anig Aspergillus niflr
Clus Candida Iusitaniae Aory AsperMJs oryzae
Dhan Debaryomyces hansenii Afum Aspergillus fumigatus
Foxy Fusarium oxysporum

 

94

 

Sccr_R
Scar_!

cyla

suit
Shay
Scar
Dhan
Chlb
Cgui
Ctro
Cpar
Cdub
Clus
’OIY
rgra
Ever
lcrl
Tree
Pans
F301
Hfir
Anid
Ater
Acla
Afla
Anig
Aory
Afun

Figure

8cor_R
Seal; 1'
Cgl a
Spar
smile
Shay
Sca-
Dhan
Ca 11:
Cgui
Ctro
Cpar
Cdub
c1 u-
Foxy
Fgra
PVC:
Nara
Troo
Pans
rs o1
N151 I
An it!
A t: -r
A c.- .1 a
A £1 a
An :1 g
Aoz-y
A fun

 

Figure Bl: SSC1 amino acid multiple sequence alignment

95

Sccr_l
Scer_i
cyla
Spar
Snik
Shay
Scar
Dhan
Chlb
Cgui
Ctro
Cpar
Cdub
Clus
bey
Igra
Iver
Ncra
Tree
Pans
r101
Nfis
Anid
Ater
Acla
Afla
Anig
“I?
Afum

ligu

Sear_R
Scar_Y
cyla
Spar
Salk
Shay
Scan
Dhan
Chlb
cgui
Ctro
Cpar
Cdub
Clus
Pbxy
tyre
Nor
Nbra
Tra-
Pan-
I501
Nfi-
Anid
Ater
Acla
Afla
Anig
Aory
Afum

 

Figure Bl: SSC1 amino acid multiple sequence alignment (continued)

96

140 150 160 170 180

 

 

 

 

 

 

VVT¥PAYFND QRQHTKDAGilvGLNVLRV¥NEPTAAALAYGLE
VVT¥PAYFND QRQHTKDAGEI¥GLNVLRV¥NEPTAAALAYGLE
‘VVT‘PAYFND QRQHTKDAGOI GLNVLRV EPTAAALAYGLE
VVT‘PAYFND QRQHTKDAGO.I‘GLNVLRV‘NEPTAAALAYGLE
T PAYFND QRQHTKDAGII‘GLNVLRV‘NEPTAAALAYGLE
1"VVT PAYFND QRQ TKDAGO:I GLNVLRV‘NEPTAAALAYGLE
'1 'VVT PAYFND QRQHTKDAGEI I¥GLNVLRV NEPTAAALAYGLE
P ‘ VVTIPAYFND QRQHTKDAGSI GLNVLRV NEPTAAALAYGLE
KVNS VVTOPAYFNDHQRQHTKDAGBI GLNVLRVENEPTAAALAYGLE
P CI‘VVTGPAYFNDHQRQHTKDAth NEPTAAALAYGLE

i
KVNS VVTEPAYFNDHQRQHTKDAGHIa
i
ii

QC)
r'r'
"2'3
FL"
”E?
Z
[‘1
'11
g
L“
3:!
K:
O
L"
D1

KINS VVTiPAYFNDHQRQHTKDA «I
KVNS VVTOPAYFND QRQ- TKDAGHI
P VVTEPAYFND' QRQH TKDAG'I GLNVLRV *NEPTAAALAYGLE
PIC -VVT§PAYFND QRQ- TKDAGOI GLNVLRV¥ NEPTAAALAYGLE
P11 VVT PAYFND QRQ TKDAGoI GLNVLRV¥NEPTAAALAYGLE
PI! -VVT¥PAYFND QRQ TKDAGOI:GLNVLRV¥NEPTAAALAYGLE
P 1WT¥PAYFND QRQ :TKDAGOIHGLNVLRV¥NEPTAAALAYGLE
P .VVT¥PAYFND- QRQ :TKDAGIIHGLNVLRV NEPTAAALAYGLE
P VVT PAYFND QRQ TKDAG-I GLNVLRV¥NEPTAAALAYGLE
PI VVT¥PAYFND§QRQETKDAGOI GLNVLRV EPTAAALAYGLE
P vvr¥ PAYFND QRQH TKDAG-I GLNVLRV NEPTAAALAYGLE
P . WT¥ PAYFND QRQRTKDAGOIHGLNVLRvfiNEPTAAALAYGLE
P HTKDAGOIHGLNVLRV¥

P

P

P a

P

P

C)

‘ *PAYFND QRQ yNEPTAAALAYGLE
:PAYFND QRQHTKDAGE.IHGLNVLRV¥NEPTAAALAYGLE

¢"W :PAYFND QRQHTKDAGEIHGLNVLRV¥NEPTAAALAYGLE
-VVT¥ PAYFND QRQHTKDAGE.IRGLNVLRV¥NEPTAAALAYGLE
VVT‘ PAYFND QRQHTKDAGEIHGLNVLRV

*NEPTAAALAYGLE
VVT¥ PAYFND‘ ‘QRQHTKDAGOIHGLNVLRV

NEPTAAALAYGLE

 

 

Figure B1: SSC1 amino acid multiple sequence alignment (continued)

97

Sco;_l
Sce;_l
Cpl:
Spar
Suit
Shay
Scan
Dhan
Cblb
Cgui
Ctro
Cpar
CUub
Clus
Fbxy
Pgra
Ivor
Hera
Tree
Pans
P301
Nfia
Anid
Ate:
Acla
Afl:
lnig
Aer]
Afu:

“m

190 200 210 220 230 240

Sccz;R
Sco:_Y
cyla
Spar
Smik
Shay
Sca-
Dhan
Calb
cgui
Ctro
Cpar
Cdub
Clu-
Foxy
tyre
Ivor
Nara
Tron
Pan-
1101
Nfil
Anid
Atar
Acla
Afla
Anig
Aory
Afum

 

Figure Bl: SSC1 amino acid multiple sequence alignment (continued)

98

250 260 270 280 290 300

Scar;R
Sco;_Y
cyla
Spar
Salk
Shay
sca-
Dhan
Chlb
cyui
Ctro
Cpar
Cdub
Clus
Pbxy
ryra
Ever
Nara
Tree
Pan-
Incl
Nfil
Anid
Ater
Acla
Ail.
Anig
Lory
Afum

 

Figure Bl: SSC1 amino acid multiple sequence alignment (continued)

99

310 320 330 340 350 360

5co:_R
Sco;_Y
cyla
Spar

Shay
Sca-
Dhan
Chlb
cyui
Ctro
Cpar
Cdub
Clus
foxy
Pyra
Ivor
Nara
Tree
Pan-
Peal
Nfia
Anid
Ater
Acla
Afla
Anig
Aory
Afum

 

Figure B1: SSC1 amino acid multiple sequence alignment (continued)

100

370 380 390 400 410 420

Ctro
Cpar
Caub
C1 in
Foxy
1'ng
Nor
Nara
Tree
Pans
F501
NH-
Anid
Ater
Acla
Afla
Anig
Aory
Afum

 

Figure B1: SSC1 amino acid multiple sequence alignment (continued)

101

430 440 450 460 470 480

Sc¢:;R
Sco:_r
cyla
spar
Snik
shay
Sca-
Dhan
Chlb
cyui
Ctro
Cpar
cdub
Clu-
bey
tyra
Ivor
Nara
Tron
Pan-
1501
Nil-
Anid
Ater
Acla
Afla
Anig
Aory
Afum

 

Figure B1: SSC1 amino acid multiple sequence alignment (continued)

102

Scor;R
Scog_r
cyla

SC.‘

Nara

P101
N115
Anid
Ater
AC1.
Afla
Anig
Lory
Alum

 

Figure Bl: SSC1 amino acid multiple sequence alignment (continued)

103

550 560 570 580 590 600

Scor;R
Scor;Y
cyla
Spar
Snik
shay
Scan
Dhan
Chlb
Cgui
Ctro
Cpar
Cdub
Clus
Foxy
Iyra
IVor
Hora
True
Pan-
Fool
Nfil
Anid
Ater
Acla
Afla
Anig
Lory
Afum

 

Figure Bl: SSC1 amino acid multiple sequence alignment (continued)

104

Sco;_R
8co;_¥
Cgla
Spar
8m1k
Shay
Scan
Dhan
Chlh
cyui
Ctro
Cpar
Cduh
Clus
taxy
Pyra
Ivor
Nora
fro.
Pan:
Fool
Nfia
Anid
Ater
Acla
Afla
Anig
Jory
Afum

 

Figure Bl: SSC1 amino acid multiple sequence alignment (continued)

105

 

8801 Spar
SSQl Smik
8801 Shay
$801 Seas
SSQl Cgla
SSQl Chlb
3501 Ctro
SSQl Cpar
.9391 Cgui
SSQl Cdub
8301 Clus
$501 Dhan

 

Figure B2: SSC1 and SSQl combined amino acid multiple sequence alignment

106

SSC1 Scot Y
SSCI Scor R
SSC1 Spar

SSC1 Shay
SSC1 Scan
SSC1 cyla

SSC1 Tron

SSC1 Ater

3801 Scar Y
R

8801 Scan
8801 cyla
SSQI Calb
SSQI Ctro
SSQl Cpar
SSQl cyui
8801 Cdub
SSQl Clu-
SSQl Dhan

 

Figure B2: SSC] and $801 combined amino acid multiple sequence alignment
(continued)

107

110 120 130 140 150

 

 

 

cool ....

8881 Scar Y
8801 Seer R
8881 Spar
SSC1 Snik
88C1 Shay
SSC1 Scan
SSC1 cyla
88C1 Chlb
88C1 Ctro

8801 Soc:
8801 Sea:
8801 Spar
8801 Smik
8801 Shay
8801 Scan
8801 cyla
8801 Calh
8801 Ctro
8801 Cpar
8801 cyui
8801 Cduh
8801 Clus
8801 Dhan

N N

 

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

108

160 170 180 190 200

88C! Scar Y
8881 Scar R
88C1 Spar

88C1 Shay
88C1 Scan

88C1 Ater

SSC1 Afla
88C! Anig

88C1 Afum
8801 Scar
8801 8a.:
8801 Spar

N N

8801 Shay
8801 Scan
8801 cyla
8801 Chlh
8801 Ctro
8801 Cpar
8801 cyui
8801 Cdub
8801 Clu-
8801 Dhan

 

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

109

88c1 Scar Y

8801 Calb
8801 Ctro

9391 Cgui
.9391 cduh
3.901 cm-
8801 Dhan

 

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

110

260 270 280 290 300
................|||

 

 

 

 

8881 Scar Y
8881 Scar R
8881 8par

ssc1 Shay
8881 Scan
8881 cyla
888! Chlb
8881 Ctro
8881 Cpar
ssc: Cgui

8881 81n-
8881 Dhan

8881 ryra
8881 rye:
8881 Pool
8881 Nora
8881 Traa

8881 Nfis
888! Anid
8881 Ater
8881 Lola
8881 Afla
ssc1 Anig
8881 Aory
8881 Afum
8801 Scot
8801 Scar
8801 Spar
8801 8m1k
8801 Shay
8801 Scan
8801 cyla
8801 Chlh
8801 Ctro
8801 Cpar
8801 cyui
8801 Cduh
8801 Clus
8801 Dhan

Y
R

 

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

111

310 320 330 340 350

 

 

8881 Scor Y
8881 Scar R
8881 8par
8881 8m1k
8881 Shay
8881 Scan
8881 891a

88C1 Clua

 

8801 Dhan

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

112

360 370 380 390 I 400

8801 Seat
8801 Scar
8801 Spar
8801 831k
8801 Shay
8801 Seas
8801 cyl-
8801 8a1h
8801 Ctro
8801 Cpar
8801 cyui
8801 Cduh
8801 Clu-
8801 Dhan

N N

 

Figure 82: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

113

410 420 430 440 450

oaaa aa-a'uaaa|aao-|aaoo

 

 

8881 Scot Y
8881 Scot R
8881 Spar
8881 Salk
8881 Shay
8881 Scan
8881 8y1a

8881 Lila
8881 Anig
8881 Lory
8881 Afum
8801 Scar
8801 Scar
8801 Spar
8801 Snik
8801 Shay
8801 Sea-
8801 cyla
8801 Chlh
8801 Ctro
8801 Cpar
8801 cyui
8801 8duh
8801 Clus
8801 Dhan

a":

 

Figure 32: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

114

460 470 480 490 500

8881 Scar Y
8881 Seat R
8881 Spar
8881 81:1):
8881 Shay
8881 Scan
8881 8g1a
8881 8a1h
8881 Ctro
8881 Cpar
8881 Cgui
8881 8duh
8881 Clus

8801 Cth
8801 Ctro

8801 Cgui
8801 8duh
5.991 Clus
5591 Dhan

 

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

115

510 520 530 540 550

8801 81ua
8801 Dhan

 

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

116

560 570 580

 

 

 

8881 Seat Y
8881 Seat R
8881 Spar

8881 Shay

8881 Ctro

SSC1 Anid
88C1 Atcr

8881 Afla
8881 Anig
8881 Jory
8881 Afum
8801 Scar
8801 Scar
8801 Spar
8801 Salt
8801 Shay
8801 Scan
8801 cyla
8801 8a1h
8801 Ctro
8801 Cpar
8801 cyui
8801 8duh
8801 Clus
8801 Dhan

N N

 

Figure B2: SSC1 and SSQ1 combined amino acid multiple sequence alignment
(continued)

117

 

 

 

. :HoH ..Omm 13.qu

0mm ..an 19491:. on

 

Eon—IE? 85.58 2.51:: E: oEEa “Dag ova—o muoafiosgueom "mm PEME

«Hon 35.
neon 3%
83m. n38
5am 35..
3% 35.
Moon «38
noon BE.

 

ObH

 

MN

 

 

ounc‘-.-o

 

 

 

 

 

 

 

         

  

 

 

 

 

 

 

  
  

 

 

   

 

 

   

 

 

   

 

   

 

 

    

omH ova on." can OH." OOH on
52:28 .mmoa .oam 239.5mxm90 .o ofiuflom..mmeummqm:n>ng>azqa 3.60 .85..
Ha gang“? quxmiaw. q .30. .38]umme m.o>r:_:.l:5mme. .91qu 9.5 l I gammy—E noon .85.
. ommméqu my? 8381.95: .1 _ ".3. 39:31:; qnqxflifliinfi know H05”.
in ..xqamwfilzgammofaam." cacaomwmxmq. WA? 5:113: mmHflmmqm>nqnq..1...i:...._.>a a: Mflfifi. ”UFO
0mm ..thxqamwflafiammo .95.... oqafimxma. m3? hi“:ugmmaflmmqmwnqnqnflnnzze Noun H015

 

hay-ntdufl _ N H00“ HU‘H.
h404$>3ufl3 N “.0“ HU‘h.

 

 

 

     

 

0mm. .4m.[mqamw¢fizqammofomm_.
mmo

   

 

 

 

 

OH

118

E2:—mum? 8:258 2.51:: Eva 95:5 ~O<n ova—o 3333.0 "3— 9.53m

EFF>Q4

IF .3an

 

 

 

 

 

 

 

 

 

 

 

 

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

      

        

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

H m>¢MZSHU nubumfikch:IME2>qmu one «Ema mmdmfi> . : Hana HUND
HmMH4>< I 4::n m .u1u_z:saiz um.<.mqumm<m. cash H016
nChnflu-I ..._...EI. . ... u lint-I...— . II“; I: _".!l uln. . 3.:an HU‘h.
BH‘QBHAMD?HIIIIII!.I :x _ Ila-.13:- uulnl. _fifiouv~w _. lad. o “mmdml. o a: . .HIQU HU‘h.
MH‘SJHH?H> i n... Ila-... ...: :15:— ENG an. qummmdm OHuU HUSH.
MH‘SAMN?HL 1 .._.-Mr“;— _ 325': #:4? ..._DE mg. 0 1mm4m _ . QHQU HUSH

.... .... .... .... .... .... .... .... .... .... .... I...._.... .... .... ....

OOH OMH OvH OMH ONH OHH OOH Om

xwmunmmmoma HdeI $.an . 4 . I nml“. _Qnmm_. ”.mOQ m fiZJ. .MQZH mu _Qmom..n~b MN an“ Hugh:
.HONQKWNHNNM .Hd , m Mrmza: . I :2!!- _Dm$.u ”.mOdJmnmth IMEM n— ame mam n ...—N UUHU HU‘N.
SWEWKKO HG. mawmzu Jddugidemi ..moqmwumtmq “EH Comm... ... 2mm I1!!! >5.— gUU HU‘H.
qmmmmqfiomo ‘ :rhifliomm 3.qu m ELF mu REG .2. .: gnu n36
..Hmam H 95 84“”an .mqémzH _ ... E. my hand 35.
zmzmxmoommuomumflfiuin .32. an. . H Dam. ..moq IIme ..mq mean u» 0.30 "DE.
Imtflfin£33u3{=; «Hdeame-dfi . HHQmm.:mOsz m._mq7KDZ>H Gamma Jam Iaiilph AHUU HUN5

 

 

....._l..._....l..

on or O» om ov on on OH

 

 

 

 

 

 

 

 

 

 

 

 

 

119

2555:: 8:253. 23:5:— 23 2:8: ~U<n ova—u Sarawak "mm 235

 

 

 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. .H . . , . . ._ .. .. . _Q>. anaqwommja aqua HUSH.
...... . . A .. . . ‘ .u . .u u . _ .. flamquommqqm ooh.“ no.8

‘ . ; . . . . . V . . . ., . qzaqwommqqm who: H05.
:55: :nl:.= m4. . ”...H :m . r ‘ . . . 4 . a . HOIh HUSH.
Ina! FFEIR m<. . o . NO: HUGH.
Iva-du— _i!_l:m¢_. . ”HUN HU‘h.
Ina! .mlln mfi . NkON HU‘h.

am." am: ova

 

 

 

 

 

 

120

9:08:33 3:253. 23:58 Eva 2:8: “Dag. ecu—o “Sauuummv‘ 6m 9.3mm

 

 

 

m>umam<_
1

m>0m¢m<

 

 

           

 

N.DQODM¢H>QHamemqu04m

 

 

l huu< ubtfi
u. oqwomflgamm . mmquoam 33¢ HOE.
M..oqoomfiBoHomjmmqq>o¢m.. 3.6.2 .35.

 

 

 

 

         

              

‘mmmUKmQHanHHW_ZH.Qmmmd..m ESN1 HUtS

4&0 afimw<mZHL4mq.m¢.oxn.

AQOAHm><MZHJ¢mQHm<€oXI.Gflm xQQSHB mmmw‘mAvamHHwfiZB Dmmm.fl.m ENO‘ NONE
Jan.Ahmw¢mZHLqulm<€onmOdmaxammHQMO: ‘ mmmOKmAH€;mHH>[ZH DammmA_m Mflfi< HUSH
qmflndhm><mZHI¢mqfim<doxv.OumnxommHB mmmUKQQHrmmHHVrZP Qmmm.fl_m IHNQ HUQS

rqufim¢.ox.m
«ma md..xnn n. momm A .mommmofimqhy.mth 29

 

.mxammwxquanHHNnZH.Qmmmd>_m lHU‘ HU‘S
mommde_m Houfl H016

 

 

 

 

 

 

 

on Oh on on ov on on OH

 

 

121

Fungal Mitochondrial Heat Shock Protein Phylogenetic Gene Tree Input Topologies

Apgndix C

for cod eml Evolutionary Rate Analysis

Tree topologies input into codem'l represent composite structures of highly supported
relationships from trees inferred using the following sequence partitions: 1‘“, 2nd, 3rd, lSt
and 2"d nucleotide positions, all nucleotides, and amino acids. Maximum Parsimony,
Maximum Likelihood, and Bayesian Inference methods were used. Branches were
manually collapsed if bootstrap support or posterior probabilities were below 90% or 0.9,
respectively. All topologically unique composite trees are shown.
Taxon name abbreviations used are listed in the table below:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Taxon Taxon

Abbreviation Fungal Species Abbreviation Fungal Species

Sce r_Y Saccharomyces cerevisiae RM11 Fgra Fusarium graminearum

Sce r_R Saccharomyces cerevisiae YJ M789 Fver Fusarium verticilliodes

Spar Saccharomycesgaradoxus Fsol Fusarium solani

Smik Saccharomyces mikatae Ncra Neurospora crassa

Sbay Saccharomyces bayanus Tree Tn'choden‘na reesei

Scas Saccharomyces caste/Iii Pans Podospora anserina
igla Candida glabrata Nfis Neosartoola fischeri

Calb Candida albicans Anid AjpeiMus nidulans

Ctro Candida tropicalis Ater Aspergillus terreus

Cpar Candida parapsilosis Acla Aspergillus clavatus
_C_gui Candida guilliermondii Afla Aspergillus flavus

Cdub Candida dub/iniensis Anig Aspergillus Liger

Clus Candida Iusitaniae Aory Aspergillus oryzae

Dhan Debaryomyces hansenii Afum Asperg'IIus fumigatus

Foxy Fusarium oxysporum

 

 

 

 

 

122

 

 

Cgkl
Seas
Sbay
Snfik
Spar

 
   
 
   
  

lus
Cpar
Cunt
Can:
Cdub
Dhan
Cgui
Nona
Pans
Tuna
Fsol
an:
Fver
Foxy
Ankl
flung
Iuer
Anny
Amh
Ackl
Ahun
Nfis

Figure C1: SSC1 Bayesian
Inference Tree 1

ScefiY
Seed!

Cgkl
Seas
Sbay
Snfik
Spar
SceFV
Seed?
(flus
Cpar
Cuni
Can:
Cdub
Dhan
Cgui
Ncnl
Pans
Tn»;
Fsol
ana
Fver
Foxy
Ask:
rung
Iuer
Ankl
AUunI
Nfis
Anny
Afla

 

Figure 62: SSC1 Bayesian
Inference Tree 2

123

 

Cgla
Seas
Sbay
Smik
Spar
SeerY
ScerR
Dhan
Cpar
Ctro
Calb
Cdub
Cgui
Clus
Nera
Pans
Tree
Fsol
Fgra
Fver
Foxy
Anid
Anig
Ater
Aory
Afla
Acla
Afum
Nfis

Cgkl
Seas
Sbay
Snfik
Spar
SeerY
Seen?
Dhan
Cpar
Chci
Can:
Cdub
Cgui
(”us
New:
Pans
Tux;
Fsol
anl
Fver
Foxy
Ackl
Auk;
IMer
Ankl
Ahun
Nfis
Anny
Afla

 

Figure 03: SSC1 Bayesian Figure C4: SSC1 Bayesian
Inference Tree 3 Inference Tree 4

124

 

 

 

Cgla
Seas
Sbay
Smik
Spar

, —- SeerY
F— — SeerR
Clus
Cgui
Dhan
Cpar
Ctro
-— Cdub
— Calb
Pans
Nera
Tree
Fsol
Fgra
r— Fver
—- Foxy
Anid
Anig
Ater

...— Aory
'_"i_. Afla
Aela
F— Nfis

Figure C5: SSC1 Maximum Figure CG: SSC1 Maximum
Likelihood Tree 1 Likelihood Tree 2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

g

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

125

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Aela
{Em
Nfis

 

Dhan Cgla
Cgla Seas
~ Seas Sbay
Sbay Smik
_ Smik Spar
Spar SeerY
SeerY SeerR
ScerR Dhan
Cpar Cgui
Ct", Clus
El ECalb Cpar
Cdub Ctro
Cgui Calb
Clus CdUb
Pans Pans
I Nera Ncra
Fsol F3°'
1% Fgra Fgra
Fver Fver
‘—-' Foxy
, :27: Anid
- Ater
21:? Aela
______ Anig
23a” Nfis
Afum
Afla

Aory

Figure CB: SSC1 Maximum
Parsimony Tree 1

Figure C7: SSC1 Maximum
Likelihood Tree 3

126

 

Cgla Cgla
Seas Seas
Smik SbaY
Shay Smik
Spar Spar
SeerY SeerY
SeerR Seer R
Dhan Clus
Cgui Cgui
Clus Dhan
Cpar Cpar
Ctro Ctro
Calb Calb
Cdub Cdub
Pans Pans
Ncra Ncra
Tree Tree
Fsol Fsol
Fgra Fgr a
Fver Fver
Foxy Foxy
An id Anid
Ater Ater
Aela Aela
Anig Anig
Nfis Nfis
Afum Afum
Afla Afla
Aory Aory
Figure 09: SSC1 Maximum Figure 010: SSC1 Maximum
Parsimony Tree 2 Parsimony Tree 3

127

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure C11: SSC1 Maximum
Parsimony Tree 4

128

 

SSQ1 Cgla
SSQ1 Seas
SSQ1 Spar
SSQ1 Sbay
SSQ1 Smik
SSQ1 SeerY

 

 

 

SSQ1 ScerR
SSQ1 Clus
SSQ1 Dhan
SSQ1 Cgui
SSQ1 Cpar
SSQ1 Ctro
SSQ1 Cdub
SSQ1 Calb
SSC1 Aory
SSC1 Afla
SSC1 Anig
SSC1 Ater
SSC1 Anid
SSC1 Aela
SSC1 Afum

SSC1 Nfis
SSC1 Ncra

SSC1 Pans
SSC1 Tree
SSC1 Fsol
SSC1 Fgra
SSC1 Fver
SSC1 Foxy
SSC1 Clus
SSC1 Cpar
SSC1 Ctro
SSC1 Calb
SSC1 Cdub
L..— r—-SSC‘l Dhan
l—SSC1 Cgui

SSC1 Cgla
SSC1 Seas
SSC1 Sbay
SSC1 Smik

Figure C12: SSC1 SSC1 Spar
and SSQ1 Bayesian SSC1 SeerY
Inference Tree 1 SSC1 SCOT“

 

 

 

 

 

 

 

 

 

in:

 

 

l
i

 

 

 

 

 

 

 

rm

129

 

SSQ1Cgh
SSQ1 Seas
SSQ1 Spar
SSQ1 Sbay
SSQ1 Smik

 

SSQ1 SeerY

 

SSQ1 SeerR

 

SSC1Cgh
SSC1 Seas
SSC1 Sbay
SSC1 Smik

 

 

 

 

 

SSC1 Spar
SSC1 SeerY
SSC1 SeerR
SSC1 Clus
SSC1 Cpar
SSC1 Ctro
SSC1 Calb
SSC1 Cdub
SSC1 Dhan
SSC1 Cgui
SSC1 Ncra
SSC1 Pans

SSC1 Tree

fissm F30.

SSC1 Fgra

SSC1 Fver

___. ESSC1 Foxy
SSC1 Aory

SSC1 Afla

SSC1 Anig
SSC1 Ater

SSC1 Anid

 

SSQ1 Dhan
SSQ1 Clus
—dl $8801 Cgui
SSQ1 Cpar
SSQ1 Ctro
SSQ1 Cdub
SSQ1 Calb

 

 

 

 

 

 

 

Figure C13: SSC1 SSC1 Aela
and SSQ1 Bayesian SSC1 Afum
Inference Tree 2 SSC1 Nfis

130

SSQ1 Cgla
SSQ1 Seas
SSQ1 Spar
SSQ1 Sbay
SSQ1 Smik
SSQ1 SeerY

 

 

 

 

SSQ1 ScerR
SSQf Dhan
SSQ1 Clus
SSQ1 Cgui
SSQ1 Cpar
SSQ1 Ctro
SSQ1 Cdub
SSQ1 Calb
SSC1 Cgla
SSC1 Seas
SSC1 Sbay
SSC1 Smik
SSC1 Spar
SSC1 SeerY
SSC1 SeerR

SSC1 Dhan
SSC1 Cpar

SSC1 Ctro

SSC1 Calb

SSC1 Cdub
l—SSC1 Clus

l--—SSC1 Cgui

SSC1 Nera
SSC1 Pans
SSC1 Tree

SSC1 Fsol

SSC1 Fgra

SSC1 Fver

SSC1 Foxy
SSC1 Aory

SSC1 Afla

SC1 Anig
SC1 Ater
SSC1 Anid
SSC1 Aela
SSC1 Afum
SSC1 Nfis

 

 

 

 

 

 

 

 

 

MW

 

 

 

 

 

 

 

 

it

Figure C14: SSC1
and SSQ1 Bayesian
Inference Tree 3

ii

131

 

SSG1Cgh

 

 

 

 

SSQ1 Seas
$01 Spar
SQ1 Sbay

SSQ1 Smik

SSQ1 SeerY

SSQ1 ScerR

SSQ1 Clus
SQ1 Dhan
$01 Cgui

SSQI Cpar
SQ1Cuo
$01 Cdub

SSQ1 Calb

 

 

 

SSC1Cgh

 

   

SSC1 Seas
SSC1 Sbay
SC1 Smik

 

SSC1 SeerFI

 

 

 

SSC1 Dhan
SSC1 Cpar

SSC1CuO
SC1 Calb
SC1 Cdub
J—"SSC1 Clus
SC1 Cgui

 

 

SSC1 Nera

 

SSC1 Pans

 

 

 

SSC1 Tree
SSC1 Fsol
SC1 Fgra
SC1 Fver
SSC1 Foxy
SSC1 Aory
SSC1 Afla

SC1 Anig
SC1 Ater

 

 

Figure C15: SSC1

and SSQ1 Bayesian

Inference Tree 4

iii

SC1 Anid
SC1 Aela
SC1 Afum
SC1 Nfis

if

132

SSQ1 Cgla
Seas
Spar
Sbay
Smik
SeerY
SSQ1 ScerR
SSQ1 Clus
Dhan
Cgui
Cpar
Ctro
Cdub
SSQ1 Calb
SSC1 Cgla
1 Seas
1 Sbay
1 Smik
1 Spar
1 SeerY
SSC1 ScerR
SSC1 Dhan
1 Cpar
1 Ctro
1 Calb
1 Cdub
1 Clus
1 Cgui
SSC1 Ncra
1 Fans
Tree
1 Fsol
1 Fgra
1 Fver
1 Foxy
SSC1 Aela

1 Aory
1 Afla

1 Anig

1 Ater

1 Anid

1 Afum
Nfis

  
  
  
 
 
 
 
 
  
  
 
  
  
 
  
  
  
 
 
  
  
  
 
 
 
 
  

Figure C16: SSC1
and SSQ1 Bayesian
Inference Tree 5

133

 

 

SSQ1 Cgla
Seas
Spar
Sbay
SSQ1 Smik
SeerY
SSQ1 ScerR
SSQ1 Dhan
Clus
Cgui
Cpar
Ctro
Cdub
SSQ1 Calb
SSC1Cgh
1 Seas
1 Sbay
1 Smik
1 Spar
1 SeerY
SSC1 ScerFI

SSC1 Dhan
SSC1 Cpar

1CuO
1 Calb
1 Cdub
1 Clus
1 Cgui
SSC1 Ncra
SSC1 Fans
1 Tree
1 Fsol
1 Fgra
1 Fver
1 Foxy
SSC1 Aela
SSC1 Aory
1 Afla

1 Anig
1 Ater
1 Anid
1 Afum
1 Nfis

 
 
  
  
  
 
  
 
 
  
  
  
 
   
 
  

Figure C1 7: SSC1
and SSQ1 Bayesian
Inference Tree 6

134

 

 

Cgla
Seas
Sbay
Smik
Spar
SeerY
SeerR
Cpar
Ctro
Cdub
Calb
Cgui
Dhan
SSQ1 Clus
SSC1Cgh
1 Seas
Sbay
1 Smik
1 Spar
1 SeerY
1 Seed!
SSC1 Clus
1 Cgui
1 Dhan
1 Cpar
Ctro
1 Cdub
1 Calb
SSC1 Ncra
1 Pans
1 Tree
Fsol
1 Fgra
Fver
1 Foxy
SSC1 Ater

1 Anig
1 Anid
1 Aela

1 Aory
1 Afla

1 Afum
Nfis

  
 
 
 
 
 
 
 
  
 
 
 
 
  
 
  
 
 
 
  
   
 
  
  
  
 
  

Figure C18: SSC1
and SSQ1 Maximum
Likelihood Tree

135

 

 

 

Seas
Cgla
Spar
Smik
Sbay
SeerY
SSQ1 SeerR
SSQ1 Dhan
Clus
Cgui
Ctro
Cpar
Calb
SSQ1 Cdub
SSC1Cgh
1 Seas
1 Sbay
1 Smik
1 Spar
1 SeerY
1 SeerR
SSC1 Clus
1 Cgui
SSC1 Dhan
1 Ctro
1 Cpar
1 Calb
1 Cdub
SSC1 Ncra
SSC1 Fans
1 Tree
1 Fsol
1 Fgra
1 Fver
1 Foxy
1 Ater

1 Anig
1 Anid
1 Aela

1 Aory
1 Afla

1 Afum
1 Nfis

   
  
  
  
  
  
  
  
  
  
 
  
 

Figure C19: SSC1
and SSQ1 Maximum
Parsimony Tree 1

136

SSQ1 Seas

 

 

 

 

SSQ1Cgh
SSQ1 Spar
SSQ1 Smik
SSQ1 Sbay
SSQ1 SeerY
SSQ1 ScerR
SSQ1 Dhan

 

SSQ1 Clus

 

 

 

SSQ1 Cgui
SSQ1 Ctro

 

SSQ‘I Cpar
SSQ1 Calb
$501 Cdub

 

SSC1Cgh
SSC1 Seas

 

Mali

SSC1 Sbay
SSC1 Smik
SSC1 Spar
SSC1 SeerY
SSC1 ScerR

 

 

SSC1 Clus
r—SSC1 Cgui

l—SSC1 Dhan
SSC1 Ctro

 

SSC1 Cpar
SSC1 Calb
SSC1 Cdub

 

SSC1 Nera
SSC1 Pans

 

SSC1 Tree

 

SSC1 Fsol

 

 

 

ii

SSC1 Fgra
SSC1 Fver
SSC1 Foxy
SSC1 Ater
SSC1 Anig
——-SSC1 Anid
-——-SSC1 Aela

 

 

 

Figure C20: SSC1
and SSQ1 Maximum
Parsimony Tree 2

SSC1 Aory
{SSC1 Afla
, SSC1 Afum
{ssm Nfis

 

137

 

 

 

SeerY
ScerR
Spar
Sbay
Smik

 

 

 

 

 

Seas
Cgla

Figure C21 : JAC1
Saccharomyces Bayesian
Inference] Maximum
Parsimony Tree

Cdub
Calb

 

 

Ctro

 

 

 

Cpar
Dhan
Clus
Cgui

Figure 023: JAC1
Candida Bayesian
Inference Tree

SeerY
_f—CW
Spar

Smik
Sbay

Cgla

 

 

 

 

 

 

Seas

Figure C22: JAC1
Saccharomyces Maximum
Likelihood Tree

Cdub
Calb

 

 

Ctro

 

 

 

Cpar
Dhan

Cgui

[—
Lem

Figure C24: JAC1
Candida Maximum
LikelihoodTree

 

138

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Calb Pans
Cdub Nera
Cpar Tree
Ctro ‘ Fsol
Dhan Fgra
Cgui Fver
[: Clus Foxy
Figure C25: JAC1 Figure C26: JAC1 Fusarium
Candida Maximum Bayesian Inference! Maximum
Parsimony Tree Likelihood I Maximum Parsimony
Tree
Nfis Aory
Afum Afla
Aela Anid
, Anid Anig
—[: Aory h—' Ater
Afla Aela
Anig ‘ Nfis
___.: Ater Afum
Figure C27: JAC1 Figure C28: JAC1
Aspergillus Bayesian Aspergillus Maximum
Inference Tree Likelihood Tree

139

 

 

 

 

 

 

 

Figure C29: JAC1
Aspergillus Maximum
Parsimony Tree

140

 

 

 

Agmndix D

Evolutionary Rate Test Specifications Used in Control Files Used to
Run codem'l of PAML

141

 

Site-Specific Model

Model = O

Nsites = 7 (a) distribution approximated as a beta distribution)
ncatG = 3 or 10 (# of categories pre-defined in the (0 distribution)

Branch-Site Model: Model A as defined by Zhang et al. (2005 )
Model = 2

Nsites = 2 (on distribution includes sites under positive selection)
ncatG = 3 (# of categories pre-defined in the to distribution)
f1x_kappa = O (kappa to be estimated)

fix_omega = O (omega to be estimated)

null model for branch-site test

Model = 2

Nsites = 2 (on distribution includes sites under positive selection)
ncatG = 3 (# of categories pre-defined in the to distribution)
fix_kappa = 1 (kappa fixed)

kappa = 1 (fixed value of kappa)

fix_omega (omega fixed)

omega = 1 (fixed value of omega)

Clade Model: Model D as defined by Bielawski and Yang (2004)
Model = 3

Nsites = 3 (discrete to distribution)

ncatG = 3 (# of categories pre-defined in the to distribution)

null model for clade test

Model = 0 (to distribution and estimated values apply to all branches of the tree)
Nsites = 0 (one gene-wide average to estimated)

ncatG = l (# of categories pre-defined in the to distribution)

142

 

 

Appgndix E

Likelihood Ratio Tests of codem'l Evolutionary Rate Analyses

 

143

 

mmfigg

hm~-mmw.~

53-2.06
3.9-x.

Nmmvné
Hmmmomd
www-mnmfi

“.29-“.

mmmnmd

em~-wmm.m

«mmmmmd
3.9-2

~m-mom.m
mvmfinofi
mammfim
3.9-x.

MMN

mmN

:3

«23 Amnessz. e2 .9 $982. 92
8.82 3352. 92 .9 22
R222: $985 92 .9 22
32:3
32:2

@342 21982. 92 .9 $982. as.
$.32 #35qu Q2 .9 22
2382 235$ 92 .9 22
32:3
«.833

3.9: sneeze e2 .9 $982. 22
3%: 36382. 92 .9 22
3.98 $982. 92 .9 22
3:33
32:2

$.32 $985 22 .9 3"onqu e2
33: Amuosszv 92 .9 22
2.23 anesszv e2 .9 22
32:3
32:2

size“ as 222 23229.:

 

3.33?

Hwémmma-

 

_ $982. 22

_ $985 22

 

emANme-

e 005 _n

modmmmfi-

 

_ $9282. 92

_ $982. as.

 

mNdVNmH-

m 00.5 _m

3.33.?

 

_ .muocszv 92

_ 335$ 92

 

afimamma-

N «on. _m

3633.

 

F $6282. 92

_ $985 92

 

«002...;

mmfimmma-
0.2

Humomma-
92

ondvmma-
22

3:8an-
0.2

3:... 323m noo£.9=..»o. gramme 59.5 :25.

case .9» 3.8.2 286 G9. a 2822.58 to» one. 8229.: am can»

144

mmmamd

www-mmm.m

mmmmmvé
m2m>d

mm-mm~.m

owmmmmé

mvwmofim
m3_m>..n_

emmaé

mm~-mmN.H

m--mm~.~
m:_m>na

ma -mmm.v
Nmmmomé
Ommmmné

3.9-2

32.26 :3 28.2 2.5 89. a 2823.28 .92 2.2 282.95 2.32.28. 3 22¢

MMN

 

$5282. 22 .
3.82 2328222.
$.22 $98232.
32:3

382.2

3.an

2:: 2282222.
22.82 8328222.
033 2328232.
3213

325.2

8.22 3328232.
3.3: 2328232.
8.22: $28232.
32:3

3232

22.2 228222.
Rea 23523.2.
3.33 $98232.
32:3

323.2

9 $282. 22
m> OS.
9 22

9 $932. 22
m> 22
9 22

9 fiuousz. 22
m> OS—
m> OS—

9 fiuouszv 22
m> 0—2
m> OS.

8.3.5” as 2.2 25229.:

 

mmdemwfi-

omKNmmH-

 

_ 3128232

L $985 22

 

ooKeNmH-

«005 as.

emNmmmH-

 

_ Amuosszv 22

_ $9822 22

 

3.3me-

m 35 .._2

mw.mw~ma-

 

_ Amuwuszv 22

L $9822 22

 

NmKHNmH-

~ 008 .=2

wHKmNmH-

 

_ 26$qu 22

_ 2"owqu 22

 

Humbug—2

2.3me
0.2

wmdomma-
22

mméommfi-
OS.

NYmNmmH-
02

3:.-. 8.33 305.0232 0283: 33.5 :25.

145

*

 

 

3mm“ .23.: meat 8% can Umm __m 2.92m 20% “502.2: 5.2. m5 38.2.12» c. 33...». 83...; 32:32.30. 9.532
32. m5 .235 9 29... #62 23:85:92. _muoE 2.29 2: 88.2.. 3.8.. v.83

mod v a an 2.8.253 8:32-; mmuocmo ..
matommumu 3 .o 32.5: 2: u .982
u x6523. ... 82.3.2. .262 3% of m. as. ..muoE =2. 2: m. 22

mmmmmé

www-mgd

~m~-m~m.¢
m:_m>-n_

«mmmfiw
Nommmfia
Nm~-mom.w

«29-2

cmmmmé

wmmmmmd

mmNQNNd
m:_m>-n_

9390 .92 28.2 2.5 89. .0 2822.50... .2: 2.3. 8229.... u....m=2_.=.3. m. 222

mmm

«1mm

3

 

 

 

 

 

 

om.mmwoa-
o_>.

mmdmwofi-
0.2

09?me
o.)—

352 89282 22 .9 .Nuwsuz. 22

3.22 $6.82. 22 .9 22

8.28 328:. 22 .9 22 3.2.8.- 8.322-

..._...3 _ .muw.82. 22 _ $282. 22
e 3: 22 e we. .._2

3.32 8282. 22 .9 .3282. 22

8.32 .mugsz. 22 .9 22

$.23 .3282 22 .9 22 22332. 8.382-

32.3 — .muwfiz. 22 _ .3282. 22
m 8: ...2 m we. 22

3.32 $6.82 22 .9 £282. 22

3.32 .3282. 22 .9 22

2.82 36.82. 22 .9 22 3.2.3.- 2.8%..

32.3 _ .muo.sz. 22 _ .mugsz. 22
~ 8: ...2 ~ 3.. ...2

3.9.23" as 0:2 .9823... ..._.... macs 28.2.9232 9.5mm: 5&3 225.

146

mmémwd

oo+mood

oo+mood
«emf.

Réwwé

mwmémmd

oo+wood
m:_m>un_

ow-mwm.m

oo+mood

oo+mood
m:_m>ua

wnmvmfi

oo+mood

oo+mood
w:_m>um

mnamfim

oo+wood

oo+mood
magi

 

8.2"... .3282. 22 .9 .3882. 22
$.83 .3282. 22 .9 9..
3:: .Nuufiz. 22 .9 22
.._....3
m 8: ...
«33 36.82. 22 .9 .3382. 22
8.28. .muofiz. 22 .9 22
8.8: .Nuosuz. 22 .9 22
32.3
e 8: ... .
83m .mugsz. 22 .9 36.82. 22
3.3: 86.82. 22 .9 22
8.25 .Nnozuz. 22 .9 22
32.3
... 8: a
21.3 36.85 22 .9 .3282. 22
$2.8 86.82. 22 .9 22
2.3: .3282. 22 .9 22
32.8
N 8.. ...
3.2;. .muu.82. 22 .9 .3382 22
8.2.2. .muo.82. 22 .9 22
2.2.: .Nuofiz. 22 .9 22
......E
n a!» a
8.3.5“ 93 2:: .8229...

 

medmNom-

 

_ .muofiz. 22

wmdmaom-

_

 

W 3282 22

3.23%.

 

T .muofiz. 22

wmdmaom-

 

m .muwusz. 22J

 

mmémfiom-

 

_ .muousz. 22 _

 

2.23m-
$2.82. 22
m 8... 3

2.28m-
$982. 22
e 8: ...

mmdwmom-
.3282. 22
m 8.. ...

2.88m-
.mugsz. 22
~ 3: ...

2.88m-
.Nuozuz. 22
u out a

mm.momfim-
0.2

mwdmmam-
22

mvémmfim-
0.2

afimmmfim-
0.2

@5523.
92

.._.... 933 82292.8. «5&2 9&8 .23.

92.50 .2: 28.2 226 89. .25 5m... ... 52.2.58 .92 25. 8229... "m 222

147

ow-mNH.m

8+mood

8500.0
~22.-.

mw-mond

oo+mood

8+mood
mag-n.

mm-wmw.H

8+wood

8+mood
mag-n.

Hw-wmmé

00.686

8+mood
mag-n.

 

 

3mm. .89.. mum... 8mm __m 9.0:... 9.8... 3003.3... .83 2... 38.3... .65 ... 32.23 83.8 uoo...9...-mo. 9.3mm:
Emu 2... 6.3... o. 2.9... .82 2.2.8.253 .32... 32. 2.. 38.3... 8.8.. .....2m

2.23 $8.82. 22 .9 38.82. 22
2.3: ...-0.32. 22 .9 22
8.23 38.82. 22 .9 22
32.3
3 8.. 22
3.8.. ...-0.82. 22 .9 38.82. 22
$8... $8.82. 22 .9 22
3.82 38.82. 22 .9 22
32.3
H 8.. 22
8.3 ...-0.82. 22 .9 38.82. 22
3.223 38.82. as. .9 22
3.2.3 38.82. 22 .9 22
32.3
8.. ..2
3.5 ...-"982. 22 .9 38.82. 22
32:3 3.8.82. 22 .9 22
8.32 38.82. 22 .9 22
32.3
o 8.. a
8.5.3. .8. 2.2 .8223...

8.3 v a 3 E8555 32?-.. $.0ch ..
8.35.8 3 .o .38.... m... u .982
u 53:89. ... 32.883 .25... 2.2. m... .... as. “.25... :2. m... m. 0.2

 

 

 

 

 

 

 

 

3.33m- 3.323. 2.22.:-
_ ...-3.82.22L 38.82.22 22
~25 22
...-.33...- 8.22.3- $.3on-
_ 88.82.on 38.82.22 22
. a... ...2
8.33. 3.3.3- 3.3.3.
_ 38.82.32 . 38.82.32 22
3.. ..2
2333- 333..- 3.82m-
_ 36.82.32 _ 38.82. 22 22
may...»

32.. 9.3. ....2...9....§ 9...»... .....so 22....

9.5.3 .3. .322 ......u 3mm .25 89. ... 82.2.28 .9» 2.... 302...... “3.2.2.2... m 22..

148

Table E3: Likelihood Ratio Test Comparison of
JAC1 Site-Specific Model Test Output

 

 

149

 

PAML output
negafive log-likelihood scores Hnl.) Likelihood ratio test statistics
Socchammyres clade
MP/Bl tree MPIBI tree
Number of 0.) Categories
3 10 2A(-ln L) df P-value
4248.93 -2249.55 1.25 7 9.906-01
Ml. tree Ml. tree
Number of w Categories
3 l 10 2A(-lnl.) df P-value
-2253.53 -2283.81 60.55 7 1.17E-10 "
Candida clade
Bl tree Bl tree
Number of w Categories
3 10 2A(-lnL) df P-value
-2767.48 -2765.66 3.63 7 8216-01
Ml. tree Ml. tree
Number of w Categories
3 10 2A(~lnL) df P-value
-2797.S3 -2796.00 3.06 7 8.79E-01
MP tree MP tree
Number of w Categories
3 10 2A(-lnL) df P-value
-2791.03 -2787.95 6.16 7 5.21E-01
Fusarium clade
MP/Bl/Ml. tree
Number of w Categories 2A(-lnL) df P-value
3 10 14.00 7 5.125-02
-2786.48 -2789.61

 

 

Table £3 (continued) : Likelihood Ratio Test Comparison of
JAC1 Site-Specific Model Test Outputs

PAML output
i i
negative log-likelihood scores (-lnL) Likelihood ratio test stat st cs

Aspergillus clade

 

 

 

 

MP Tree MP/Bl/ML tree
Number of w Categories
3 10 2A(-lnL) df P-value
-2508.72 _ -2508.73 0.03 7 1.00E+00 *
Bl tree Bl tree I
Number of w Cate ories
3 I 10 I 2A(-lnL) df P-value
-2801.61 -2790.07 23.08 7 1656-03 “‘
ML tree ML tree l
Number of w Categories
3 [ 10 J 2A(-lnL) df P-value
-2856.26 -2847.00 18.51 7 9.886-03

"' Denotes P-values significant at P < 0.05
Black boxes indicate the number of in categories in the site-specific model that was

significantly most likely to predict the data
Negative log-likelihood values shaded in gray indicate the overall best likelihood score

for the given clade obtained among all site-specific tests

150

 

 

Tabe E4: Likelihood Ratio Test Comparison of $501 Branch-Site Model Test Outputs

PAML output negative log-likelihood scores I-InL)

 

 

 

 

 

 

 

 

Bl tree 1
L Model A, estimated in I Model A, fixed to = 1
-31029.34 31129.88
Bl tree 2
I Model A, estimated in I Model A, fixed in = 1
-31023.33 7“ -31128.51
Bl tree 3
rModel A, estimated nu Model A, fixed a) = 1
81037.84 31139.44
Bl tree 4
I Model A, estimated in I Model A, fixed on = 1
31043.68 31124.20
Bl tree 5

 

I Model A, estimated in I Model A, fixed in = 1

 

34199.23 -34292.48

PAML output negative log-likelihood scores (-lnL)

BltreeG

 

L Model A, estimated in J Model A, fixed in = 1

 

-31088.51 -31166.09

ML tree

 

I Model A, estimated to I Model A, fixed (1) = 1

 

-31158.86 -31254.19

MPtreel

 

I Model A, estimated in I Model A, fixed in = 1

 

-31374.70 -31463.91

151

Likelihood ratio test statistics
Bl tree 1
Model A, estimated in vs. Model A, fixed in = 1
2A(-lnL) df P-value
201.07 2 2.18E-44
8i tree 2
Model A, estimated (.0 vs. Model A, fixed 10 = 1
2A(-lnL) df P-value
210.35 2 2116-46
Bl tree 3
Model A, estimated in vs. Model A, fixed in = 1
2A(-lnL) df P-value
203.21 2 7.48E-45
Bl tree 4
Model A, estimated to vs. Model A, fixed in = 1
2A(-lnL) df P-value
161.03 2 1.08E-35
Bl tree 5
Model A, estimated in vs. Model A, fixed to = 1
2A(-lnL) df P-value
186.50 2 3.18E-41
Likelihood ratio test statistics
Bl tree 6
Model A, estimated to vs. Model A, fixed in = 1
2A(-|nL) df P-value
155.15 2 2.04E-34
ML tree
Model A, estimated in vs. Model A, fixed in = 1
2A(-InL) df P-value
190.67 2 3.95E-42
MP tree 1
Model A, estimated 0.) vs. Model A, fixed in = 1
2A(-lnL) df P-value
178.43 2 1.80E-39

- q-.-“ :-

 

Table E4 (continued): Likelihood Ratio Test Comparison of $501
Branch-Site Model Test Outputs

 

 

PAML output negative log-likelihood scores (-lnL) Likelihood ratio test statistics
MP tree 2 MP tree 2
L Model A, estimated on ] Model A, fixed to = 1 Model A, estimated (.0 vs. Model A, fixed to = 1
-31351.81 -31458.31 2A(-|nL) df P-value
213.00 2 5.59E-47 *

‘ Denotes P-values significant at P < 0.05
Black boxes indicate the significantly model most likely to predict the data
Negative log-likelihood values shaded in gray indicate the overall best likelihood score obtained

among all branch-site PAML tests

152

 

REFERENCES

Abascal F, Zardoya R, Posada D (2005) ProtTest: selection of best-fit models of protein
evolution. Bioinformatics 21(9): 2104-2105.

Andrew A], Dutkiewicz R, Knieszner H, Craig EA, Marszalek J (2006) Characterization
of the interaction between the J -protein J aclp and the scaffold for Fe-S cluster
biogenesis, Isulp. J Biol Chem 281(21): 14580-14587.

Baumann F, Milisav I, Neupert W, Herrmann J M (2000) EcmlO, a novel hsp70 homolog
in the mitochondrial matrix of the yeast Saccharomyces cerevisiae. FEBS Lett
487(2): 307-312.

Benedict MQ, Cockburn AF, Seawright J A (1993) The Hsp70 heat-shock gene family of
the mosquito Anopheles albimanus. Insect Mol Biol 2(2): 93-102.

Bettencourt BR, Feder ME (2002) Rapid concerted evolution via gene conversion at the
Drosophila hsp70 genes. J MolEv0154(5): 569—586.

Bielawski JP, Yang Z (2003) Maximum likelihood methods for detecting adaptive
evolution after gene duplication. J Struct Funct Genomics 3(1-4): 201-212.

Bielawski JP, Yang Z (2004a) Maximum Likelihood Methods for Detecting Adaptive
Protein Evolution.

Bielawski JP, Yang Z (2004b) A maximum likelihood method for detecting functional
divergence at individual codon sites, with application to gene family evolution. J
Mol Evol 59(1): 121-132.

Boorstein WR, Ziegelhoffer T, Craig EA (1994) Molecular evolution of the HSP70
multigene family. J Mol Evol 38(1): 1-17.

Brown C], Todd KM, Rosenzweig RF (1998) Multiple duplications of yeast hexose
transport genes in response to selection in a glucose-limited environment. Mol
Biol Evol 15(8): 931-942.

Bucciantini M, Giannoni E, Chiti F, Baroni F, Formigli L et al. (2002) Inherent toxicity
of aggregates implies a common mechanism for protein misfolding diseases.

Nature 416(6880): 507-51 1.

Bukau B, Horwich AL (1998) The Hsp70 and Hsp60 chaperone machines. Cell- 92(3):
351-366.

153

Cheetham ME, Caplan AJ (1998) Structure, function and evolution of DnaJ: conservation
and adaptation of chaperone function. Cell Stress Chaperones 3(1): 28-36.

Conant GC, Wagner A (2003) Asymmetric sequence divergence of duplicate genes.
Genome Res 13(9): 2052-2058.

Craig EA (1989) Essential roles of 70kDa heat inducible proteins. Bioessays 11(2-3): 48-
52.

Davis J C, Petrov DA (2004) Preferential duplication of conserved proteins in eukaryotic
genomes. PLoS Biol 2(3): E55.

DeBry RW (1999) Maximum likelihood analysis of gene-based and structure-based
process partitions, using mammalian mitochondrial genomes. Syst Biol 48(2):
286-299.

Des Marais DL, Rausher MD (2008) Escape from adaptive conflict after duplication in an
anthocyanin pathway gene. Nature 454(7205): 762-765.

Diamond ME, Dowhanick JJ, Nemeroff ME, Pietras DF, Tu CL et al. (1989) Overlapping
genes in a yeast double-stranded RNA virus. J Virol 63(9): 3983-3990.

Drummond DA, Wilke CO (2008) Mistranslation-induced protein misfolding as a
dominant constraint on coding-sequence evolution. Cell 134(2): 341-352.

Drummond DA, Bloom JD, Adami C, Wilke CO, Arnold PH (2005) Why highly
expressed proteins evolve slowly. Proc Natl Acad Sci U S A 102(40): 14338-
14343.

Endo T, Ikeo K, Gojobori T (1996) Large-scale search for genes on which positive
selection may operate. Mol Biol Evol 13(5): 685-690.

Feder ME, Cartano NV, Milos L, Krebs RA, Lindquist SL (1996) Effect of engineering
Hsp70 copy number on Hsp70 expression and tolerance of ecologically relevant
heat shock in larvae and pupae of Drosophila melanogaster. J Exp Biol 199(Pt 8):
1837-1844.

Felsenstein J (1978) Cases in which parsimony or compatibility methods can be
positively misleading. Syst Zool 27: 401-419.

Felsenstein J (1981) Evolutionary trees from DNA sequences: a maximum likelihood
approach. J Mol Evol 17(6): 368-376.

Felsenstein J (2003) Inferring Phylogeneies. United States: Sinauer Associates.

154

Fitzpatrick DA, Logue ME, Stajich JE, Butler G (2006) A fungal phylogeny based on 42
complete genomes derived from supertree and combined gene analysis. BMC
Evol Biol 6: 99.

Force A, Lynch M, Pickett FB, Amores A, Yan YL et al. (1999) Preservation of duplicate
genes by complementary, degenerative mutations. Genetics 151(4): 1531-1545.

Gerrnaniuk A, Liberek K, Marszalek J (2002) A bichaperone (Hsp70-Hsp78) system
restores mitochondrial DNA synthesis following thermal inactivation of Miplp
polymerase. J Biol Chem 277(31): 27801-27808.

Golding GB, Dean AM (1998) The structural basis of molecular adaptation. Mol Biol
Evol 15(4): 355-369.

Goldman N, Yang Z (1994) A codon-based model of nucleotide substitution for protein-
coding DNA sequences. Mol Biol Evol 11(5): 725-736.

Gribaldo S, Lumia V, Creti R, de Macario EC, Sanangelantoni A et al. (1999)
Discontinuous occurrence of the hsp70 (dnaK) gene among Archaea and sequence
features of HSP70 suggest a novel outlook on phylogenies inferred from this
protein. J Bacteriol 181(2): 434-443.

Guindon S, Gascuel O (2003) A simple, fast, and accurate algorithm to estimate large
phylogenies by maximum likelihood. Syst Biol 52(5): 696-704.

Gupta RS (1999) Hsp70 sequences and the phylogeny of prokaryotes. Mol Microbiol
31(3): 1007-1009.

Gupta RS, Singh B (1992) Cloning of the HSP70 gene from Halobacterium marismortui:
relatedness of archaebacterial HSP70 to its eubacterial homologs and a model for
the evolution of the HSP70 gene. J Bacteriol 174(14): 4594—4605.

Gupta RS, Singh B (1994) Phylogenetic analysis of 70 kD heat shock protein sequences
suggests a chimeric origin for the eukaryotic cell nucleus. Curr Biol 4(12): 1104-
1114.

Hakes L, Lovell SC, Oliver SG, Robertson DL (2007) Specificity in protein interactions
and its relationship with sequence diversity and coevolution. Proc Natl Acad Sci
U S A 104(19): 7999-8004.

Han W, Christen P (2003) Mechanism of the targeting action of DnaJ in the DnaK
molecular chaperone system. J Biol Chem 278(21): 19038-19043.

Hennig W (1966) Phylogenetic Systematics. Urbana, IL: Univ. of Illinois Press.

155

Herrmann J M, Stuart RA, Craig EA, Neupert W (1994) Mitochondrial heat shock protein
70, a molecular chaperone for proteins encoded by mitochondrial DNA. J Cell
Biol 127(4): 893-902.

Hittinger CT, Carroll SB (2007) Gene duplication and the adaptive evolution of a classic
genetic switch. Nature 449(7163): 677-681.

Holder M, Lewis PO (2003) Phylogeny estimation: traditional and Bayesian approaches.
Nat Rev Genet 4(4): 275-284.

Hurles M (2004) Gene duplication: the genomic trade in spare parts. PLoS Biol 2(7):
E206.

Itoh T, Matsuda H, Mori H (1999) Phylogenetic analysis of the third hsp7O homolog in
Escherichia coli; a novel member of the Hsc66 subfamily and its possible co-
chaperone. DNA Res 6(5): 299-305.

Kellis M, Birren BW, Lander ES (2004) Proof and evolutionary analysis of ancient
genome duplication in the yeast Saccharomyces cerevisiae. Nature 428(6983):
617-624.

Kiley PJ, Beinert H (2003) The role of Fe-S proteins in sensing and regulation in
bacteria. Curr Opin Microbiol 6(2): 181-185.

Kondrashov FA, Rogozin IB, Wolf Y1, Koonin EV (2002) Selection in the evolution of
gene duplications. Genome Biol 3(2): RESEARCHOOOS.

Lange M, Macario AJ, Ahring BK, Conway de Macario E (1997) Heat-shock response in
Methanosarcina mazei S-6. Curr Microbiol 35(2): 116-121.

Langkjaer RB, Cliften PF, Johnston M, Piskur J (2003) Yeast genome duplication was
followed by asynchronous differentiation of duplicated genes. Nature 421(6925):
848-852.

Li WH (1980) Rate of gene silencing at duplicate loci: a theoretical study and
interpretation of data from tetraploid fishes. Genetics 95(1): 237-258.

Li WH, Yang J, Gu X (2005) Expression divergence between duplicate genes. Trends
Genet 21(11): 602-607.

Lill R, Muhlenhoff U (2008) Maturation of iron-sulfur proteins in eukaryotes:
mechanisms, connected processes, and diseases. Annu Rev Biochem 77: 669-700.

Lim EH, Brenner S (1999) Short-range linkage relationships, genomic organisation and

sequence comparisons of a cluster of five HSP70 genes in Fugu rubripes. Cell
Mol Life Sci 55(4): 668-678.

156

Lim J H, Martin F, Guiard B, Pfanner N, Voos W (2001) The mitochondrial Hsp70-
dependent import system actively unfolds preproteins and shortens the lag phase
of translocation. Embo J 20(5): 941-950.

Lindquist S, Craig EA (1988) The heat-shock proteins. Annu Rev Genet 22: 631-677.

Lutz T, Westermann B, Neupert W, Herrmann J M (2001) The mitochondrial proteins
Ssql and J acl are required for the assembly of iron sulfur clusters in
mitochondria. J Mol Biol 307(3): 815-825.

Macario AJ, Du gan CB, Conway de Macario E (1991) A dnaK homolog in the
archaebacterium Methanosarcina mazei S6. Gene 108(1): 133-137.

Messier W, Stewart CB (1997) Episodic adaptive evolution of primate lysozymes. Nature
385(6612): 151-154.

Muhlenhoff U, Lill R (2000) Biogenesis of iron-sulfur proteins in eukaryotes: a novel
task of mitochondria that is inherited from bacteria. Biochim Biophys Acta
1459(2-3): 370-382.

Neupert W (1997) Protein import into mitochondria. Annu Rev Biochem 66: 863-917.

Nielsen R, Yang Z (1998) Likelihood models for detecting positively selected amino acid
sites and applications to the HIV-1 envelope gene. Genetics 148(3): 929-936.

Nikolaidis N, Nei M (2004) Concerted and nonconcetted evolution of the Hsp70 gene
superfamily in two sibling species of nematodes. Mol Biol Evol 21(3): 498-505.

Nimura K, Yoshikawa H, Takahashi H (1996) DnaK3, one of the three DnaK proteins of
cyanobacterium Synechococcus sp. PCC7942, is quantitatively detected in the
thylakoid membrane. Biochem Biophys Res Commun 229(1): 334-340.

N uin PA, Wang Z, Tillier ER (2006) The accuracy of several multiple sequence
alignment programs for proteins. BMC Bioinformatics 7: 471.

Ohno S (1970) Evolution by gene duplication. New York: Springer-Verlag.

Ota T, Nei M (1994) Divergent evolution and evolution by the birth-and-death process in
the immunoglobulin VH gene family. Mol Biol Evol 11(3): 469-482.

Outten FW, Djaman O, Storz G (2004) A suf operon requirement for Fe-S cluster
assembly during iron starvation in Escherichia coli. Mol Microbiol 52(3): 861-
872.

157

Pal C, Papp B, Lercher MJ (2006) An integrated view of protein evolution. Nat Rev
Genet 7(5): 337-348.

Pelham HR (1984) Hsp70 accelerates the recovery of nucleolar morphology after heat
shock. Embo J 3(13): 3095-3100.

Philippe H, Budin K, Moreira D (1999) Horizontal transfers confuse the prokaryotic
phylogeny based on the HSP70 protein family. Mol Microbiol 31(3): 1007-1010.

Renner T, Waters ER (2007) Comparative genomic analysis of the Hsp70s from five
diverse photosynthetic eukaryotes. Cell Stress Chaperones 12(2): 172-185.

Ritossa F (1962) A new puffing pattern induced by heat shock and DNP in Drosophila.
Experientia 18: 571-573.

Ronquist F, Huelsenbeck JP (2003) MrBayes 3: Bayesian phylogenetic inference under
mixed models. Bioinformatics 19(12): 1572-1574.

Rouault TA, Tong WH (2005) Iron-sulphur cluster biogenesis and mitochondrial iron
homeostasis. Nat Rev Mol Cell Biol 6(4): 345-351.

Sahi C, Craig EA (2007) Network of general and specialty J protein chaperones of the
yeast cytosol. Proc Natl Acad Sci U S A 104(17): 7163-7168.

Scannell DR, Wolfe KH (2008) A burst of protein sequence evolution and a prolonged
period of asymmetric evolution follow gene duplication in yeast. Genome Res
18(1): 137-147.

Schilke B, Williams B, Knieszner H, Pukszta S, D'Silva P et al. (2006) Evolution of
mitochondrial chaperones utilized in Fe-S cluster biogenesis. Curr Biol 16(16):
1660-1665.

Schilke B, Forster J, Davis J, James P, Walter W et al. (1996) The cold sensitivity of a
mutant of Saccharomyces cerevisiae lacking a mitochondrial heat shock protein
70 is suppressed by loss of mitochondrial DNA. J Cell Biol 134(3): 603-613.

Schmidt S, Strub A, Rottgers K, Zufall N, Voos W (2001) The two mitochondrial heat
shock proteins 70, Sscl and Ssql, compete for the cochaperone Mgel. J Mol Biol
313(1): 13-26.

Suzuki Y, Nei M (2001) Reliabilities of parsimony-based and likelihood-based methods
for detecting positive selection at single amino acid sites. Mol Biol Evol 18(12):
2179-2185.

Swofford DL (2000) PAUP*, Phylogenetic Analysis Using Parsimony, Version 4.0b10.

158

Takahashi Y, Tokumoto U (2002) A third bacterial system for the assembly of iron-sulfur
clusters with homologs in archaea and plastids. J Biol Chem 277(32): 28380-
28383.

Tavaria M, Gabriele T, Kola I, Anderson RL (1996) A hitchhiker's guide to the human
Hsp70 family. Cell Stress Chaperones 1(1): 23-28.

Thompson JD, Higgins DG, Gibson TJ (1994) CLUSTAL W: improving the sensitivity
of progressive multiple sequence alignment through sequence weighting,
position-specific gap penalties and weight matrix choice. Nucleic Acids Res
22(22): 4673-4680.

Tissieres A, Mitchell HK, Tracy UM (1974) Protein synthesis in salivary glands of
Drosophila melanogaster: Relation to chromosome puffs. J Mol Biol 85(3): 389-
398.

Tuschl T, Eckstein F (1993) Hammerhead ribozymes: importance of stem-loop II for
activity. Proc Natl Acad Sci U S A 90(15): 6991-6994.

Voisine C, Schilke B, Ohlson M, Beinert H, Marszalek J et al. (2000) Role of the
mitochondrial Hsp70s, Sscl and Ssql, in the maturation of thl. Mol Cell Biol
20(10): 3677-3684.

Voisine C, Cheng YC, Ohlson M, Schilke B, Hoff K et al. (2001)]ac1, a mitochondrial J -
type chaperone, is involved in the biogenesis of Fe/S clusters in Saccharomyces
cerevisiae. Proc Natl Acad Sci U S A 98(4): 1483-1488.

Walter L, Rauh F, Gunther E (1994) Comparative analysis of the three major
histocompatibility complex-linked heat shock protein 70 (Hsp70) genes of the rat.
Immunogenetics 40(5): 325-330.

Wang R, Prince JT, Marcotte EM (2005) Mass spectrometry of the M. smegmatis
proteome: protein expression levels correlate with function, operons, and codon
bias. Genome Res 15(8): 1118-1126.

Wang TF, Chang J H, Wang C (1993) Identification of the peptide binding domain of
hsc70. l8-Kilodalton fragment located immediately after ATPase domain is
sufficient for high affinity binding. J Biol Chem 268(35): 26049-26051.

Ward-Rainey N, Rainey FA, Stackebrandt E (1997) The presence of a dnaK (HSP70)
multigene family in members of the orders Planctomycetales and

Verrucomicrobiales. J Bacteriol 179(20): 6360-6366.

Yang Z ( 1997) PAML: a program package for phylogenetic analysis by maximum
likelihood. Comput Appl Biosci 13(5): 555-556.

159

Yang Z (1998) Likelihood ratio tests for detecting positive selection and application to
primate lysozyme evolution. Mol Biol Evol 15(5): 568-573.

Yang Z (2007) PAML 4: phylogenetic analysis by maximum likelihood. Mol Biol Evol
24(8): 1586-1591.

Yang Z, Swanson WJ (2002) Codon-substitution models to detect adaptive evolution that
account for heterogeneous selective pressures among site classes. Mol Biol Evol

19(1): 49-57.

Zhang J, Rosenberg HF (2002) Complementary advantageous substitutions in the
evolution of an antiviral RNase of higher primates. Proc Natl Acad Sci U S A
99(8): 5486-5491.

Zhang J, Nielsen R, Yang Z (2005) Evaluation of an improved branch-site likelihood
method for detecting positive selection at the molecular level. Mol Biol Evol
22(12): 2472-2479.

Zhang P, Gu Z, Li WH (2003) Different evolutionary patterns between young duplicate
genes in the human genome. Genome Biol 4(9): R56.

Zheng L, Cash VL, Flint DH, Dean DR (1998) Assembly of iron-sulfur clusters.
Identification of an iscSUA-hscBA-fdx gene cluster from Azotobacter vinelandii.
J Biol Chem 273(21): 13264-13272.

Zheng L, White RH, Cash VL, Jack RF, Dean DR (1993) Cysteine desulfurase activity

indicates a role for NIFS in metallocluster biosynthesis. Proc Natl Acad Sci U S A
90(7): 2754-2758.

160