the coevolutionary genetics of medicago truncatula … · the coevolutionary genetics of medicago...
TRANSCRIPT
The Coevolutionary Genetics of Medicago truncatula and its
Associated Rhizobia
by
Amanda Gorton
A thesis submitted in conformity with the requirements
For the degree of Masters of Science
Department of Ecology and Evolutionary Biology
University of Toronto
© Copyright by Amanda Gorton 2011
ii
The Coevolutionary Genetics of Medicago truncatula and its
Associated Rhizobia
Amanda Gorton
Masters of Science
Department of Ecology and Evolutionary Biology
University of Toronto
2011
Abstract
Contrary to the predictions of numerous theoretical models, variation in partner quality
continues to persist in mutualisms, including in the symbiosis between legumes and rhizobia.
One potential explanation for the maintenance of this genetic diversity is genotype × genotype
interactions, however it is unknown which genetic regions might underlie these interactions. To
investigate this question, I performed a quantitative trait loci mapping experiment with two
different rhizobium strains to locate potential regions of the genome influencing genotype ×
genotype interactions between the legume Medicago truncatula and its symbiont Sinorhizobium
meliloti. I found no evidence for genotype × genotype or QTL × rhizobium interactions, however
some of the QTLs colocalized with genes involved in the symbiosis signaling pathway,
suggesting variation in these genes could potentially affect plant performance and fitness traits.
These findings have important implications for the evolutionary interactions between legumes
and rhizobia, and the genetic architecture of Medicago truncatula.
iii
Acknowledgments
First and foremost, I want to thank my supervisor, John Stinchcombe, for his guidance,
support, and enthusiasm, without which my thesis would not exist. I have learned a great deal
from him, and I am truly grateful for the opportunities he has given me. I want to also thank the
rest of the Stinchcombe Lab. I was very fortunate to be part of a great lab, full of excellent
scientists. I want to thank Brandon Campitelli, Emily Josephs, Young Wha Lee, and Anna
Simonsen for their help, advice and friendship. In addition, a special thank you is needed for
Katy Heath, who not only performed much of the work upon which my thesis is based, but was
also an invaluable source of information and advice.
There are many people who helped me plant and harvest my experiment. In particular I
want to thank Theresa Chow, Leila Kent, and Nikki Scodras, who dried, weighed and counted
many hundreds of plants. I am very grateful for their assistance. None of my molecular work
would have been possible without the help of David Maj, who taught me everything I know
about the magic of PCR. Many thanks to David for his patience, endless troubleshooting, and
comic relief.
I want to thank Stephen Wright and Megan Frederickson, both of whom offered
invaluable advice and insight as members of my supervisory committee. I also want to thank
Asher Cutter for kindly agreeing to sit on my exam committee. Many thanks to Bruce Hall and
Andrew Petrie for their help and resourcefulness with greenhouse work.
During my time in the Department of Ecology and Evolutionary Biology, I was lucky to
meet and befriend a number of smart, kind and supportive graduate students. In particular, I want
iv
to thank Emily Austen, Lucia Kwan, Alison Parker, and Jane Ogilvie for their kindness and
encouragement.
I am incredibly grateful for the love and encouragement of my family and friends. Thank
you to my friends Leigha Dimitroff, Susannah Ireland, Christine Rentschler for always being
supportive and understanding. I want to thank my sister, Penelope Gorton, for our many chats,
and my brother, Andrew Gorton, for his invaluable computer skills. Thank you to my dad, Mike
Gorton, for spending many late hours in the greenhouse planting and picking fruits with me, and
to my mom, Pamela Gorton, for making sure I always had plenty of snacks and treats to keep me
going. Last but not least, I want to thank Mike le Riche, for keeping me focused, taking me out
for much needed breaks, and always cheering me up.
v
Table of Contents
Acknowledgements……………………………….……………………………………………....iv
List of Tables……………………………………………………………………………………..ix
List of Figures……………………………………………………………………………………..x
List of Appendices………………………………………………………………………………..xi
Chapter 1: General Introduction……………………………………………………………….1
Mutualism stability and the maintenance of genetic variation……………………………………2
G × G interactions…………………………………………………………………………………6
Nodulation signaling pathway…………………………………………………………………….9
QTL mapping…………………………………………………………………………………….10
Study species……………………………………………………………………………………..13
Proposed experiment……………………………………………………………………………..14
Chapter 2: Resolving the genetic basis of genotype by genotype interactions between plants
and their mutualists…………………………………………………………………………….15
Materials and methods…………………………………………………………………………...19
Study species and mapping population…………………………………………………..19
Rhizobium strain isolation……………………………………………………………….20
Greenhouse experiment………………………………………………………………….20
vi
Planting design…………………………………………………………………..20
Rhizobium strain inoculation…………………………………………………….23
Trait measurement……………………………………………………………….24
Candidate gene sequencing………………………………………………………………25
Statistical analyses……………………………………………………………………….27
Quantitative genetic analysis…………………………………………………….27
Linkage map construction………………………………………………………..27
QTL analysis……………………………………………………………………..29
Results……………………………………………………………………………………………31
Survival and control plants………………………………………………………………31
Quantitative genetics……………………………………………………………………..32
QTL mapping…………………………………………………………………………….32
QTL by rhizobium strain interactions…………..………………………………………..40
Epistatic interactions……………………………………………………………………..40
QTL mapping across strain treatments….……………………………………………….41
Mapping of symbiotic signaling genes…………………………………………………..41
Branching orthologs in M. truncatula……………………………………………………45
Discussion……………………………………………………………………………………..…45
No evidence for a genetic basis of G × G interactions…………………………………..46
QTL architecture of phenotypic traits……………………………………………………48
vii
Epistatic interactions……………………………………………………………………..50
Colocalization of symbiotic signaling genes…………………………………………….53
Evidence for branching pattern orthologs………………………………………………..54
Conclusion……………………………………………………………………………….55
Chapter 3: Conclusions and future directions………………………………………………..57
Implications for G × G interactions……………………………………………………………...58
The QTL architecture of M. truncatula………………………………………………………….61
Genetic variation in Nod factor signaling genes…………………………………………………63
Future directions and final conclusions………………………………………………………….64
Literature cited…………………………………………………………………………………...68
Appendix…………………………………………………………………………………………81
viii
List of Tables
Table 1. Results of a mixed model ANOVA partitioning variation among RILs into genetic and
phenotypic variance……………………………………………………………………………...34
Table 2. QTL identified for plant traits collected on M. truncatula grown with the rhizobium
strain Naut………………………………………………………………………………………..36
Table 3. QTL identified for plant traits collected on M. truncatula grown with the rhizobium
strain Sals………………………………………………………………………………………...37
Table 4. QTL identified for plant traits collected on M. truncatula averaged over both rhizobium
strains…………………………………………………………………………………………….42
Table A-1. Information on the parental lines of the LR03 RIL mapping population……………83
Table A-2. Primer sequences used for sequencing Nod factor signaling genes in parental lines
and RILs………………………………………………………………………………………….84
Table A-3. PCR reagent concentration and volumes (in µL) used for sequencing Nod factor
signaling genes in parental lines and RILs………………………………………………………85
Table A-4. Correlations between RIL least-square means for all traits in the Naut and Sals
treatments………………………………………………………………………………………...86
Table A-5. Correlations between RIL least-square means for all traits in the across rhizobium
strains analysis…………………………………………………………………………………...87
Table A-6. Arabidopsis thaliana branching genes and description with the corresponding M.
truncatula orthologs……………………………………………………………………………...88
ix
List of Figures
Figure 1. Depiction of how negative frequency dependence can maintain genetic variation in
mutualisms………………………………………………………………………………………...8
Figure 2. Experimental set-up in greenhouse……………………………………………………22
Figure 3. Back-to-back histograms of trait least-square means of the M. truncatula RILs grown
with Naut and Sals……………………………………………………………………………….33
Figure 4. Genomic locations of significant QTL detected for the phenotypic traits of M.
truncatula when grown with rhizobium strains Naut and Sals…………………………………..38
Figure 5. Genomic locations of significant QTL detected for the phenotypic traits of M.
truncatula averaged across rhizobium strains……………………………………………………43
Figure 6. Epistatic QTL detected between markers EM2265 and EM17407 in Naut…………...44
Figure A-1. Preliminary leaf count data collected on the parental lines of the LR03 RIL mapping
population grown with rhizobium strains Naut and Sals………………………………………...81
Figure A-2. Linkage map of the LR03 RIL mapping population………………………………..82
x
List of Appendices
Appendix…………………………………………………………………………………………81
Chapter 1: General introduction 1
Chapter 1:
General Introduction
Interspecific interactions are one of the central processes influencing evolution and
adaptation. Thompson (1988) expressed that just as variation within populations is required for
the evolution of species, variation in the outcome of interactions is required for the evolution of
species interactions. The outcome of interspecific interactions may be dependent on the
environment and can also vary with the age, size, and genotype of the interacting individuals
(e.g., Moran, 1981; Price, 1987; Pyke, 1987). For example, the effects of herbivory on grasses
can vary from reduced growth and reproduction to fatal depending on the age of the plant (Pyke,
1987).
Mutualisms – reciprocally beneficial interactions between species – are ubiquitous
throughout all organismal kingdoms and in all ecosystems (Boucher, 1985; Smith & Douglas,
1987). Organisms use mutualisms to obtain resources or services which they cannot obtain by
other means or produce themselves (Bronstein, 2001), and this exchange of goods is favoured
when the benefits obtained through the mutualism are greater than the cost of reciprocating (see
Leigh, 2010 for a review). As in other interspecific interactions, the outcome of mutualisms can
vary depending on the abiotic and biotic conditions in which the interactions take place
(Bronstein, 1994). For example, plants eliminate their connections with mycorrhizal fungi when
phosphorous is added to the soil because the benefits they receive from the symbiosis are no
longer needed in the presence of a cost-free alternative (Johnson, 2010).
Chapter 1: General introduction 2
My thesis focuses on the effect of genetic variation on the outcome of biotic interactions,
which in turn can affect the outcome of mutualisms. Using the model legume Medicago
truncatula and its associated rhizobia, I attempted to locate specific regions of the genome
influencing interactions between different partner genotypes. I conducted a quantitative trait loci
mapping experiment with two different rhizobium strains to identify loci affecting the outcome
of biotic interactions and plant fitness traits. In this chapter, I provide necessary background
information on a number of topics which are relevant to my thesis including: the mechanisms of
mutualism stability, an introduction to genotype by genotype interactions, a brief description of
the signaling process involved in the initiation of nodulation between legumes and rhizobia, and
an overview of the process of quantitative trait loci mapping. I have also included a brief
description of my study organism and a summary of my experiment.
Mutualism stability and the maintenance of genetic variation
The evolutionary dilemma of mutualisms is that while an individual can benefit from
cooperating, it can achieve higher fitness by exploiting the cooperative efforts of others (Axelrod
& Hamilton, 1981; Soberon Mainero & Martinez del Rio, 1985; Bull & Rice, 1991). Given this,
theoretical models predict that mutualisms should be highly susceptible to invasion by cheaters,
individuals that exploit the host but fail to reciprocate benefits, ultimately leading to mutualism
breakdown (Trivers, 1971; Bull & Rice, 1991; Denison, 2000; Sachs et al., 2004). However,
mutualisms continue to persist and although cheating individuals are present in most mutualisms,
complete shifts to parasitism appear to be rare (Sachs & Wilcox, 2006; Sachs & Simms, 2006),
therefore stabilizing mechanisms must exist. In particular, three mechanisms have been
proposed: partner fidelity feedback, partner choice, and host sanctions.
Chapter 1: General introduction 3
In partner fidelity feedback, the two partners are associated for extended series of
exchanges, ultimately leading to coupling of their fitness due to repeated interactions (Trivers,
1971; Axelrod & Hamilton, 1981; Bull & Rice, 1991; Sachs et al., 2004; Foster & Wenseleers,
2006). Therefore, if a partner – for example a microbial mutualist – fails to cooperate or cheats, it
negatively affects its own fitness because its host plant’s fitness loss feeds back as a fitness loss
to the microbial partner (Bull & Rice, 1991; Sachs et al., 2004). Partner fidelity is expected to
occur in vertically transmitted symbionts or horizontally transmitted symbionts where the
dispersal abilities of the partners are limited (Sachs et al., 2004; Foster & Wenseleers, 2006).
An alternative mechanism is host control over uncooperative individuals via partner
choice or host sanctions (Bull & Rice, 1991; Eshel & Cavalli-Sforza, 1982; Denison, 2000;Sachs
et al., 2004; Foster & Wenseleers, 2006). The host plant can assess the cooperative ability of
several individuals of the microbial mutualist and only select or reward those which are the most
cooperative. In contrast to partner fidelity, partners do not need to interact continuously for
mutualism to be maintained. Furthermore, this mechanism can operate either at the initiation of
the symbiosis (―partner choice‖, Eshel & Cavalli-Sforza, 1982; Noë, 1990; Noë & Hammerstein,
1994; Bull & Rice, 1991, e.g., Simms & Taylor, 2002; Mueller, 2004; Heath & Tiffin, 2009;
Gubry-Rangin et al., 2010) or during the symbiosis via punishment of uncooperative partners –
i.e., host sanctions (Denison, 2000; e.g., Kiers et al., 2003; Jander & Herre, 2010). Both
mechanisms are expected to occur in mutualisms with horizontal transmission of symbionts and
where the host has multiple partners from which to choose (Bull & Rice, 1991).
A recent model by Weyl et al. (2010) concluded that many of the examples of
punishment or host sanctions are more likely examples of partner fidelity feedback. Using a
Chapter 1: General introduction 4
principle-agent (host-symbiont) model, they found that some of the described examples of host
sanctions (e.g., Kiers et al., 2003; Jander & Herre, 2010) are more likely general adaptations of
hosts to environmental damage – i.e., a response of partner fidelity feedback. For example, in the
mutualism between yucca plants and yucca moths, the host plant aborts flowers with an excess
of moth eggs and low pollen loads (Pellmyr & Huth, 1994). If host sanctions are occurring,
punishment should be greater when moth eggs are present, whereas partner fidelity feedback
predicts equal levels of floral abortion whether the damage is due to cheaters or environmental
factors. Empirical evidence indicates that the latter is occurring: mechanical damage of the
ovules is sufficient to cause floral abortion, and the experimental addition of moth eggs does not
influence the host plant response (Richter & Weis, 1995; Marr & Pellmyr, 2003). I think the
model and logic presented by Weyl et al. (2010) is sound, and suggests that partner fidelity
feedback may be more common than previously thought. I also agree with Weyl et al. (2010)
that host sanctions are more likely in mutualisms where the symbionts have greater cognitive
power, as in the mutualism between cleaning fish and client fish (Bshary, 2002; Bshary &
Grutter, 2005).
Given these mechanisms of removing uncooperative individuals, it is expected that only
highly mutualistic partners will remain, and in turn there will be minimal intraspecific genetic
variation in partner quality (May, 1976; Bull & Rice, 1991; Denison, 2000). However, empirical
evidence for genetic variation in partner quality has been found in mutualisms, including the
symbiosis between legumes and rhizobia. In this symbiosis, soil bacteria fix nitrogen into a
plant-usable form in exchange for plant photosynthates and shelter inside specialized organs
known as root nodules. Studies have reported variation in plant fitness and performance traits
Chapter 1: General introduction 5
such as seed weight and plant biomass when grown with different rhizobium strains (e.g.,
Mytton et al., 1977; Parker, 1995; Burdon et al., 1999; Mhadhbi et al., 2005; Thrall et al., 2011),
i.e., there is genetic variation for the benefits the plants receive from the symbiosis. Similarly,
variation among rhizobium strains in the number of nodules and ability to nodulate has been
found in response to different legume cultivars and species (e.g., Lieven-Antoniou & Whittam,
1997; Robinson et al., 2000; Mhadhbi et al., 2005; Abi-Ghanem et al., 2010).
There are at least three non-exclusive hypotheses to explain the existence of such genetic
variation in partner quality in mutualistic interactions, as summarized by Heath & Tiffin (2007):
1) Selection processes, such as host sanctions (Denison, 2000; Kiers et al., 2003) may not be
completely effective at removing less beneficial partners, allowing them to remain in the
population, 2) The best partner is dependent on the abiotic or biotic environment (e.g., West et
al., 2002; Piculell et al., 2008; Heath et al., 2010; Heath & Lau, 2011) therefore it is expected to
change with varying conditions and/or 3) The fitness both the host and symbiont are dependent
on the genotype of its interacting partner – i.e., genotype by genotype interactions (G × G)
(Heath & Tiffin, 2007; Heath, 2010).
These three hypotheses have varying levels of evidence in the literature. Kiers and
Denison (2008) provided several suggestions for the continued persistence of less-effective
partners in plant-rhizosphere interactions given the existence of host sanctions. One explanation
is that some species may have weaker sanctions, although empirical evidence for variation in
mutualist quality being explicitly maintained due to ineffective host sanctions is lacking. In terms
of the influence of biotic and abiotic conditions on the outcome of mutualisms, a number of
studies have found the fitness of both plant hosts and microbial mutualists can vary in response
Chapter 1: General introduction 6
to a range of factors such as soil community composition, herbivory, nutrient availability, and
light availability (e.g., Heath & Tiffin, 2007; Piculell et al., 2008; Ehinger et al., 2009; Davitt et
al., 2010; Heath et al., 2010; Heath & Lau, 2011). However, there appears to be the most
evidence for G × G interactions: it has repeatedly been found that the fitness benefits both
partners obtain from the mutualism can vary in response to both the plant host genotype and the
genotype of the interacting symbiont (e.g., Mytton et al., 1977; Parker, 1995; Hoeksema &
Thompson, 2007; Heath & Tiffin, 2007; Heath, 2010). Nevertheless, the influence of G × G
interactions and environmental conditions are not mutually exclusive: selection mosaics or
genotype × genotype × environment interactions (Thompson, 2005) – where selection on species
interactions varies with the abiotic and biotic environment – exist in mutualisms (e.g., Hoeksema
& Thompson, 2007; Piculell et al., 2008; Heath et al., 2010). However, a greater understanding
of the influence of G × G interactions on the maintenance of genetic diversity in mutualist
quality should first be resolved prior to investigating the effect of increasing levels of complexity
such as environmental and trophic interactions.
G × G interactions
G × G interactions, also known as intergenomic epistasis (Wade, 2007), are epistatic
interactions between the genomes of two interacting partners, where the effects of a gene in one
partner can vary depending on the genetic background of the other partner (e.g., Moran, 1981;
Service, 1984; Hoeksema & Thompson, 2007; Vale & Little, 2009; Heath, 2010). In mutualisms,
the phenotype of an individual can change depending on the genome of its interacting partner: a
symbiont may be a cooperator with one individual, but it may be suboptimal partner or even a
cheater with another individual.
Chapter 1: General introduction 7
The presence of such G × G interactions can maintain genetic variation in mutualist
quality by frequency dependent selection, where the fitness of both the host and symbiont are
dependent on genotypes and frequencies of their interacting partners in the population (Bever,
1999; Molofsky et al., 2001; Heath & Tiffin, 2007). Bever (1999) generated a model with two
hosts and two symbionts to show that negative frequency dependent selection can maintain
genetic variation in mutualist quality. For example, if a particular host plant obtains the highest
fitness with one microbial mutualist, however the same microbial mutualist has higher fitness
with a different host genotype, eventually the frequency of the host will feed back negatively on
its own fitness through its symbiont (Figure 1). As such, no one host genotype or symbiont
genotype can dominate the population, leading to genetic variation in partner quality. In
contrast, in positive frequency dependent selection, the optimal partner matches for the host and
symbiont, therefore the frequency of the host feeds back positively on its own fitness (Bever,
1999; Molofsky et al., 2001). In well-mixed systems, positive frequency dependence is predicted
to lead to loss of species diversity (May, 1974), however it can maintain genetic variation in
partner populations if the populations are spatially structured (Molofsky et al., 2001). When
positive frequency dependent interactions occur over small spatial scales – such as interactions
between plants and soil communities – stable clusters of coexisting species are created, which
can be further maintained with increasing physical complexity of the environment (Molofsky et
al., 2001).
In the legume Medicago truncatula and its nitrogen-fixing bacteria Sinorhizobium
melliloti G × G interactions have been previously found (Heath & Tiffin, 2007; Heath, 2010). G
× G interactions are evidence that genome level variation is influencing the outcome of legume-
Chapter 1: General introduction 8
Host plant Host plant
genotype A genotype B
Microbial mutualist Microbial mutualist
genotype X genotype Y
Figure 1. Depiction of how negative-frequency dependence can maintain genetic variation in
mutualisms (Modified from Figure 1 in Bever, 1999). The direction of the arrows represents the
direction of fitness effects, and the width of the arrows indicates the relative fitness benefits
obtained from the interaction, with wider arrows meaning greater fitness. Black arrows indicate
fitness effects and benefits for host plants, white arrows indicate fitness effects and benefits for
the microbial mutualists.
Chapter 1: General introduction 9
rhizobium interactions, however the exact location of the loci and genes involved remain
unknown. The next step is to locate the regions of the genome causing G × G interactions.
Although the genes involved in the legume-rhizobium symbiosis pathway have been well
characterized (see Kouchi et al., 2010 for a review), little is known about the genomic regions
influencing natural variation in the fitness benefits both partners obtain from the mutualism –
i.e., G × G interactions. It is important to locate the genetic bases of this symbiotic variation, as it
is this genetic variation that is under coevolutionary selection. Ultimately, locating these regions
is an essential step towards identifying the genetic mechanisms of coevolution. Furthermore,
once this information is obtained, future studies can determine whether particular genomic
regions have similar or different effects on the outcome of the mutualism when exposed to
varying biotic or abiotic conditions. A promising place to look for potential functional genetic
variation is in the plant genes involved in the initiation of the signaling pathway between
legumes and rhizobia.
Nodulation signaling pathway
The initiation and subsequent formation of nodules in the legume-rhizobium symbiosis
involves a complex molecular signaling dialogue between the two partners (see Jones et al.,
2007; Ferguson et al., 2010 for reviews). The host plant releases flavonoids and isoflavonoids
into the rhizosphere, which stimulate the expression of nodulation (nod) genes in the rhizobia.
Nod genes encode proteins involved in the synthesis and secretion of nodulation factors (Nod
factors), lipochito-oligosaccharides which induce morphological changes in the host, leading to
rhizobial infection of the plant root. Nod factor structure varies between rhizobia species and
also determines host specificity. Nod factors are recognized by the root hairs of the host, leading
Chapter 1: General introduction 10
to root hair curling, calcium spiking, growth of infection threads, and ultimately root nodule
formation. Once inside the nodule, rhizobia differentiate into nitrogen-fixing bacteroids.
In the past decade, a number of host legume genes involved in Nod factor perception,
symbiotic signaling, bacterial infection and nodule formation have been identified (see Kouchi et
al., 2010 for list). Among these genes, five of them are involved in the earliest stages of the Nod
factor recognition and signaling pathway in M. truncatula: NFP, DMI1, DMI3, NORK and NIN
(see Jones et al., 2010, Figure 2 for pathway diagram). NFP is a lysine motif (LysM)-receptor-
like kinase and is the candidate for the Nod factor receptor gene (Amor et al., 2003; Arrighi et
al., 2006); DMI1 encodes a calcium gated ion channel located on the nuclear membrane (Ané et
al., 2004); DMI3 encodes a calcium-calmodulin-dependent kinase that can control gene
expression (Lévy et al., 2004); NORK or DMI2 encodes a receptor kinase that is also involved in
rhizobial Nod factor perception (Endre et al., 2002); and NIN is a putative transcription factor
and is required for the formation of the plant infection threads (Schauser et al., 1999). Loss of
function mutations in of any of these five genes results in no rhizobial infection or formation of
nodules (Schauser et al., 1999; Catoira et al., 2000; Amor et al., 2003), illustrating the
importance of these genes to the nodulation pathway. Given that these genes are essential to the
formation of the symbiosis, they are also promising candidate genes to investigate whether
sequence level variation affects the outcome of G × G interactions.
QTL mapping
Quantitative trait loci (QTL) mapping involves the construction of a genetic map
followed by subsequent searching for a statistical association between a genetic marker and a
phenotypic trait (Liu, 1998). A significant association between the trait and the markers indicates
Chapter 1: General introduction 11
the presence of a QTL, i.e., a region of the genome that explains variation in the quantitative trait
of interest (Doerge, 2002).
The first step in QTL mapping is to construct a mapping population from homozygous,
inbred parental lines (Mauricio, 2001; Doerge, 2002). The parental lines do not have to be
different in the mean phenotypic value of the trait, however they must display allelic variation
(Mauricio, 2001). These parental lines are then mated to produce F1 lines, which are
heterozygous at all markers. From here, a number of different crossing designs are used
(backcross, F2, recombinant inbred lines) to develop the population which will be used for QTL
mapping. Once the mapping population has been created, each line must be genotyped for
neutral markers spread evenly across the genome, and these marker genotypes are then used to
assemble the genetic map.
A range of techniques are used to perform QTL mapping: single-marker mapping
(Weller, 1986; Beckmann & Soller, 1988), interval mapping (Lander & Botstein, 1989),
composite interval mapping (Zeng, 1993; Zeng, 1994; Jansen, 1993), multiple interval mapping
(Kao et al., 1999; Zeng et al., 1999) and multiple trait mapping (Jiang & Zeng, 1995). All of
these techniques test for associations between the marker genotypes and the measured
phenotypic traits, ultimately leading to the detection of QTLs.
Single-marker mapping involves performing a t-test or an ANOVA at each marker to
detect QTL (Weller, 1986; Beckmann & Soller, 1988), however it provides no information about
the location of the loci and suffers from the problem of multiple testing. Interval mapping solves
this problem by using an estimated genetic map, which is searched in increments to determine if
Chapter 1: General introduction 12
a QTL is likely to be found within intervals defined by ordered pairs of markers (Lander &
Botstein, 1989). Composite interval mapping utilizes the same approach, however it also controls
for the influence of genetic background by including additional markers outside the interval as
cofactors in the model (Zeng, 1993; Janesen, 1993; Zeng, 1994). Multiple interval mapping uses
multiple marker intervals simultaneously to fit multiple QTL into one model, and also detects
epistasis between QTLs (Kao et al., 1999; Zeng et al., 1999). Finally multiple-trait mapping
extends the composite interval mapping method to multiple correlated traits (Jiang & Zeng,
1995).
Recombinant inbred lines (RILs) are generated either by taking the F1 generation through
multiple generations of selfing or brother-sister matings (Lynch & Walsh, 1998). The RILs
which are created are genetically identical within a line, but have significant between-line
genetic variance, as each line is a mixture of the two parental genomes (Lynch & Walsh, 1998;
Broman, 2005). RILs offer a number of advantages for QTL mapping for several reasons: lines
only need to be genotyped once; environmental, individual and measurement variability can be
reduced by phenotyping multiple individuals per line; they offer greater mapping resolution due
to increased recombination events in comparison to other crossing designs; and they are
excellent for measuring QTL × environment interactions because the same lines can be used in
different environments (Lynch & Walsh, 1998; Broman, 2005). For these reasons, I used RILs
for my experimental work.
Study species
Medicago truncatula (Fabaceae), commonly known as the barrel medic, is a small,
annual legume native to the Mediterranean region. It has trifoliate leaves, small inflorescences
Chapter 1: General introduction 13
with one to five yellow flowers and is found mainly in open areas (Bataillon & Ronfort, 2006). It
has recently been split into three subspecies, largely based on the characteristics of the fruit pod:
ssp. truncatula, ssp. tricycla, and ssp. longeaculata (Bataillon & Ronfort, 2006). It is also grown
in rotation with cereal crops in some areas of Australia to improve soil conditions (Barker et al.,
1990; Young et al., 2005; Bataillon & Ronfort, 2006). Populations of this legume display
substantial phenotypic variation for a range of traits, including flowering time, symbiont
specificity and disease resistance (Cook, 1999). Medicago truncatula forms indeterminate
nodules with the nitrogen-fixing bacteria Sinorhizobium medicae and Sinorhizobium meliloti
(Rhizobiaceae) (Bailly et al., 2006).
Medicago truncatula has been developed as the model legume for studying the genetics
of plant-mycorrhizal and plant-rhizobium symbioses due to its short generation time, high selfing
rate (~97.5%, Bonnin et al., 1996) and small diploid genome (2n = 16; 550 Mb);(Cook, 1999;
Ané et al., 2008; Young et al., 2005). To date, its genome has been 84% sequenced and
assembled, there is a HapMap in progress, and there are mutant collections, mapping
populations, cDNA libraries, a gene expression atlas and a genome browser which are available.
It is a member of the cool season legumes (Galegoid clade), and has high levels of synteny with
closely related crops like pea (Pisum sativum)(Choi et al., 2004; Aubert et al., 2006), lentils
(Lens culinaris)(Phan et al., 2007), and alfalfa (Medicago sativa)(Julier et al., 2003; Choi et al.,
2004).
Experiment
To investigate the impact of genetic architecture on the outcome of G × G interactions
between Medicago truncatula and Sinorhizobium meliloti, I performed a large quantitative trait
Chapter 1: General introduction 14
loci (QTL) mapping experiment with two genotypically distinct rhizobium strains (Chapter 2).
The goal of the experiment was to locate regions of the Medicago genome that influence G × G
interactions and plant fitness. As a second component, I was also interested in determining if
genetic variation in plant nodulation genes affects phenotypic traits. Specifically, I wanted to
address the following questions: (1) Which regions of the M. truncatula genome contribute to
variation in plant fitness? (2) Do the number and effect size of additive and epistatic QTL differ
in two genotypic distinct rhizobia treatments, i.e., are QTL differentially expressed depending on
the rhizobium strain? and (3) Does sequence level variation in known symbiosis genes have an
effect on plant fitness and the ecological outcome of the symbiosis?
I found numerous QTLs that appear to affect plant fitness traits, but no evidence for G ×
G interactions or differences in QTL detection between the rhizobium strain treatments. Some of
the plant fitness QTLs co-localized with the nodulation signaling genes, suggesting that variation
in these genes may potentially influence phenotypic plant traits. The implications of these results
for G × G interactions are included in Chapter 2, and Chapter 3 reviews the main findings of my
thesis and suggests avenues for future work.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 15
Chapter 2:
Resolving the Genetic Basis of Genotype by Genotype Interactions
Between Plants and their Mutualists
Introduction
Mutualisms, or interspecific cooperation, are reciprocally beneficial interactions between
species (Bronstein, 2001) that affect many ecological and evolutionary processes. Organisms
from every kingdom participate in mutualisms, from mitochondria within cells, to mycorrhizal
fungi and plants, to cleaner fish and their clients (Boucher, 1985; Smith & Douglas, 1987;
Bronstein, 1994). Despite their prevalence, theoretical models predict that mutualisms should be
characterized by low levels of genetic variation: either from the fixation of the best mutualists
(May, 1976) or from the invasion of cheaters (individuals who benefit from the goods exchange
but do not reciprocate; Trivers, 1971; Axelrod & Hamilton, 1981; Bull & Rice, 1991; Sachs et
al., 2004), or from the suite of mechanisms that can stabilize mutualisms against cheating
(Trivers, 1971; Bull & Rice, 1991; Denison, 2000; Simms & Taylor, 2002). In contrast to these
predictions from a diverse set of theoretical models, a persistent empirical observation is that
substantial genetic variation in mutualist partner quality exists (e.g., Heath & Tiffin, 2007;
Hoeksema & Thompson, 2007). One explanation for this genetic variation is genotype by
genotype interactions, where the fitness of both partners is dependent on the genotype of its
interacting partner (Bever, 1999; Heath & Tiffin, 2007; Heath, 2010). Although these
interactions have the potential to explain variation in mutualism quality and outcome, the
genomic regions influencing these interactions remain unknown. Here I used quantitative trait
loci mapping to investigate genomic regions that influence the outcome of possible genotype by
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 16
genotype interactions between the legume Medicago truncatula and its associated rhizobia, and
whether variation in known mutualism signaling genes affects plant fitness traits.
Models predict that genes underlying mutualisms should have minimal variation in
partner quality – either due to selection for the best partners, the invasion of cheating individuals,
or mechanisms preventing the spread of cheaters – typically assume the fitness of each partner is
constant regardless of the genotype of its interacting host or symbiont (e.g., Trivers, 1971; Bull
& Rice, 1991). In contrast, a number of experimental studies have demonstrated the opposite: the
fitness of individuals involved in coevolutionary interactions can be dependent on the genotype
of both partners – i.e., genotype × genotype interactions (G × G) for fitness (Moran, 1981;
Service, 1984; Salvaudon et al., 2005; Hoeksema & Thompson, 2007; Heath, 2010). G × G
interactions, also referred to as intergenomic epistasis (Wade, 2007), occur when the effects of a
gene in one partner has variable effects depending on the genetic background of the other
partner. Applied to mutualisms, this means the phenotype of an individual can change depending
on the genome of its interacting partner: a cheater with one individual may be a cooperator with
another partner (Heath & Tiffin, 2007; Heath, 2010). Such fluctuating interactions could
generate a form of frequency dependent selection, where the fitness of both the host and
symbiont are dependent on genotypes and frequencies of their interacting partners in the
population (Bever, 1999; Molofsky et al., 2001). In turn, these interactions could help maintain
genetic variation in partner quality and provide an explanation for the stability of cooperation
between species (Bever, 1999; Heath & Tiffin, 2007) .
One system where G × G interactions have been found is in the model legume Medicago
truncatula and its nitrogen-fixing bacteria Sinorhizobium melilioti (Heath & Tiffin, 2007; Heath,
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 17
2010). In this symbiosis, the rhizobia fix atmospheric nitrogen into a plant-usable form in
exchange for shelter inside specialized root nodules and plant photosynthates. For example,
Heath (2010) found that some rhizobium strains were highly beneficial mutualists with certain
plant genotypes, but provided little or no benefit when matched with different plant genotypes.
Despite the variety of genetic resources available in M. truncatula (e.g., genome ~85%
sequenced and assembled, HapMap in progress, gene expression atlas, available RIL mapping
populations), the genetic basis of these G × G interactions remains almost totally unknown.
In addition to the genomic resources which have been developed, several genes involved
in the signaling pathway between rhizobia and legumes have also been identified (Schauser et
al., 1999; Catoira et al., 2000; Endre et al., 2002; Ane et al., 2004; Lévy et al., 2004; Arrighi et
al., 2006). Briefly, nodule formation and host specificity are controlled by mutual signaling
between the two partners. Legumes release flavanoids into the soil, which trigger rhizobia to
produce Nod factors. Nod factors are lipochito-oligosaccharide compounds which in turn induce
multiple downstream responses in the host, ultimately leading to the formation of root nodules
(see Jones et al., 2007 for a review). Recently, five genes involved in the earliest stages of the
Nod factor recognition and signaling pathway have been characterized in M. truncatula: DMI1,
DMI3, NFP, NORK and NIN (see Figure 1 in De Mita et al., 2007 and Figure 2 in Jones et al.,
2007 for diagrams of the signaling pathway). DMI1 encodes an ion channel located on the
nuclear membrane (Ané et al., 2004); DMI3 encodes a calcium-calmodulin-dependent kinase
that can control gene expression (Lévy et al., 2004); NFP is a lysine motif (LysM)-receptor-like
kinase and is the candidate for the Nod factor receptor gene (Arrighi et al., 2006); DMI2 or
NORK encodes a receptor kinase and is involved in rhizobial Nod factor perception (Endre et al.,
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 18
2002); and NIN is required for the formation of the plant infection threads and has homologies to
transcription factors (Schauser et al., 1999). Mutants with non-functional copies of any of these
genes cannot be infected by rhizobia and therefore produce no nodules (Schauser et al., 1999;
Catoira et al., 2000; Amor et al., 2003). Given that these genes are essential to the formation of
the symbiosis, they are promising candidates to investigate whether sequence level variation
affects the outcome of G × G interactions.
Quantitative trait loci (QTL) mapping is an excellent method for locating regions of the
genome affecting ecologically and evolutionarily important traits. Previous QTL mapping
studies in M. truncatula have focused on agronomically important traits such as resistance to
pathogens (Djébali et al., 2009; Hamon et al., 2010), seed mineral concentrations (Sankaran et
al., 2009), seed germination (Dias et al., 2011) and flowering time (Pierre et al., 2008).
However, thus far, no one has examined QTL architecture of plant fitness traits in response to
two different rhizobium genotypes (mapping natural variation in G × G interactions), or the
contribution of known signaling genes to variation in phenotypic traits.
Using a recombinant inbred line (RIL) mapping population of M. truncatula, I examined
the QTL architecture of a range of plant fitness traits when grown with two genotypically distinct
S. meliloti strains. Specifically, I wanted to address the following questions: (1) Which regions of
the M. truncatula genome contribute to variation in plant fitness? (2) Do the number and effect
size of additive and epistatic QTL differ in two genotypic distinct rhizobia treatments, i.e., are
QTL differentially expressed depending on the rhizobium strain? and (3) Does sequence level
variation in known symbiosis genes have an effect on the ecological outcome of the symbiosis?
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 19
Materials and methods
Study species and mapping population
Medicago truncatula (Fabaceae), commonly known as the barrel medic, is a small,
annual legume native to the Mediterranean region. It is found in symbiosis with the nitrogen-
fixing bacteria Sinorhizobium medicae and S. meliloti (Rhizobiaceae) (Bailly et al., 2006). I
tested for G × G interations with S. meliloti. Medicago truncatula has trifoliate leaves, small
inflorescences with one to five yellow flowers and is found mainly in open areas (Bataillon &
Ronfort, 2006). It has been developed as the model legume for studying the genetics of plant-
mycorrhizal and plant-rhizobium symbioses due to its short generation time, high selfing rate
(~97.5%, Bonnin et al., 1996) and small diploid genome (2n = 16; 550 Mb);(Cook, 1999; Ané et
al., 2008; Young et al., 2005). Furthermore M. truncatula belongs to the cool season legumes
(Galegoid clade), and is closely related to important crops like pea (Pisum sativum), lentils (Lens
culinaris) and alfalfa (Medicago sativa)(Young & Udvardi, 2009).
Seeds from the M. truncatula recombinant inbred line (RIL) LR03 mapping population
were provided by L’Insitut National de Reserche Agronomique (INRA) in Montpellier, France.
The lines were created by manually crossing two M. truncatula inbred line accessions,
F803005.5 (530, female parent, origin: France) and DZA045.5 (735, male parent, origin:
Algeria) to produce an F1 generation, which was later selfed. The parental genotypes have been
previously found to be divergent in a number of traits (Appendix, Table A-1), and preliminary
results indicated that they exhibit evidence of G × G interactions with the two focal rhizobium
strains used in this experiment (ANOVA for leaf number, rhizobium strain × parental line, p =
0.02. See Appendix, Figure A-1). The RILs (n=177) were derived from the F2 generation by self-
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 20
fertilization and single seed descent for five generations, creating the F6 RIL mapping population
that I used.
Rhizobium strain isolation
Both S. meliloti strains used in this experiment (Sals b and Naut a, hereafter referred to as
Naut and Sals) were previously isolated and genotyped from soil samples collected on the
Mediterranean coast of southern France by Heath (Heath, 2010); full details are given there, and
only a brief summary is provided here. Strain Sals was isolated from the soil of the wild M.
truncatula Population Salses II, which is located near the town of Salses-le-Chateau and strain
Naut was isolated from Population Nautique, which is located just east of the town of La
Nautique (rhizobium strain locations approximately ~50 km apart). Wild M. truncatula seeds
taken from the two populations were grown in the greenhouse and inoculated with a mixture of
soil taken from that area and sterile water. After 2 months of growth, nodules from each plant
were crushed and streaked on plates with modified arabinose-gluconate (MAG) media. Once
rhizobia colonies were visble, one isolate from each plate was randomly selected and grown in 2
mL liquid MAG media for 2 days at 30°C. To ensure that the strains were not simply clonal
replicates, the strains were genotyped using rep-PCR.
Greenhouse experiment
Planting design
I grew 8 replicates of each RIL with the two rhizobium strain treatments. To minimize
contamination, the following design was used. One seedling per line was planted into Ray-Leach
SC10 UV-stabilized cone-tainers (Stuewe and Sons Inc, Tangent, OR); I then placed four racks
of cone-tainers into flow-trays (large plastic ―bins‖ that provide sub-irrigation) each of which
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 21
received its own rhizobium strain inoculation. Within each bin, there were 177 genotyped RILs,
4 replicates of each of the parental lines, and an additional 10 RILs that lacked genetic markers
(N=195 plants per bin). I used 8 replicate bins per rhizobium treatment for a total of 16 bins,
arranged in the greenhouse in a checkboard array of 8 blocks (one bin of each strain per block) to
eliminate any potential relationships between rhizobium treatment and spatial position. Within
rhizobium treatments, the design is a randomized blocked design; for the entire experiment, the
rhizobium treatment (which was applied to bin of 195 plants at a time) is a plot-level effect. An
additional 49 plants were not inoculated and were included as controls to test for contamination
within the greenhouse (see Figure 2 for greenhouse set-up).
All seeds were scarified with a razor blade then surface sterilized for 30 seconds in 95%
ethanol, followed by 7 minutes in full bleach. Following sterilization, seeds were imbibed in
ddH20 for 20 mins and plated on 0.08% agar. The plates were sealed with parafilm, wrapped in
aluminum foil and stratified at 4°C for 2 weeks. After 2 weeks, the seedlings were transplanted
into 164 mL cone-tainers containing a mixture of 1:1 Sunshine Mix 2 (Sun Gro Horticulture,
Bellvue, WA) and Turface (Profile Products, LLC Buffalo Grove, IL). I used Sunshine Mix 2
because it contains no added fertilizer, as it has been previously shown that nitrogen can affect
the outcome of the symbiosis (Heath & Tiffin, 2007; Heath et al., 2010). Prior to transplanting,
the soil mixture was steam sterilized at 121°C for 30 mins in small bags to kill any contaminants.
I planted one seed per line in a randomized design within each bin and conetainers were placed
in every other space in each 98 cell rack. Planting was completed on a bin by bin basis from Jan
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 22
Figure 2. Experimental set-up in the greenhouse. The numbers 1-16 represent the bins. The colour of the bin indicates the rhizobium
strain it was inoculated with (grey = Naut; white= Sals). The control bin in the bottom left corner was not inoculated with any strain in
order to assess greenhouse level contamination.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 23
1-6 2010. Seeds that failed to establish were replaced with a second set of plants (when seeds for
that line were available) using the same methods described above on Jan 29-31 2010.
For two weeks following transplanting, I misted all seedlings twice daily with water to
prevent desiccation. Once plants were sufficiently large to have established roots, I bottom
watered the plants every second day. I fertilized the plants only once with N-free Fahreus
solution (Somasegaran & Hoben, 1994) and grew the plants under 16 hour days at an average
greenhouse temperature of 22°C.
Rhizobium strain inoculation
Three days before planting, I streaked agar petri plates containing tryptone yeast (TY)
media with thawed rhizobia culture stocks that were previously stored at -20°C. I incubated
plates at 30°C, and allowed the rhizobia cultures to grow for one week. Afterwards, I selected an
isolate from a single bacteria colony for each strain and inoculated sterile liquid TY growth
media. The liquid cultures were grown for three days at 30°C and were mixed continuously at a
rate of 200 rpm.
One week after planting, I inoculated all plants with the rhizobia cultures. Prior to
inoculation, I diluted both cultures to a density of 0.1 OD600. I inoculated each plant with 1 mL
of either Sals or Naut, which was determined based on bin number. The following day, I poured
25 mL of sterile water on the soil of each cone-tainer to help distribute the rhizobia cells
throughout the soil.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 24
Trait measurement
I recorded a total of eight different traits on each plant as measures of plant growth, or
size, or fitness: leaf counts at 2 weeks, leaf counts at 6 weeks, date of first flower, number of
fruits, fruit weight, dry shoot weight, dry root weight and number of primary branches. Fruit
weight is positively correlated with seed number (r = 0.75, p < 0.0001, N = 100; Gorton, unpub.
data). For leaf counts, I counted the number of true leaves present on each plant at two and six
weeks. Early leaf count could be an important trait in the wild where seedlings are in competition
with their neighbours for access to light (Heath & Tiffin, 2007), and I used the second leaf count
as a measure of plant growth. I checked all plants daily for signs of flowering. Once the plants
began to set fruit, I collected mature fruit daily and kept them in separate coin envelopes for each
plant. The fruit fall off the plant easily once they are mature (personal observation), thus
occasionally a few fruit were missed. To estimate this loss, I collected all loose fruits in the
greenhouse (from the floor, drains, greenhouse benches, etc) to determine the percentage of fruits
successfully collected, which corresponded to 95.5% (Total number of fruits collected = 25,001;
total number of fruits lost = 1122).
The plants were harvested in sequential order based on planting date approximately 15
weeks after the first day of planting. Prior to harvesting, all fruits were removed for later
counting. At harvest, I separated above-ground and below-ground biomass. Once plants were
harvested, I weighed all the fruits per plant, and then calculated a new variable, ―average fruit
weight‖. Both the above-ground and below-ground biomass were dried at 55°C for 3 days prior
to weighing them. After harvesting, I counted the number of primary branches on each plant
based on the description in Moreau et al.(2006).
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 25
Unfortunately, it was not possible to record any traits for rhizobium fitness, such as
nodule counts, weight or branching, on the majority of the plants. Once I harvested all of the
plants, it was apparent that the roots were very dense and contained numerous nodules, which
would be impossible to process in a timely manner. Before eliminating collection of nodule data
altogether, I counted all nodules on every parental plant that was included in the experiment. I
also randomly selected a subset of 15-20 nodules from each parental plant and placed them in 1.5
mL tubes with Drierite to dry out the nodules for weighing. The analysis of the parental lines
indicated that there was no effect of rhizobium strain on nodule number (p = 0.53) or weight (p =
0.59), nor was there a strain × line interaction for either trait (p = 0.32, p = 0.31), therefore I
abandoned any further collection of rhizobia traits on the remaining plants.
Candidate gene sequencing
The parental lines were initially sequenced from polymerase chain reaction (PCR)
products for five genes involved in the Nod factor signaling pathway: DMI1, DMI3, NIN, NFP
and NORK. A separate set of F6 progeny and parental lines were grown exclusively for DNA
purposes. DNA was extracted from 100 mg of frozen leaf tissue according to DNAeasy Plant
Mini Kits (Qiagen), with one modification to the protocol: the final DNA elution used Tris HCl
in the place of Buffer AE and included a 30-min hot elution at 65°C. The PCR primer sequences
for DMI1 were taken from De Mita et al. (2007), and the remaining primer pairs were designed
from M. truncatula genomic DNA sequences for each gene, which are publicly available from
the National Centre for Biotechnology Information (www.ncbi.nlm.nih.gov). The primers were
designed using Integrated DNA Technologies (IDT) PrimerQuestSM
(www.idtdna.com/Scitools/Applications/Primerquest). Once a primer set was chosen, I used
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 26
NCBI’s BLASTn to ensure they only matched the gene of interest and would not amplify other
parts of the M. truncatula genome (See Appendix, Table A-2 for primer sequences).
The PCR reaction and protocol was optimized for each gene (modified from the
suggested protocol provided by Froggabio) and was performed using Froggabio I-TaqTM
DNA
polymerase and associated ingredients (see Appendix, Table A-3 for PCR mix details). All PCR
products were checked prior to sequencing using 1% agarose gel electrophoresis. I cleaned PCR
products by adding 0.2 µL Exonuclease I (Fermentas) and 2 µL Shrimp Alkaline Phosphatase
(New England BioLabs) to each sample. Samples were incubated at 37°C for 30 min then heated
to 80°C for 15 min. I sequenced the products with BigDye Terminator cycle sequencing kit
(v.3.1, Applied Biosystems) using the forward PCR primers, which were then run on an ABI
3730 DNA Analyzer by the Centre for the Analysis of Genome Evolution and Function
(University of Toronto). All RILs and parental lines were sequenced twice and I aligned the
DNA sequences for each gene using ClustalW in BioEdit (Hall, 1999).
Only DMI1, DMI3 and NFP contained one or more SNPs that differed between the
parents, therefore they were chosen to be fully sequenced in the RILs to develop markers. In
DMI1 (5815 bp), a fifth of the gene was successfully sequenced, and the fragment contained
three polymorphisms: one was a nonsynonymous change, resulting in the substitution of
isoleucine for threonine, and two were in introns. The entire NFP gene (1788 bp) was sequenced
and three polymorphisms were found, two of which were synonymous changes and one was a
nonsynonymous change, resulting in the substitution of tryptophan for serine in the protein
sequence. In DMI3 (6700 bp), a fifth of the gene was successfully sequenced, and this segment
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 27
contained five polymorphisms within introns. One SNP per gene was chosen to be scored across
all RILs and used as a molecular marker for QTL mapping.
Statistical analyses
Quantitative genetic analysis
I tested for genetic variation in plant traits in separate analyses for each rhizobium strain
treatment by using a mixed model ANOVA with restricted maximum likelihood (PROC
MIXED, SAS v. 9.2). For each trait as a response variable, I included block and planting set as
fixed effects and line as a random effect. The significance of the line effect was tested by running
the models with and without the line effect; the difference in the -2 log likelihoods of the models
was compared to chi-square distribution with 1 degree of freedom. To formally test for line ×
rhizobium genotype interactions, I also used mixed model ANOVA. For each trait as a response
variable, I included block, planting set, and rhizobium genotype as fixed effects, and line and
line × rhizobium genotype as random effects. As described above, I tested the significance of the
random effects with likelihood ratio tests by comparing the differences in -2 log likelihoods to a
chi-square distribution.
Linkage map construction
The RILs were genotyped at 204 markers, of which the majority are simple sequence
repeats (SSRs) and amplified fragment length polymorphisms (AFLPs). 201 of the marker
genotypes were provided by Marie-Laure Pilet and Alain Baranger from INRA. Some of the
markers have previously been published in Djébali et al. (2009) and Mun et al. (2006), and a
description of the remaining markers is in progress (Avia et al. in prep). A subset of the markers
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 28
from Mun et al. (2006) were designed from M. truncatula bacterial artificial chromosomes
(BAC) accessions, anchored to the integrated genetic-physical map
(http://www.medicago.org/genhome/map.php, University of Minnesota 2006). These markers are
indicated in bold on the linkage map. The remaining three markers (DMI1441, NFP1697 and
DMI3427) are single nucleotide polymorphisms which were designed from sequence data from
the symbiotic signaling pathway genes DMI1, NFP, and DMI3 using the method described
above.
I used JoinMap 4.0 (Van Ooijen, 2006) to determine the linkage map of the population
(N= 204 markers). Although I was provided with an ordered linkage map including 201 of the
markers (Avia et al. in prep, see Hamon et al., 2010 for partial map), I reordered all of them for
this experiment because I added the extra markers described above, and I did not have
phenotypic data for one of the RILs included in the original map ordering. I determined the
linkage groups using a LOD score between 4 and 8, and the marker order and distances between
markers were calculated using maximum likelihood mapping. All of the markers localized to the
same linkage groups where they were previously mapped, however in some groups the order of
the markers changed. Furthermore, for some of the linkage groups multiple marker orders were
generated. In these cases, I investigated the raw genotype data for evidence of improbable double
recombinants or transcript reading errors. I excluded a small number of problematic markers
from the analysis if they were located in regions of reasonable marker density (i.e., ~ 8-10 cM
between markers), and if removing them did not create a large gap in the linkage group.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 29
QTL analysis
Although I found no evidence of significant line × rhizobium strain interactions for any
of the phenotypic traits described (see Results below), I still performed QTL mapping separately
for each strain treatment. The explanation for this approach is conditional neutrality: strain
specific QTL may exist, however they may exhibit conditional neutrality (sensu Mackay 2001)
where they affect phenotypes in one strain treatment, but have no effect in the other. If the QTL
which display conditional neutrality differ between the two rhizobium strains, it is possible for
different QTL to affect traits in the different strain treatments, even in the absence of a
significant line × rhizobium strain treatment interaction. To evaluate this possibility, I performed
QTL mapping in each rhizobium strain treatment separately.
For the QTL mapping, I estimated the least-square means of each plant trait using a
mixed model ANOVA (PROC MIXED, SAS v 9.2), which included block, plant set, and line as
independent variables. I did these estimations and all QTL mapping analyses described below
separately for each rhizobium strain treatment. I chose not to use BLUPs (best linear unbiased
predictors) or the random effects solutions of the mixed models described above because BLUPs
can have poor properties when used as data in regression-based analyses (Hadfield et al., 2010).
QTLs were detected using the composite interval mapping procedure (CIM; Zeng, 1993;
Jansen, 1993; Zeng, 1994) in QTL Cartographer v. 2.5 (Wang et al., 2011). The CIM procedure
tests the hypothesis that a QTL is located within an interval defined by two markers, while
accounting for genetic background (i.e., possible linked QTLs) by including additional markers
outside the interval as cofactors. The cofactors used in each CIM were selected using forward-
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 30
backward stepwise regression (p = 0.05) under the standard model (Model 6). All QTL analyses
were performed on the calculated least-square means of each RIL and tests were performed at 2
cM intervals, with a window size of 10 cM. For each trait, the genome-wide threshold values
(likelihood ratio test statistic (LR); p = 0.05) for declaring a significant QTL were determined
through 1000 permutations of the data set (Churchill & Doerge, 1994). For significant QTL, I
calculated 2-LOD support intervals as the nearest markers on either side of the QTL peak where
the LR dropped by 9.22 (Van Ooijen, 1992). All significant QTL were drawn on the linkage map
using MapChart (Voorrips, 2002).
To determine if QTLs were differentially expressed depending on the rhizobium
genotype (i.e., QTL × strain treatment or QTL × E), I performed a multi-way ANOVA for each
trait (PROC GLM, SAS v. 9.2). I selected the markers closest to the peak of each significant
QTL in the two strain treatments, and I included all of these markers as well as two-way marker
× rhizobium strain interactions as main effects in the model. A formal statistical test is needed
because it is possible for a QTL to have a significant phenotypic effect in one environment, but
have little or no effect in the other environment and thus remain undetected due to low statistical
power.
I also tested for epistatic interactions between markers (epistatic QTLs) for each trait in
both strain treatments using Epistacy (Holland, 1998). This program tests for the effect of all
possible pairwise combinations of the markers on each trait using SAS (PROC GLM). To
account for the problem of multiple testing and determine which results were truly significant, I
used a p-value threshold of 0.99 in Epistacy and implemented a false discovery approach using
QVALUE (Storey & Tibshirani, 2003) with the false discovery rate set at 0.05. The program
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 31
QVALUE assigns a q-value to each p-value, where the q-value indicates the expected percentage
that the result is a false discovery, given that test is interpreted as statistically significant. For
significant epistatic QTL, I also tested for QTL×QTL×E by including the two significant
markers, the two-way marker × marker interaction, and the three-way marker × marker ×
rhizobial strain interaction.
In addition to the strain-specific QTL mapping, I also mapped QTL for each plant trait on
the line least-square means estimated from a mixed model ANOVA that included both strain
treatments (i.e., QTL mapping on phenotypes averaged across both strain treatments). This
model included block, strain, plant set and line as fixed effects, and line × strain, strain × block
and line × strain × block as the random effects. The last two random effects were included
because the rhizobia strain treatments were applied half a block at a time, rather than to
individual plants.
Results
Survival and control plants
In total 2078 plants survived, with a mean and median of five plants per line in each
strain treatment. According to Keurentjes et al. (2007), in Arabidopsis thaliana, QTL detection
levels off at 4 replicates per line with 161 RILs. Therefore despite mortality, there was still
sufficient replication to perform QTL mapping in this experiment. Out of the 50 uninoculated
control plants, 30 survived and only one plant nodulated (indicative of contamination). The
remaining plants were small, pale green, did not flower or set seed, and had no nodules,
indicating that greenhouse level contamination was low.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 32
Quantitative genetics
I detected significant genetic variation in all phenotypic traits in both rhizobium strain
treatments, with the exception of leaf number at two weeks in the Naut treatment (Table 1). I did
not find any significant line × rhizobium strain interaction (p = 0.5 for all traits), and a
comparison of the RIL trait least-square means when grown with Naut and Sals (Figure 2)
indicated that their distributions were very similar. All of the phenotypic traits displayed
evidence of transgressive segregation; i.e., the range of trait values of the RILs exceeded that of
the parental lines (Figure 3).
Broad sense heritability was in general always higher when estimated from RILs grown
in the Sals treatment (mean H2
= 0.29, range = 0.08-0.59), in comparison to RILs grown in the
Naut treatment (mean H2
= 0.19, range = 0.01-0.33) (Table 1). However, for the majority of the
phenotypic traits measured, the coefficient of genetic variation did not differ substantially
between the two rhizobium treatments (Table 1), suggesting that the differences in heritability
were due to differences in phenotypic variation or environmental effects. For example, the broad
sense heritability of average fruit weight was 0.14 higher in Sals, however the CVg is barely
larger; this can be attributable to the lower phenotypic variance in the Sals treatment. One
exception was shoot weight: both heritability and the coefficient of genetic variation were
substantially higher in the Sals treatment (Table 1).
QTL mapping
The final linkage map used for the QTL mapping analyses consisted of 184 markers
spanning over eight linkage groups and covering 1,215 cM (average of 6.6 cM/marker; for
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 33
Figure 3. Back-to-back histograms of trait least-square means of the M. truncatula RILs grown with rhizobium strains Naut (grey bars)
and Sals (white bars). Parental trait means are indicated by arrows (P1= female parent, line 530, P2= male parent, line 735).
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 34
Table 1. Mixed model ANOVA results partitioning variation among RILs into genetic (Vg) and phenotypic (Vp) variance. Broad sense
heritability (H2) and the coefficient of genetic variation (CVg) are also included.
Vg Vp H2 CVg Line significance
Trait Naut Sals Naut Sals Naut Sals Naut Sals Naut Sals
Leaf number – 2 weeks 0.006 0.05 0.66 0.68 0.01 0.08 4.36 13.10 0.75 <0.001
Leaf number – 6 weeks 8.21 10.52 51.31 55.43 0.10 0.19 15.82 17.84 <0.001 <0.001
Days to flowering 14.23 23.52 91.44 118.42 0.15 0.20 9.69 12.15 <0.001 <0.001
Fruit number 8.37 12.04 31.63 41.38 0.26 0.29 22.36 26.76 <0.001 <0.001
Average fruit weight 0.0003 0.0003 0.001 0.0008 0.33 0.47 19.01 19.25 <0.001 <0.001
Shoot weight 0.041 0.10 0.13 0.17 0.31 0.59 39.44 58.39 <0.001 <0.001
Root weight 0.014 0.02 0.08 0.06 0.17 0.30 22.23 25.78 <0.001 <0.001
Primary branch number 0.081 0.09 0.44 0.43 0.18 0.21 20.77 23.11 <0.001 <0.001
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 35
complete marker orders with genetic distances see Appendix, Figure A-2). All the significant
QTLs detected in the Sals and Naut rhizobium treatments are listed in Table 2 and 3. Due to the
lack of among-line variance for leaf number at two weeks (Table 1), I did not perform QTL
mapping on this trait. QTLs which had overlapping 2-LOD support intervals were considered to
be the same QTL, as indicated by the QTL number column in Table 2 and 3. In total, I identified
8 separate QTLs in the Naut rhizobium treatment, explaining between 3.5 and 10% of the total
genetic variation, depending on the trait examined (Table 2, Figure 4A). Three of these QTLs
affected multiple phenotypic traits: NQTL1-2 had a significant effect on leaf number, number of
fruits and shoot weight; NQTL3-2 had a significant effect on leaf number and number of fruits;
and NQTL8 had a significant effect on leaf number and average fruit weight. The direction of
NQTL3-2 was opposite for number of fruits and leaf number: the 530 allele increased the former
but decreased the latter (Table 2). All the other QTLs which affected multiple traits in Naut had
effects in the same direction. No QTLs were identified for days to flowering or root weight in
this treatment.
In the Sals rhizobium treatment, fewer QTLs were identified, with only 4 unique QTLs,
two of which affected multiple phenotypic traits: SQTL1-1 influenced leaf number, number of
fruit and root weight, and SQTL1-2 influenced number of fruit and shoot weight (Table 3, Figure
4B). These QTLs were of small effect, explaining between 6.1 and 12.7% of the total genetic
variation. In this treatment, no QTL were detected for days to flowering or primary branch
number.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 36
Table 2. QTL identified for plant traits collected on M. truncatula grown with the Naut rhizobium strain. The significant QTL for each
trait are listed along with the chromosome, position, marker directly below the QTL peak, LOD score, 2-LOD support interval,
percent variance explained by each QTL (R2), and additive effect (a0, positive values indicate that 530 alleles increase trait means).
QTL with overlapping intervals share the same QTL number and indicate those which affected multiple traits.
QTL
Number
Chr
Position
(cM)
Marker
LOD
2-LOD
Interval (cM)
R2
a0 Trait
Leaf number - 6 weeks NQTL1-1 1 30.4 MTIC448 3.1 19.9- 43.8 5.8 1.114
NQTL1-2 1 113.7 MTIC064 4.9 91.5-119.7 10 1.375
NQTL3-2 3 151.8 MTB122 3.4 141.4-159.8 8.4 -1.362
NQTL8 8 52 DMI3427 3.7 39.9-74.9 7.4 1.197
Number of fruits NQTL1-2 1 94.3 EM4.333 3.3 75.3-103.6 7.2 1.052
NQTL3-2 3 158.3 MTIC044 4.1 141.8-159.8 8.8 1.268
Average fruit weight NQTL3-1 3 130.2 MTB6 5 124.6-134.8 8.9 -0.007
NQTL5 5 28.1 MTIC148 4.3 12-43.4 7.7 -0.007
NQTL8 8 84.9 MTB333 2.8 74.5-109.9 6.6 0.006
Shoot weight NQTL1-2 1 100.2 MTIC146 2.9 87.5-111.7 6.2 0.065
NQTL4-1 4 42.6 GO3.350 2.9 30.6-54.1 6.3 -0.065
Primary branch number NQTL4-2 4 122.3 MTIC186 4.8 112.1-133.4 3.5 0.048
Note: QTL were named as follows: rhizobium strain prefix (N = Naut), and the chromosome number as a suffix, with
an additional number depending if there were multiple QTL per chromosome.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 37
Table 3. QTL identified for plant traits collected on M. truncatula grown with the Sals rhizobium strain. The significant QTLs for each
trait are listed along with the chromosome, position, marker directly below the QTL peak, LOD score, 2-LOD support interval,
percent variance explained by each QTL (R2), and additive effect (a0, positive values indicate that 530 alleles increase trait means).
QTL with overlapping intervals share the same QTL number and indicate those which affected multiple traits.
QTL
Number
Chr
Position
(cM)
Marker
LOD
2-LOD
Interval(cM)
R2
a0 Trait
Leaf number - 6 weeks SQTL1-1 1 30.4 MTIC448 2.9 17-43.6 6.1 1.302
Number of fruits SQTL1-1 1 26.3 MTB269 5.7 17.9-32.4 12.7 1.786
SQTL1-2 1 110.1 MTIC285 5 103.6-117.7 10.2 1.566
SQTL3 3 155.9 MTIC371 3.4 141.8-159.8 6.7 1.353
Average fruit weight SQTL5 5 12.2 MTIC078 4 2-28.1 7 -0.01
Shoot weight SQTL1-2 1 111.7 MTIC064 4.8 103.6-119.7 10.9 0.147
Root weight SQTL1-1 1 24.3 MTB269 3.1 13-34.4 6.9 0.049
Note: QTL were named as follows: rhizobium strain prefix (S = Sals), and the chromosome number as a suffix,
with an additional number depending if there were multiple QTL per chromosome.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 38
A)
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 39
B)
Figure 4. Genomic locations of significant QTL detected for the phenotypic traits of M.
truncatula when grown with rhizobium strains A) Naut and B) Sals. Chromosome number is
listed across the top of the linkage groups. The scale on the left indicates the genetic distance
between markers in centimorgans (cM). Each horizontal line represents the position of one
genetic marker. Phenotypic trait names are listed beside the estimated QTL positions, and the
length of the QTL represents the 2-LOD support interval. The vertical line in each QTL indicates
where the highest peak was located.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 40
QTL by rhizobium strain interactions
A superficial comparison of the QTLs found in the two rhizobium strain treatments
indicated that four QTLs influenced the same traits in both treatments: NQTL1-1 and SQTL1-1
for leaf number, NQTL3-2 and SQTL3 for number of fruits, NQTL5 and SQTL5 for average
fruit weight, and NQTL1-2 and SQTL1-2 for shoot weight (Table 2, 3). However some of these
common QTLs differed in the number of phenotypic traits they affected: for example, NQTL1-1
only affected leaf number in Naut, but the same QTL influenced leaf number, fruit number, and
root weight in Sals. Furthermore, QTL for basal branch number were found only in Naut, and
QTL for root weight were found only in Sals. Despite these apparent differences in QTL between
the two strain treatments, no significant QTL×E (or QTL × rhizobium strain) interactions were
detected for any of the measured phenotypic traits (p = 0.23-0.97)
Epistatic interactions
In the genome-wide scan of epistatic interactions between markers, I detected one
significant epistatic QTL in Naut which influenced root weight. It was found on chromosome 6
between markers EM17.407 and EM2.265, and had a partial-r2 of 13% (Figure 6). These markers
are located 8.9 cM apart (see Appendix, Figure A-2 for genetic distances).The percentage of
each allelic class were as follows (where P1 = parental allele 530 and P2 = parental allele 735):
EM17.407-P1 × EM2.265-P1 = 36%, EM17.407-P1 × EM2.265-P2 = 6%, EM17.407-P2 ×
EM2.265-P2 = 53%, EM17.407-P2 × EM2.265-P1 = 5%. Neither marker was found within the
support intervals of the detected additive QTL, nor was there evidence of QTL×QTL×E (p =
0.38).
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 41
QTL mapping across strain treatments
In the QTL analyses performed using data averaged across the strain treatments, I found
10 main-effect QTLs, 4 of which affected multiple traits (Table 4, Figure 5). These QTLs
explained between 5 and 13.5% of the total genetic variation. No QTL were detected for days to
flowering. Many of the QTLs detected in this analysis were previously found in Naut and Sals
QTL mapping, however several new QTLs were found: QTL4-1 and QTL7 significantly affected
number of fruits, QTL3-1 affected average fruit weight, QTL5-2 affected plant weight and QTL6
influenced basal branch number.
Mapping of symbiotic signaling genes
The markers DMI1441, NFP1697 and DMI3427 corresponding to the symbiotic pathway
signaling genes DMI1, NFP and DMI3, respectively, all localized to where they are found on the
M. truncatula genome (http://www.medicagohapmap.org). DMI1 mapped at 17.6 cM on
chromosome 2, NFP mapped at 20.5 cM on chromosome 5, and DMI3 mapped at 46 cM on
chromosome 8 (See Appendix, Figure A-2 for positions).
A few of the QTLs colocalized with the symbiotic signaling genes DMI3 and NFP. In
Naut, DMI3427 was located directly below the highest point of one the QTL which affected leaf
number (NQTL8, Table 1) and NFP1660 was located within the interval of a QTL which
influenced average fruit weight (NQTL5, Table 1). The latter QTL was also detected in Sals (for
average fruit weight) and in the across strain analysis (average fruit weight and number of fruit).
The marker DMI1441 was not found to be significantly associated with any QTL.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 42
Table 4. QTL identified for plant traits collected on M. truncatula average over both rhizobium strains. The significant QTLs for each trait
are listed along with the chromosome, position, marker directly below the QTL peak, LOD score, 2-LOD support interval, percent
variance explained by each QTL (R2), and additive effect (a0, positive values indicate that 530 alleles increase trait means). QTL with
overlapping intervals share the same QTL number and indicate those which affected multiple traits.
Trait
QTL
number
Chr
Position
(cM)
Marker
2-LOD
Interval (cM)
R2
a0
Leaf number - 6 weeks QTL1-1 1 30.4 MTIC448 24.3-41.6 7.3 1.146
QTL1-2 1 110.1 MTIC285 100.2-119.7 7.9 1.123
Number of fruits QTL1-1 1 28.3 MTB269 17.9-34.4 10.7 1.268
QTL1-2 1 110.1 MTIC285 103.6-119.7 7.8 1.044
QTL3-2 3 156.9 MTIC237 143.8-159.7 6.7 1.043
QTL4-1 4 10.7 MTIC033 0-18.7 5.2 -0.86
QTL5-1 5 28.1 MTIC148 20.5-43.4 5.1 0.849
QTL7 7 14.9 MTIC147 0-70.3 6.3 0.945
Average fruit weight QTL3-1 3 41.3 MTIC124 29.5-52.2 5.3 -0.005
QTL3-2 3 158.3 MTIC044 147.8-159.7 5 -0.006
QTL5-1 5 26.5 NFP1660 12-36.8 8.7 -0.007
Shoot weight QTL1-2 1 100.2 MTIC146 98.3-103.6 11.9 0.135
QTL5-2 5 107.8 MTB310 93.4-129.8 7.7 -0.86
Root weight QTL1-1 1 21.9 MTB46 11.5-30.3 8.4 0.047
Primary branch number QTL4-2 4 124.3 MTIC186 110.1-139.4 13.5 0.134
QTL6 6 58.5 G06.1400 29.1-70.4 5.2 0.085
Note: QTL were named as follows: chromosome number as a suffix, with an additional number depending if there were
multiple QTL per chromosome.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 43
Figure 5. Genomic locations of significant QTL detected for the phenotypic traits of M.
truncatula averaged across rhizobium strains. Chromosome number is listed across the top of the
linkage groups. The scale on the left indicates the genetic distance between markers in
centimorgans (cM). Each horizontal line represents the position of one genetic marker.
Phenotypic trait names are listed beside the estimated QTL positions, and the length of the QTL
represents the 2-LOD support interval. The vertical line in each QTL indicates where the highest
peak was located.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 44
Figure 6. Epistatic QTL detected between markers EM2265 and EM17407 in Naut. The least-
square means of each genotype combination are shown with + SE.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
530 735
Root
wei
ght
(g)
Marker allele at EM17407
EM2265-530
EM2265-735
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 45
Branching gene orthologs in M. truncatula
The genes involved in branching pattern have been well characterized in Arabidopsis
thaliana (Ehrenreich et al., 2007), and numerous candidate genes exist. I used BLASTn to
identify orthologs of 36 candidate genes from Arabidopsis that showed >200 bp sequence
identity in the M. truncatula genome. Significant sequence matches were further investigated
using the Medicago truncatula genome BLAST
(http://www.medicago.org/genome/cvit_blast.php ) to determine their location and if there were
any candidate genes in the region. Out of the 36 genes investigated (see Ehrenreich et al.,2007
for descriptions), three of the genes mapped to either chromosome 4 or 6 on the M. truncatula
genome, i.e., on the same chromosomes as the primary branch number QTLs. The genes MAX2
and PID mapped to two separate regions on the lower portion of chromosome 4, and were
located farther down the chromosome compared to the physically linked marker, MTIC089. Both
genes have orthologs which have been previously annotated: for MAX2 a cyclin-like Fbox is
located in the matching M. truncatula sequence and for PID there is a tyrosine protein kinase.
The gene PIN1 matched to sequence data on chromosome 6, and was also located farther down
the chromosome compared to the physically linked marker, MTIC153. PIN1 also had an
annotated ortholog, another auxin efflux carrier protein (Appendix, Table A-6)
Discussion
Genotype by genotype interactions between legumes and rhizobia have been previously
found to influence the benefits each partner receives from the symbiosis (Mhadhbi et al., 2005;
Laguerre et al., 2007; Heath & Tiffin, 2007; Rangin et al., 2008; Heath et al., 2010; Heath,
2010), and have been proposed to maintain genetic variation in mutualist quality (Bever, 1999;
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 46
Molofsky et al., 2001; Heath & Tiffin, 2007; Heath, 2010). To date, no one has attempted to
locate regions of the genome that influence G × G interactions. Here, I performed QTL mapping
in two genotypically distinct rhizobium treatments to map additive or epistatic QTL for plant
fitness traits and determine if they were differentially expressed depending on the rhizobium
genotype. Three major results emerged from the experiment: 1) I detected no evidence for G × G
interactions at the line × rhizobium strain level or at the QTL × rhizobium strain level 2) I
detected numerous QTL that appeared to affect multiple phenotypic traits, suggestive of
pleiotropy or tight linkage, and 3) QTL for plant fitness and growth traits co-localized with
previously described signaling genes. The implications of the results are discussed below.
No evidence for a genetic basis of G × G interactions
Despite preliminary data that suggested otherwise, I found no evidence for G × G
interactions for plant fitness traits in this population. It is possible to detect QTL × rhizobium
genotype interactions without a significant G × G interaction if multiple different QTLs exhibit
conditional neutrality. However no significant QTL × rhizobium strain interaction or RIL ×
rhizobium strain interaction were found in this experiment. One implication of this experiment is
that it illustrates the difficulties associated with trying to locate the genetic basis of context-
dependent traits. Both Naut and Sals have previously displayed varying partner quality when
grown with different plant genotypes (Heath, 2010) and a preliminary experiment found
evidence of G × G interactions when they were grown with the parental lines of the LR03 RIL
population (Appendix, Figure A-1). Nevertheless, in this experiment, there was no interaction
between plant line and rhizobium strain in the parents or in the RILs.
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 47
There are two potential hypotheses to explain the lack of G × G interaction detected in
this experiment: 1) The parental lines do not exhibit any form of G × G interactions, thus
accordingly neither do the RILs or 2) G × G interactions are small enough to only be detected
with multiple plant × rhizobium genotype combinations. Previous studies which have found
evidence of G × G interactions in the legume-rhizobia symbiosis investigated substantially more
unique genotype × genotype combinations in comparison to the number used in this experiment
(e.g., Heath & Tiffin, 2007; Laguerre et al., 2007; Rangin et al., 2008; Heath, 2010; Heath et al.,
2010) . Although each RIL differs from the others, they are essentially just an assortment of the
two parental genomes (Broman, 2005). Therefore, from a simplified view, my experiment was a
2 × 2 factorial design, represented by the two parental lines and the rhizobium genotypes Naut
and Sals. When RILs are used to detect G × G interactions, only large effect loci underlying such
interactions will be found because there are only a few genotype - genotype combinations. In
comparison, in experiments with multiple plant families and rhizobium genotypes, there are
considerably more genotypic combinations, and there is likely more statistical power to detect G
× G interactions. However, if these population-wide G × G interactions are due to multiple small
effect loci which are distributed throughout the plant and rhizobium genomes, the probability of
detecting such G × G interactions in traditional QTL mapping experiments using RILs is very
low.
An alternative approach to isolating the genomic regions influencing G × G interactions
would be to perform nested association mapping (NAM) (Yu et al., 2008). NAM combines the
advantages of using experimental crosses in traditional QTL mapping with the high resolution of
association mapping to resolve quantitative traits to their causal loci. In maize, 5000 RILs have
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 48
been created from separate crosses of a reference line with 25 genetically different inbred lines.
Each individual RIL is a mosaic of chromosome segments derived from either the reference line
or one of the founding inbred lines. In comparison to the design used in the present study, NAM
would allow significantly more legume × rhizobium genotypes to be tested and provide greater
statistical power to detect potential small effect loci influencing G × G interactions. The time and
cost of such a mapping design is considerable, however the potential outcomes would be
invaluable to the study of G × G interactions in the legume-rhizobium symbiosis.
QTL architecture of phenotypic traits
I detected numerous QTL in the two rhizobium strain treatments (Table 2 and 3). These
QTL were of small to moderate effect, explaining between 3.5 and 12.7% of the total genetic
variation, with additive effect sizes of 0.007-1.78. Despite significant genetic variation in
flowering time, no QTL were detected for this phenotypic trait, suggesting that the observed
genetic variation in days to flowering is due to many QTL of small effect rather than a few major
QTL. These small QTLs were probably undetectable with the mapping population size used in
this experiment (Lynch & Walsh, 1998).
All of the phenotypic traits displayed evidence of transgressive segregation (Figure 3),
suggesting that these traits are controlled by many genes of small effect. The most likely
explanation for the observed trangressive segregation is antagonistic QTLs, i.e., QTLs with
effects that are in the opposite direction to the parental differences for that trait (Rieseberg et al.,
2003). Antagonistic QTLs are commonly found in studies where transgressive phenotypes are
detected, and are more frequent in studies using plants, intraspecific crosses or investigating
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 49
morphological traits (Rieseberg et al., 1999; Rieseberg et al., 2003). Between the two strain
treatments, most of the measured traits had at least one antagonistic QTL (Figure 3, Table 2, 3),
with the exception of number of fruits, root weight and primary branch number. It is probable
that some QTL of small effect were not detected due to low power, which may explain the
missing antagonistic QTLs for those traits.
In all QTL analyses, there were QTL which influenced multiple traits: leaf number, fruit
number, average fruit weight, shoot weight and root weight all colocalized with at least one other
listed trait (Table 2, 3, 4; Figure 4 and 5). These regions could be indicative of pleiotropic genes
acting on multiple traits or numerous genes in tight linkage. For example, in Naut, plant weight,
fruit number and leaf number all colocalized to the end of chromosome 1. The QTL effects of
NQTL1-2 match the broadscale patterns evident in the line means: plant weight was positively
correlated with fruit number and fruit number is positively correlated with leaf number (see
Appendix, Table A-4 for correlation matrix), i.e., larger plants make more fruit. These regions
may indicate locations for QTL controlling a trait common to those phenotypes (e.g., growth rate
or plant size or plant fitness).
There were a few QTL which influenced multiple traits, however in opposite directions in
the two parental lines. NQTL3-2 in Naut increased number of fruit and decreased leaf number
(Table 2), which was surprising as these two traits are positively correlated, as mentioned above.
One possible explanation is that this region may contain multiple tightly linked QTL, which have
opposite effects on number of fruit and leaf number. Fine-mapping of QTLs often show that
single QTLs will break into multiple closely linked QTLs, which in turn frequently act in
opposite directions (e.g., Steinmetz et al., 2002; Kroymann & Mitchell-Olds, 2005). An
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 50
alternative explanation is that a single locus is influencing both fruit number and leaf number,
but in opposite directions, creating a genetic trade-off (i.e., antagonistic pleiotropy). Either tight
linkage between multiple QTLs or pleiotropy may account for the opposite fitness effects of
NQTL3-2, however additional generations of recombination or fine-mapping would be required
to distinguish between the two mechanisms.
Two additional QTLs had opposite effects in the across strain analysis: QTL3-2 and
QTL5-1 increased number of fruit but decreased average fruit weight (Table 3), which is
expected given their negative correlation (Appendix, Table A-5). The RILs appear to be adopting
two different life history strategies for producing fruit: either making fewer, larger fruit (quality),
or many, smaller fruit (quantity). It is possible that these differences in resource allocation
translate into variation in above or below ground biomass. For example, lines which produce the
small, lighter fruit may in turn have more resources to invest in vegetative growth, which could
be important for plants in natural habitats where competition with neighbours for access to light
is essential.
Epistatic interactions
To fully understand the genetic architecture of any trait, it is necessary to understand
epistasis, which is defined as the nonadditive interactions between alleles of different genes (see
Phillips (2008) for a review). Epistatic interactions influence a myriad of areas in evolutionary
genetics including the maintenance of genetic diversity, the evolution of sexual reproduction,
speciation, developmental canalization, and inbreeding depression (Fenster et al., 1997; Wolf et
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 51
al., 2000; Phillips, 2008). Given its importance to these evolutionary processes, accurate and
powerful tests for epistasis are essential.
In QTL mapping, ignoring epistatic interactions can lead to underestimating genetic
variance and overestimating individual QTL effects (Carlborg & Haley, 2004). Due to their
importance, QTL studies are increasingly investigating the presence of epistatic QTLs, i.e.,
marker by marker interactions which have significant effects on phenotypic traits (e.g., Leips &
Mackay, 2000; Weinig, et al., 2003; Malmberg et al., 2005; Brock et al., 2009). Epistatic QTLs
can be detected in one of two ways: either searching for epistatic interactions between pre-
existing QTLs which already have a significant main effect or performing all possible two-way
interactions between markers. The problem with the first approach is that it can miss epistatic
interactions where only one of the loci is involved in a main-effect QTL or additional epistatic
QTL between markers which have no main effect alone (Holland, 1998; Malmberg et al., 2005).
In this experiment, I tested all pairwise combinations of markers and found only one
epistatic QTL for root weight in Naut between markers EM2265 and EM17407 on chromosome
6 (Figure 6). In general, the low frequency of epistatic QTLs indicates that epistasis is relatively
rare in this mapping population, at least when plants are grown under greenhouse conditions.
There was no main-effect QTL detected in this region because the average effect of the EM2265
alleles over EM17407 was the same. However, the epistatic QTL explained a moderate amount
of the variation (13 %, in a model of root weight = EM2265 + EM17407 + EM2265 × EM17407
+ ε), therefore it is evident that variation at this genomic region has a significant effect on root
weight. Alternatively, it is possible that the effect is overestimated due to the small number of
lines in the recombinant allele class – i.e., the Beavis effect (Beavis, 1998). Nevertheless, the
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 52
QTL displayed an interesting crossing reaction norm between the two markers: RILs which have
alleles from the same parents for both markers have much lower root weight than those which
have alleles from alternate parents. Although this epistatic QTL displayed no evidence for QTL
× QTL × E, it would be interesting to see if the effect of the interaction changed with alternate
rhizobium genotypes, since root weight is likely to be correlated with rhizobia fitness traits such
as nodule number.
To account for the problem of multiple testing in Epistacy, Holland (1998) recommended
dividing the desired p-value by g(g-1)/2, where g is the number of linkage groups. Although this
approach is better than the naïve application of α = 0.05, it has two shortcomings. First, Epistacy
tests for epistatic interactions between all markers –i.e., within and between linkage groups, not
just between linkage groups–and second it does not account for the number of markers in the
analysis, which directly influences the number of statistical tests performed. In contrast, I suggest
that a better approach is to test all possible marker by marker interactions (number of tests = (n ×
n -1)/2, where n = number of markers) and then implement false discovery rate and q-value
testing to control for Type I errors in multiple testing (Benjamini & Hochberg, 1995; Storey,
2002).
As an example, using the correction suggested by Holland (1998), Epistacy detected 26
significant epistatic QTLs for leaf number in the Naut treatment. Analyzing the same data using
FDR, it seems likely that there are no significant epistatic QTLs – even interactions with p =
0.0002 had q-values of 0.75, suggesting a high chance of false discovery. Furthermore, plotting a
histogram of the p-values from the epistasis tests for leaf number suggests the null hypothesis:
there was a uniformly even distribution of p-values between zero and one, indicative of no
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 53
statistically significant interactions. Therefore, in future studies, I recommend implementing
FDR for detecting epistatic QTLs in Epistacy.
Colocalization of symbiotic signaling genes
Two of the symbiotic signaling genes, NFP and DMI3, colocalized with QTLs (Figure 4
and 5; see Appendix, Figure A-2 for positions). NFP was found in the interval of a QTL which
influenced average fruit weight on chromosome 5 in all treatments (NQTL, SQTL5, QTL5-1;
Table 2, 3, and 4). The effect sizes of these QTLs were small, however the QTL was consistently
detected across all analyses, suggesting that it indeed affects fitness. Similarly, DMI3 colocalized
with a QTL affecting leaf number in Naut only (NTQL8, Table 2); despite its small additive
effect, DMI3427 was the marker directly underlying the QTL peak, suggesting that DMI3 may
contribute to leaf number variation.
The colocalization of NFP and DMI3 with average fruit weight and leaf number indicates
that variation in nodulation signaling genes can potentially influence plant performance and
fitness traits. Legumes require rhizobia to survive and reproduce, and the formation of root
nodules is an essential component to the establishment of the symbiosis. NFP and DMI3 are both
among the genes which are required for the initiation of the nodulation pathway, and lab induced
mutations in either of these genes prevents the formation of nodules altogether (Catoira et al.,
2000; Ben Amor et al., 2003). Furthermore, NFP and DMI3 have both been found to be under
strong purifying selection in M. truncatula (De Mita et al., 2007). It has been hypothesized that
purifying selection is acting on these genes to maintain high specificity of recognition and
remove mutations which may be deleterious to the signaling molecules involved in establishment
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 54
of the symbiosis, and indeed this has been found in some populations of M. truncatula (De Mita
et al., 2006; De Mita et al., 2007). Given this information, it seems plausible that mutations in
Nod factor signaling genes will in turn affect plant fitness.
The colocalization of NFP and DMI3 with average fruit weight and leaf number QTLs
strongly supports this hypothesis: variation in signaling genes can affect plant traits. In addition,
these QTLs suggest that one of the parental lines is better able to attract and incorporate rhizobia
partners in root nodules, which in turn translates to obtaining higher fitness. The potential
mechanism leading to a better plant signaler is unknown, but the presence of the QTLs in these
regions suggests that variation in Nod factor signaling efficiency can exist between genetically
different individuals. Nevertheless, these results should be interpreted with caution without
further investigation of the loci underlying the QTL intervals. Colocalization of the QTLs with
NFP and DMI3 does not imply causality: QTL intervals often contain multiple genes and
millions of basepairs. NFP and DMI3 may simply be in tight linkage with the loci influencing
the plant fitness traits, therefore they both colocalized to the same QTL interval.
Evidence for branching pattern orthologs
The largest QTL was detected for primary branch number: in the across strain analysis,
QTL4-2 explained 13.5% of the total genetic variation (Table 4). This QTL was also detected in
the Naut treatment (NQTL4-2, Table 2), and an additional QTL was identified on chromosome 6
in the across strain analysis (QTL6). I found that primary branch number was positively
correlated with flowering date across all analyses (Appendix, Table A-4, A-5), confirming what
has been previously found in natural accessions (Julier et al., 2007).
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 55
The sequence comparison with Arabidopsis thaliana yielded three potential gene
orthologs of the branching pattern genes MAX2, PID and PIN1 in M. truncatula (Appendix,
Table A-5). MAX2, an Fbox protein, and PID, a serine/threonine kinase both matched to
segments of sequence data located on the lower end of chromosome 4, where the largest effect
QTL was detected for primary branch number. Located in these regions were two potential
orthologs: a cyclin-like Fbox and a tyrosine protein kinase. The other gene PIN1, an auxin efflux
carrier protein, mapped to sequence segments on chromosome 6, in the area of the other primary
branch QTL. In this region of the M. truncatula genome, there was an auxin efflux carrier
protein, an exact match for the gene function of PIN1.
It is difficult to determine the physical proximity of these orthologous genes to the branch
number QTLs as there is no concrete conversion of cM to kb in M. truncatula. Furthermore, one
potential problem is that the functions of these genes were likely annotated based on sequence
homology with Arabidopsis thaliana. Nevertheless, they are promising candidate genes for
branch morphology in M. truncatula, and future studies which anchor these genes into linkage
maps would help determine if they influence variation in branch morphology.
Conclusion
Mutualisms are predicted to have low genetic variation, yet empirical studies consistently
indicate that substantial genetic variation in partner quality exists. One potential mechanism
which may be maintaining this variation is genotype by genotype interactions between the host
and symbiont. I found no evidence for G × G interactions at the line or QTL level in this
population, but future studies should focus on using experimental designs which allow a greater
Chapter 2: Resolving the genetic basis of genotype by genotype interactions 56
variety of genotype × genotype interactions to be investigated. In addition, I found a number of
QTLs influencing multiple traits, including some which colocalized with plant nodulation
signaling genes, suggesting these genes may potentially affect plant performance and fitness.
Chapter 3: Conclusions and future directions 57
Chapter 3:
Conclusions and Future Directions
The goal my thesis was to locate regions of the genome affecting the outcome of
interactions between legumes and rhizobia and to identify loci influencing variation in
phenotypic plant traits. Contrary to the predictions of numerous theoretical models (Trivers,
1971; May, 1976; Bull & Rice, 1991; West et al., 2002), variation in partner quality of both the
hosts and symbionts continues to persist in mutualisms (e.g., Bever et al., 1996; van der Heijden
et al., 1998; Mhadhbi et al., 2005). One potential explanation for the maintenance of this genetic
diversity is G × G interactions (Wade, 2007), where an individual’s optimal mutualist is
dependent on the genome of its interacting partner. Previous work on the legume-rhizobium
mutualism found that G × G interactions can affect the fitness benefits both partners receive from
the symbiosis (e.g., Heath & Tiffin, 2007; Rangin et al., 2008; Heath, 2010). However the
specific loci or genes underlying these interactions have yet to be identified. I conducted a QTL
mapping experiment with two different rhizobium strains to locate potential regions of the
genome influencing G × G interactions between M. truncatula and S. meliloti.
There were three major findings of my thesis. First, there were no G × G interactions
between the M. trunctula LR03 RIL population and the rhizobium strains Naut and Sals, nor
were there different QTL affecting plant traits in the two strain treatments, at least under the
experimental conditions I used. Second, there were numerous small to moderate affect QTLs
spread across the genome which affected multiple plant phenotypic traits, indicative of either
pleiotropy or tight linkage between genes. Lastly, two of 13 QTLs detected for plant fitness
components colocalized with genes involved in the nodulation pathway, suggesting that variation
Chapter 3: Conclusions and future directions 58
in these genes can potentially affect plant performance and fitness traits. These findings have
important implications for the evolutionary interactions between legumes and rhizobia, and the
genetic architecture of Medicago truncatula.
Implications for G × G interactions
Genotype by genotype interactions between legumes and rhizobia are known to influence
the outcome of the mutualism (Mytton et al., 1977; Parker, 1995; Heath & Tiffin, 2007; Rangin
et al., 2008; Heath, 2010), therefore there must be genomic regions underlying these interactions.
In contrast, the results of my thesis suggest that there were no G × G interactions between the
RIL population and the strains Naut and Sals, and consequently no genomic regions influencing
the direction of partner genotype interactions were located. The generality of these conclusions
for future work on G × G interactions is potentially limited, as I only investigated one plant
population and two rhizobium strains. However, it does indicate that the presence of G × G
interactions is highly context-dependent since G × G interactions were previously found in Heath
(2010) using the same rhizobium strains (albeit with different plant lines) and preliminary data
suggested the parental lines of the RIL population I used displayed G × G for plant performance
traits (Appendix, Figure A-1).
One aspect of the experimental design was not discussed in Chapter 2: the parental lines
and rhizobium strains were from different locations, i.e., they were foreign partners. I omitted the
discussion of this topic from Chapter 2 as the effect of foreign vs native rhizobia on plant fitness
are mixed (e.g., Parker, 1995; Wilkinson & Parker, 1996; Burdon et al., 1999; Thrall et al.,
2000), and no firm conclusion can be drawn. Nevertheless, it has interesting implications which
are worth briefly addressing.
Chapter 3: Conclusions and future directions 59
The rhizobium strains Naut and Sals were isolated from soil in south of France,
approximately 300 and 700 km away from the collection sites of the French and Algeria parents,
respectively (Appendix, Table A-1). Since they are geographically mismatched, there is no
known history of interaction between the partner genotypes. However, using foreign partners
instead of native partners could potentially affect the estimation of G × G interactions. If G × G
interactions are maintaining genetic variation in partner quality through frequency dependent
selection, plant genotypes with the highest fitness should change depending on the rhizobium
strain, resulting in negative correlations for plant fitness traits across rhizobium strains (Bever,
1999; Molofsky et al., 2001; Heath & Tiffin, 2007). In my thesis, fitness was positively
correlated across strain treatments, indicating no evidence of a genetic trade-off between
rhizobium genotypes. However, it is possible that there are a few loci with antagonistic effects in
Naut and Sals whose effects were masked by using a novel environment: some plant and
rhizobium genotypes may have initially high fitness in new environments due to chance alone,
which can lead to systematic bias towards positive correlation among life history traits and
fitness components (Service & Rose, 1985; Fry, 1993). The net effect of this might potentially
underestimate genetic trade-offs or small effect G × G interactions. Alternatively, foreign
individuals may bring new genetic variation to the population, thereby increasing the chance of
detecting G × G interactions. In future studies, to remove any effects of using novel genotype
combinations, it would be useful to grow the plant lines and rhizobium strains of interest together
for few generations in the greenhouse prior to beginning any experiments. Alternatively, this
issue could be avoided altogether by obtaining plant lines and rhizobium strains from the same
location.
Chapter 3: Conclusions and future directions 60
One limitation of my thesis is that I was unable to record any rhizobium fitness traits.
Although G × G interactions did not influence any of the plant fitness traits in this population,
the same conclusion cannot be drawn concerning rhizobium fitness traits, such as nodule
number. As described in Chapter 2, a preliminary analysis of nodule traits on the parental plants
in the experiment yielded no support for plant line × rhizobium genotype interactions. Based on
this data, I abandoned any further collection of rhizobia traits and focused on recording other
phenotypic traits and developing molecular markers that seemed more promising. However,
given the results I obtained for the plant traits, and the results of previous studies, there are
several hypotheses and implications for rhizobium fitness traits.
Upon completion of the experiment, it was apparent that the RILs displayed transgressive
segregation for all traits, i.e., more extreme phenotypes than either of the parents (Figure 2).
Furthermore, even in cases where parental line means were close in value, at least one QTL was
detected for each trait, indicating that the parents displayed allelic variation in those traits. It is
entirely possible that similar distributions might have been found for rhizobium fitness traits
such as nodule number and weight, and in turn, significant QTL may have been detected. In
addition, nodule number has repeatedly been found to be influenced by G × G interactions
(Heath & Tiffin, 2007; Laguerre et al., 2007; Rangin et al., 2008; Heath, 2010; Heath et al.,
2010), whereas G × G for plant fitness traits such as fruit number and seed number appear to be
less often detected (Parker, 1995; Heath & Tiffin, 2007; Heath, 2010; Heath et al., 2010). These
two pieces of information together suggest that rhizobium fitness traits are promising avenues for
future studies trying to isolate the genomic regions influencing G × G interactions.
Chapter 3: Conclusions and future directions 61
The QTL architecture of Medicago truncatula
QTL mapping is an excellent first step towards characterizing the genetic architecture of
a quantitative trait. I did not find any QTL × rhizobium strain interactions, however my thesis
nevertheless makes a significant contribution to the field as a novel approach and perspective to
QTL mapping. Specifically, it was novel because I mapped traits in response to changing
genotype interactions, not simply different environmental conditions. In addition, I identified
QTLs influencing the same traits in both strain treatments and found evidence for loci affecting
multiple traits, both of which can be used for future studies investigating the genetic basis of
phenotypic traits in the legume-rhizobium mutualism.
In comparison to previous QTL mapping studies which have investigated QTL ×
environment interactions (e.g., Fry et al., 1998; Juenger et al., 2005; Gardner & Latta, 2006;
Brock et al., 2010), my thesis presents a novel approach because I used separate genotypes of an
interacting species as different environments. I used the methodology for examining QTL ×
environment interactions, however I was testing for QTL × genotype interactions, a unique
analysis which has not been performed before. Although there are a number of QTL studies
which have mapped plant traits in response to interactions with other organisms, such as insects,
ectomycorrhizal fungi, and pathogens (e.g., Weinig et al., 2003; Tagu et al., 2005; Hamon et al.,
2010; Tetard-Jones et al., 2011), none of these investigate plant responses to more than one
genotype of the interacting species. I did not find any significant QTL × rhizobium genotype
interactions, but my approach could potentially be used in future experiments trying to locate the
genetic basis of G × G interactions between interacting species, such as herbivory, where G × G
interactions are known to exist (Moran, 1981; Service, 1984; Tetard-Jones et al., 2007).
Chapter 3: Conclusions and future directions 62
As I described in Chapter 2, a number of the QTLs affected multiple traits, which is
indicative of either one locus acting pleiotropically or tight linkage between different loci acting
on each trait. Here I discuss the evolutionary implications of the two aforementioned
mechanisms. If variation in each trait is controlled by trait-specific loci, then each trait should be
able to evolve independently, unless they are in linkage disequilibrium. Alternatively, if
pleiotropy is occurring, it can have significant effects on the evolutionary trajectory of the traits
involved (Lande, 1979). For example, QTL1-2 in the across strain analysis influenced leaf
number, fruit weight and shoot weight in the same direction (Table 4), all of which are positively
correlated (Appendix, Table A-5). If this QTL is a true pleiotropic locus, these traits are expected
to evolve in the same direction, producing the common phenotype of larger plants having more
fruits and more leaves. However, pleiotropy can also create a situation where selection to
maintain one trait may constrain the evolvability of another – i.e., antagonistic pleiotropy
(Anderson et al., 2011). Such a situation could arise if the traits influenced by the pleiotropic
locus are either negatively correlated or influenced by the locus in opposite directions. As an
example, NQTL3-2 in Naut had opposite effects on leaf number and fruit weight (Table 2),
which could create a potential genetic trade-off between reproduction and vegetative growth.
However, as I mentioned in Chapter 2, ultimately determining if pleiotropy is truly occurring
requires fine-mapping of the QTLs or additional generations of recombination to break up any
potential linkage between genes.
Genetic variation at the Nod factor signaling genes
The signaling genes involved in the early stages of the Nod factor signaling pathway are
essential to the formation of the symbiosis between legumes and rhizobia (Jones et al., 2007;
Chapter 3: Conclusions and future directions 63
Ferguson et al., 2010). Two of these genes, NFP and DMI3, were found within the intervals of
QTLs affecting leaf number and average fruit weight, suggesting that variation in these genes
may potentially affect plant performance and plant fitness. If NFP and DMI3 are the causal loci
of these QTLs, there are a number of interesting implications of these findings.
As described in Chapter 2, these results suggest that one of the parental lines must be
better signaler at some point in the nodulation pathway. The next step is to determine at both the
genomic and phenotypic level specifically which changes in the Nod factor signaling pathway
result in particular plant genotypes obtaining higher fitness. The nonsynonymous SNP in NFP is
a promising place to begin. The amino acid substitution may change the conformational structure
of the Nod factor receptor, allowing rhizobium Nod factors to be recognized and processed more
rapidly by the host plant. The functional changes in DMI3 are more difficult to conceive as all of
the SNPs I found are located in non-coding regions and this gene is involved later in the
signaling pathway (Catoira et al., 2000; Lévy et al., 2004). There are two possibilities: either
there is a mutation in the non-coding region upstream of DMI3 which results in differential gene
expression between the two parents, or there is a nonsynonymous mutation which is currently
undetected. I was only able to sequence a third of DMI3, therefore it is possible that some of the
genetic polymorphism may have been missed.
Under natural conditions, the parental allele which confers better signaling ability and
higher fitness should eventually spread to fixation. Alternatively, genetic variation in signaling
ability could be maintained if the allele or genotype which is the better signaler depends on the
rhizobial strain – i.e., NFP or DMI3 × rhizobium strain interaction, a form of G × G interaction.
Although I found no evidence that this occurring in my thesis, I think it is worthwhile
Chapter 3: Conclusions and future directions 64
investigating further with additional rhizobium genotypes for two reasons: 1) The parental line
which had higher fitness depending on SNP at NFP and DMI3 is from a different country than
the rhizobium strains, suggesting some sort of foreign plant genotype advantage, and 2) There
may be an allele that results in an optimal plant signaler with all compatible rhizobium strains. In
other words, plant hosts with this allele would always obtain higher fitness in comparison to
those individuals with alternate alleles, regardless of the interacting rhizobium strain. Additional
investigations with more rhizobium strains from multiple locations would help determine if
either of these two observations are general patterns in the Medicago-Sinohizobium mutualism.
Future directions and final conclusions
There are a number of additional studies which would provide valuable insights into the
G × G interactions between legumes and rhizobia and their genetic basis. As I have mentioned
repeatedly throughout the discusson of my findings, I believe it is worthwhile to conduct similar
experiments with additional plant lines and rhizobium strains. By using a different experimental
design, such as nested association mapping (described in Chapter 2, pg. 47) or replicated F2
mapping, significantly more genotype combinations could be evaluated with greater statistical
power. Furthermore, precise strain selection could allow the impact of foreign vs native strains
on G × G interactions to be evaluated concurrently. Below I describe a few additional areas for
future studies to investigate based on the conclusions of my thesis.
The next goal in many QTL mapping studies is to perform fine-mapping in the regions
of interest either by increasing the number of markers or recombinants in the region of the
putative QTL. After an initial round of QTL mapping, the intervals of each QTL are often 10-30
cM wide, as can be seen in the QTLs I detected (Table 2, 3, 4). The problem is that these wide
Chapter 3: Conclusions and future directions 65
intervals can in turn contain hundreds to thousands of genes and millions of basepairs. Fine
mapping can reduce the QTL interval to 1-5 cM which can be sufficiently small to locate
candidate genes of interest for additional functional studies (Abiola et al., 2003). However
putative QTLs should be confirmed in additional studies prior to performing fine-mapping, as it
is time consuming process. One method to validate QTLs is to create near-isogenic lines, which
differ from the parental lines only at the location of the targeted QTL (Borevitz & Chory, 2004).
As such, any phenotypic differences between the NILs and the parental lines are likely due to the
QTL. I think this is the next direction to proceed in to locate the specific loci affecting some of
traits I mapped in my thesis. In particular I think it would be worthwhile to confirm the QTLs
found for primary branching and the plant fitness QTLs which colocalized with the Nod factor
signaling genes. Once these QTLs have been validated, additional studies with more markers in
the region of the QTLs can help narrow down the interval and the potential causal loci.
Some of the genes involved in the Nod factor signaling pathway, such as NFP, are unique
to the legume-rhizobium symbiosis, while others, such as the DMI genes, also function in the
mycorrhizal infection process (Catoira et al., 2000). If DMI3 is the causal locus influencing plant
performance in my thesis, I think an interesting complementary experiment would be determine
if genomic variation in the DMI genes also influences plant phenotypes when grown with
mycorrhizal fungi. It would help determine if the influence of these genes is unique to affecting
the host plant in the legume-rhizobium interaction, or if it is a general pattern found in plant-
microbe symbioses.
One aspect that is missing from the investigation of G × G interactions between legumes
and rhizobia is an assessment of the frequency, distribution, and probability of association
Chapter 3: Conclusions and future directions 66
between different genotypes. The underlying assumption of G × G interactions maintaining
genetic diversity in partner quality is that both the host and symbiont encounter different partner
genotypes often enough for frequency dependent selection to occur (Bever, 1999; Molofsky et
al., 2001; Heath & Tiffin, 2007). However, without explicit information on the frequency of
interactions, the actual – versus potential – importance of G × G interactions to coevolutionary
dynamics is unknown. It likely remains unanswered due to the difficulty of monitoring both
partners under natural conditions and a variety of other factors which can influenced the outcome
of G × G interactions such as herbivory, competitors and nutrient availability (e.g., Heath &
Tiffin, 2007; Heath et al., 2010; Heath & Lau, 2011).
A potential solution to help address these issues would be to perform a small, long-term
G × G controlled greenhouse experiment in a large soil flat. The experiment would start initially
with equal frequencies of two genotypes of each partner which have repeatedly shown evidence
of G × G interactions. In particular, it would be important to obtain pairings as described by
Bever (1999), where host plant genotype A obtains the highest fitness with microbial mutualist
X, but symbiont X obtains the highest fitness with host plant genotype B, and vice versa for host
plant genotype B and microbial mutualist Y. By using a large soil flat instead of individual pots,
the rhizobium strains would be able to move freely between host plants, and the outcome of G ×
G interactions within any one host would influence the frequencies of partner genotypes in the
entire experimental population. Once set up, the experiment would be left to run for two or three
plant generations, largely untouched. Inevitably, the density of plants would probably be
increasingly high after each generation, therefore some systematic thinning would be required or
the initial set-up could begin with plants at a very low density. Midway through the experiment,
Chapter 3: Conclusions and future directions 67
half of the plants and their associated rhizobia would be randomly harvested and genotyped
(most likely only subset of nodules on each plant) to assess the frequencies and associations
between different partner genotypes. The remaining plants would be left to grow for an equal
number of generations prior to also being harvested and genotyped. If G × G interactions and
frequency dependent selection are truly occurring, there should be a marked difference in partner
genotype frequencies and associations at the two harvesting stages.
The results of such an experiment would not be directly comparable to field settings, as
there are undoubtedly many more plant and rhizobium genotypes which are interacting under
natural conditions. However, it would help narrow the gap in knowledge concerning the
fluctuating frequencies of partner genotypes across multiple generations. Furthermore,
depending on the results, this experiment could serve as a proof of concept for the ability of
frequency dependent selection to maintain partner quality diversity over coevolutionary time.
Although this experiment would involve a significant amount of work, and is also a long term
study (best suited as a collaborative experiment), the results could potentially contribute to
determining if G × G interactions and frequency dependent selection can truly maintain genetic
variation in mutualist quality over multiple generations.
Literature cited 68
Literature Cited
Abi-Ghanem, R., Carpenter-Boggs, L., & Smith, J. L. 2010. Cultivar effects on nitrogen fixation
in peas and lentils. Biology and Fertility of Soils 2008.
Abiola, O., Angel, J. M., Avner, P., Bachmanov, A. A., Belknap, J. K., Bennett, B., Blankerhorn,
E. P., Blizard, D. A., Bolivar, V., Brockmann, G. A., Buck, K. J., Bureaus, J.-F., Caley, W.
L., Chesler, E. J., Cheverud, J. M., et al. 2003. The nature and identification of quantitative
trait loci. Nature Reviews Genetics 4: 911-916.
Amor, B. B., Shaw, S. L., Oldroyd, G. E. D., Maillet, F., Penmetsa, R V., Cook, D., Long, S. R.,
Dénarié, J., & Gough, C. 2003. The NFP locus of Medicago truncatula controls an early
step of Nod factor signal transduction upstream of a rapid calcium flux and root hair
deformation. The Plant Journal 34: 495-506.
Anderson, J. T., Willis, J. H., & Mitchell-Olds, T. 2011. Evolutionary genetics of plant
adaptation. Trends in Genetics : 27: 258-266.
Ané J.-M., Kiss, G. B., Riely, B. K., Penmetsa, R. V.,, Oldroyd, G. E. D., Ayax, C., Levy, J.,
Debelle, F., Baek, J.-M., Kalo, P., Rosenberg, C., R., B. A., Long, S. R., Denarie, J., &
Cook, D. R. 2004. Medicago truncatula DMI1 required for bacterial and fungal symbioses
in legumes. Science 303: 1364-1367.
Ané, J.-M., Zhu, H., & Frugoli, J. 2008. Recent Advances in Medicago truncatula Genomics.
International Journal of Plant Genomics 2008: 256597.
Arrighi, J.-F., Barre, A., Ben Amor, B., Bersoult, A., Soriano, L. C., Mirabella, R., Carvalho-
Niebel, F. de, Journet, E.-P., Ghérardi, M., Huguet, T., Geurts, R., Dénarié, J., Rougé, P., &
Gough, C.. 2006. The Medicago truncatula lysin [corrected] motif-receptor-like kinase gene
family includes NFP and new nodule-expressed genes. Plant Physiology 142: 265-79.
Aubert, G., Morin, J., Jacquin, F., Loridon, K., Quillet, M. C., Petit, a, Rameau, C., Lejeune-
Hénaut, I., Huguet, T, & Burstin, J. 2006. Functional mapping in pea, as an aid to the
candidate gene selection and for investigating synteny with the model legume Medicago
truncatula. Theoretical and Applied Genetics 112: 1024-41.
Axelrod, R., & Hamilton, W. D. 1981. The evolution of cooperation. Science 211: 1390-6.
Bailly, X., Olivieri, I., De Mita, S., Cleyet-Marel, J.-C., & Béna, G. 2006. Recombination and
selection shape the molecular diversity pattern of nitrogen-fixing Sinorhizobium sp.
associated to Medicago. Molecular Ecology 15: 2719-34.
Literature cited 69
Barker, D. G., Bianchi, S., Blondon, F., Dattée, Y., Duc, G., Essad, S., Flament, P., Gallusci, P.,
Génier, G., Guy, P., Muel, X., Tourneur, J., Dénarié, J,, & Huguet, T. 1990. Medicago
truncatula, a model plant for studying the molecular genetics of theRhizobium-legume
symbiosis. Plant Molecular Biology Reporter 8: 40-49.
Bataillon, T., & Ronfort, J. 2006. Evolutionary and Ecological Genetics of Medicago truncatula.
Medicago truncatula Handbook: 1-12.
Beavis, W. D. 1998. QTL analyses: Power, precision and accuracy. In A. H. Patterson (Ed.),
Molecular Dissection of Complex Traits. New York: CRC Press (pp. 145-162).
Beckmann, J. S., & Soller, M. 1988. Detection of linkage between marker loci and loci affecting
quantitative traits in crosses between segregating populations. Theoretical and Applied
Genetics 76: 228-236.
Benedito, V. A., Torres-Jerez, I., Murray, J. D., Andriankaja, A., Allen, S., Kakar, K., Wandrey,
M., Verdier, J., Zuber, H., Ott, T., Moreau, S., Niebel, A., Frickey, T., Weiller, G., He, J., et
al. 2008. A gene expression atlas of the model legume Medicago truncatula. The Plant
Journal 55: 504-13.
Benjamini, Y., & Hochberg, Y. 1995. Controlling the false discovery rate: a practical and
powerful approach to multiple testing. Journal of the Royal Statistical Society, Series B 57:
289-300.
Bever, J. D. 1999. Dynamics within mutualism and the maintenance of diversity: inference from
a model of interguild frequency dependence. Ecology Letters 2: 52-61.
Bever, J. D., Morton, J. B., Antonovics, J., Schultz, P. A., & Schultz, A. 1996. Host-dependent
sporultation and species diversity of arbuscular mycorrhizal fungi in a mown grassland.
Journal of Ecology 84: 71-82.
Bonnin, I., Prosperi, J.-marie, & Olivieri, I. 1996. Genetic markers and quantitative genetic
variation in Medicago truncatula (Leguminosae): A comparative analysis of population
structure. Genetics 143: 1795-1805.
Borevitz, J. O., & Chory, J. 2004. Genomics tools for QTL analysis and gene discovery. Current
Opinion in Plant Biology 7: 132-6.
Boucher, D. B. 1985. The biology of mutualism (D. B. Boucher, Ed.). Sydney: Croom Helm.
Brock, M T, Stinchcombe, J. R, & Weinig, C. 2009. Indirect effects of FRIGIDA : floral trait (
co ) variances are altered by seasonally variable abiotic factors associated with flowering
time. Journal of Evolutionary Biology 22: 1826-1838.
Literature cited 70
Brock, M. T., Dechaine, J. M., Iniguez-Luy, F. L., Maloof, J. N., Stinchcombe, J. R., & Weinig,
C. 2010. Floral genetic architecture: an examination of QTL architecture underlying floral
(co)variation across environments. Genetics 186: 1451-65.
Broman, K. W. 2005. The genomes of recombinant inbred lines. Genetics 169: 1133-46.
Bronstein, J. 2001. The costs of mutualism. Integrative and Comparative Biology 839: 825-839.
Bronstein, J. L. 1994. Our Current Understanding of Mutualism. The Quarterly Review of
Biology 69: 31-51.
Bshary, R. 2002. Asymmetric cheating opportunities and partner control in a cleaner fish
mutualism. Animal Behaviour 63: 547-555.
Bshary, R., & Grutter, A. S. 2005. Punishment and partner switching cause cooperative
behaviour in a cleaning mutualism. Biology Letters 1: 396-9.
Bull, J. J., & Rice, W. R. 1991. Distinguising mechanisms for the evolution of co-operation.
Journal of Theorectical Biology 149: 63-74.
Burdon, J. J., Gibson, A. H., Suzette, D. S., Woods, M J, & Brockwell, J. 1999. Variation in the
Effectiveness of Symbiotic Associations between Native Rhizobia and Temperate
Australian Acacia : Within-Species Interactions. Journal of Applied Ecology 36: 398-408.
Carlborg, O., & Haley, C. S. 2004. Epistasis: too often neglected in complex trait studies? Nature
Reviews Genetics 5: 618-25.
Catoira, R., Galera, C., Billy, F. de, Penmetsa, R V, Journet, E. P., Maillet, F, Rosenberg, C,
Cook, D, Gough, C, & Dénarié, J. 2000. Four genes of Medicago truncatula controlling
components of a nod factor transduction pathway. The Plant Cell 12: 1647-66.
Choi, H.-K., Kim, D., Uhm, T., Limpens, E., Lim, H., Mun, J.-H., Kalo, P., Penmetsa, R. V.,
Seres, A., Kulikova, O., Roe, B. A., Bisseling, T., Kiss, G. B., & Cook, D. R. 2004. A
sequence-based genetic map of Medicago truncatula and comparison of marker colinearity
with M. sativa. Genetics 166: 1463-502.
Choi, H.-K., Mun, J.-H., Kim, D.-J., Zhu, H., Baek, J.-M., Mudge, J., Roe, B., Ellis, N., Doyle,
J., Kiss, G. B., Young, N. D., & Cook, D. R. 2004. Estimating genome conservation
between crop and model legume species. Proceedings of the National Academy of Sciences
101: 15289-94.
Churchill, G. A, & Doerge, R W. 1994. Empirical threshold values for quantitative trait
mapping. Genetics 138: 963-71.
Literature cited 71
Cook, D. 1999. Medicago truncatula — a model in the making!Commentary. Current Opinion
in Plant Biology 2: 301-304.
Davitt, A. J., Stansberry, M., & Rudgers, J. A. 2010. Do the costs and benefits of fungal
endophyte symbiosis vary with light availability? The New phytologist 188: 824-34.
De Mita, S., Ronfort, J., McKhann, H. I., Poncet, C., El Malki, R., & Bataillon, T.. 2007.
Investigation of the demographic and selective forces shaping the nucleotide diversity of
genes involved in nod factor signaling in Medicago truncatula. Genetics 177: 1-3.
De Mita, S., Santoni, S, Hochu, I., Ronfort, J, & Bataillon, T. 2006. Molecular evolution and
positive selection of the symbiotic gene NORK in Medicago truncatula. Journal of
Molecular Evolution 62: 234-244.
Denison, R. F. 2000. Legume Sanctions and the Evolution of Symbiotic Cooperation by
Rhizobia. The American Naturalist 156: 567-576.
Dias, P. M. B., Brunel-Muguet, S., Dürr, C., Huguet, T., Demilly, D., Wagner, M.-H., & Teulat-
Merah, B. 2011. QTL analysis of seed germination and pre-emergence growth at extreme
temperatures in Medicago truncatula. Theoretical and Applied Genetic 122: 429-44.
Djébali, N., Jauneau, A., Ameline-Torregrosa, C., Chardon, F., Jaulneau, V., Mathé, C., Bottin,
A., Cazaux, M., Pilet-Nayel, M.-L., Baranger, A., Aouani, M. E., Esquerré-Tugayé, M.-T.,
Dumas, B., Huguet, T., & Jacquet, C. 2009. Partial resistance of Medicago truncatula to
Aphanomyces euteiches is associated with protection of the root stele and is controlled by a
major QTL rich in proteasome-related genes. Molecular Plant-Microbe Iinteractions 22:
1043-55.
Doerge, R. W. 2002. Mapping and analysis of quantitative trait loci in experimental populations.
Nature Reviews Genetics 3: 43-52.
Ehinger, M., Koch, A. M., & Sanders, I. R. 2009. Changes in arbuscular mycorrhizal fungal
phenotypes and genotypes in response to plant species identity and phosphorus
concentration. The New Phytologist 184: 412-23.
Ehrenreich, I. M., Stafford, P. A, & Purugganan, M. D. 2007. The genetic architecture of shoot
branching in Arabidopsis thaliana: a comparative assessment of candidate gene associations
vs. quantitative trait locus mapping. Genetics 176: 1223-36.
Endre, G., Kereszt, A., Kevei, Z., Mihacea, S., Kalo, P, & Kiss, G. B. 2002. A receptor kinase
gene regulating symbiotic nodule development. Nature 417: 962-966.
Eshel, I., & Cavalli-Sforza, L. L. 1982. Assortment of encounters and evolution of
cooperativeness. Proceedings of the National Academy of Sciences 79: 1331-1335.
Literature cited 72
Fenster, C., Galloway, L. F., & Chao, L. 1997. Epistasis and its consequences for the evolution
of natural populations. Trends in Ecology & Evolution 12: 282-286.
Ferguson, B. J., Indrasumunar, A., Hayashi, S., Lin, M.-H., Lin, Y.-H., Reid, D. E., & Gresshoff,
P. M. 2010. Molecular analysis of legume nodule development and autoregulation. Journal
of Integrative Plant Biology 52: 61-76.
Foster, K. R., & Wenseleers, T. 2006. A general model for the evolution of mutualisms. Journal
of Evolutionary Biology 19: 1283-93.
Fry, J. D. 1993. The ― General Vigor ‖ Problem : Can Antagonistic Pleiotropy be Detected when
Genetic Covariances are Positive ? Evolution 47: 327-333.
Fry, J. D., Nuzhdin, S. V., Pasyukova, E. G., & Mackay, T. F. 1998. QTL mapping of genotype-
environment interaction for fitness in Drosophila melanogaster. Genetical Research 71:
133-41.
Gardner, K. M., & Latta, R. G. 2006. Identifying loci under selection across contrasting
environments in Avena barbata using quantitative trait locus mapping. Molecular Ecology
15: 1321-33.
Gubry-Rangin, C., Garcia, M., & Béna, G. 2010. Partner choice in Medicago truncatula-
Sinorhizobium symbiosis. Proceedings of the Royal Society B277: 1947-1951.
Hadfield, J. D., Wilson, A. J., Garant, D., Sheldon, B. C., & Kruuk, L. E. B. 2010. The misuse of
BLUP in ecology and evolution. The American Naturalist 175: 116-25.
Hall, T. M. 1999. BioEdit: a user-friendly biological sequence alignment editos and analysis
program for Windows 95/98/NT. Nucleic Acids Symposium Series 41: 95-98.
Hamon, C., Baranger, A., Miteul, H., Lecointe, R., Goff, I. Le, Deniot, G., Onfroy, C., Moussart,
A., Prosperi, J.-M., Tivoli, B., Delourme, R., & Pilet-Nayel, M.-L. 2010. A complex genetic
network involving a broad-spectrum locus and strain-specific loci controls resistance to
different pathotypes of Aphanomyces euteiches in Medicago truncatula. Theoretical and
Applied Genetics 120: 955-70. Heath, Katy D, & Tiffin, Peter. 2009. Stabilizing
mechanisms in a legume-rhizobium mutualism. Evolution 63: 652-62. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/19087187
Heath, K. D., & Tiffin, P. 2007. Context dependence in the coevolution of plant and rhizobial
mutualists. Proceedings of the Royal Society B 274: 1905-1912.
Heath, K. D., & Tiffin, P. 2009. Stabilizing mechanisms in a legume-rhizobium mutualism.
Evolution 63: 652-662.
Literature cited 73
Heath, K. D., Stock, A. J., & Stinchcombe, J R. 2010. Mutualism variation in the nodulation
response to nitrate. Journal of Evolutionary Biology 23: 1-7.
Heath, K. D. 2010. Intergenomic Epistasis and Coevolutionary Constraint in Plants and
Rhizobia. Evolution 64: 1446-1458.
Heath, K. D., & Lau, J. A. 2011. Herbivores alter the fitness benefits of a plant–rhizobium
mutualism. Acta Oecologica 37: 87-92.
Heijden, M. G. A. van der, Klironomos, J. N., Ursic, M., Moutoglis, P., Streitwolf-engel, R.,
Boller, T., Wiemken, A., & Sanders, I. R. 1998. Mycorrhizal fungal diversity determines
plant biodiversity, ecosystem variability and productivity. Nature 74: 69-72.
Hoeksema, J D, & Thompson, J N. 2007. Geographic structure in a widespread plant-
mycorrhizal interaction: pines and false truffles. Journal of Evolutionary Biology 20: 1148-
63.
Holland, J. B. 1998. EPISTACY: a SAS program for detecting two-locus epistatic interactions
using genetic marker information. Journal of Heredity 89: 374-375.
Jander, K. C., & Herre, E. A. 2010. Supplementary Material. Proceedings Of The National
Academy Of Sciences 7: 1-3.
Jansen, R. C. 1993. Interval mapping of multiple quantitative trait loci. Genetics 135: 205-211.
Jiang, C., & Zeng, Z. B. 1995. Multiple trait analysis of genetic mapping for quantitative trait
loci. Genetics 140: 1111-1127.
Johnson, N. C. 2010. Resource stoichiometry elucidates the structure and function of arbuscular
mycorrhizas across scales. The New Phytologist 185: 631-47.
Jones, K. M., Kobayashi, H., Davies, B. W., Taga, M. E., & Walker, G. C. 2007. How rhizobial
symbionts invade plants: the Sinorhizobium-Medicago model. Nature Reviews.
Microbiology 5: 619-33.
Juenger, T. E., Sen, S., Stowe, K. A., & Simms, E. L. 2005. Epistasis and genotype-environment
interaction for quantitative trait loci affecting flowering time in Arabidopsis thaliana.
Genetica 123: 87-105.
Julier, B., Flajoulot, S., Barre, P., Cardinet, G., Santoni, S., Huguet, T., & Huyghe, C. 2003.
Construction of two genetic linkage maps in cultivated tetraploid alfalfa ( Medicago sativa )
using microsatellite and AFLP markers. BMC Plant Biology 19: 1-19.
Literature cited 74
Julier, B., Huguet, T., Chardon, F., Ayadi, R., Pierre, J.-B., Prosperi, J.-M., Barre, P., & Huyghe,
C. 2007. Identification of quantitative trait loci influencing aerial morphogenesis in the
model legume Medicago truncatula. Theoretical and Applied Genetics 114: 1391-406.
Kao, C.-H., Zeng, Z-B., & Teasdale, R. D. 1999. Multiple interval mapping for quantitative trait
loci. Genetics 152: 1203-1216.
Keurentjes, J. J. B., Bentsink, L., Alonso-Blanco, C., Hanhart, C. J., Blankestijn-De Vries, H.,
Effgen, S., Vreugdenhil, D., & Koornneef, M. 2007. Development of a near-isogenic line
population of Arabidopsis thaliana and comparison of mapping power with a recombinant
inbred line population. Genetics 175: 891-905.
Kiers, E. T,, Rosseau, R. A., West, S. A., & Denison, R F. 2003. Host sanctions and the legume-
Rhizobium mutualism. Nature 425: 78-81.
Kiers, E. T., & Denison, R. F. 2008. Sanctions, Cooperation, and the Stability of Plant-
Rhizosphere Mutualisms. Annual Review of Ecology, Evolution, and Systematics 39: 215-
236.
Kouchi, H., Imaizumi-Anraku, H., Hayashi, M., Hakoyama, T., Nakagawa, T., Umehara, Y.,
Suganuma, N., & Kawaguchi, M. 2010. How many peas in a pod? Legume genes
responsible for mutualistic symbioses underground. Plant & Cell Physiology 51: 1381-97.
Kroymann, J., & Mitchell-Olds, T. 2005. Epistasis and balanced polymorphism influencing
complex trait variation. Nature 435: 95-8.
Laguerre, G., Depret, G., Bourion, V., & Duc, G. 2007. Rhizobium leguminosarum bv. viciae
genotypes interact with pea plants in developmental responses of nodules, roots and shoots.
New Phytologist 176: 680-690.
Lande, R. 1979. Quantitative Genetic Analysis of Multivariate Evolution , Applied to Brain :
Body Size Allometry. Evolution 33: 402-416.
Lander, E. S., & Botstein, D. 1989. Mapping Mendelian factors underlying quantitative traits
using RFLP linkage maps. Genetics 121: 185-199.
Leigh, E. G. 2010. The evolution of mutualism. Journal of Evolutionary Biology 23: 2507-2528.
Leips, J., & Mackay, T. F. 2000. Quantitative trait loci for life span in Drosophila melanogaster:
interactions with genetic background and larval density. Genetics 155: 1773-88.
Lieven-Antoniou, C. A., & Whittam, T. S. 1997. Specificity in the symbiotic association of Lotus
corniculatus and Rhizobium loti from natural populations. Molecular Ecology 6: 629-639.
Liu, B. H. 1998. Statistical genomics: linkage, mapping and QTL analysis. Florida: CRC Press.
Literature cited 75
Lynch, M., & Walsh, B. 1998. Genetics and analysis of quantitative traits. Sunderland: Sinauer
Associaties, Inc Publishers.
Lévy, J., Bres, C., Geurts, R., Chalhoub, B., Kulikova, O., Duc, G., Journet, E.-P., Ané, J.-M.,
Lauber, E., Bisseling, T., Dénarié, J., Rosenberg, C., & Debellé, F. 2004. A putative Ca2+
and calmodulin-dependent protein kinase required for bacterial and fungal symbioses.
Science 303: 1361-4.
Mackay, T. F. C. 2001. Quantitative trait loci in Drosophila. Nature Reviews Genetics 2: 11-20.
Malmberg, R. L., Held, S., Waits, A., & Mauricio, R. 2005. Epistasis for fitness-related
quantitative traits in Arabidopsis thaliana grown in the field and in the greenhouse.
Genetics 171: 2013-27.
Marr, D. L., & Pellmyr, O. 2003. Effect of pollinator-inflicted ovule damage on floral abscission
in the yucca-yucca moth mutualism: the role of mechanical and chemical factors. Oecologia
136: 236-43.
Mauricio, R. 2001. Mapping quantitative trait loci in plants: uses and caveats for evolutionary
biology. Nature Reviews Genetics 2: 370-81.
May, R. M. 1976. Model for two interacting populations. In R. M. May (Ed.), Theoretical
Ecology: Principles and Applications. Philidelphia: W. B. Saunders Company (pp. 49-70).
May, R. M. 1974. Stability and complexity in model ecosystems. Baskerville: Princeton
University Press.
Mhadhbi, H., Jebara, M., Limam, F., Huguet, T., & Aouani, M. E. 2005. Interaction between
Medicago truncatula lines and Sinorhizobium meliloti strains for symbiotic efficiency and
nodule antioxidant activities. Physiologia Plantarum 124: 4-11.
Molofsky, J., Bever, J D, & Antonovics, J. 2001. Coexistence under positive frequency
dependence. Proceedings of the Royal Society B 268: 273-7.
Moran, N. 1981. Intraspecific variability in herbivore performance and host quality: a field study
of Uroleucon caligatum (Homoptera: Aphididae) and its Solidago hosts (Asteraceae).
Ecological Entomology 6: 301-306.
Moreau, D., Salon, C., & Munier-Jolain, N. 2006. Using a standard framework for the
phenotypic analysis of Medicago truncatula: an effective method for characterizing the
plant material used for functional genomics approaches. Plant, Cell and Environment 29:
1087-1098.
Mueller, U. G. 2004. Symbiont choice in a fungus-growing ant (Attini, Formicidae). Behavioral
Ecology 15: 357-364.
Literature cited 76
Mun, J.-H., Kim, D.-J., Choi, H.-K., Gish, J., Debellé, F., Mudge, J., Denny, R., Endré, G.,
Saurat, O., Dudez, A.-M., Kiss, G. B., Roe, B., Young, N. D., & Cook, D. R. 2006.
Distribution of microsatellites in the genome of Medicago truncatula: a resource of genetic
markers that integrate genetic and physical maps. Genetics 172: 2541-55.
Mytton, L. R., El-Sherbeeny, M. H., & Lawes, D. A. 1977. Symbiotic variability in Vicia faba. 3.
Genetic effects of host plant, rhizobium strain and of host strain interaction. Euphytica 26:
785-791.
Noë, R. 1990. A Veto game played by baboons: a challenge to the use of the Prisoner’s Delimma
as a paradigm for reciprocite and cooperation. Animal Behaviours 39: 78-90.
Noë, R., & Hammerstein, P. 1994. Biological markets: supply and demand determine the effect
of partner choice in cooperation, mutualism and mating. Behavioral Ecology and
Sociobiology 35: 1-11.
Parker, M. P. 1995. Plant Fitness Variation Caused by Different Mutualist Genotypes. Ecology
76: 1525-1535.
Pellmyr, O, & Hurth, C. J. 1994. Evolutionary stability of mutualism between yucca and yucca
moths. Nature 17: 257-260.
Phan, H. T. T., Ellwood, S. R., Hane, J. K., Ford, R., Materne, M., & Oliver, R. P. 2007.
Extensive macrosynteny between Medicago truncatula and Lens culinaris ssp. culinaris.
Theoretical and Applied Genetics 114: 549-58.
Phillips, P. C. 2008. Epistasis--the essential role of gene interactions in the structure and
evolution of genetic systems. Nature Reviews Genetics 9: 855-67.
Piculell, B. J., Hoeksema, J. D., & Thompson, J. N. 2008. Interactions of biotic and abiotic
environmental factors in an ectomycorrhizal symbiosis, and the potential for selection
mosaics. BMC Biology 6: 23.
Pierre, J.-B., Huguet, T., Barre, P., Huyghe, C., & Julier, B. 2008. Detection of QTLs for
flowering date in three mapping populations of the model legume species Medicago
truncatula. Theoretical and Applied Genetics 117: 609-20.
Price, T. 1987. Diet Variation in a Population of Darwin ’ s Finches. Ecology 68: 1015-1028.
Pyke, D. A. 1987. Demographic Responses of Bromus tectorum and Seedlings of Agropyron
spicatum to Grazing by Small Mammals : The Influence of Grazing Frequency and Plant
Age. Journal of Ecology 75: 825-835.
Literature cited 77
Rangin, C., Brunel, B., Cleyet-Marel, J.-C., Perrineau, M.-M., & Béna, G. 2008. Effects of
Medicago truncatula genetic diversity, rhizobial competition, and strain effectiveness on the
diversity of a natural sinorhizobium species community. Applied and Environmental
Microbiology 74: 5653-61.
Richter, K. S., & Weis, A. E. 1995. Differential abortion in the yucca. Nature 376: 557-558.
Rieseberg, L. H., Archer, M. A., & Wayne, R. K. 1999. Transgressive segregation , adaptation
and speciation. Journal of Heredity 83.
Rieseberg, L. H., Widmer, A., Arntz, a M., & Burke, J. M. 2003. The genetic architecture
necessary for transgressive segregation is common in both natural and domesticated
populations. Philosophical transactions of the Royal Society of London B 358: 1141-7.
Robinson, K., Beyene, D., Berkum P, van, Knight-Mason, R., & Bhardwaj, H. 2000. Variability
in plant-microbe interaction between Lupinus lines and Bradyrhizobium strains. Plant
Science 159: 257-264.
Sachs, J. L., Mueller, U G, Wilcox, T P, & Bull, J. J. 2004. The evolution of cooperation. The
Quarterly Review of Biology 79: 135-160.
Sachs, J. L., & Simms, E. L. 2006. Pathways to mutualism breakdown. Trends in Ecology &
Evolution 21: 585-92.
Sachs, J. L., & Wilcox, T. P. 2006. A shift to parasitism in the jellyfish symbiont Symbiodinium
microadriaticum. Proceedings of the Royal Society B 273: 425-9.
Salvaudon, L., Héraudet, V., & Shykoff, J. A. 2005. Parasite-host fitness trade-offs change with
parasite identity: genotype-specific interactions in a plant-pathogen system. Evolution 59:
2518-24.
Sankaran, R. P., Huguet, T., & Grusak, M. A. 2009. Identification of QTL affecting seed mineral
concentrations and content in the model legume Medicago truncatula. Theoretical and
Applied Genetics 119: 241-53.
Schauser, L., Roussis, A., Stiller, J., & Stougaard, J. 1999. A plant regulator controlling
development of symbiotic root nodules. Nature 402: 191-195.
Service, P. 1984. Genotypic interactions in an aphid-host plant relationship : Uroleucon
rudbeckiae and Rudbeckia laciniata. Oecologia 61: 271-276.
Service, P. M., & Rose, M. R. 1985. Genetic Covariation Among Life-History Components : The
Effect of Novel Environments. Evolution 39: 943-945.
Literature cited 78
Simms, E. L., & Taylor, D. L. 2002. Partner choice in nitrogen-fixation mutualisms of legumes
and rhizobia. Integrative and Comparative Biology 42: 369-380.
Smith, D. C., & Douglas, A. E. 1987. The biology of symbiosis (D. C. Smith & A. E. Douglas,
Eds.). London: Edward Arnold.
Soberon Mainero, J., & Martinez del Rio, C. 1985. Cheating and taking advantage in mutualistic
associations. In D. B. Boucher (Ed.), The Biology of Mutualism. Sydney: Croom Helm (pp.
192-216).
Somasegaran, P., & Hoben, H. J. 1994. Handbook for rhizobia: Methods in Legume-rhizobium
technology. New York: Springer-Verlag.
Steinmetz, L. M., Sinha, H., Richards, D. R., Spiegelman, J. I., Oefner, P. J., McCusker, J. H., &
Davis, R. W. 2002. Dissecting the architecture of a quantitative trait locus in yeast. Nature
416: 326-30.
Storey, J. D. 2002. A direct approach to false discovery rates. Journal of the Royal Statistical
Society B 64: 479-498.
Storey, J. D., & Tibshirani, R. 2003. Statistical significance for genomewide studies.
Proceedings of the National Academy of Sciences 100: 9440-5.
Tagu, D., Bastien, C., Faivre-Rampant, P., Garbaye, J., Vion, P., Villar, M., & Martin, F. 2005.
Genetic analysis of phenotypic variation for ectomycorrhiza formation in an interspecific F1
poplar full-sib family. Mycorrhiza 15: 87-91.
Tétard-Jones, C., Kertesz, M. A., Gallois, P., & Preziosi, Richard F. 2007. Genotype-by-
genotype interactions modified by a third species in a plant-insect system. The American
Naturalist 170: 492-9.
Tétard-Jones, C., Kertesz, M. A., & Preziosi, R. F. 2011. Quantitative trait loci mapping of
phenotypic plasticity and genotype-environment interactions in plant and insect
performance. Philosophical Transactions of the Royal Society B 366: 1368-1379.
Thompson, J N. 1988. Variation in Interspecific Interactions. Annual Review of Ecology and
Systematics 19: 65-87.
Thompson, J N. 2005. The geographic mosaic of coevolution (J N Thompson, Ed.). Chicago:
University of Chicago Press.
Thrall, P. H., Burdon, J.J., & Woods, Matthew J. 2000. Variation in the effectiveness of
symbiotic associations between native rhizobia and temperate Australian legumes:
interactions within and between genera. Journal of Applied Ecology 37: 52-65.
Literature cited 79
Thrall, P. H., Laine, A.-L., Broadhurst, L. M., Bagnall, D. J., & Brockwell, J. 2011. Symbiotic
Effectiveness of Rhizobial Mutualists Varies in Interactions with Native Australian Legume
Genera. PLoS ONE 6: e23545.
Trivers, R. L. 1971. The evolution of reciprocal altruism. The Quarterly Review of Biology 46:
35-57.
Vale, P. F., & Little, T. J. 2009. Measuring parasite fitness under genetic and thermal variation.
Journal of Heredity 103: 102-9.
Van Ooijen, J. 1992. Accuracy of mapping quantitative trait loci in autogamous species.
Theoretical and Applied Genetics 84-84: 803-811.
Van Ooijen, J. 2006. JoinMap 4, Software for the calculation of genetic linkage maps in
expeirmental populations.
Voorrips, R. E. 2002. MapChart: Software for the Graphical Presentation of Linkage Maps and
QTLs. Journal of Heredity 93: 77-78.
Wade, M. J. 2007. The co-evolutionary genetics of ecological communities. Nature Reviews
Genetics 8: 185-95.
Wang, S., Basten, C. J., & Zeng, Z-B. 2011. Windows QTL Cartographer 2.5.
Weinig, C., Stinchcombe, J. R., & Schmitt, J. 2003. QTL architecture of resistance and tolerance
traits in Arabidopsis thaliana in natural environments. Molecular Ecology 12: 1153-63.
Weller, J. I. 1986. Maximum likelihood techniques for the mapping and analysis of quantitative
trait loci. Biometrics 42: 627-640.
West, S. A., Kiers, E. T., Pen, I., & Denison, R. F. 2002. Sanctions and mutualism stability:
when should less beneficial mutualists be tolerated. Journal of Evolutionary Biology 15:
830-837.
Weyl, E. G., Frederickson, M. E., Yu, D. W., & Pierce, N. E. 2010. Economic contract theory
tests models of mutualism. PNAS 107: 15712-15716.
Wilkinson, H. H., & Parker, M. A. 1996. Symbiotic specialization and the potential for genotypic
coexistence in a plant-bacterial mutualism. Oecologia 108: 361-367.
Young, N. D., Cannon, S. B., Sato, S., Kim, D., Cook, D. R., Town, C. D., Roe, B. A., & Tabata,
S. 2005. Sequencing the Genespaces of Medicago truncatula and Lotus japonicus. Plant
Physiology 137: 1174-1181.
Literature cited 80
Young, N. D., & Udvardi, M. 2009. Translating Medicago truncatula genomics to crop legumes.
Current Opinion in Plant Biology 12: 193-201.
Yu, J., Holland, J. B., McMullen, M. D., & Buckler, E. S. 2008. Genetic design and statistical
power of nested association mapping in maize. Genetics 178: 539-51.
Zeng, Z-B. 1994. Precision Mapping of Quantitative Trait Loci. Genetics 136: 1457-1468.
Zeng, Z-B. 1993. Theoretical basis for separation of multiple linked gene effects in mapping
quantitative trait loci. Proceedings of the National Academy of Sciences 90: 10972-10976.
Zeng, Z-B., Kao, C. H., & Basten, C. J. 1999. Estimating the genetic architecture of quantitative
traits. Genetical Research 74: 279-89.
Appendix 81
Appendix
Figure A-1. Preliminary leaf count data collected on the parental lines of the LR03 RIL mapping
population grown with rhizobium strains Naut and Sals (Heath, unpublished). White = female
parent, line 530; black = male parent, line 735. Means are shown + SE (n= 20 for each plant
line). Rhizobium strain × parental line = 0.02.
Appendix 82
Figure A-2. Linkage map of the LR03 RIL mapping population, constructed in JoinMap 4.0.
Number of markers =184. Bolded markers are anchored to the M. truncatula genome. The
genetic distances (in cM) between markers are indicated on the left side of the linkage groups.
Appendix 83
Table A-1. Information on the parental lines of the LR03 RIL mapping population. From
www.montpellier.inra.fr/BRC-MTR
Parental lines
Information F833005-5 DZA045-5
Line number
Origin; altitude
L000530
France, Salernes; 261 m
L00735
Algeria, Annaba; 100 m
Latitude, longitude 43.5 N, 6.23 E 36.9 N, 7.7 E
Weight of 1000 seeds (mg)
Weight of 100 pods (mg)
4238
6104
4340
10574
Pod spines Short Long
Flowering time Later than A17 Later than A17
Frost Tolerant Sensitive
Appendix 84
Table A-2. Primer sequences used for sequencing Nod factor signaling genes in parental lines and RILs All primers were initially used on
the parental lines of the LR03 RIL population to look for SNPs. Only those primers indicated with as asterisk were sequenced in the RILs.
Fragments Forward primer Reverse primer Source
DMI1
DMI1-MT1* AATACATACACATAAAAGGAATC CATCTACCATATAAGCAACTCT De Mita et al. 2007
DMI3
DMI3.3* TCTTGAGCTTTGTTCCGGTGGTGA AGATGTGAGCTACCGTGTTCCCAA This study
DMI3.4 CGTGATGGAACAGTTGACATGCGT TGTGTGCATTACCCTGAGCATGGA This study
NFP
NFP-1* TTACATGCCCTGTGGATTCTCCTC ATCTGCAGTCTCGGACGATGAAGT This study
NFP-2 AAGTACTTCATCGTCCGAGACTGC CTGCCAAAGAAGCCAAACTTAGAGC This study
NIN
NIN-1 TTGAGGAGCTGTTGGGAGAAGGTT TCCATTATCTCGTTCACCGCTGCT This study
NIN-2 GTGGTGCATCAGGTTGTGGTGTTT TACTGCTCTGATCATGCTGCTGCT This study
NIN-4 GTGAAGGCAACTTTCGCGGATGAA CTGCTGTTGCGGAAAGTGTTTGGA This study
NORK
NORK-1 GGTTGATATTGTCCGCGAGT TGACAAGGTTTGGGTTGTGA This study
NORK-2 ATATTGTCCGCGAGTTGGAG ATCTCGGTTGAGGGTGTGAC This study
NORK-3 CCCCTTTTGAATGCCTATGA TTTTCTTGGTTGTGCAGCAG This study
NORK-4 AACTCAGGGAACCCGAGAAT GTAACCCAGAAG AGGCACCA This study
Appendix 85
Table A-3. PCR reagent concentrations and volumes (in µL) used for sequencing Nod factor
signaling genes in parental lines and RILs. All reagents were obtained from Froggabio.
Reagent
DMI1
DMI3
NFP
NORK
NIN
ddH20 4.2 7.8 7.8 4.2 7.8
Betaine (5 M) 4 4 4 4 4
MgCl2 (25 mM) 2 2 2 2 2
10x PCR buffer 2 2 2 2 2
Forward primer 2 0.2 0.2 2 0.2
Reverse primer 2 0.2 0.2 2 0.2
dNTP (10 mM) 1.6 1.6 1.6 1.6 1.6
iTaq (5U/µL) 0.2 0.2 0.2 0.2 0.2
DNA template 2 2 2 2 2
Appendix 86
Table A-4. Correlations between RIL least-square means for all traits in Naut and Sals treatments. Data presented are the pearson
correlation coefficients between least-square line means calculated using PROC CORR in SAS (v.9.2). P< 0.0001 = ****, P< 0.001 =
***, P<0.01 = **, P<0.05 = *. Correlations above the diagonal are in Naut, correlations below are in Sals.
Leaf
number (6
weeks)
Days to
flowering
Fruit
number
Average
fruit
weight
Shoot
weight
Root
weight
Primary branch
number
Leaf number (6 weeks) -0.26 *** 0.36**** 0.29**** 0.07 0.14 0.29****
Days to flowering -0.36**** 0.08 -0.02 0.43**** 0.31**** 0.34****
Fruit number 0.35**** 0.14 -0.31**** 0.52**** 0.40**** 0.21 **
Average fruit weight 0.27** -0.08 -0.19** -0.11 -0.04 0.10
Shoot weight -0.08 0.59**** 0.49**** -0.15 0.66**** 0.35****
Root weight 0.11 0.50**** 0.51**** 0.05 0.70**** 0.38****
Primary branch number 0.33**** 0.24** 0.23** 0.09 0.18** 0.30****
Appendix 87
Table A-5. Correlations between RIL least-square means for all traits in the across rhizobium strains analysis. Data presented are the
pearson correlation coefficients between least-square line means calculated using PROC CORR in SAS (v.9.2). P< 0.0001 = ****, P<
0.001 = ***, P<0.01 = **, P<0.05 = *.
Leaf
number (6
weeks)
Days to
flowering
Fruit
number
Average
fruit
weight
Shoot
weight
Root
weight
Primary branch
number
Leaf number (6 weeks) -0.29**** 0.37**** 0.29**** -0.008 0.18 ** 0.33****
Days to flowering 0.16 * -0.04 0.57**** 0.46**** 0.39****
Fruit number -0.32**** 0.49**** 0.51**** 0.27 ***
Average fruit weight -0.13 0.02 0.09
Shoot weight 0.70**** 0.31****
Root weight 0.42****
Primary branch number
Appendix 88
Table A-6. Arabidopsis thaliana branching genes and descriptions with the matching Medicago
truncatula orthologs. DNA sequence alignment score obtained from BLASTn.
Gene Description Alignment score
with Medicago
Location in
Medicago genome
Gene
orthologs
MAX2
More
Axillary
Growth 2
F-box protein 320 Chr. 4 Cyclin-like
Fbox
PID
Pinoid
Serine/threonine
kinase
509 Chr. 4 Tyrosine
protein
kinase
PIN1
Pinformed1
Auxin efflux
carrier protein
549 Chr. 6 Auxin
efflux
carrier