Copyright © 2005 by the Genetics Society of AmericaDOI: 10.1534/genetics.104.031948

Mapping Quantitative Trait Loci in Multiple Populations of Arabidopsis thalianaIdentifies Natural Allelic Variation for Trichome Density

V. Vaughan Symonds,*,1 A. Veronica Godoy,*,2 Teresa Alconada,† Javier F. Botto,†

Thomas E. Juenger,‡ Jorge J. Casal† and Alan M. Lloyd*

*Section of Molecular, Cell and Developmental Biology, Institute for Cellular and Molecular Biology and ‡Section of Integrative Biology,University of Texas, Austin, Texas 78712 and †IFEVA, Universidad de Buenos Aires, 1417-Buenos Aires, Argentina

Manuscript received July 13, 2004Accepted for publication December 1, 2004

ABSTRACTThe majority of biological traits are genetically complex. Mapping the quantitative trait loci (QTL) that

determine these phenotypes is a powerful means for estimating many parameters of the genetic architecturefor a trait and potentially identifying the genes responsible for natural variation. Typically, such experimentsare conducted in a single mapping population and, therefore, have only the potential to reveal genomicregions that are polymorphic between the progenitors of the population. What remains unclear is howwell the QTL identified in any one mapping experiment characterize the genetics that underlie naturalvariation in traits. Here we provide QTL mapping data for trichome density from four recombinant inbredmapping populations of Arabidopsis thaliana. By aligning the linkage maps for these four populations ontoa common physical map, the results from each experiment were directly compared. Seven of the nineQTL identified are population specific while two were mapped in all four populations. Our results showthat many lineage-specific alleles that either increase or decrease trichome density persist in naturalpopulations and that most of this genetic variation is additive. More generally, these findings suggest thatthe use of multiple populations holds great promise for better understanding the genetic architecture ofnatural variation.

THE genetic and molecular bases of complex traits regarding how well any one mapping study capturesthis global genetic architecture.are poorly understood. A common approach to this

problem is the use of whole-genome scans to identify Our best initial opportunities to address this issue liewith highly variable traits in model organisms, wherepolygenes or quantitative trait loci (QTL; Barton and

Keightley 2002; Abiola et al. 2003). The results of such strong genetic infrastructures have been established(e.g., see Frary et al. 2000; El-Din El-Assal et al. 2001;analyses provide estimates of several genetic parameters

that underlie phenotypic variation, including the number Robin et al. 2002). To this end, trichome density inArabidopsis thaliana is one such trait (Figure 1, data col-of loci, the type and magnitude of their effects, interactions

between genes (epistasis), and gene-by-environment inter- lected in this study). Trichomes (plant hairs) are presentactions—collectively referred to as the genetic architec- on nearly all land plants and are known to play impor-ture of a trait. However, these parameters are largely tant roles in plant protection, specifically against insectpopulation specific (Lynch and Walsh 1998) and most herbivory (Poos 1929; Gilbert 1971; Kennedy 2003),species show some degree of population genetic struc- drought (Smith and Nobel 1977; Sandquist andture (Hamrick and Godt 1996; Bohonak 1999). There- Ehleringer 2003), and UV radiation (Karabourni-fore, any description of a trait’s genetic architecture otis and Manetas 1995; Liakoura et al. 1997). Themade from a single population, natural or experimen- near ubiquity of trichomes and their morphological di-tal, likely describes only a small part of what might be versity argue for a strong role for selection in theirthought of as the “global genetic architecture”—all of maintenance and ultimate diversification (see Levinthe loci, their effects, and potential interactions that 1973 for review).contribute to standing variation for a trait within a spe- In A. thaliana, the single-celled trichome has becomecies. Although it is certain to vary among traits and a popular system for the study of cellular differentia-organisms, in general, we have very little information tion and development processes (Szymanski et al. 2000;

Larkin et al. 2003). Many of the genes involved in tri-chome initiation and development have been identified

1Corresponding author: Section of Molecular, Cell and Developmental through forward genetics (Koornneef 1981; Koorn-Biology, University of Texas, 2500 Speedway, MBB 1.448b, Austin, TX neef et al. 1982; Hulskamp et al. 1994; Zhang et al.78712. E-mail: vsymonds@mail.utexas.edu

2003) and have led to hypothetical genetic models for2Present address: Instituto de Investigaciones Biologicas, UniversidadNacional de Mar del Plata, 7600-Mar del Plata, Argentina. the pathways that control trichome development (Lar-

Genetics 169: 1649–1658 (March 2005)

1650 V. V. Symonds et al.

population, five replicates of each RIL were randomized acrosskin et al. 2003; Zhang et al. 2003). Despite this body ofPRO72 multicell flats filled with Promix BX soil, vernalizedwork, little is known about the genetics that underliefor 11 days at 4�, and ultimately maintained under 14:10

standing variation for trichome density, a character (light:dark) days in a greenhouse at Brackenridge Field Sta-known to be of selective importance in natural popula- tion in Austin, Texas until time of data collection. All experi-

ments were completed between March and May of 2003; al-tions of A. thaliana (Mauricio 1998). The only pub-though growing each population concurrently is ideal, thelished study of the quantitative genetics of trichomescale of each mapping experiment precluded this. At �24density in A. thaliana (Larkin et al. 1996) provided map-days postgermination, the fifth true leaf from each plant was

ping results from one recombinant inbred line (RIL) removed and the number of trichomes was counted within apopulation and identified a single QTL of very large 25-mm2 area midway between the midrib and leaf edge at the

widest point on the adaxial leaf surface.effect (73% of within-population variation). It is highlyLinkage map construction: Because of differences in markerunlikely, however, that a single gene is responsible for

use between this study and others, all linkage maps were regen-the 20-fold range of natural variation observable in tri-erated from raw marker data. Markers for each population

chome density within this species (Figure 1). To more were selected from all available data sets to optimize genomethoroughly explore the genetic basis of trichome density coverage and reduce missing genotypes. Linkage maps were

constructed using JoinMap 3.0 (Van Ooijen and Voorripsvariation and the potential of multiple-population map-2001) and were considered complete only when unambiguousping, we have mapped QTL for this trait in four RILmarker order was attained.populations derived from six A. thaliana accessions. Our

QTL mapping: The mean trichome density score was calcu-mapping results identify nine QTL, termed trichome lated from replicates of each RIL and used for mapping. QTLdensity loci 1–9. Here we provide mapping results, com- were mapped for each population independently. To identify

regions of the genome that possess significant QTL, we usedparative analyses, and interpretations of our findings.interval mapping and multiple QTL model (MQM) mapping(akin to composite interval mapping) as implemented by theMapQTL 4.0 software package (Van Ooijen et al. 2002). QTLMATERIALS AND METHODS significance thresholds were determined empirically for eachmapping population by permuting the trait data over individu-Natural variation survey for trichome density: To screenals while holding the marker data fixed. One thousand itera-natural variation for trichome density within A. thaliana, 79tions provided a null distribution, from which a 95% signifi-single-seed-descent accessions that represent populationscance threshold was attained. Mapping was conducted via anfrom throughout the species’ range were acquired from theiterative method. Interval mapping was performed first toArabidopsis Biological Resource Center (ABRC). Three repli-identify QTL with significant main effects. The markers mostcates of each accession were planted in 6.35-cm pots (oneclosely associated with each QTL were then selected as cofac-plant per pot) in Promix BX soil (Hummert International),tors for MQM mapping. If new QTL were identified underrandomized across flats, and vernalized for 7 days at 4�. Flatsthe new model, their associated markers were incorporatedwere then moved to 20� under 24-hr light. At 21 days postger-as cofactors for the next round of mapping until no new QTLmination, the fully expanded third true leaf of each plantwere identified. MQM mapping performs the same way aswas removed and all trichomes on the adaxial surface wereinterval mapping while statistically controlling for the effect(s)counted under a dissecting microscope at �35 magnification.of other QTL (cofactors). The only exception to the MapQTLEach leaf was then pressed flat for 24–48 hr, digitally photo-manual-suggested method of mapping was that when estimat-graphed, and the total leaf blade area was measured usinging the LOD score and confidence intervals for any one QTL,ImageJ v1.32 (NIH Image). The number of trichomes permarkers flanking the QTL of interest were deselected as cofac-square centimeter was calculated for each leaf, and the meantors. The effect of each QTL was estimated in MapQTL anddensity for each line was determined. Total phenotypic vari-reflects the percentage of variance within a population thatance was partitioned into genotypic and environmental (er-is explained by the QTL. Hereafter, when QTL “effects” areror) components by ANOVA.discussed, they refer to this measure.Mapping populations: Four RIL mapping populations were

A screen for QTL that interact epistatically was performedused in this study. Seed for three populations, Ler-0 � Col-4using the Pseudomarker multiple-QTL framework presented(Lister and Dean 1993), Ler-2 � CVI (Alonso-Blanco et al.by Sen and Churchill (2001). This methodology employs a1998), and Bay � Sha (Loudet et al. 2002), were acquiredMonte Carlo imputation algorithm to simulate multiple ver-from the ABRC. The fourth RIL population, Ler-2 � No-0,sions of complete genotype information on a dense genome-was developed jointly by the Lloyd (University of Texas, Aus-wide grid; these imputed genotypes are referred to as “pseu-tin) and Casal (Universidad de Argentina) laboratories. Thisdomarkers.” The pseudomarker grid was scanned using bothpopulation was genotyped using microsatellite loci and bothone- and two-QTL models at each position across the genomegenotyping data and seed will be made publicly available inand evidence for a QTL or for QTL-QTL interactions (epista-the near future. All four populations were genotyped in the F8

generation or later, yielding lines that are �99% homozygous; sis) was determined using robust one- and two-dimensionalpermutation tests. Details of model selection and QTL map-genotyping details for each population can be found in the

original descriptions cited above and at The Arabidopsis Infor- ping for epistasis detection followed that of Juenger et al.(2005).mation Service (http://www.arabidopsis.org) and Nottingham

Arabidopsis Stock Center (http://nasc.nott.ac.uk) websites. The Power analyses: To determine whether significant differ-ences in the power to detect QTL exist among the mappingnumbers of RILs and markers used for mapping in each popu-

lation are listed in Table 1. This set of four populations has populations used, a power analysis was performed followinga modification from Falconer and Mackay (1996) and Sokalthe additional benefit of a testcross design. Because three of

the populations share the Ler accession as a parent, their and Rohlf (1997): n � 2(z� � z2�)2/(�/W)2. This analysis isdesigned to determine how many samples (lines) are requiredresults can be directly compared in reference to a tester line.

Scoring the trichome density phenotype: For each mapping to detect a QTL with a particular size effect using single-

1651Genetic Architecture of Trichome Density

marker analysis. However, when rearranged, it can also be usedto identify how small of an effect can be detected given a certainsample size (number of lines). We used standard �- (0.05) and�-(0.10) values (probabilities of type I and type II errors,respectively), with the number of lines and empirical standarddeviation estimates as variables; the model assumes completelinkage (sufficient marker coverage) between markers andQTL. Because the test described by Falconer and Mackay(1996) is based on an F2 population, sample sizes used inthese calculations were adjusted (from number of RILs toequivalent number of F2) according to Soller and Beckman(1990).

Map alignments: Differences in linkage map length betweenpopulations are generally expected due to nonoverlappingmarkers, different marker densities, and presumed variationin recombination rates (Sanchez-Moran et al. 2002). As aconsequence, comparing mapping results from different pop-ulations is difficult. To align the four linkage maps per chro-mosome that were generated here, the linkage positions ofmarkers in each population were aligned with their corre-sponding physical positions on the Col physical map (Arabi-dopsis Genome Initiative 2000). To more closely compareQTL positions across mapping populations, we also estimated Figure 1.—Trichome density variation of the third true leaf.the physical positions of QTL on the basis of the coarse linear (A) Distribution of trichome density on the third true leaf forrelationship between linkage map positions and physical posi- 79 natural accessions of A. thaliana. (B) Scanning electrontions for each marker using a least-squares regression ap- micrographs of the adaxial surface of the third true leaf fromproach. Although there is certain to be intermarker variation 3 natural accessions of A. thaliana that represent the range offor recombination rates (Copenhaver et al. 2002), QTL physi- trichome density, from completely glabrous to densely hairy.cal position estimates are bound by known flanking marker The numbers below each leaf refer to ABRC accession num-alignment positions. For ease of interpretation, QTL mapped bers.in the Ler-0 � Col-4 population were plotted directly on thephysical map. QTL mapped in different populations wereinferred to be the same locus when the physical positionsof markers within 2-LOD support intervals overlapped (see similar spread, but the entire distribution was shiftedresults for assumption details). higher by �10 trichomes.

Linkage maps and alignments: Separate linkage mapswere constructed for each mapping population. The

RESULTS use of different markers, numbers of markers, and pre-sumed differences in recombination rates between pop-Natural variation for trichome density: To examineulations led to considerable variation in total linkagethe distribution and nature of trichome density varia-map lengths, which ranged from 364 to 490 cM (Tabletion within A. thaliana, 79 natural accessions were1). Because of this variation, all linkage maps for a givenscreened. ANOVA revealed highly significant among-chromosome were aligned using the physical positionsline variance (P 0.0005), indicating that a strongof markers for each population (see below).genetic component underlies variation for trichome

Mapping results: Three to five QTL with significantdensity within A. thaliana. The broad-sense heritabilityeffects on trichome density were identified in each map-estimate (VG/VP) for trichome density in this collectionping population. Initially, a total of 15 QTL wereof lines was 0.703. Excluding completely glabrous lines,mapped, with individual effects that explained from 6.8the results of this screen showed a roughly 20-fold rangeto 25% of within-population variation (Table 2). Modelsof variation for trichome density that has a broad, nor-including gene-by-gene interaction showed the pres-mal to bimodal distribution (Figure 1).ence of only weak epistasis (4% effect) between two lociGenetic analysis of mapping populations: ANOVAin the Ler-2 � CVI population (Figure 3). Although theshowed significant variation for trichome density amongtwo parents for the Bay � Sha population failed toRILs within each of the four mapping populations (P germinate, thus preventing an analysis of transgressive0.0001 for each population). Consistent with this, herita-segregation, transgressive phenotypes were observed inbility estimates (VG/VP) were relatively high in all map-the other three populations. Strong genetic evidenceping populations, ranging from 0.795 to 0.896 (Tablefor transgressive segregation (based on allelic effects at1). These heritabilities are also consistent with thosemapped QTL) was clear only in the Ler -2 � CVI andestimated from natural accessions. The frequency distri-Ler -2 � No-0 populations (Table 2), where both parentbutions for trichome density were relatively normal foraccessions possess high and low alleles for trichomeeach mapping population (Figure 2); the three popula-density. In the Ler -0 � Col-4 and Bay � Sha populations,tions that have Ler as a parent show roughly the same

distribution, while the Bay � Sha population had a QTL effects were unidirectional; one parent possesses

1652 V. V. Symonds et al.

TABLE 1

Mapping population details

Linkage group e

Mapping population RILs a Markers b H 2 c QTL d 1 2 3 4 5

Ler -0 � Col-4 92 125 0.795 3 114 68 71 82 100Ler -2 � CVI 103 242 0.896 5 127 82 83 83 115Ler -2 � No-0 96 46 0.877 4 70 88 93 67 83Bay � Sha 144 38 0.892 3 85 63 68 72 76

a Number of recombinant inbred lines used.b Number of markers used in a given population.c Broad-sense heritability estimate.d Number of QTLs identified.e Linkage group lengths in centimorgans for chromosomes I–V.

the low allele at all loci (Ler-0 and Bay) and the other in two ways. First, estimates obtained from the analysesrevealed that QTL with effects as low as 8.4–9.5% (ofparent has all high alleles (Col-4 and Sha).

Power analyses: The power analyses were informative within-population variation) should be detectable in allfour populations. This translates into additive (2a) phe-notypic effects of approximately two trichomes. Second,under the assumption of complete linkage, the differ-ence in power to detect QTL among the four popula-tions is very small; the largest difference between popu-lations is �1.1% effect. These estimates appear to befairly accurate, as the smallest effect detected among allpopulations was 6.8%. It should be noted that thesepower analyses are conservative estimates because theyare based upon single-marker association tests.

QTL overlap between populations: To investigatewhether the same or different QTL were mapped inthe four RIL populations, linkage map positions werealigned with known physical positions for markers andwith estimated physical positions for QTL (Figure 4).In general, there is strong overlap for QTL positionsand 2-LOD support intervals at two loci across all fourpopulations (bottom of chromosome II and top of chro-mosome IV). Although in a different context (mappingQTL for different traits in the same population), theoccurrence of overlapping support intervals is often in-terpreted as evidence either that the same QTL wasmapped or that genes contributing to trait variance arepositionally clustered. Because of the observed overlapand unidirectional allelic effects (Ler has the “low” alleleat both loci that show overlap across populations), itseems likely that the same QTL was mapped in all fourpopulations (considered further in discussion). Theremaining 7 QTL demonstrate no positional overlapamong populations. Thus, the initial 15 QTL were con-densed to 9 independent QTL, trichome density loci(TDL) 1–9.

DISCUSSIONFigure 2.—Trichome density phenotype distributions for

Although the mapping of quantitative trait loci inRILs of each mapping population. Letters above graphs indi-experimental populations has become a commoncate the mean phenotype for the two parents of each popula-

tion. method by which many parameters of a trait’s genetic

1653Genetic Architecture of Trichome Density

TA

BL

E2

Tri

chom

ede

nsit

yQ

TL

deta

ils

Map

pin

gpo

pula

tion

Ler

�C

olL

er�

CV

IL

er�

No-

0B

ay�

Sha

Add

itiv

eA

ddit

ive

Add

itiv

eA

ddit

ive

Ch

rom

osom

ea

QT

Lb

Posi

tion

cPE

def

fect

eL

erf

Posi

tion

PEef

fect

Ler

Posi

tion

PEef

fect

Ler

Posi

tion

PEef

fect

Bay

fC

andi

date

g

1T

DL

136

.211

.63.

0↓

2T

DL

251

.621

.04.

6↓

68.4

19.4

5.2

↓63

.211

.52.

8↓

46.6

12.7

4.2

↓T

TG

23

TD

L3

36.1

22.1

5.6

↓G

L1

3T

DL

451

.011

.84.

1↓

4T

DL

53.

59.

13.

0↓

2.0

12.1

3.0

↓11

.011

.43.

0↓

1.0

25.0

6.0

↓4

TD

L6

48.5

16.5

4.1

↑4

TD

L7

60.2

11.0

2.9

↑5

TD

L8

32.4

7.0

2.6

↑T

TG

15

TD

L9

68.9

6.8

2.4

↓G

L3

QT

Lpo

siti

ons

init

alic

type

indi

cate

loci

that

epis

tati

cally

inte

ract

.a

Ch

rom

osom

eon

wh

ich

QT

Lw

asm

appe

d.bQ

TL

iden

tifi

er.

cM

appo

siti

onin

cen

tim

orga

ns.

dPe

rcen

tage

ofva

riat

ion

wit

hin

map

pin

gpo

pula

tion

expl

ain

edby

the

QT

L.

eA

ddit

ive

effe

ctof

QT

L(2

a).

fD

irec

tion

ofef

fect

ontr

ich

ome

den

sity

ofei

ther

the

Ler

orth

eB

ayal

lele

(dep

endi

ng

onth

epo

pula

tion

)fo

rea

chQ

TL

.gC

andi

date

gen

eba

sed

onph

ysic

alpr

oxim

ity

toQ

TL

posi

tion

.

1654 V. V. Symonds et al.

21% of within-population variation explained) is muchsmaller than that estimated for RTN (73%). In additionto the TDL 2/RTN locus, our mapping results from theLer-0 � Col-4 population identify two other QTL fortrichome density (Table 2; Figure 4). The differencesbetween the two studies in the number of QTL mappedand their estimated effects may be due to differencesin the phenotype scored [number of trichomes on firsttrue leaves (Larkin et al. 1996) vs. trichome densityon fifth true leaves (this study)], methods of analysis,different environmental conditions, or some combina-tion of these. Although any QTL mapping study canidentify only a minimum number of loci that affecta trait, our findings highlight the power afforded bymapping in multiple populations. Our results from theother three populations identify six more QTL, for atotal of eight not previously described for trichome den-

Figure 3.—QTL with significant epistatic interactions insity, several of which are of relatively large effect.the Ler-2 � CVI population. Plotted points represent two-

Overlapping QTL: The map positions and 2-LODlocus genotype means �SE for TDL 2 � TDL 6.support intervals for QTL in only two genomic regions[bottom of chromosome II (TDL 2) and top of chromo-some IV (TDL 5)] show strong overlap across all four

architecture are characterized (Abiola et al. 2003), the populations studied (Figure 4). As QTL mapped hereextent to which these specific parameters differ across have 2-LOD support intervals that span an average ofpopulations remains largely unexplored. Here, we have �1100 genes, positionally overlapping QTL may havebroadly surveyed natural lineages of A. thaliana for tri- been mapped due to variation at different loci or varia-chome density and detected a roughly 20-fold range tion within the same locus. If we posit the latter, still manyof variation. To better characterize the genetics that scenarios may explain this result; at each locus, there mayunderlie this variation and to investigate genetic archi- be as many as six or as few as two alleles segregatingtecture heterogeneity, we mapped QTL in four recombi- among the parental accessions. The most parsimoniousnant inbred line populations. The results from each explanation for overlapping QTL across these popula-population were compared by aligning the linkage maps tions is that the Ler accession possesses an allele thatfor each chromosome across all four populations. Our differs in effect significantly from the Col-4, CVI, andresults identify at least nine QTL with significant effects No-0 alleles, all of which may or may not possess theon trichome density; two QTL were mapped in all four same allele at each locus. This scenario is supported bypopulations and seven were mapped in one population the fact that the Ler allele confers lower trichome den-each. These data allow us to (1) characterize the nature sity at TDL 2 and TDL 5 in all three populations forof quantitative genetic variation for trichome density in which it is a parent. Whether the alleles segregatingA. thaliana, (2) compare the results of mapping QTL in in the Bay � Sha population are the same as thosemultiple populations derived from different individuals, segregating in the other three populations is not clear.and (3) initiate a search for candidate genes that may However, the effect of TDL 5 in the Bay � Sha popula-play a role in standing variation for the trait. tion is more than double that of any of the other three

Trichome density QTL study comparison: A total of (Table 2), suggesting that, if it is the same locus, thennine QTL that affect trichome density were identified the allele(s) present are different or the effects of thehere, with at least one QTL occurring on each of the alleles are conditional on genetic background.five A. thaliana chromosomes (Table 2; Figure 4). These Population-specific QTL: The genetic architectureresults are in sharp contrast to those reported by Larkin elucidated for any trait is expected to be populationet al. (1996) in the only related effort to map QTL for specific. While among-population variation for genetictrichome density in A. thaliana. Their study focused on architecture has been detected by classic quantitativethe number of trichomes that develop on first true leaves genetic approaches (e.g., Donohue et al. 2000; Morganin a single RIL population, Ler -0 � Col-4, and identified et al. 2001; Ashman 2003), little has been described onone QTL (reduced trichome number, RTN). The re- exactly how genetic architecture for a trait varies amongported map position of RTN overlaps with a QTL for natural or experimental populations; specifically, wetrichome density identified here, TDL 2. This locus was would like to know which components of genetic archi-mapped in all four populations examined, including tecture vary and in what way(s). Although the experi-the same Ler -0 � Col-4 population studied by Larkin mental design employed here precludes an examination

of every aspect of genetic architecture, our analyses re-et al. (1996), but the effect estimated in this study (11.5–

1655Genetic Architecture of Trichome Density

Figure 4.—Linkage map alignments for chromosomes I–V. Linkage maps for the Ler -2 � CVI (LCV), Ler -2 � No-0 (LNo),and Bay � Sha (BSh) populations are represented by open bars and are aligned with the physical map of Col (solid bar). Small,black horizontal lines within bars indicate QTL positions and the gray area around each QTL represents the 2-LOD supportinterval; QTL mapped in the Ler -0 � Col-4 population are shown on the Col physical map. The linkage positions of markersused in each population are aligned with their corresponding physical position on the Col map (gray lines). Dashed lines wereused to distinguish lines connecting markers to the Ler -2 � No-0 population from those connecting to the Bay � Shah population.To compare the positions of QTL mapped in different populations, each was connected to its estimated physical position onthe physical map (red lines). The positions of candidate genes are shown in yellow, with their corresponding gene name at theright.

veal the strongest evidence of genetic architecture varia- segregation of meaningful genetic variation among nat-ural lineages. Because population-specific QTL of rela-tion by the mapping of different QTL in different popu-

lations (Table 2; Figure 4). Differences in the QTL tively small effects were mapped in all four populations,it does not appear that RIL population size or markeridentified for a trait among RIL populations may be

due to differences in the power to detect QTL, environ- coverage contributed significantly to the mapping ofdifferent QTL in different populations. To investigatemental variation between mapping experiments, or the

1656 V. V. Symonds et al.

the potential for a significant gene-by-environment com- is additive in nature; however, it should be noted thatthe Pseudomarker analyses performed are very conser-ponent, subsamples of each population were grown con-

currently several months after the initial experiments and vative and may underestimate the epistatic contributionto trichome density variation.scored in the same manner as before (data not shown).

Comparisons between initial and subsequent RIL means Transgressive segregation: Recently it was argued thatthe genetic architecture necessary for transgressive seg-showed no significant differences (paired t-test, P �

0.355). Taken together, these analyses suggest that the regation is common in both natural and domesticatedpopulations (Rieseberg et al. 2003). Our data show thatpopulation-specific QTL detected are due to natural

genetic variation for trichome density that is partitioned in the three populations for which parent phenotypeswere observed, transgressive segregation was evident foramong accessions of A. thaliana. This is further sup-

ported by the broad distribution of natural trichome trichome density (Figure 2). However, only two of thefour mapping populations showed strong genetic evi-density phenotypes and a high among-accession herita-

bility estimate (Figure 1). dence of transgressive segregation (Ler -2 � CVI andLer -0 � No-0); that is, for only two populations doesWhat do these mapping population-specific QTL

mean? QTL are identified on the basis of a significant each parent possess at least one “high” and one “low”trichome density allele among the detected QTL. Thedifference in the mean phenotype of RILs possessing

alternative alleles. Therefore, any QTL identified in only observation of transgressive phenotypes in populationswhere QTL allelic effects are unidirectional may reflectone of the four mapping populations is likely due to

an accession-specific allele, otherwise the polymorphism QTL of relatively small effects that were undetected.Although all populations studied here are experimen-also would have been mapped in other populations.

This hypothesis and the origin of novel alleles are most tal, they were derived from (mostly) natural lineagesof A. thaliana and these results provide some insighteasily explored by, again, comparing mapping results

from the three populations that share the Ler accession regarding the potential for transgressive segregation innatural populations. Specifically, relative to one an-as a parent; these three populations are, effectively, a

set of testcrosses. For example, TDL 3 was mapped only other, at least three of the six parent accessions possessa genetically unique way of increasing trichome densityin the Ler -0 � Col-4 population (Figure 4). Because this

locus was not mapped in the other two populations, and at least three possess a genetically unique way ofdecreasing trichome density. So, even though all fourthe Ler allele at this locus is functionally, perhaps also

practically, the same allele as those possessed by CVI experimental populations do not show strong geneticevidence of transgressive segregation, the finding ofand No-0. Therefore, the Col-4 accession must possess

an allele at TDL 3 that is not present in either of the several lineage-specific alleles that alter trichome den-sity in the same way shows that sufficient intraspecificother mapping populations. By this reasoning, any QTL

mapped in one of these populations and not the other genetic variation exists to generate this phenomenonin natural populations of A. thaliana.two represents an allele sampled only in the non-Ler

accession of the population in which the QTL was Candidate loci and phenotype regulation: The com-plete genome sequence of A. thaliana and a rich tri-mapped. Because the Bay � Sha population does not

share a parent with any other mapping population, re- chome development literature enabled the search forcandidate genes near mapped QTL. For four of the ninesults from this population cannot be compared in the

same way as above. However, TDL 4 was mapped only QTL mapped here, strong candidate genes lie within orvery near ascribed confidence intervals (Table 2; Figurein this population, so at least one of these two accessions

must possess an allele not sampled in the other mapping 4). One of the QTL maps near a gene for which knock-outs have known quantitative effects on trichome den-populations. In all, seven population-specific QTL were

mapped, each of which reveals an accession-specific al- sity (TDL 9), and three map near known regulators oftrichome initiation (TDL 2, 3, and 8). Notably, all oflele that either increases (TDL 1, 3, and 9) or decreases

(TDL 6, 7, and 8) the trichome density phenotype rela- the A. thaliana genes identified by conventional geneticsthat affect trichome initiation, including these four can-tive to Ler ; the effect of the novel allele(s) at TDL 4

cannot be inferred from the data. didate loci, encode either proteins with transcriptionfactor motifs or a protein that directly interacts withEpistasis: Among the QTL identified here, significant

gene-by-gene interaction (deviation from an additive transcription factors (Larkin et al. 2003; Zhang et al.2003). Although striking, this finding is in accord withmodel) was detected between only two loci in only one

population (Ler -2 � CVI); this interaction is responsible early and recent speculation that quantitative trait varia-tion results from regulatory variants more often thanfor 4% of within-population variation, which is less than

that of any main effect QTL. Interestingly, one of these from coding sequence variants (King and Wilson 1975;Mackay 2001; Korstanje and Paigen 2002); theseQTL was mapped in all four populations (TDL 2) and

the other was identified only in the Ler -2 � CVI popula- suppositions focus mainly on the contrast between muta-tions within regulatory regions vs. mutations within cod-tion (TDL 6). In general, it seems that the majority of

genetic variation for trichome density within A. thaliana ing regions. However, recent evidence shows that pro-

1657Genetic Architecture of Trichome Density

tein-coding change in transcription factors may play a maps and that a multiple mapping population approachlarger role in pathway evolution than previously thought has the potential to much better characterize the spe-(Hsia and McGinnis 2003). Therefore, this contrast cies-wide genetic architecture for a trait. Fortunately, formay be quite naturally extended to include mutations several model systems, the development of new mappingin regulatory regions and mutations within coding re- populations is under way, which will provide opportuni-gions of regulatory genes vs. coding mutations in struc- ties to more broadly characterize the genetic architec-tural genes alone. Indeed, of the 10 QTL thus far cloned ture of complex traits and eventually aid in identifyingfrom plants (Paran and Zamir 2003), at least one-half the molecular variants responsible for natural variation.are transcription factors or the phenotype has been We thank Ulrich Mueller, Don Levin, and two anonymous reviewersascribed to altered gene expression levels. Of course, for helpful comments on the manuscript and John Mendenhall for

technical assistance. This work was supported by National Scienceprecisely determining the genetic sources of variationFoundation grants DEB-0206610 and MCB-0114976.in trichome density will require the identification and

molecular characterization of mapped QTL, and eachof the candidate loci identified here provides an oppor-tunity to do just this. LITERATURE CITED

Conclusions: There are many hypotheses regardingAbiola, O., J. M. Angel, P. Avner, A. A. Bachmanov, J. K. Belknapthe genetic architecture of natural trait variation, and et al., 2003 The nature and identification of quantitative trait

it seems clear that the correct hypothesis will depend loci: a community’s view. Nat. Rev. Genet. 4: 911–916.Alonso-Blanco, C., A. J. Peeters, M. Koornneef, C. Lister, C. Deanon the particular trait and organism under study. Very

et al., 1998 Development of an AFLP based linkage map of Ler,generally, standing variation for quantitative traits may Col, and Cvi Arabidopsis thaliana ecotypes and construction of abe due to few genes with many mutations, each with Ler/Cvi recombinant inbred line population. Plant J. 14: 259–271.

Arabidopsis Genome Initiative, 2000 Analysis of the genome se-their own distribution of effects; several genes with fewquence of the flowering plant Arabidopsis thaliana. Nature 408:or many segregating mutations with intermediate to796–815.

small effects; or some combination of these. As these Ashman, T. L., 2003 Constraints on the evolution of males andsexual dimorphism: field estimates of genetic architecture ofparameters apply directly to the potential for selectionreproductive traits in three populations of gynodioecious Fragariaresponse and morphological evolution in general (Bar-virginiana. Evolution 57: 2012–2025.

ton and Keightley 2002), understanding the frequen- Barton, N. H., and P. D. Keightley, 2002 Understanding quantita-tive genetic variation. Nature Rev. 3: 11–21.cies at which different genetic architecture hypotheses

Bohonak, A. J., 1999 Dispersal, gene flow, and population structure.hold true is an important goal. Among the most com-Q. Rev. Biol. 74: 21–45.

mon methods for estimating parameters of genetic ar- Copenhaver, G. P., E. A. Housworth and F. W. Stahl, 2002 Cross-over interference in Arabidopsis. Genetics 160: 1631–1639.chitecture is QTL mapping. However, the majority of

Donohue, K., E. H. Pyle, D. Messiqua, M. S. Heschel and J.QTL studies in natural systems utilize a single popula-Schmitt, 2000 Density dependence and population differentia-

tion derived from only two individuals (see Long et al. tion of genetic architecture in Impatiens capensis in natural envi-1995 and Gurganus et al. 1999 for notable exceptions) ronments. Evolution 54: 1969–1981.

El-Din El-Assal, S., C. Alonso-Blanco, A. J. Peeters, V. Raz andand an obvious, although typically understated, caveatM. Koornneef, 2001 A QTL for flowering time in Arabidopsisto these studies is that they characterize only some por- reveals a novel allele of CRY2. Nat. Genet. 29: 435–440.

tion of the genetic architecture that underlies trait varia- Falconer, D. S., and T. F. Mackay, 1996 Introduction to QuantitativeGenetics. Prentice Hall, Singapore.tion. Importantly, we do not yet know how well the

Frary, A., T. C. Nesbitt, S. Grandillo, E. Knaap, B. Cong et al.,results from any one mapping population describe the 2000 fw2.2: a quantitative trait locus key to the evolution ofgenetics that underlie standing variation within a spe- tomato fruit size. Science 289: 85–88.

Gilbert, L., 1971 Butterfly-plant coevolution: has Passiflora adeno-cies. If we intend to use QTL mapping to understandpoda won the selectional race with Heliconiine butterflies? Sci-the range of genetic variation that is responsible forence 172: 585–586.

standing trait variation, a broader approach is needed. Gurganus, M. C., S. V. Nuzhdin, J. W. Leips and T. F. M Ackay,1999 High-resolution mapping of quantitative trait loci forTo explore the extent of interpopulation variation forsternopleural bristle number in Drosophila melanogaster. Geneticsgenetic architecture, we have mapped QTL for trichome152: 1585–1604.

density in four RIL populations of A. thaliana. Where Hamrick, J. L., and M. J. W. Godt, 1996 Effects of life history traitsonly a single QTL had been described for this trait on genetic diversity in plant species. Philos. Trans. R. Soc. Lond.

B Biol. Sci. 351: 1291–1298.previously, our results identify a total of nine QTL. Com-Hsia, C. C., and W. McGinnis, 2003 Evolution of transcription factorparative analyses reveal that much of the genetic varia- function. Curr. Opin. Genet. Dev. 13: 199–206.

tion detected for trichome density is partitioned among Hulskamp, M., S. Misra and G. Jurgens, 1994 Genetic dissectionof trichome cell development in Arabidopsis. Cell 76: 555–566.lineages of A. thaliana. Specifically, these data reveal

Juenger, T., S. Sen, K. Stowe and E. Simms, 2005 Epistasis andmultiple accession-specific alleles that confer similar di-genotype-environment interaction for quantitative trait loci af-

rectional change; whether or not this pattern extends fecting flowering time in Arabidopsis thaliana. Genetica 123: 83–101.to a broader sampling of A. thaliana lineages will be an

Karabourniotis, G., and Y. Manetas, 1995 Trichome density andinteresting line of inquiry. More generally, our findingsits protective potential against ultraviolet-B radiation damage dur-

show that QTL mapping results can be strongly depen- ing leaf development. Can. J. Bot. 73: 376–383.Kennedy, G. G., 2003 Tomato, pests, parasitoids, and predators:dent upon the particular population(s) in which one

1658 V. V. Symonds et al.

tritrophic interactions involving the genus Lycopersicon. Annu. molecular markers in a subdivided species. Evolution 55: 1753–1761.Rev. Entomol. 48: 51–72.

King, M., and A. Wilson, 1975 Evolution at two levels in humans Paran, I., and D. Zamir, 2003 Quantitative traits in plants: beyondthe QTL. Trends Genet. 19: 303–306.and chimpanzees. Science 188: 107–116.

Koornneef, M., 1981 The complex syndrome of ttg mutants. Arabi- Poos, F., 1929 Leafhopper injury to legumes. J. Econ. Entomol. 22:146–153.dopsis Inf. Serv. 18: 45–51.

Koornneef, M., W. Dellaert and J. van der Veen, 1982 EMS- Rieseberg, L. H., A. Widmer, A. M. Arntz and J. M. Burke, 2003The genetic architecture necessary for transgressive segregationand radiation-induced mutation frequencies at individual loci in

Arabidopsis thaliana (L.). Mutat. Res. 93: 109–123. is common in both natural and domesticated populations. Philos.Trans. R. Soc. Lond. B Biol. Sci. 358: 1141–1147.Korstanje, R., and B. Paigen, 2002 From QTL to gene: the harvest

Robin, C., R. F. Lyman, A. D. Long, C. H. Langley and T. F. Mackay,begins. Nat. Genet. 31: 235–236.2002 hairy: a quantitative trait locus for Drosophila sensoryLarkin, J. C., N. Young, M. Prigge and M. D. Marks, 1996 Thebristle number. Genetics 162: 155–164.control of trichome spacing and number in Arabidopsis. Develop-

Sanchez-Moran, E., S. J. Armstrong, J. L. Santos, F. C. Franklinment 122: 997–1005.and G. H. Jones, 2002 Variation in chiasma frequency amongLarkin, J. C., M. L. Brown and J. Schiefelbein, 2003 How do cellseight accessions of Arabidopsis thaliana. Genetics 162: 1415–1422.know what they want to be when they grow up? Lessons from

Sandquist, D., and J. Ehleringer, 2003 Population- and family-epidermal patterning in Arabidopsis. Annu. Rev. Plant Biol. 54:level variation of Brittlebush (Encelia farinosa, Asteraceae) pubes-403–430.cence: its relation to drought and implications for selection inLevin, D., 1973 The role of trichomes in plant defense. Q. Rev.variable environments. Am. J. Bot. 90: 1481–1486.Biol. 48: 3–15.

Sen, S., and G. Churchill, 2001 A statistical framework for quantita-Liakoura, V., M. Stefanou, Y. Manetas, C. Cholevas and G. Kara-tive trait mapping. Genetics 159: 371–387.bourniotis, 1997 Trichome density and its UV-B protective

Smith, W., and P. Nobel, 1977 Influences of seasonal changes inpotential are affected by shading and leaf position on the canopy.leaf morphology on water-use efficiency for three desert broadleafEnviron. Exp. Bot. 38: 223–229.shrubs. Ecology 58: 1033–1043.Lister, C., and C. Dean, 1993 Recombinant inbred lines for map-

Sokal, R. R., and F. J. Rohlf, 1997 Biometry. W. H. Freeman, Newping RFLP and phenotypic markers in Arabidopsis thaliana. PlantYork.J. 4: 745–750.

Soller, M., and J. S. Beckman, 1990 Marker-based mapping ofLong, A. D., S. L. Mullaney, L. A. Reid, J. D. Fry, C. H. Langleyquantitative trait loci using replicated progenies. Theor. Appl.et al., 1995 High resolution mapping of genetic factors affectingGenet. 80: 205–208.abdominal bristle number in Drosophila melanogaster. Genetics

Szymanski, D. B., A. M. Lloyd and M. D. Marks, 2000 Progress in139: 1273–1291. the molecular genetic analysis of trichome initiation and morpho-Loudet, O., S. Chaillou, C. Camilleri and D. Bouchez, 2002 Bay- genesis in Arabidopsis. Trends Plant Sci. 5: 214–219.0 � Shahdara recombinant inbred line populations: a powerful Van Ooijen, J., and R. Voorrips, 2001 JoinMap 3.0, Software for thetool for the genetic dissection of complex traits in Arabidopsis. Calculation of Genetic Linkage Maps. Plant Research International,Theor. Appl. Genet. 104: 1173–1184. Wageningen, The Netherlands.

Lynch, M., and B. Walsh, 1998 Genetics and Analysis of Quantitative Van Ooijen, J., M. Boer, R. Jansen and C. Maliepaard,Traits. Sinauer Associates, Sunderland, MA. 2002 MapQTL 4.0, Software for the Calculation of QTL Positions on

Mackay, T. F., 2001 The genetic architecture of quantitative traits. Genetic Maps. Plant Research International, Wageningen, TheAnnu. Rev. Genet. 35: 303–339. Netherlands.

Mauricio, R., 1998 Costs of resistance to natural enemies in field Zhang, F., A. Gonzalez, M. Zhao, C. T. Payne and A. Lloyd, 2003populations of the annual plant Arabidopsis thaliana. Am. Nat. A network of redundant bHLH proteins functions in all TTG1-151: 20–28. dependent pathways of Arabidopsis. Development 130: 4859–69.

Morgan, K. K., J. Hicks, K. Spitze, L. Latta, M. E. Pfrender et al.,2001 Patterns of genetic architecture for life-history traits and Communicating editor: V. Sundaresan