Copyright � 2007 by the Genetics Society of AmericaDOI: 10.1534/genetics.106.068338

The Genetic Architecture of Reproductive Isolation in LouisianaIrises: Flowering Phenology

Noland H. Martin,*,1 Amy C. Bouck† and Michael L. Arnold*

*Department of Genetics, University of Georgia, Athens, Georgia 30602 and †Department of Biology, Duke University,Durham, North Carolina 27708

Manuscript received November 15, 2006Accepted for publication January 5, 2007

ABSTRACT

Despite the potential importance of divergent reproductive phenologies as a barrier to gene flow, weknow less about the genetics of this factor than we do about any other isolating barrier. Here, we report onthe genetic architecture of divergent flowering phenologies that result in substantial reproductiveisolation between the naturally hybridizing plant species Iris fulva and I. brevicaulis. I. fulva initiates andterminates flowering significantly earlier than I. brevicaulis. We examined line crosses of reciprocal F1 andbackcross (BC1) hybrids and determined that flowering time was polygenic in nature. We further definedquantitative trait loci (QTL) that affect the initiation of flowering in each of these species. QTL analyseswere performed separately for two different growing seasons in the greenhouse, as well as in two fieldplots where experimental plants were placed into nature. For BCIF hybrids (BC1 toward I. fulva), 14 of 17detected QTL caused flowering to occur later in the season when I. brevicaulis alleles were present, whilethe remaining 3 caused flowering to occur earlier. In BCIB hybrids (BC1 toward I. brevicaulis), 11 of 15detected QTL caused flowering to occur earlier in the season when introgressed I. fulva alleles werepresent, while the remaining 4 caused flowering to occur later. These ratios are consistent withexpectations of selection (as opposed to drift) promoting flowering divergence in the evolutionary historyof these species. Furthermore, epistatic interactions among the QTL also reflected the same trends, withthe majority of epistatic effects causing later flowering than expected in BCIF hybrids and earlierflowering in BCIB hybrids. Overlapping QTL that influenced flowering time across all four habitat/treatment types were not detected, indicating that increasing the sample size of genotyped plants wouldlikely increase the number of significant QTL found in this study.

EARLY stages of speciation usually involve the evo-lution of numerous prezygotic and postzygotic

isolating barriers that act in concert to reduce thechance for gene flow between genetically and pheno-typically divergent populations. Recent, detailed experi-ments in which the relative contributions of numerousprezygotic and postzygotic barriers were quantified con-sistently revealed prezygotic barriers to be the strongestimpediments to interspecific gene flow (Ramsey et al.2003; Husband and Sabara 2004; Hurt et al. 2005;Whiteman and Semlitsch 2005; Martin and Willis

2007). If divergence between organisms often occurssimultaneously with gene flow (see Arnold 2006 fora review), the strength of such prezygotic barriers maybe fundamentally important for the establishment ofnew evolutionary lineages. Given (i) the sequential na-ture of reproductive barriers and (ii) that phenologicalisolation (isolation due to the timing of reproductionof the two species) occurs first in the sequence (e.g.,Martin and Willis 2007), this factor has the potentialto be the most important barrier preventing the

formation of viable hybrids. However, despite this im-portance, Coyne and Orr (2004, p. 21) concluded ‘‘Weknow less about the genetics of temporal [phenological]isolation than about the genetics of any other isolatingbarrier.’’

Quantitative trait locus (QTL) mapping studies havebeen utilized extensively as a means for dissecting thegenetic architecture of a number of postzygotic barriers(e.g., Slotman et al. 2004; Martin et al. 2005; Moyle

and Graham 2005; Moehring et al. 2006a), due largelyto the fact that hybrid inviability and hybrid sterilityare much more tractable under laboratory conditionsthan are most prezygotic barriers. Fewer studies of thegenetic architecture of prezygotic isolation have beencarried out, since prezygotic barriers must often be stud-ied using more labor-intensive field conditions. Excep-tions to this include studies of mating displays involved insexual isolation in the laboratory between insect species(e.g., Drosophila: Gleason et al. 2002; Gleason andRitchie 2004; Moehring et al. 2004, 2006b), as well aslaboratory studies examining divergent floral displaysbelieved to be important in promoting pollinator iso-lation in flowering plants (e.g., Mimulus: Bradshaw

et al. 1995, 1998; Fishman et al. 2002; Aquilegia: Hodges

1Corresponding author: Department of Biology, Texas State University,San Marcos, TX 78666. E-mail: nm14@txstate.edu

Genetics 175: 1803–1812 (April 2007)

et al. 2002). However, we have been unable to find anyreports of studies examining the genetic architecture ofphenological isolation.

In this study, we examined F1 and reciprocal back-cross populations of two closely related, and naturallyhybridizing, species of Iris, Iris fulva and I. brevicaulis, todetermine whether or not the phenotypic expression offlowering time was polygenic in nature and to test forQTL responsible for this flowering difference. Thesetwo species are broadly sympatric throughout the NorthAmerican Mississippi River drainage system (Viosca

1935). In sympatric populations, divergence of flower-ing phenology acts as a major barrier for F1 hybridformation (Cruzan and Arnold 1994). In this regard,I. fulva initiates flowering �1 month earlier than I.brevicaulis in southern Louisiana populations (Cruzan

and Arnold 1994). Consistent with this observation,few F1 genotypes have been found in areas of sympatry.However, when F1 crosses are made by artificial means,the resulting progeny are vigorous and completelyfertile, suggesting that prezygotic barriers (includingflowering phenology) are largely responsible for thepaucity of F1 hybrid production in natural populations.Interestingly, the rare formation of F1 hybrids has beenevolutionarily significant, as reflected by numerous hy-brid zones containing a wide array of later-generationhybrids (Arnold 1993; Cruzan and Arnold 1993;Johnston et al. 2001).

Using two reciprocal backcross 1 (BC1) mappingpopulations, we performed genome scans to searchfor QTL that affected the timing of flowering in twoseparate years under greenhouse conditions, as well asin two experimental plots established within the natu-rally occurring ranges of both species. This allowed us todefine the genetic architecture of the flowering phe-nology of both species. Specifically, we asked:

1. How many genomic regions were responsible foraffecting flowering phenology?

2. Did the identified QTL interact epistatically?3. Did the same QTL affect flowering phenology in two

different greenhouse years and in the two differentexperimental plots?

Addressing these questions allowed us to define thegenomic regions that affect a large proportion of theprezygotic isolation between these two naturally hybrid-izing plant species.

MATERIALS AND METHODS

Construction of mapping populations: One wild-collectedindividual each from I. fulva (If174, collected from Terre-bonne Parish, Louisiana) and I. brevicaulis (Ib72, collectedfrom St. Martinville Parish, Louisiana) was used to makereciprocal, interspecific BC1 populations. The two field siteswhere the pure species were found were markedly different interms of microhabitat, with the I. fulva and I. brevicaulis geno-types occurring along the margins of a bayou and an open

hardwood forest, respectively (M. L. Arnold, unpublisheddata). To minimize the within-species genetic variation, clonesof these parental individuals were used to produce the F1

parents and were then utilized as the recurrent backcrossparents. F1 plants were produced using Ib72 as the maternalparent and If174 as the pollen parent. Two F1 genotypes,designated as F1(2) and F1(3), were used as pollen parents toproduce the BC1 hybrids. F1(2) was used to pollinate severalclones of If174, and F1(3) was used as the pollen donor forseveral clones of Ib72. Several hundred I. fulva backcross(BCIF) and I. brevicaulis backcross (BCIB) hybrid seeds wereultimately produced in 1999. Seeds were germinated and theresulting plants were maintained in the Department of PlantBiology greenhouses at the University of Georgia by trans-planting into 6-in. azalea pots shortly after germination.During the fall of all subsequent years, a single rhizome wasbroken off of each BC1 plant and transplanted into 8-in. azaleapots. All BC1 plants were assigned to random positions in thegreenhouse. Replicates that survived throughout these 5 yearsin the greenhouse were planted into the field as describedbelow.

Construction of linkage maps: Two independent linkagemaps were constructed using the Iris retroelement (IRRE)transposon display marker system designed by Kentner et al.(2003) and the linkage mapping program Mapmaker 3.0(Lander et al. 1987; Lincoln et al. 1992). Bouck et al. (2005)provide a detailed description of the initial map constructionprotocol. In short, BCIB genotypes were determined for 414IRRE transposon display markers (N ¼ 230 plants), and BCIFgenotypes were determined for 309 IRRE markers (N ¼ 120plants). For the BCIF map, 63 additional markers were addedto the marker data set originally utilized in the study by Bouck

et al. (2005). The two reciprocal linkage maps are independentbecause transposon display markers possess dominant inheri-tance. The presence or absence of dominant I. fulva markersin BCIB individuals was ascertained, and these markers wereused to create the BCIB map. Likewise, the presence or ab-sence of dominant I. brevicaulis markers in BCIF individualswas also recorded and utilized for creating the BCIF map.Linkage groups (LGs) identified in both maps were labeledin order of the largest to the smallest lengths (in centimor-gans, LG1–LG22). Shared designations between LGs fromthe two maps do not imply homology. The BCIB map con-sists of 142 framework markers and 22 LGs, with an averagemarker spacing of 12 cM. The BCIF map consists of 108 frame-work markers and 22LGs, with an average marker spacing of13 cM. The BCIF map presented here has been modifiedsomewhat relative to the map presented in Bouck et al. (2005)due to the addition of new, informative markers identified inthis study.

Assaying flowering phenology in the greenhouse: The twogreenhouse experiments were performed in the Departmentof Plant Biology facilities in the spring of 2002 and of 2003 andwere part of a larger study examining a number of floral-traitQTL (Bouck 2004). Data were collected from clones of thesame individuals used to create the BCIF and BCIB maps. Thetotal number of BCIF individuals that were used to mapflowering phenology QTL was 67 and 57 during 2002 and2003, respectively. The total number of BCIB genotypes thatwere scored for flowering phenology was 176 and 145 during2002 and 2003, respectively. For each individual, the date atwhich the first flower appeared was recorded and assigneda numerical score (1 for the first day that a flower was ob-served, 2 for the second day, etc.). Phenological plots for thegreenhouse plants are not presented here, as F1 and pure-species plants were not included in the study, and only the dateof first flowering was recorded for the BCIF and BCIBindividuals.

1804 N. H. Martin, A. C. Bouck and M. L. Arnold

Assaying flowering phenology in the field: The field experi-ment was performed in a cypress–mixed hardwood forest inwhich natural populations of I. brevicaulis and I. fulva aretypically found (Viosca 1935; Cruzan and Arnold 1993;Johnston et al. 2001). Two experimental plots were estab-lished �1 km away from one another near the ChoupiqueBayou located in the U. S. Army Corps of Engineers Atch-afalaya Basin Floodway in Louisiana. The elevation at both ofthese sites is �5.0 m above mean sea level, and the elevationalchange within each plot differed by ,0.5 m. It should be notedthat these plots are not the same as those described previouslyby Martin et al. (2006), as extensive and extended floodinglargely destroyed the latter. Throughout this study, we willdifferentiate the two plots as either the ‘‘dry’’ or the ‘‘wet’’ site.We performed no soil moisture analysis, but we observed thatduring periods of heavy rainfall, much of the wet plot becameinundated with �0.5 m of water. Furthermore, following thecessation of rainfall, several days passed before the water re-ceded completely. In contrast, within the dry site, no standingwater was ever observed for more than a single day during thefield season.

In October 2005, we collected three to four rhizomes fromeach BCIF and BCIB plant (172 and 243 individuals, re-spectively), along with 43 I. fulva rhizomes, 62 I. brevicaulisrhizomes, and 47 F1 rhizomes. We transplanted one to two ofthese rhizomes into each of the two plots. Within each plot, werandomly assigned positions to the various genotypes, spacingthem 0.5 m apart. We noted the date at which each floweropened, and the date at which each flower wilted to the pointwhere it was unattractive to pollinators (�2 days post open-ing). For the majority of flowers, we were present to record thedate of opening and wilting. However, for a few genotypes,we were not present when a flower opened, and in those in-stances, we estimated the date based upon floral morphology.Flowers are normally open for 2 days, with dramatic morpho-logical differences apparent between 1-day-old and 2-day-oldflowers. For example, stigmas do not become exposed topollinators until day 2. Flowering phenology for the two fieldplots are given in Figure 1.

We utilized a two-way analysis of variance (ANOVA) to testwhether there were significant differences among the meanflower initiation dates for the five different cross types. A ‘‘crosstype’’ main effect (I. fulva, BCIF, F1, BCIB, and I. brevicaulis), a‘‘habitat’’ main effect (wet or dry), and a ‘‘cross type 3 habitat’’interaction effect were included in the model (Table 1). Wealso used posthoc Tukey honestly significant differences (HSD)tests (corrected for multiple comparisons) to test whetherthere were significant differences among the cross types.

Under a purely additive model of genetic inheritance, theexpected mean flowering date of F1 hybrids would be theaverage of the mean flowering dates of I. fulva and I. brevicaulis.Planned linear contrasts were used to test whether the meanof F1 hybrids differed significantly from this additive model’sexpectation by comparing the F1 hybrid class to that of thetwo parental species. Further, under a purely additive modelof genetic inheritance, the expected mean flowering dates ofthe backcross hybrids would be 0.75 3 the mean flowering datesof the recurrent parent 1 0.25 3 the mean flowering date ofthe other parent. Thus, we used two planned linear contraststo test whether the mean of both BCIF and BCIB hybridsdiffered significantly from this additive model’s expectation.

QTL analysis of flowering phenology: We mapped flowerinitiation dates (day 1 for the date at which the first flower wasobserved, day 2 for the second date, etc.) in both BCIB andBCIF mapping populations planted in the greenhouse and innature. In the greenhouse analyses, only a single copy of eachgenotype was present each year. For the field data, up to twocopies of each BC1 genotype were planted into each plot, and

in instances where both copies flowered at a site, we used themean of the two initiation dates of the two replicates toperform QTL mapping. No transformations were performedon the data. Separate genome scans for QTL affecting flower-ing phenology were performed for each of the two BC1 maps,first using composite interval mapping (CIM) (Zeng 1993,1994) followed by refinement with multiple interval mapping(MIM) methods (Kao et al. 1999; Zeng et al. 1999). Analyseswere performed separately for the 2002 and 2003 greenhousedata sets and separately for the two plots used in the single2006 field season. CIM tests were performed at 1-cM steps, witha 10-cM window size to exclude closely linked cofactors. TheCIM program through forward and backward regression chosethree cofactors for each QTL analysis. Genomewide thresholdvalues were determined at the 5% level by running 1000permutations for each map (Churchill and Doerge 1994;Doerge and Churchill 1996). MIM tends to have morepower and precision for detecting QTL (Kao et al. 1999), sowe used QTL detected in CIM analyses as an initial modelfor MIM mapping. If no QTL were found with CIM, we usedforward and backward regression selection on all markerswith partial r2 probability set at 0.01 as the initial model forMIM. MIM was then used to (i) refine QTL positions, (ii) searchfor additional QTL, (iii) search for epistatic effects amongdetected QTL, and (iv) estimate individual QTL effects and

Figure 1.—Flowering phenology for experimental I. fulva(red), BCIF (pink), F1 (yellow), BCIB (purple), and I. brevi-caulis (blue) assessed on a daily basis at two separate sites.The census number (y-axis) is expressed here as a percentageof the maximum number of flowers open throughout the cen-sus period. (A) Dry site. A total of 146 I. fulva, 2005 BCIF, 470F1, 2115 BCIB, and 51 I. brevicaulis flowers were censusedthroughout the entire flowering period. (B) Wet site. A totalof 233 I. fulva, 2351 BCIF, 489 F1, 2469 BCIB, and 151 I. bre-vicaulis flowers were censused throughout the entire flower-ing period.

Phenological Isolation in Iris 1805

the proportion of phenotypic variance explained by each ofthe QTL and significant interactions. Tests in MIM wereperformed at 1-cM steps, with Bayesian Information Criterionmodel selection and the penalty function set to c(n) ¼ ln(n).Two-LOD support limits for each QTL were calculated aroundthe most likely QTL position, as illustrated on the maps inFigures 3 and 4. The positions, effect size and directions, andthe proportion of phenotypic variance explained by each QTLare presented in Tables 2 and 3. To test whether selectionmight have been important in causing flowering divergence,we utilized sign tests in each mapping population to deter-mine whether the directionality (1 or �) of nonoverlappingQTL effects significantly differed from the null expectation ofequal proportions (i.e., consistent with a model of neutrality).

RESULTS

The mean flowering-initiation dates (61 SE) at thetwo different sites were as follows: dry site: I. fulva—March 30 (61.4 days); BCIF—April 10 (60.4 days);F1—April 14 (60.7 days); BCIB—April 23 (60.4 days); I.brevicaulis—May 2 (60.4 days); wet site: I. fulva—March30 (61.4 days); BCIF—April 9 (60.4 days); F1—April 11(60.7 days); BCIB—April 22 (60.4 days); I. brevicaulis—May 1 (0.7 days). Results of two-way ANOVA indicatedthat there was a main effect for ‘‘cross type,’’ no maineffect for ‘‘site,’’ and no ‘‘site 3 cross type’’ interaction(Table 1). Examining the significant main effect (crosstype) further, posthoc Tukey HSD tests accounting formultiple comparisons revealed that all cross types werestatistically different from one another with respect toflower initiation (corrected P-values ,0.05 for all 10comparisons).

We tested whether the flowering-date phenotypes ofF1 and reciprocal backcross hybrids deviated signifi-cantly from an additive model of genetic inheritance.Under a purely additive model, the expected mean flo-wering date of F1 hybrids would be the mean floweringdate of both pure-species parents, and F1 hybrids diddeviate from this additive model, but only with marginalsignificance (linear contrasts, F¼ 2.91, 1 d.f., P¼ 0.088,Figure 2). For BCIF hybrids, given strict additivity, theexpected mean flowering date of BCIF hybrids would be0.75 3 the mean flowering date of I. fulva 1 0.25 3 themean flowering date of I. brevicaulis. The BCIF hybridsdid in fact deviate significantly from this null additivemodel, with BCIF hybrids flowering only 3 days earlier

than F1 hybrids, on average, but 11.5 days later, onaverage, than pure I. fulva species (linear contrast, F ¼4.44, 1 d.f., P ¼ 0.035, Figure 2). The expected meanflowering date of BCIB hybrids, given a purely additivegenetic model of inheritance, would be 0.75 3 the meanflowering date of I. brevicaulis 1 0.25 3 the meanflowering date of I. fulva, and the BCIB hybrids didnot significantly deviate from this model (linear con-trast, F ¼ 0.063, 1 d.f., P ¼ 0.802, Figure 1).

Quantitative trait locus analysis: Using CIM followedby refinement with MIM, we identified 17 QTL in theBCIF mapping population that affected flowering timein one or more field habitats or greenhouse years (Ta-ble 2, Figure 3). In this backcross population, LG1 pos-sessed four QTL: three of the QTL had negative effects,meaning that the introgressed I. brevicaulis alleles causedflowering time to occur earlier. The confidence intervalsof these three QTL were overlapping, suggesting thatthe genes causing earlier flowering times could be thesame across the three separate habitats (greenhouseseason 2002, wet and dry field sites; Table 2, Figure 3).A fourth QTL was detected at the topmost portion ofthe linkage group, and this QTL had a positive effect,meaning that the introgressed I. brevicaulis alleles causedflowering time to occur later (in the dry field site; Table2, Figure 3). LG12 was the only other linkage groupin the BCIF mapping population that revealed a QTLwith negative effect (i.e., introgressed I. brevicaulis allelescaused flowering to initiate earlier in the dry field site).In contrast, a QTL that caused a later flowering timein the wet field site was also detected on LG12. Thesetwo QTL, given their opposite effects, are unlikely to bedue to the same genes. The remaining 11 QTL detectedin the BCIF mapping population caused flowering toinitiate later. These loci included single QTL (i.e., affect-ing flowering time in only one of the four greenhouse orfield habitats) found on LG2, LG6, LG8, LG9, and LG13and two overlapping QTL (affecting flowering time intwo of the four greenhouse or field habitats) found onLG5, LG7, and LG11. Four epistatic interactions were

TABLE 1

Results of a two-way analysis of variance on ‘‘date of firstflower’’ conducted in two separate field sites

Factor d.f. Mean squares F-ratio P-value

Cross 4 11394.2291 233.75 ,0.0001Site 1 114.24128 2.3437 0.1261Site 3 cross 4 29.2575705 0.6002 0.6626

Main effects include ‘‘cross’’ and ‘‘site’’ and a ‘‘site 3 cross’’interaction. Significant effects are in italics.

Figure 2.—Observed vs. expected mean flower initiationdates (62 SE) for I. fulva, I. brevicaulis, F1, BCIF, and BCIBhybrids. The diagonal line denotes the expected mean flower-ing date under an additive model of gene action.

1806 N. H. Martin, A. C. Bouck and M. L. Arnold

detected between QTL using MIM methodologies: be-tween QTL 2 and 3 in the 2002 greenhouse study;between QTL 2 and 3 and 3 and 4 in the dry site; and be-tween QTL 1 and 6 in the wet site (Table 2). In allinstances, epistatic interactions between two alleles re-sulted in flowering time occurring later than expected,given a purely additive model.

In the BCIB mapping population, we utilized CIMfollowed by MIM and found a total of 15 QTL thataffected flowering time (Table 3, Figure 4). Only 4 of the15 detected QTL had positive effects (i.e., caused laterflowering when I. fulva alleles were present). Two ofthese QTL were detected in the 2002 greenhouse studyand were located on LG6 and LG15. The other twopositive QTL were detected in the wet field site and werelocated on LG9 and LG13. All other QTL in the BCIBmapping population were negative; i.e., they causedearlier flowering when introgressed I. fulva alleles werepresent. Two overlapping, negative QTL were locatedon LG2 and were detected in the 2002 greenhousesample and the dry field site. Another two overlappingQTL were located on LG7 and were detected in the 2002and 2003 greenhouse studies. The remaining negativeQTL were found singly (i.e., affected flowering time inonly one of the four greenhouse or field habitats) onLG4, LG5, LG10, LG16, LG17, LG20, and LG21. Threeepistatic interactions were detected in the BCIB map-ping population: between QTL 2 and 3 and betweenQTL 4 and 5 in the 2002 greenhouse study and betweenQTL 1 and 4 in the wet field site (Table 3). The epistaticinteractions detected in the greenhouse study caused anearlier flowering time when both I. fulva alleles at thedifferent QTL were present. In contrast, the epistaticinteraction detected in the wet site caused floweringtime to occur later than expected in comparison to apurely additive model.

DISCUSSION

Patterns of flowering phenology: As with many otherclosely related taxa that have the potential to hybridize,a number of reproductive isolating mechanisms interactto prevent F1 hybrid formation in I. brevicaulis/I. fulvahybrid zones (see Arnold 1997 and 2006 for reviews).Divergent flowering phenologies between I. fulva and I.brevicaulis, with I. fulva initiating flowering earlier in theseason, have been shown to be an important prezygoticbarrier to hybridization (Cruzan and Arnold 1994). Infact, this divergence of flowering time has the potentialto be the most important barrier to gene flow betweensympatric populations, since the two species must obvi-ously overlap in their reproductive times for later-actingbarriers to come into play. In this study, we showed thatin two field sites planted in natural conditions, I. fulvaand I. brevicaulis genotypes were nonoverlapping in theirflowering times, with I. brevicaulis initiating flowering

9–10 days after the last I. fulva flowers were observed(Figure 1). In the absence of other results, the findingsfrom this study would suggest that flowering phenologyalone is sufficient to prevent gene flow between thesetwo species. However, previous studies have shown someamount of overlap in natural, sympatric populations ofI. fulva and I. brevicaulis (Cruzan and Arnold 1994).Given this, along with the detection of numerous natu-ral hybrid zones between the two species (e.g., Arnold

1993; Cruzan and Arnold 1993; Johnston et al. 2001),we can assume that flowering phenology is not an im-permeable reproductive barrier. Assuming that thereis intraspecific variability for the timing of flowering,an experiment involving larger sample sizes of the two

TABLE 2

QTL for flowering initiation in two field sites (dry and wet)and in two greenhouse seasons (2002 and 2003)

detected in a BC1 (BCIF) population

QTL no.Linkagegroup Positiona Effectb PVEc

Greenhouse 2002 (n ¼ 67)1 LG1 94.2 �12.76 15.62 LG5 37.1 8.87 5.63 LG6 32.9 11.58 8.14 LG13 3 7.14 6.82 3 3 24.14 13.5

Total: 49.6

Greenhouse 2003 (n ¼ 57)1 LG7 39 15.91 5

Dry field site (n ¼ 45)1 LG1 0 8.55 21.32 LG1 50.3 �8.65 21.73 LG7 39.1 3.24 2.34 LG8 0 2.19 3.95 LG11 16 6.53 21.76 LG12 41.4 �4.95 112 3 3 7.55 7.63 3 4 4.3 2.4

Total: 91.9

Wet field site (n ¼ 52)1 LG1 108.1 �3.33 3.52 LG2 93.4 4.99 21.63 LG6 52.8 2.18 4.34 LG11 19 5.67 18.55 LG9 47.2 2.69 7.56 LG12 17.1 5.14 17.91 3 6 6.37 4.8

Total: 78.1

Results were obtained by MIM. QTL previously identifiedby CIM were used as the initial model for MIM analysis.The number of genotypes used in the QTL analyses is in pa-rentheses (n ¼ x). Epistatic interactions between QTL are de-noted with a ‘‘3’’.

a QTL position in centimorgans.b Effect of the QTL on date of flowering (in number of days).c The proportion of the phenotypic variance explained by

the QTL.

Phenological Isolation in Iris 1807

Fig

ur

e3.

—L

inka

gem

apo

fd

om

inan

tI.

brev

icau

lis

IRR

Ere

tro

tran

spo

son

dis

pla

ym

arke

rsse

greg

atin

gin

the

F1

hyb

rid

use

dto

pro

du

ceB

CIF

hyb

rid

s.Si

gnifi

can

tQ

TL

for

flo

wer

ing

ph

eno

logy

are

den

ote

d(w

ith

2-L

OD

con

fid

ence

inte

rval

s)to

the

righ

to

fth

eli

nka

gegr

ou

ps.

Red

bar

sre

pre

sen

tre

gio

ns

wh

ere

intr

ogr

esse

dI.

brev

icau

lis

alle

les

cau

sed

flo

wer

ing

toin

itia

teea

rlie

r,w

hil

eb

lue

bar

sre

pre

sen

tre

gio

ns

wh

ere

intr

ogr

esse

dI.

brev

icau

lis

alle

les

cau

sed

flo

wer

ing

toin

itia

tela

ter.

QT

Lan

alys

esw

ere

per

form

edd

uri

ng

the

two

gree

nh

ou

seye

ars

2002

and

2003

asw

ell

asin

two

fiel

dp

lots

(dry

and

wet

)in

2006

.

1808 N. H. Martin, A. C. Bouck and M. L. Arnold

species might reveal an overlap in flowering times [as,for example, was found in the much larger, naturalpopulation studied by Cruzan and Arnold (1994)].Regardless, flowering time is apparently an extremelystrong (in an absolute sense) and important prezygoticbarrier between I. fulva and I. brevicaulis. Furthermore,the observation of the nonoverlapping of floweringphenologies of the I. fulva and I. brevicaulis suggests thatthe formation of the natural hybrid populations is likelylimited by the episodic formation of F1 hybrids betweenthese two Iris species.

An examination of the flowering times of the threehybrid classes planted into the wild revealed the F1

hybrids to be intermediate between the two species.Furthermore, as predicted for quantitative traits, BCIFwere intermediate between I. fulva and F1 hybrids, whileBCIB hybrids were intermediate between I. brevicaulisand F1 hybrids. These data are consistent with the detec-tion of numerous natural hybrid zones between thesespecies. Thus, although the divergent flowering timesof I. fulva and I. brevicaulis greatly limit F1 hybrid for-mation, these rare events produce effective intermedi-aries for gene flow between the two species. Furthermore,we would predict that the later-generation hybrids couldalso be conduits for gene flow, but primarily with thespecies with which they were backcrossed.

The genetic architecture of flowering phenology: Byexamining the phenotypic expression of floweringtimes of the three hybrid classes used in this study, andcomparing those flowering times to the parental geno-types, we can gain an initial estimate of the underlyinggenetic architecture responsible for the divergentphenologies of I. fulva and I. brevicaulis. If the divergentflowering times were due solely to a single gene thatacted in a completely dominant/recessive fashion, thenthe F1 flowering phenology would mirror that of oneof the parents. Here, F1 hybrids are intermediate be-tween the two parents, suggesting that this simple modelcannot explain the divergent flowering phenologies.In addition, by examining the flowering times of bothreciprocal backcross hybrids (BCIF and BCIB), we seethat the mean flowering times of those hybrids are in-termediate between the F1 and the recurrent paren-tal species. Furthermore, the flowering phenotypes ofmany backcross individuals are indistinguishable fromthe recurrent parents. This suggests that the divergentflowering phenologies of I. fulva and I. brevicaulis aredue to multiple genes. Furthermore, not all of the genesact in a purely additive fashion. Thus, the mean flower-ing phenology of BCIF hybrids is significantly later thanwould be expected, given a purely additive model ofgene action (Figure 2).

We further examined the polygenic nature of thiscomplex quantitative trait by performing QTL map-ping of flower initiation in each of the BCIF and BCIBmapping populations in both greenhouse and fieldconditions. We identified 16 nonoverlapping QTL that

affected flowering phenology in both BCIF and BCIBhybrids in at least one of the four greenhouse/fieldhabitats. Our results indicate that these phenologicalQTL, depending on the habitats being examined, havesmall-to-large effect sizes. We would add a cautionarynote that QTL mapping tends to overestimate the effectmagnitude of QTL, especially with smaller sample sizessuch as those found in our BCIF mapping population.Most importantly, the power to detect QTL of smallereffect is also dependent upon the sample size. Thenumber of genotyped individuals that flowered in thisstudy was relatively small. Thus, the number of QTLdetected should be considered a minimum. Further-more, the limited sample sizes may have contributed tothe observation of no overlapping QTL detected acrossall four of the habitats (greenhouse and field) exam-ined. Athough it is not unusual to detect QTL that are

TABLE 3

QTL for flowering initiation in two field sites (dry and wet)and in two greenhouse seasons (2002 and 2003)

detected in a BC1 (BCIB) population

QTL no.Linkagegroup Positiona Effectb PVEc

Greenhouse 2002 (n ¼ 176)1 LG2 28.7 �13.53 12.32 LG4 80.3 �13.88 14.93 LG6 26.4 6.59 2.34 LG7 55.6 �10.84 8.55 LG15 13 14.32 12.56 LG21 12.1 �6.6 3.64 3 5 �15.46 3.32 3 3 �12.68 4.4

Total: 61.8

Greenhouse 2003 (n ¼ 146)1 LG7 60 �5.77 1.62 LG16 43.2 �13.43 10.2

Total: 11.8

Dry field site (n ¼ 98)1 LG2 25.7 �6.4 25.8

Wet field site (n ¼ 101)1 LG5 17.4 �4.26 11.82 LG9 15.9 4.3 14.23 LG10 55.8 �2.74 3.24 LG13 10.3 2.84 4.85 LG17 0 �5.55 22.16 LG20 10.9 �3.34 9.21 3 4 5.54 3.1

Total: 68.4

Results were obtained by MIM. QTL previously identifiedby CIM were used as the initial model for MIM analysis.The number of genotypes used in the QTL analyses is in pa-rentheses (n ¼ x). Epistatic interactions between QTL are de-noted with a ‘‘3’’.

a QTL position in centimorgans.b Effect of the QTL on date of flowering (in number of days).c The proportion of the phenotypic variance explained by

the QTL.

Phenological Isolation in Iris 1809

Fig

ur

e4.

—L

inka

gem

apo

fd

om

inan

tI.

fulv

aIR

RE

retr

otr

ansp

oso

nd

isp

lay

mar

kers

segr

egat

ing

inth

eF

1h

ybri

du

sed

top

rod

uce

BC

IBh

ybri

ds.

Sign

ifica

nt

QT

Lfo

rfl

ow

er-

ing

ph

eno

logy

are

den

ote

d(w

ith

2-L

OD

con

fid

ence

inte

rval

s)to

the

righ

to

fth

eli

nka

gegr

ou

ps.

Red

bar

sre

pre

sen

tre

gio

ns

wh

ere

intr

ogr

esse

dI.

fulv

aal

lele

sca

use

dfl

ow

erin

gto

init

iate

earl

ier,

wh

ile

blu

eb

ars

rep

rese

nt

regi

on

sw

her

ein

tro

gres

sed

I.fu

lva

alle

les

cau

sed

flo

wer

ing

toin

itia

tela

ter.

QT

Lan

alys

esw

ere

per

form

edd

uri

ng

the

two

gree

n-

ho

use

year

s20

02an

d20

03as

wel

las

intw

ofi

eld

plo

ts(d

ryan

dw

et)

in20

06.

1810 N. H. Martin, A. C. Bouck and M. L. Arnold

unique across habitat types, it is striking that none of theQTL detected in this study affected flowering phenol-ogy in all habitats examined. This is especially strikingsince ‘‘habitat’’ apparently did not significantly affectflowering time of plants in the field (Table 1). We there-fore attribute this observation largely to our relativelylow sample sizes, as it is difficult to argue that the dif-ferent habitats are themselves responsible for the lackof detection of overlapping QTL. Readers should thusinterpret these results with the following caveat: lackof detection of a QTL in one particular environmentdoes not rule out the possibility that a QTL could beaffecting the phenotype in that environment. However,if a QTL is detected in any particular environment, theQTL reflects a significant phenotype/genotype asso-ciation. These data clearly show that multiple QTL,distributed throughout the genome, are responsible forthe divergent flowering times between I. fulva and I.brevicaulis and should be viewed as ‘‘preliminary’’ only inthe sense that an increased sample size will likely revealeven more overlapping QTL, as the power to detect themincreases.

A consideration of the directionality of the detectedQTL suggests that divergence in flowering time likelyhas an adaptive explanation. In the BCIF population,only 4 of the 17 QTL caused flowering time to occurearlier when I. brevicaulis (i.e., the later-flowering spe-cies) alleles were present. A sign test (assuming all of theQTL are independent) indicates that this significantlydiffers from the expected proportion if no selection wasacting on flowering time (i.e., a neutral model: 1 QTL13, � QTL 4, P ¼ 0.049). Assuming that overlappingQTL (i.e., those that affected flowering time in multiplehabitats, in the same direction, and whose confidenceintervals overlapped) are the same also results in asignificant sign test result (1 QTL 10, � QTL 2, P ¼0.039). The opposite trend was detected in the BCIBbackcross hybrids, with only 4 of the 15 detected QTLincreasing flowering time when I. fulva (i.e., the earlier-flowering species) alleles were present. However, a signtest (assuming all QTL are independent) suggests onlymarginal significance (P ¼ 0.12). Furthermore, assum-ing overlapping QTL are the same, no significantdifference is detected (P ¼ 0.27). Yet, combining boththe BCIB and BCIF data sets reveals that the majority ofQTL are in the expected direction (24/32), with resultsfrom a sign test on this proportion (P ¼ 0.007) beingconsistent with the contribution of directional selectionin promoting divergence in reproductive phenology.Furthermore, epistatic interactions among QTL pre-dominantly affected phenology in the predicted di-rection. Thus, all four interactions between I. brevicaulisalleles in the BCIF plants caused later flowering times.Similarly, two of three epistatic interactions among I.fulva alleles in the BCIB population were associated withearlier flowering. Thus, not only does flowering phe-nology cause large-scale reproductive isolation between

these two species, but also the genetic architecture fits amodel of adaptive divergence.

We thank S. Wessler, R. Mauricio, and E. Kentner for contributionsto the linkage mapping projects that form the basis for these analyses.We thank Y. Sapir, D. Wills, S. Taylor, and N. Sherman for helpfulcomments on previous versions of the manuscript. R. H. Martin and Y.Z. Martin provided invaluable assistance in the field. B. Weckerleyprovided invaluable statistical advice. We thank the Army Corps ofEngineers Atchafalaya Basin Floodway System for field sites and fieldassistance. This study was supported by two National Science Founda-tion grants to M.L.A. (DEB-0074159 and DEB-0345123) and by a giftfrom M. Caillet.

LITERATURE CITED

Arnold, M. L., 1993 Iris nelsonii: origin and genetic composition ofa homoploid hybrid species. Am. J. Bot. 80: 577–583.

Arnold, M. L., 1997 Natural Hybridization and Evolution. OxfordUniversity Press, New York.

Arnold, M. L., 2006 Evolution Through Genetic Exchange. OxfordUniversity Press, New York.

Bouck, A. C., 2004 The Genetic Architecture of Reproductive Isolationin Louisiana Irises. Ph.D. Dissertation, University of Georgia,Athens, GA.

Bouck, A. C., R. Peeler, M. L.Arnold and S. R. Wessler, 2005 Geneticmapping of species boundaries in Louisiana irises using IRREretrotransposon display markers. Genetics 171: 1289–1303.

Bradshaw, H. D., Jr., S. M. Wilbert, K. G. Otto and D. W.Schemske, 1995 Genetic mapping of floral traits associatedwith reproductive isolation in monkeyflowers (Mimulus). Nature376: 762–765.

Bradshaw, H. D., Jr., K. G. Otto, B. E. Frewen, J. K. McKay andD. W. Schemske, 1998 Quantitative trait loci affecting differen-ces in floral morphology between two species of monkeyflower(Mimulus). Genetics 149: 367–382.

Churchill, G. A., and R. W. Doerge, 1994 Empirical threshold val-ues for quantitative trait mapping. Genetics 138: 963–971.

Coyne, J. A., and H. A. Orr, 2004 Speciation. Sinauer Associates,Sunderland, MA.

Cruzan, M. B., and M. L. Arnold, 1993 Ecological and geneticassociations in an Iris hybrid zone. Evolution 47: 1432–1445.

Cruzan, M. B., and M. L. Arnold, 1994 Assortative mating and nat-ural selection in an Iris hybrid zone. Evolution 48: 1946–1958.

Doerge, R. W., and G. A. Churchill, 1996 Permutation tests formultiple loci affecting a quantitative character. Genetics 142:285–294.

Gleason, J. M., and M. G. Ritchie, 2004 Do quantitative trait loci(QTL) for a courtship song difference between Drosophila simu-lans and D. sechellia coincide with candidate genes and intraspe-cific QTL? Genetics 166: 1303–1311.

Gleason, J. M., S. V. Nuzhdin and M. G. Ritchie, 2002 Quantita-tive trait loci affecting a courtship signal in Drosophila melanogaster.Heredity 89: 1–6.

Fishman, L., A. J. Kelly and J. H. Willis, 2002 Minor quantitativetrait loci underlie floral traits associated with mating system diver-gence in Mimulus. Evolution 56: 2138–2155.

Hodges, S. A., J. B. Whitall, M. Fulton and J. Y. Yang, 2002 Genet-ics of floral traits influencing reproductive isolation betweenAquilegia formosa and A. pubescens. Am. Nat. 159: S51–S60.

Hurt, C. R., M. Farzin and P. W. Hedrick, 2005 Premating, notpostmating, barriers drive genetic dynamics in experimentalhybrid populations of the endangered Sonoran topminnow.Genetics 171: 655–662.

Husband, B. C., and H. A. Sabara, 2004 Reproductive isolation be-tween autotetraploids and their diploid progenitors in fireweed,Chamerion angustifolium (Onagraceae). New Phytol. 161: 703–713.

Johnston, J. A., R. A. Wesselingh, A. C. Bouck, L. A. Donovan andM. L. Arnold, 2001 Intimately linked or hardly speaking? Therelationship between genotype and environmental gradients in aLouisiana Iris hybrid population. Mol. Ecol. 10: 673–681.

Kao, C. H., Z-B. Zeng and R. D. Teasdale, 1999 Multiple intervalmapping for quantitative trait loci. Genetics 152: 1203–1216.

Phenological Isolation in Iris 1811

Kentner, E. K., M. L. Arnold and S. R. Wessler, 2003 Characteri-zation of high-copy-number retrotransposons from the large ge-nomes of the Louisiana Iris species and their use as molecularmarkers. Genetics 164: 685–697.

Lander, E. S., P. Green, J. Abrahamson, A. Barlow, M. J. Daly et al.,1987 MAPMAKER: an interactive computer package for con-structing primary genetic linkage maps of experimental andnatural populations. Genomics 1: 174–181.

Lincoln, S., M. Daly and E. S. Lander, 1992 Constructing geneticmaps with MAPMAKER/EXP 3.0. Whitehead Institute TechnicalReport, Ed. 2. Whitehead Institute, Cambridge, MA.

Martin, N. H., and J. H. Willis, 2007 Ecological divergence asso-ciated with mating system causes nearly complete reproductiveisolation between sympatric Mimulus species. Evolution 61: 68–82.

Martin, N. H., A. C. Bouck and M. L. Arnold, 2005 Loci affectinglong-term hybrid survivorship in Louisiana Irises: implicationsfor reproductive isolation and introgression. Evolution 59:2116–2124.

Martin, N. H., A. C. Bouck and M. L. Arnold, 2006 Detectingadaptive trait introgression between Iris fulva and I. brevicaulisin highly selective field conditions. Genetics 172: 2481–2489.

Moehring, A. J., J. Li, M. D. Schug, S. G. Smith, M. deAngelis et al.,2004 Quantitative trait loci for sexual isolation between Drosoph-ila simulans and D. mauritiana. Genetics 167: 1265–1274.

Moehring, A. J., A. Llopart, S. Elwyn, J. A. Coyne and T. F. C.Mackay, 2006a The genetic basis of postzygotic reproductiveisolation between Drosophila santomea and D. yakuba due to hybridmale sterility. Genetics 173: 225–233.

Moehring, A. J., A. Llopart, S. Elwyn, J. A. Coyne and T. F. C.Mackay, 2006b The genetic basis of prezygotic reproductiveisolation between Drosophila santomea and D. yakuba due to matingpreference. Genetics 173: 215–223.

Moyle, L. C., and E. B. Graham, 2005 Genetics of hybrid incompat-ibility between Lycopersicon esculentum and L. hirsutum. Genetics169: 355–373.

Ramsey, J., H. D. Bradshaw and D. W. Schemske, 2003 Com-ponents of reproductive isolation between the monkeyflowersMimulus lewisii and M. cardinalis (Phrymaceae). Evolution 57:1520–1534.

Slotman, M., A. D. Torre and J. R. Powell, 2004 The genetics ofinviability and male sterility in hybrids between Anopheles gambiaeand An. arabiensis. Genetics 167: 275–287.

Viosca, P. J., 1935 The Irises of southeastern Louisiana: a taxonomicand ecological interpretation. Bull. Am. Iris Soc. 57: 3–56.

Whiteman, H. H., and R. D. Semlitsch, 2005 Asymmetric repro-ductive isolation among polymorphic salamanders. Biol. J. Linn.Soc. 86: 265–281.

Zeng, Z-B., 1993 Theoretical basis of separation of multiple linkedgene effects on mapping quantitative trait loci. Proc. Natl. Acad.Sci. USA 90: 10972–10976.

Zeng, Z-B., 1994 Precision mapping of quantitative trait loci.Genetics 136: 1457–1468.

Zeng, Z-B., C-H. Kao and C. J. Basten, 1999 Estimating the geneticarchitecture of quantitative traits. Genet. Res. 4: 279–289.

Communicating editor: S. W. Schaeffer

1812 N. H. Martin, A. C. Bouck and M. L. Arnold