Genetics of Local Adaptation in the Laboratory: Flowering Time Quantitative Trait Loci under Geographic and Seasonal Conditions in Arabidopsis

Flowering time in Arabidopsis thaliana is controlled by a large number of genes and various environmental factors, such as light and temperature. The objective of this study was to identify flowering time quantitative trait loci (QTL) under growth conditions simulating seasonal conditions from native geographic locations. Our growth chambers were set to simulate the spring conditions in Spain and Sweden, with appropriate changes in light color, intensity and day length, as well as temperature and relative humidity. Thus the Sweden-like spring conditions changed more dramatically compared to Spain-like spring conditions across the duration of our experiment. We have used these conditions to map QTL responsible for flowering time in the Kas-1/Col-gl1 recombinant inbred lines (RILs) across two replicate blocks. A linkage map from 96 RILs was established using 119 markers including 64 new SNPs markers. One major QTL, mapping to the FRIGIDA (FRI) locus, was detected on the top of chromosome 4 that showed significant gene×seasonal environment interactions. Three other minor QTL also were detected. One QTL mapping near FLOWERING LOCUS M (FLM) showed an epistatic interaction with the QTL at FRI. These QTL×environment and QTL×QTL interactions suggest that subtle ecologically relevant changes in light, temperature, and relative humidity are differentially felt by alleles controlling flowering time and may be responsible for adaptation to regional environments.


INTRODUCTION
Flowering time is a critical step in the life of annual species such as Arabidopsis thaliana. The proper timing of reproduction to coincide with suitable environments is important for its survival and interactions with other ecological factors. The major environmental cues that affect flowering time are light and temperature, which change daily through seasons and vary with geographic locations. The diversity of flowering time and responses to environmental variables in wild accessions of A. thaliana [1][2][3] suggest that A. thaliana has precise and diversified mechanisms controlling flowering time across geographic locations and seasons.
A. thaliana provides a great opportunity to determine the genetic basis of flowering time and response to environments. Many genes controlling flowering time have been identified (see the flowering web http://www.salk.edu/LABS/pbio-w/flower_web.html and reviews by [4][5][6][7]), and a genetic network has been outlined which includes four main pathways: photoperiod, vernalization, autonomous, and gibberellin [8]. Longer days act via the photoperiod pathway and an extended period of winter-like temperature act via the vernalization pathway to accelerate flowering time by releasing repression caused by the floral inhibitor Flowering Locus C (FLC). Functional alleles of another gene, FRIGIDA (FRI), promote the accumulation of FLC mRNA, which delays flowering. The autonomous pathway also promotes flowering by negatively regulating FLC. Ambient growth temperature may affect flowering time through an autonomous pathway [9]. In addition to FLC, another MADS (MCM1/AGAMOUS/DEFICIENS/SRF1)-box gene, FLOWERING LOCUS M (FLM), similar in amino-acid sequence to FLC, also acts as an inhibitor of flowering in the Ws accession [10]. The Col allele at FLM delays flowering compared to the null flm, including the natural null allele in the Nd accession [11].
Natural genetic variation in A. thaliana for flowering response under different environmental conditions has been observed including photoperiod, vernalization and ambient temperature [1][2][3]12,13]. A latitudinal cline in flowering time was observed in the European accessions of A. thaliana when grown over winter in a common garden in Rhode Island. This cline however was only seen in accessions that lacked common FRI deletion polymorphisms [3]. This may be due to a latitudinal cline in vernalization sensitivity: the southern accessions were more sensitive to vernalization than the northern accessions [2]. Since environmental factors such as light and temperature are correlated with latitude, the latitudinal clines for flowering time and vernalization sensitivity are suggestive of natural selection acting on flowering time. Direct tests are difficult as field studies are influenced by many random factors. It may be difficult to see modest genetic effects of evolutionary significance and when identified may correspond to the particulars of a given year rather than the standard local conditions. Generally, flowering time is accelerated by longer days, higher temperature [14], and vernalization, conditions associated with the more favorable conditions of spring and summer. How plants integrate signals from these environmental cues to decide when to flower is essential to survival. In natural conditions, a major environmental cue is the continuous seasonal changes of light, including the sunrise and sunset times perceived as changes in light quality and intensity. Such dynamic changes are distinct from traditional studies of fixed photoperiod, light quality and intensity. In nature, temperature also cycles during the day and changes predictably during the season, in coordination with light. These seasonal changes in light and temperature are a unique fingerprint of a particular geographic location. Standard lab experiments of long day vs. short day, with and without vernalization at constant temperature do not capture the changing seasonal patterns unique to local conditions. Thus to identify the loci involved in local adaptation to predictable seasonal environmental cues of day length, temperature, light color, and humidity, experiments should be performed under controlled settings simulating these multivariate conditions.
Our experiments aim to recreate seasonal conditions to study the flowering time QTL and their interactions with geographic locations within the spring season. This general approach of simulating natural conditions in the lab provides a compromise between simple environmental contrasts in the lab and the unpredictable complexity of field studies. However, the demanding dual strategy of lab and field studies is also warranted [15,16]. We used Kas-1/Col-gl1 recombinant inbred lines (RILs) (Somerville lines) in this study as a test case before moving to local populations. The parents Kas-1 and Col-gl1 differ in flowering time [17]. The results confirmed the quantitative nature of genetic variation in flowering time and the major role of polymorphisms in the FRI gene in flowering time variation. QTL6environment interactions suggested that seasonal changes affect flowering time differentially via alleles known to control flowering time. We propose that such interactions may be responsible for adaptation to regional environments. Additionally, epistasis between the loci near FLOWERING LOCUS M (FLM) and FRI suggest a novel interaction between loci known to control natural variation of flowering time in A. thaliana.

Growth Chambers
Experiments were conducted within two walk-in growth chambers (AR-916, Percival Scientific, Inc. Perry, IA) that were programmed to simulative ideal seasonal conditions in Madrid, Spain and Stockholm, Sweden, respectively (Fig. 1). Sunrise and sunset, light spectrum, temperature and relative humidity were programmed to cycle throughout the day and the season according to 30 year seasonal averages (modeled as described in the supplemental data http://naturalvariation.org/KasCol). The experiment began on May 1 st settings in both Spain and Sweden. In brief, sunrise and sunset changed every day throughout the season, light intensity and spectral mixture changed every six minutes throughout the day as did temperature and relative humidity. Sunrise and sunset times were generated from mathematical models as a function of latitude, longitude, time zone, and day of year (http://www.sci.fi/,benefon/sun.php3). The light quality spectrum and intensity were recreated using far-red (700-750 nm), red (630-700 nm), cool-white (520-600 nm), and blue (400-500 nm) fluorescent bulbs on electronically dimmable ballasts based on the solar spectrum model. The ratio of blue light to red light varied from 0.2 at sunrise to 1.1 at noon and the ratio of red light to far-red light varied from 0.95 at sunrise to 0.65 at noon. Daily temperature variation was modeled based on a general empirical profile fitted over several days. Temperature increased at sunrise, reached maximum shortly after noon and decreased again until sunrise. The amplitude of the curve spans the daily minimum and maximum temperature for the location of interest. Data on minimum and maximum daily temperatures were obtained from Weather.com. The relative humidity changed with temperature maintaining constant total humidity.

Plant Materials
A set of 128 F 6 Kas-1/Col-gl1 recombinant inbred lines (RILs) and the two parental lines (Somerville lines) were obtained from the Arabidopsis Biological Resource Center (Columbus, OH). Fourteen lines were reordered from the Arabidopsis Biological Resource Center (ABRC) due to an inconsistency between previous studies [17,18] and 96 lines were included in the final data set.

Experimental design and flowering time measurements
The two walk-in growth chambers mimiced two geographic locations, Sweden and Spain. There were two blocks in each location labeled Sweden 1, Sweden 2, Spain 1, and Spain 2. Seeds were stratified for 24 hours at 4uC. Initially one to two seeds were planted in each pot of the 469-well flat and thinned to a single seedling after germination. Within each block, 12 lines were planted per 469-cell flat, with 3 pots for each line. Lines were arranged randomly across flats. In total, 12 flats contained 142 lines (128 plus the 14 duplicate lines). This set of 12 flats was replicated an additional 3 times so that four sets of 12 flats were placed in the four blocks. Flats were rotated within the block at a rate of approximately 2 flats per week. Individual pots in each flat were also occasionally rotated within the flat. Plants were watered daily by hand and additional water was provided using an automatic watering system. Each single plant in a pot was bar-coded with a unique identifier which allowed individual tracking of flowering time measured as days to flowering (DTF) i.e. the number of days from sowing until the appearance of the first floral buds monitored almost daily.

DNA extraction and SNP genotyping
Genomic DNA was extracted using MagAttract 96 DNA Plant Kit (Qiagen Inc., Valencia, CA) and KingFisher 96 (Thermo Electron Corporation, Waltham, MA). One plant from each line of the RIL set including the parents and the duplicate lines were genotyped at 65 single nucleotide polymorphism markers by Sequenom (San Diego, CA). Markers were a subset of those described at http:// naturalvariation.org/geno.html.

Preparation of data files
After SNP genotyping, one marker was dropped from further analysis because of severe segregation distortion likely due to duplicate amplification. After cluster analysis (hclust in R, complete linkage method) of the genotypic data generated from 64 SNP markers, lines with genotypes inconsistent with previous reports [17,18] were removed and those with identical genotypes were merged. 18 lines had inconsistent genotypes or were likely to be mislabeled and were excluded from the further analysis: CS84873, C84875, CS84902, CS84904, CS84907, CS84910, CS84912, CS84914, CS84924, CS84925, CS84930, CS84947, CS84970, CS84974, CS84981, CS84991, CS84992, and CS84993. Two lines had identical genotypes (CS84943 = CS84945). The phenotypes for identical lines (including the five duplicate lines and the two identical genotype lines) were merged together. Three lines with less than six measurements of the phenotypic data were also excluded from further analysis.

Genetic map
The genotypic data for the 96 RILs from the 64 SNP markers and the previous 55 markers [18] was analyzed together using GMendel 3.0 [19]. Most of the marker orders were consistent with the physical map (AGI sequence map) of the Col sequence (see supporting information). Map distances were obtained from GMendel 3.0 using the Kosambi map function. The final map was rendered using MapChart 2.1 [20].

QTL analyses
QTL analyses were performed in bQTL, R package (http:// hacuna.ucsd.edu/bqtl). Interval mapping and analysis of posterior probabilities using Bayes model averaging over different, multigene QTL models were carried out separately in the four blocks (Table 1). We present the results of the multigene linear Bayes models with 2-cM scanning resolution. The false discovery rate (FDR) posterior odds threshold was obtained by sorting the results of 400 genotype phenotype permutations scans [21]. Four QTL were detected at 1% FDR, SNP21607030 (Chr1), SNP21607175 (Chr3), MSAT4.39 (Chr4), and SNP44607955 (Chr4). These were included as main effect background markers in subsequent complete pairwise epistasis scans performed in each of four blocks. Significant epistasis was identified in Spain 2 at a P,0.05 genomewise threshold set by 500 permutations. In the final analysis, QTL, environment, QTL6QTL, and QTL6environment interactions were included as fixed effects with block and flat as nested random effects in a single linear mixed-effects model testing each plant separately. RIL was not treated as a random factor in this model. The R scripts, raw data, and other supporting materials are available at http://naturalvariation.org/KasCol.

Genetic map in Kas-1/Col-gl1 RILs
A linkage map (Fig. 2) was established using 119 markers including 64 new SNP loci and 55 previously genotyped markers [18] among 96 Kas-1/Col-gl1 RILs. The total genetic distance covered by these 119 markers was 492.9 cM with the largest gap of 22.5 cM on chromosome 3. Forty-six (39%) markers showed segregation distortion (P,0.05), especially on three regions: the top of chromosomes I and IV, and the middle of chromosome III, where more lines with Col-gl1 alleles than Kas-1 alleles in these three regions. Significant segregation distortion of some regions in the Kas-1/Col-gl1 RILs seen here, were also detected in previous studies [17,18]. Segregation distortions of markers have also been seen in other A. thaliana RIL populations [13,22,23]. This distortion could be due to inadvertent selection of the lines in this mapping population, biases in genotyping error, and/or due to loci containing underlying sterility or incompatibility factors [24][25][26][27].

Flowering-time variation in Kas-1/Col-gl1 RILs
The histograms of the RIL flowering-time reveal continuous, quantitative genetic variation with greater variation in the Spain environment (Fig. 3). The parental line Kas-1 flowered later than Col-gl1 (63.665.5 vs. 35.562.9 in days) across the four experimental blocks. An analysis of variance showed that the floweringtime was significantly different among the 96 Kas-1/Col-gl1 RILs (P,0.01) but the main effect of two ''geographic locations'', Sweden and Spain (P.0.05) was not. Significant interactions of RILs6geographic locations and RILs6blocks were detected (P,0.01). Since the block effect was significant, QTL were separately mapped under the four block environments in the following analyses.

Identification of flowering-time QTL, gene6environment interactions, and epistasis
BQTL mapping was carried out independently in four blocks since blocks within Sweden and Spain environments had significant effects. Four QTL were detected using Linear Bayes analyses at 1% FDR (Fig. 4), and their effects were estimated separately in the four individual blocks (Table 1) or together in the full model ( Table 2, analysis details described in the methods). Kas alleles delayed flowering at two QTL (chr4. 89659 bp and chr4. 5591486 bp) while promoted flowering at the other two QTL (chr1. 27650179 bp and chr3. 5140894 bp) compared to Col alleles. We next tested all possible marker and 1 cM marker interval pairs for interactions with experiment-wise threshold of P = 0.05 by 500 permutations [28]. A significant epistasis was identified between a marker at the QTL on chromosome 1, SNP 21607030 which is 1.3 Mbp above FLM, and a marker at the first chromosome 4 QTL, MSAT 4.39 which is 0.18 Mbp above the FRI locus.
The QTL and environment main effects, QTL6QTL interaction, and all QTL6environment interactions were tested in the final linear mixed-effects model with block/flat as random factors ( Table 2). The four QTL were still significant in the joint analysis (P,0.05). The major QTL near FRI on chromosome 4 showed a strong QTL6environment and minor QTL6QTL interaction ( Table 2). Col is known to carry a deletion at FRI [13]. In Figure 5, three models are presented showing that FRI's effect can be detected as a main effect ( fig 5A) but also responds to both environment (Fig. 5B) and genotype (Fig. 5C). The likely functional Kas allele at this QTL delayed flowering-time by 17.4 days in Spain as compared to the Col loss of function allele ( Table 2, Fig 5A). In the colder Sweden-like environment the effect of the functional Kas allele was reduced by 6.6-days. Thus the magnitude of FRI's effect is environment-dependant, and Kas FRI's effect is reduced in the colder Sweden conditions (Fig. 5B). Additionally, the presence of an effect of FLM on flowering time was dependant on FRI genotype (Fig. 5C). Kas FLM promotes flowering in a Col-fri null background, but has little effect when in a functional Kas-FRI background.

DISCUSSION
In this study, simulated seasonal environments resembling Sweden and Spain with variable diurnal light, temperature, and relative humidity were set up in two walk-in growth chambers. Under these conditions, four QTL responsible for the difference in spring flowering-time between Kas-1 and Col-gl1 were mapped in 96 Kas-1/Col-gl1 RILs using 119 molecular markers. The Kas allele of a large effect QTL at the FRIGIDA (FRI) locus, delayed flowering time as did another minor QTL, while at two other minor QTL Col alleles delayed flowering time. A significant QTL6environment interaction was detected at FRI. The effect was dampened in the cooler Sweden-like environment implying that subtle changes in light quality, rate of change of day length, and temperature are differentially integrated at the genetic level and thus may be responsible for adaptation to regional abiotic cues. The Kas-1 FRI allele does not carry common deletions while Col-gl1 FRI carries a loss of function deletion polymorphism [13].
It is very likely that the FRI indel polymorphism is the molecular basis of the first chromosome 4 QTL (Figure 4).
The significant epistasis between markers SNP 21607030 and MSAT 4.39 suggests a potential interaction between the alleles of FLM and FRI, which has not been reported before. FLM is a MADS-domain gene on chromosome 1 and delays flowering time similar to it close homologue the FLC MADS-domain gene [10]. Our earlier study [11] showed that Col FLM delays flowering compared to null flm alleles, such as found in the Nd ecotype. In the present study, the Kas FLM allele lacks floral repressor function compared with Col FLM, similar to the Nd null allele. This allelic variation however is only seen in the Colfri null allele background which differs from the well known positive effect of FRI on FLC. FLM is a candidate gene for the chromosome 1 QTL; however, on the Col reference sequence, the NGA692 marker is slightly closer to FLM (,120 kb below) compared to SNP 21607030 marker, but is less tightly linked with the chromosome 1 QTL. The epistasis between markers linked to the FLM and FRI loci is suggestive and warrants further study. Creating near isogenic lines that differ for the Kas and Col alleles at FLM in functional and non-functional FRI backgrounds would allow us to confirm the predicted genetic basis of the epistasis identified in this study. If confirmed, the epistatic relation described here between FRI and FLM reveals a novel regulatory evolution that is similar to the FRI-FLC pair [29,30].
The QTL and QTL6environment interaction detected in this study suggest that natural-like environments can reveal potentially ecologically important variation in relevant traits such as flowering time. In addition to contrasting geographic locations at ranges of latitudes as we have done here, the approach can be applied to other contrasts. For example seasons (e.g. spring and fall) can be compared thus allowing experimentation addressing the basis of the life history strategies of winter and summer annual growth habits. We believe that abiotic seasonal and geographic cues are the major predictable local environmental signals to which A. thaliana must adapt in order to succeed. Thus, quantitative trait loci identified under such conditions are more likely to be ecologically relevant and reflect the genetic basis of differences in life history between locally adapted populations.

ACKNOWLEDGMENTS
We appreciate the help from Nicole Morrison with collecting phenotypic data. Kurt Spokas at USDA consulted with aspects of seasonal light and temperature conditions and has developed SolarCalc, an open source java program to implement these conditions.  The first three columns indicate the nearest marker locus, chromosome position in bp, and the genetic distance calculated using Kosambi map function in GMendel. b The fixed QTL effects, environment effects, QTL6environment, and QTL6QTL6environment interactions were tested using linear mixed-effects model with block/flat as random factors.