Do large effect QTLs fractionate? A case study at the maize domestication QTL teosinte branched1

Quantitative trait loci (QTL) mapping is a valuable tool for studying the genetic architecture of trait variation. Despite the large number of QTL studies reported in the literature, the identified QTL are rarely mapped to the underlying genes and it is usually unclear whether a QTL corresponds to one or multiple linked genes. Similarly, when QTL for several traits colocalize, it is usually unclear whether this is due to the pleiotropic action of a single gene or multiple linked genes, each affecting one trait. The domestication gene teosinte branched1 (tb1) was previously identified as a major domestication QTL with large effects on the differences in plant and ear architecture between maize and teosinte. Here we present the results of two experiments that were performed to determine if the single gene tb1 explains all trait variation for its genomic region or if the domestication QTL at tb1 fractionates into multiple linked QTL. For traits measuring plant architecture, we detected only one QTL per trait and these QTL all mapped to tb1. These results indicate that tb1 is the sole gene for plant architecture traits that segregates in our QTL mapping populations. For most traits related to ear morphology, we detected multiple QTL per trait in the tb1 genomic region including a large effect QTL at tb1 itself plus one or two additional linked QTL. tb1 is epistatic to two of these additional QTL for ear traits. Overall, these results provide examples for both a major QTL that maps to a single gene, as well as a case in which a QTL fractionates into multiple linked QTL. A. J. Studer and J. F. Doebley 4 Quantitative trait loci (QTL) mapping studies have become widely used to elucidate the genetic architecture of trait variation in many organisms (Mackay et al. 2009). A common observation in these studies is that QTL of large effect are often detected. Noor et al. (2001) have questioned whether such large effect QTL represent single genes or groups of tightly linked genes. These authors have suggested that such large effect QTL, upon closer examination, might fractionate into multiple linked small effect QTL, representing multiple genes. A single QTL does not necessarily equal a single gene. Even in cases where QTL effects have been fine mapped to a specific gene, the research may not formally exclude the presence of additional linked genes that contribute to the overall QTL effect for that genomic region. Doebley and Stec (1991, 1993) identified a QTL of large effect on the long arm of maize chromosome 1 controlling the differences in plant and ear architecture between maize and teosinte. These authors proposed that tb1, a known mutant of maize, was the gene underlying this QTL because tb1 falls within the 1 LOD support interval for the QTL, and because the tb1 mutant and the QTL affect the same suite of traits. Subsequently, Doebley et al. (1995) used a complementation test that showed that the teosinte allele of the QTL fails to complement the tb1 mutant of maize, indicating that they are alleles of the same gene. However, complementation tests do not provide formal proof because of the potential for non-allelic non-complementation. Additional support for the hypothesis that tb1 is the gene underlying the major domestication QTL was obtained after the cloning of tb1 (Doebley et al. 1997). With a knowledge of the molecular identity of tb1, Doebley et al. (1997) showed that the maize allele of this gene is expressed at twice the level of the teosinte allele in the developing branch and in immature ears. Thus, a change in gene regulation was hypothesized to be the causative difference between maize and teosinte. Finally, Clark et al. (2006) provided formal proof that tb1 is the A. J. Studer and J. F. Doebley 5 QTL by fine-mapping the QTL to a 12 kb “control region” located ~58-69 kb upstream of the tb1 open reading frame. These authors further demonstrated that this control region contains a factor that acts as a cis-regulatory element with the maize allele conditioning a higher level of tb1 expression than the teosinte allele. However, their data does not address the possibility of additional QTL linked to tb1, and indeed some of their data suggest that such additional linked QTL may exist, i.e. that tb1 fractionates into multiple linked QTL. In this paper, we report two experiments performed to address whether there are additional QTL closely linked to tb1. In Experiment I, we analyzed a mapping population in which the tb1 control region identified by Clark et al. (2006) is fixed for the teosinte haplotype, but the regions flanking it are segregating for maize vs. teosinte chromosomal segments. If there are additional QTL linked to the control region, then there should be phenotypic effects associated with the segregating maize vs. teosinte chromosomal segments despite the fact that the tb1 control region is not segregating. Conversely, if the control region alone explains all phenotypic effects, then there should be no phenotypic effects associated with the flanking chromosomal regions. In Experiment II, we analyzed a set of nearly isogenic recombinant inbred lines (NIRILs) for the tb1 genomic region to see if we could detect any QTL other than tb1. This experiment has more power than a standard QTL analysis to detect closely linked QTL because the NIRILs have an isogenic background and the NIRILs were grown in replicate to obtain better estimates of QTL effects. Based on these two experiments, we confirm that tb1 is a large effect QTL contributing to the differences in plant and ear architecture between maize and teosinte. In fact, tb1 is the only QTL for plant architecture traits that we detected. However, we identify four additional QTL affecting ear architecture. One of these additional QTL is located only 6 cM upstream of tb1. A. J. Studer and J. F. Doebley 6 Two of these additional QTL have significant epistatic interactions with tb1. Thus, our results provide examples for both a major QTL that maps to a single gene as shown for plant architecture, as well as a case in which a QTL fractionates into multiple QTL as shown for ear architecture. MATERIALS AND METHODS Plant materials: Segments of the long arm of chromosome 1 from teosinte were introgressed into a maize inbred W22 background for both Experiments I and II. For Experiment I, a segment of the long arm of chromosome 1 from a teosinte (Zea mays ssp. mexicana; collection Wilkes-Panindicuaro) was introgressed into W22 via six generations of backcrossing (Figure 1). A BC6S1 line (I01) that was homozygous for the teosinte alleles at markers bnlg615 and bnlg1671, which flank tb1, was recovered. I01 was then crossed to W22 and the F2 progeny of this cross were screened for crossovers near tb1. A plant with one of the newly identified recombinants was itself crossed to W22, and the F2 progeny of this cross were screened for crossovers near tb1. From this process, a homozygous introgression line (I16) containing an ~69 kb segment of teosinte chromosome which encompasses the tb1 upstream control region and part of the ORF was recovered (Clark et al. 2006). Homozygous I01 and I16 lines were crossed and the resulting F1 plants were selfed to produce an F2 population for Experiment I. For Experiment II, a segment of the long arm of chromosome 1 (T1L) from a teosinte (Zea mays ssp. parviglumis; Iltis and Cochrane collection 81) was introgressed into W22 via six generations of backcrossing (Figure 1). During the backcrossing process, molecular markers were used both to follow the target segment surrounding the QTL on the long arm of chromosome 1, as well as to eliminate teosinte chromosome segments at other major A. J. Studer and J. F. Doebley 7 domestication QTL identified by Doebley and Stec (1993) (Table S1). Six separate BC6 plants heterozygous for the target segment were selfed to give six BC6S1 families (designated families A-E). These six families were selfed an additional five generations to produce a set of 153 homozygous nearly isogenic recombinant inbred lines (NIRILs). These 153 lines were distributed among the six families as follows: A: 24, B: 31, C: 39, D: 25, E: 19, F: 15. These lines possess a set of maize-teosinte recombinant chromosomes for the tb1 genomic region in the W22 genetic background. These 153 lines make up the QTL mapping population of Experiment II. Molecular markers and linkage map: Plants in Experiment I were genotyped using a PCR-based indel marker, GS3, previously described by Clark et al. (2006). GS3 is located in the coding region of tb1 and segregates in the I01 × I16 F2 population. Plants in Experiment II were genotyped using a set of 25 PCR-based markers: 16 SSRs, 6 insertion or deletion (indel), and 3 markers scored for the presence/absence of a PCR product (Figure 3). Marker information is available at either Panzea (www.panzea.org) or MaizeGDB (www.maizegdb.org). There were a total 174 crossovers among the 153 lines, averaging 1.1 crossovers per line. The distribution of crossovers among lines was as follows: 0 (46 lines), 1 (52 lines), 2 (44 lines), 3 (10 lines) and 4 (1 line). A genetic map was constructed using the Kosambi map function and a genotyping error rate of 0.0001 as parameter values for the “est.map” command in the R/qtl module of the R statistical computing package (Broman et al. 2003). Phenotypic data collection: The plants for Experiment I were grown at the University of Wisconsin West Madison Agricultural Research Station, Madison, WI, USA during summer 2006. F2 seed from three ears (A, B, and C) generated by three separate I01 × I16 crosses was planted in a randomized complete block design using a grid with 0.9 meter spacing between A. J. Studer and J. F. Doebley 8 plants in both dimensions. This spacing minimized the degree to which plants shaded their neighbors. The following five traits were phenotyped for Experiment I: cupules per rank (CUPR; number of cupules in a single rank from base to the tip of the ear), ear diameter (ED; diameter, in mm, of the midsection of each ear), lateral branch internode length (LBIL; mean internode length, in cm, of the uppermost lateral branch), tillering (TILL; the ratio of the sum of tiller heights/plant height), and tiller number (TILN; the number of tillers per plant). CUPR and ED were both measured on the uppermost, well-formed lateral inflorescence (ear) of each plant. The NIRILs for Experiment II, along with the backcross parent W22, were grown using a randomized complete block design at the University of Wisconsin West Madison Agricultural Research Station, Madison, WI, USA during summer 2008. The design included three replicates (blocks A, B, and C) with a single 10-plant plot of each NIRIL per replicate. Each plot was 3.7 m long and 0.9 m wide. The plots within each block were arranged in a grid with row and column designations so that position effects could be included during data analysis. Three plants were phenotyped per plot. In addition to the five traits measured in Experiment I, the following three traits were evaluated: 10-kernel length (10KL; length, in mm, of 10 consecutive kernels in a single rank along the ear), ear length (EL; distance, in cm, from the base to the tip of the ear), and percent staminate spikelets (STAM; percentage of male spikelets in the inflorescence). 10KL, CUPR, ED, EL, and STAM were all measured on the uppermost, well-formed lateral inflorescence (ear) of each plant. Data analysis: For Experiment I, we used the GLM procedure of SAS (Littel et al. 1996) to compare the effects of the I01 and I16 introgression segments on phenotypes. Genotype (homozygous I01, homozygous I16, or heterozygous) and ear parent (A, B or C) were considered as fixed effects. The general linear model used was A. J. Studer and J. F. Doebley 9 Yijk = μ + ai + bj + eijk where Yijk is the trait value for the k plant from the j ear parent with i genotype, μ is the overall mean of the experiment, ai is the genotype effect, bj is the ear parent effect, and eijk is the sampling error. Using this model, the effects of the different introgressions (I01 vs. I16) were evaluated. For Experiment II, we obtained least-squares means for each NIRIL using the MIXED procedure of SAS (Littel et al. 1996). The NIRIL (or parental) lines and families (A-E) were considered fixed effects while blocks (A, B, and C) and plot coordinates were treated as random effects. The linear model used was Yhijklm = μ + ah(bi) + bi + cj + dk + fl + ehijkl + ghijklm where Yhijklm is the trait value for the m plant at l column and k row in the j block of the h NIRIL nested in the i family, μ is the overall mean of the experiment, ah is the NIRIL (or parental) line effect, bi is the family effect, cj is the block effect, dk is the row effect, fl is the column effect, and ehijkl is the experimental error (random variation among plots), and ghijklm is the (within-plot) sampling error. All fixed effects were significant and were included in the model for the calculation of the least-squares means. The random effects of this full model were subjected to the Likelihood Ratio Test for significance for each trait. Effects that were not significant were dropped from the model on a trait by trait basis. The least-squares means estimates were used for QTL mapping in Experiment II, which was conducted in the R/qtl module of the R statistical computing package (Broman et al. 2003). For each trait, an initial QTL scan was performed using simple interval mapping with a 0.25 cM step (Lander and Botstein 1989) and the position of the highest LOD score was recorded. Statistical significance of the peak LOD score was assessed using 10,000 permutations of the A. J. Studer and J. F. Doebley 10 data (Doerge and Churchill 1996). Then, the position and effect of the QTL was refined using the Haley-Knott Regression method (Haley and Knott 1992) by executing the “calc.genoprob” command (0.25 cM step size and assumed genotyping error rate of 0.001), followed by the “fitqtl” command. To search for additional QTL, the “addqtl” command was used. If a second QTL was detected, then “fitqtl” was used to test a model containing both QTL and their interaction effect. If both QTL remained significant, the “refineqtl” command was used to reestimate the QTL positions based on the full model including both QTL. Finally, each QTL was removed from the model and then added back using the “addqtl” command to re-confirm its significance and position. Approximate confidence intervals for the locations of the QTL were obtained via 1.5 LOD support intervals to each side of the position of the LOD maximum. We calculated broadsense heritabilities (H) for Experiment II on a line mean-basis