Floral phenology

Previous studies have shown that flowering time in natural populations of I. fulva and I. brevicaulis acts as a strong pre-zygotic isolating barrier, with only a small proportion of the latest flowering I. fulva overlapping with the earliest flowering I. brevicaulis for approximately 2 weeks in late-April and early May [19, 21]. Across both maps, we found 13 QTL for flowering time. At all eight loci detected in the BCIB population that were associated with variation in flowering time, individuals with introgressed I. fulva alleles flowered earlier than those with the I. brevicaulis allele, and conversely, for all 5 flowering time QTL detected in the BCIF population, introgression of the I. brevicaulis alleles resulted in an increase in the time to flowering. Two pairs of these QTL overlap on homologous linkage groups 4 and 13 in both backcross maps (Figure1, Table 1). These overlapping QTL suggest the potential for allelic differences at the same locus affecting flowering time between I. brevicaulis and I. fulva, although the confidence intervals for each QTL are large, encompassing many genes, leaving the possibility that each overlapping QTL may actually represent different loci affecting flowering time. Within each backcross population, several QTL from different years or treatments overlapped, indicating that there may be loci responding to specific developmental factors (e.g. first vs. second year post transplant) or environmental cues (e.g. wet vs. dry). For example, QTL detected in the dry site in both 2006 and 2007 overlap in both mapping populations (LG 1 in the BCIF population and LG 4 in the BCIB population; Figure1), suggesting that these loci may function to control flowering time in dry conditions. Another pair of overlapping QTL on linkage group eight in the BCIB mapping population were identified using data collected from both the wet and the dry plots in 2006 that may represent either a common locus responding to particular environmental conditions experienced in 2006 or, perhaps, a locus that controls variation in flowering during the first year.

Genetic studies in model systems have shown that flowering time is a complex trait responding to both endogenous and environmental cues, with loci that promote and delay flowering interacting to establish proper timing [32]. The 2006 phenological data using a dominant Iris retroelement (IRRE) marker system was previously analyzed [4] and although most of the QTL that were identified in that study had effect directions consistent with what is found in this study (i.e. I. brevicaulis alleles cause later floral transition, I. fulva alleles cause earlier floral transition), several QTL with opposite effect directions (e.g. I. brevicaulis alleles that cause earlier floral transition, I. fulva alleles that cause later floral transition) were identified [4] that we were unable to detect in this study. This may be attributed to the fact that their study had slightly larger sample sizes and identified a greater overall number of flowering time QTL [4].

Flood tolerance and long-term survival

The habitat commonly associated with Louisiana Irises fluctuates dramatically both throughout the year, and year-to-year, as water levels and temperatures fluctuate. Under changing conditions, it is expected that plants will have environment dependent responses that may appear under stressful conditions (e.g. flood, drought). We evaluated data for both long-term survival in mildly fluctuating conditions as well as survival in extreme flooding conditions in the backcross mapping populations using data from a transplanted field plot that experienced standard environmental fluctuations after 3 years and another plot that experienced abnormally strong flooding. Only two survival QTL were detected: one QTL associated with variation in flood tolerance and one QTL associated with long-term survival. As would be predicted from the habitat associations of the two species [20, 33], introgression of the I. brevicaulis allele into the I. fulva genetic background at either of these loci resulted in decreased survivorship. Two QTL linked to increased survival in the BCIF mapping population that were identified in a previous study were not recovered here [9]. As in the previous studies that analyzed both survival in the greenhouse and the flood survival data using dominant markers, no loci affecting survival were identified in the BCIB populations in this study [9].

Sterility

The BCIB mapping population exhibits higher pollen sterility (32.3 % mean sterility; range 0.58 – 100 % sterile) relative to the BCIF population (7.56 % mean sterility; range 0.18-66.7 % sterile). These BCIB values contrast with the parents used to generate the crosses as Ib 25, If 174, and the F1 hybrid all had pollen sterility less than 10 %. We detected two QTL in the BCIB mapping population in which introgression of the heterospecific (I. fulva) allele resulted in a decrease in the proportion of sterile pollen grains and one QTL in the BCIF mapping population in which introgression of the heterospecific (I. brevicaulis) allele resulted in an increase in pollen sterility. The location of the QTL on BCIB LG 3 associated with a decrease in sterility also corresponds to a region with significant transmission ratio distortion (TRD) whereby I. fulva alleles were overrepresented in the BCIB mapping population [30], consistent with heterozygosity being favored in this region. The other two QTL associated with sterility do not correspond with regions exhibiting TRD. A potential explanation for the increase in fertility in the BCIB individuals associated with the introgression of either of the two I. fulva loci is that homozygosity in I. brevicaulis may have negative effects (i.e. due to inbreeding depression). Both iris species demonstrate a mixed-mating reproductive strategy, but I. brevicaulis demonstrates lower levels of inbreeding than does I. fulva[21, 34]. This pattern of mating would be predicted to result in more heterozygosity in I. brevicaulis, and indeed this species does demonstrate higher levels of heterozygosity relative to I. fulva[30, 35]. Given that a proportion of the heteroyzgosity in these species involves deleterious recessive alleles, we would predict that our crossing design would uncover more deleterious alleles in I. brevicaulis than in I. fulva. Specifically, both backcross populations would have higher levels of homozygosity than would be present in the progeny of natural outcrossing individuals, potentially revealing loci that cause sterility when homozygous [36]. Alternatively, the fact that each backcross population was created from a different F₁ individual (F₁2 and F₁3) may contribute to differential levels of sterility in the backcross populations. Interspecific incompatibilities between the species could also explain the increase in sterility in the BCIB mapping population, although evidence supporting either negative interactions between heterospecific nuclear genes or cytonuclear incompatibilities has not been found [30]. Further crossing experiments among and within I. fulva and I. brevicaulis individuals should help to elucidate the potential causes for pollen sterility.

Growth traits affecting fitness

It is well documented that I. brevicaulis and I. fulva occur in different habitats, indicating that they vary in physiological attributes [9, 15, 19, 33]. In addition, they also differ in vegetative and floral traits that may affect fitness [26, 27, 35]. The interplay between genetic pathways controlling physiology and architecture are likely important for controlling variation in these traits [37]. For example, the ability to generate carbon and nutrient stores may affect the ability of plants to produce more branching points, but this trait is also controlled by genes important for determining the location and frequency of branch production [38]. While developmental pathways controlling branching (number of growth points produced, number of nodes per inflorescence), the transition to flowering (inflorescence present/absent), and fruit production all have unique downstream genetic components, these pathways are also dependent to some extent on physiological processes such as resource accumulation [38–41]. Therefore, it would be expected that some QTL for these traits would be independent and specific to the downstream pathway involved, but some affecting physiology may be shared.

Relative to the previous analysis using dominant markers (i.e. [26]), far fewer QTL were detected in this study (~1/3 as many for BCIB and ~1/2 as many for BCIF). Interestingly, in very few instances were overlapping QTL for the same growth trait identified across study year and habitat (BCIB LG 4, BCIF LG 5, LG 11), however in several cases, overlapping QTL for different growth traits were identified. The limited number of overlapping QTL detected in this study could potentially result from the independent genetic architecture underlying these traits. Alternatively, the small sample size for some of these traits limits our power to detect QTL, so QTL of small effect that may actually be overlapping are not detected. As such, the degree of overlap reported herein is likely a conservative estimate of QTL overlap.

In the plants examined for this study, the number of growth points produced before the flowering season increased in both backcross populations from 2006 to 2007 as the newly planted rhizomes became more established [26]. Although the increased number of growth points translated into a higher likelihood of flowering, this increase in the number of new growth points actually resulted in a lower proportion of growth points that produced an inflorescence [26]. This phenomenon may be explained by the co-occurrence of QTL on BCIF LG 8 where introgression of I. brevicaulis alleles results in a decrease in the number of growth points produced by weight and an increase in the proportion of growth points that produce an inflorescence (Figure1).

QTL for two traits, whether or not an inflorescence is formed (flowering) and the number of nodes produced per inflorescence (branching), are found in the same regions across several individual linkage groups. This can be seen on linkage groups BCIB LG 11 and BCIF LG 3. In both cases, the heterospecific allele is associated with decreases in the trait values. On LG 11, two QTL detected in the BCIB population are in a homologous position with a QTL detected in the BCIF population associated with variation in the number of floral nodes per inflorescence, indicating that I. brevicaulis has an allele in this region that increases inflorescence and floral node production in both genetic backgrounds. A similar pattern is seen on the end of LG 2 where QTL were detected in which the I. fulva allele is associated with increased inflorescence production (in the BCIB population) and higher numbers of floral nodes per inflorescence (in the BCIF population). It is possible that a few genetic regions controlling resource acquisition explain QTL for both of these traits.

One interesting pattern is that all of the QTL in this study associated with the number of flowers per node have negative additive effects, even those identified in both populations on homologous regions of LGs 1 and 19. Usually a second flower is produced at a node only after flowers at all other nodes have fully developed [42]. Therefore, the trait ‘number of flowers per node’ is likely to be affected by whether or not there is enough energy to produce flowers at all nodes and then begin to produce secondary flowers at nodes. In parental populations, I. fulva produces more flowers per node [26] and we do not have an explanation for why all of the QTL identified in this study, independent of cross direction, have negative additive effects unless heterozygosity at each of these loci results in decreased trait values.

Overall genomic architecture of pre-zygotic isolation and hybrid fitness

Introgression of traits in a hybrid zone is dependent on the genetic architecture underlying traits affecting isolation [43, 44]. If QTL underlying traits that contribute to isolation are dispersed throughout the genome, a greater proportion of the genome will be linked to these loci, decreasing the potential for the introgression of beneficial QTL while maintaining species boundaries. However, when these factors are clustered, the likelihood of introgression across the genome is increased, especially if there are positive fitness effects of the donor allele on the recurrent parent [14]. Additionally, introgression is influenced by the effect size and direction of alleles at clustered loci. Most of the QTL identified in this study are relatively evenly distributed across the genome. In this study, flowering time and sterility are the primary traits affecting isolation. The flowering time QTL are distributed throughout the genome, but introgression of any of the alleles identified in this study would have the effect of decreasing reproductive isolation between the parental taxa, weakening this key species isolation barrier. The effect of this shift on fitness is complicated and likely differs depending on other genetic and environmental variables. On BCIB LG 4, QTL for four different traits are located in a relatively small region. These traits include flowering time, sterility, and fruit set – all of which have the potential to affect reproductive isolation and fitness. In this region, introgression of the heterospecific allele increases the fitness of the introgressed individual in that it increases the number of nodes per inflorescence and fruit production while decreasing sterility, suggesting that this region may likely introgress across species boundaries in nature, although this introgression may have the added effect of decreasing isolation through flowering time. One region of the genome that would likely experience selection against introgression is on LG 9 where I. brevicaulis alleles in the I. fulva background increase sterility and are linked to additional QTL decreasing fitness (fewer inflorescences per growth point and fewer flowers per node). However, QTL increasing sterility are not widespread in this study, suggesting that this trait only restricts introgression in a small portion of the genome.

Construction of mapping populations and linkage maps

Reciprocal first generation backcross (BC₁) mapping populations were generated between wild-collected individuals of I. brevicaulis and I. fulva. The parental I. brevicaulis genotype (Ib25), used as the maternal parent, was collected from an oak hardwood forest in St. Martin Parish, Louisiana, while the parental I. fulva genotype (If174), used as the paternal parent, was collected along bayou margins in Terrebonne Parish, Louisiana. The clonal nature of these irises allowed for multiple clones of genetically identical plants to be used to generate the backcross populations. Multiple clones of two F₁ individuals (F₁3 and F₁2) were used as the pollen parents and crossed to clones of Ib25 and If174, respectively, to generate first generation backcross populations to I. brevicaulis (the ‘BCIB’ population) and I. fulva (the ‘BCIF’ population). Seeds from these crosses were planted in 1999 and were repotted from a single rhizome each subsequent year.

A subset of the backcross individuals, 94 BCIB and 92 BCIF, were used to generate linked high-density genetic linkage maps using 232 and 237 co-dominant EST-SSR markers, respectively [30]. For this study, the remaining backcross plants were genotyped at a subset of the EST-SSR markers. 239 BCIB plants were genotyped for 131 markers and 168 BCIF plants were genotyped for 123 markers (118 shared markers) distributed across each map. Genotyping of the microsatellite markers was also as described in [30]. New maps were generated from these data using Mapmaker 3.0 [45–47]. Initial framework maps for each backcross population were generated using a likelihood odds (LOD) threshold of 7.0 and a maximum recombination frequency threshold of 0.4. Decreasing the LOD threshold to 5.0 and 3.0 allowed for the remaining unlinked markers to be incorporated into the maps. Map distances (cM) were calculated using the Kosambi mapping function. The maps utilized in the current study have thus been modified slightly from those published in [30].

Assaying trait values

The following traits were examined in this study: (1) flowering time, (2) flood tolerance, (3) long-term survival, (4) sterility, (5) number of growth points (branches on the rhizome) produced before the flowering season, (6) presence/absence of an inflorescence on each plant, (7) proportion of growth points that produced inflorescences, (8) number of flowering nodes produced per inflorescence, (9) number of flowers produced per node, (10) whether or not flowering plants set fruit, and (11) proportion of flowers that set fruit. The data for most of these traits (2, 5–11) have been previously reported and QTL for these traits were mapped on non-linked genetic maps created with a dominant marker system. Some data for flowering phenology, hybrid sterility and long-term survival are presented for the first time here.

Traits 1–3, and 5–11 were all assayed in field conditions that relied on similar experimental plots located in southern Louisiana. Generally, these experimental plots were set up in a similar manner in which several ramets from the same individual of every BCIB and BCIF plant were transplanted into evenly spaced (0.5 m) randomized positions. Differences between these previous experimental plots, highlighted below, involve the specific location of each plot, the exact number of clones and genotypes present in each plot, and the years and conditions in which various traits were assayed.

Flood tolerance was assayed in an experimental field site along the edge of the Choupique Bayou in the Atchafalaya Basin Floodway in Louisiana (as described in [9]). In 2005, this field site experienced flooding of an abnormally long duration that acted as an extremely strong selective agent. Multiple clonal replicates for each of 185 BCIB genotypes and 209 BCIF genotypes, totaling 416 BCIB and 357 BCIF individuals were assayed for flood tolerance, defined as the proportion of clonal replicates that survived this flooding event.

Flowering phenology and growth traits affecting fitness (traits 5–11 above) were assayed in two field plots also located along the Choupique Bayou in the Atchafalaya Basin Floodway (the same plots described in [4, 23, 26]. These two field plots were qualified as “wet” and “dry,” with the “wet” plot remaining inundated with shallow water for a much longer duration after heavy rains. From 1–5 clones for each of 243 BCIB and 172 BCIF genotypes were planted into each plot in 2005. In 2006 and 2007, flowering phenology and the growth traits affecting fitness were measured in each plot. Flowering phenology for each plant was measured in days after the date at which the first flower was observed each year (data for 2006 have been analyzed previously in [4], data from 2007 have not been previously analyzed). The number of new growth points was counted prior to the flowering season each year (January 2006 and March 2007) and we controlled for the original weight of the planted rhizome. Plants were scored for whether or not they produced an inflorescence, the proportion of ramets that produced inflorescences, the number of flowering nodes produced per inflorescence, the number of flowers produced per node, whether or not each flowering plant set fruit, and the proportion of flowers per plant that set fruit, as described in [26]. Long-term survival was assessed in another nearby plot, located along the Choupique Bayou in the Atchafalaya Basin Floodway. A total of 1200 backcross hybrid plants (average 2.4 clones/genotype) were transplanted into this plot in October 2008 and examined for survival in the spring of 2011.

Pollen sterility was assessed in 194 BCIF and 258 BCIB genotypes in the spring of 2001. Pollen grains were collected from one anther per plant. The proportion of fertile pollen grains was assessed by examining pollen stained with a solution of lactophenol-aniline blue [48] and counting the number of both fertile and infertile pollen grains. Lactophenol-analine blue stains starch molecules of potentially viable pollen grains, while inviable pollen grains with no starch do not uptake the stain [48]. Approximately 500 pollen grains per plant were scored to estimate levels of sterility among the parental and backcross genotypes.

QTL analysis

For each trait, when data were available for multiple clones per genotype, trait values were averaged across clones in a per plot and per annum basis (i.e. trait values for all clones per genotype in the wet plot in 2006 were averaged). Averaging across clonal replicates renders nominal traits, such as flood tolerance, long-term survival, and the presence/absence of inflorescences and fruits, as continuous characters. The distribution of a quantitative trait is likely determined by several underlying QTL, each of which can affect the phenotype in a number of different ways. Therefore, applying transformations to normalize data from this “mixture distribution” is not appropriate for QTL mapping studies, and such transformations were not applied to data in this study [49, 50]. Windows QTL Cartographer version 2.5 [51] was used to conduct composite interval mapping (CIM; [52]) on the data to identify QTL. CIM was carried out using forward and backward regression at 2-cM intervals separately on the BCIB and BCIF maps. A 10-cM window was used to exclude closely linked cofactors, and the number of control markers was set at five (the default program setting). For each trait, 300 permutations were run to calculate the genome-wide significance threshold for declaring a QTL. The locations of significant QTL are reported as the map location of the point where the LR statistic is the greatest, including confidence intervals of 2-LOD on either side. When two QTL peaks for the same trait occurred in close proximity on the same linkage group, a drop below the permutation cut-off or a change in the directionality of the QTL effect was used to determine whether each peak represented different QTL. The directionality of the effects (whether or not heterospecific alleles increased or decreased trait values), additive effect sizes, and proportion of variance explained (R²) of QTL were also calculated using CIM.

Genotypic information for use in the QTL analyses was only available for plants that survived until 2009 when they were genotyped for the linkage maps. Consequently, the sample sizes in the current analyses are somewhat smaller than sample sizes in previous analyses. Calculations of the percent variation explained by a given QTL are known to be over-estimated when sample sizes are small [53], therefore, we are more interested in interpreting the directionality of the QTL and the relationship among the QTL for the traits of interest. Values for both directionality and the percentage of variance (R²) explained for each trait are given in Table 2.