We examined the level of postzygotic reproductive isolation in F1 and F2 hybrids of reciprocal crosses between the Arabidopsis lyrata subspecies lyrata (North American) and petraea (European). Our main results are: first, the percentage of fertile pollen was significantly reduced in the F1 and F2 compared to the parental populations. Second, mean pollen fertility differed markedly between reciprocal crosses: 84% in the F2 with ssp. lyrata cytoplasm and 61% in the F2 with ssp. petraea cytoplasm. Third, 17% of the F2 with ssp. petraea cytoplasm showed male sterility (produced less than 30 pollen grains in our subsample). The hybrids were female fertile. We used QTL mapping to find the genomic regions that determine pollen fertility and that restore cytoplasmic male sterility (CMS). In the F2 with ssp. lyrata cytoplasm, an epistatic pair of QTLs was detected. In the reciprocal F2 progeny, four QTLs demonstrated within-population polymorphism for hybrid male sterility. In addition, in the F2 with ssp. petraea cytoplasm, there was a strong male fertility restorer locus on chromosome 2 where a cluster of CMS restorer gene-related PPR genes have been found in A. lyrata. Our results underline the importance of cytonuclear interactions in understanding genetics of the early stages of speciation.

Established species are often separated by multiple prezygotic and postzygotic reproductive barriers (e.g., Lowry et al. 2008), and interspecific hybrids often suffer from reduced viability or fertility (for a review, see Coyne and Orr 2004). Some degree of postzygotic reproductive isolation may already be observed in early stages of divergence for example, in hybrids between highly differentiated populations (Fishman and Stratton 2004; Willet and Berkowitz 2007). Reduced fitness of hybrids is often thought to result, during allopatry, from accumulation of Bateson–Dobzhansky–Muller incompatibilities (BDMI), negative interactions between genes that have diverged between the hybridizing populations (Dobzhansky 1936; Muller 1942; Orr 1995; 1996). In the simplest scenario, different mutations accumulate at two loci in allopatric populations, creating at each of the two loci novel alleles that, upon secondary contact, turn out to be incompatible with each other.

Genes causing BDMI have been identified in Drosophila, yeast, mouse, Arabidopsis thaliana, rice, and Mimulus (reviewed in Presgraves 2010). In some cases even the interacting pair of genes is known (Brideau et al. 2006). These speciation genes commonly appear to have been subject to positive selection (Ting et al. 1998; Barbash et al. 2003; Presgraves et al. 2003; Brideau et al. 2006; Maheshwari et al. 2008; Tang and Presgraves 2009). Results of QTL mapping of speciation genes in plants, for example Mimulus (Bradshaw and Schemske 2003; Fishman and Willis 2006; Sweigart et al. 2006), Solanum (Moyle and Graham 2005; Moyle and Nakazato 2008) and Helianthus (Lai et al. 2005) suggest some differences from the patterns seen in animals (reviewed in Coyne and Orr 2004), such as a lower number of QTLs for reduced hybrid fertility. Besides this finding, Moyle et al. (2004) found a weaker relationship between the degree of divergence of the parental species and the degree of hybrid incompatibility in plants than in animals. The number of studies is, however, still too low to merit firm conclusions. In plants in particular, aside from loci influencing hybrid viability in A. thaliana (Bomblies et al. 2007; Bikard et al. 2009), reduced hybrid fertility in rice varieties (Chen et al. 2008; Long et al. 2008), and cytoplasmic male sterility in Mimulus (Case and Willis 2008; Barr and Fishman 2010), the genes underlying reduced hybrid fitness remain mostly unknown.

Not only do the genes involved in BDMI often remain unidentified, but their naturally occurring variation has also remained mostly uncharacterized. Variation, or the lack of it, could shed light on the question how hybrid incompatibility alleles accumulate to eventually establish reproductive isolation between populations. Fixation of alleles at BDMI loci within populations would imply spread by directional selection, whereas polymorphism would accumulate due to drift or balancing selection (Shuker et al. 2005). Studies on mouse (Forejt and Iványi 1974, Good et al. 2008), Mimulus (Christie and Macnair 1987; Sweigart et al. 2007; Martin and Willis 2010), Drosophila (Reed and Markow 2004; Reed et al. 2008), and Chorthippus (Shuker et al. 2005) have identified intraspecific polymorphism for hybrid incompatibility genes.

Hybrid compatibilities may develop between nuclear genes, but also between nuclear and cytoplasmic genes, causing asymmetric reduction of fertility or viability in reciprocal crosses (Tiffin et al. 2001; Levin 2003). Darwin (1859) was already aware that the progeny of reciprocal crosses often shows different levels of viability or fertility reduction (see Tiffin et al. 2001 for review). Cytoplasmic male sterility (CMS) is a common manifestation of asymmetric hybrid sterility in plants and is often seen in reciprocal crosses (Schnable and Wise 1998). It has been shown to arise from chimeric genes in mitochondria and may cause complete lack of anthers or pollen, or lower the quality of pollen grains (reviewed by Hanson 1991; Schnable and Wise 1998). Pollen production can be restored by nuclear male fertility restorer gene(s) (Rf) which often contain pentatricopeptide repeats (reviewed by Schnable and Wise 1998; Hanson and Bentolila 2004). Crosses between species may reveal CMS alleles even in fully hermaphrodite species (Fishman and Willis 2006). Asymmetric reduction of fertility or viability in reciprocal crosses may also be due to genetic maternal effects, and in plants due to pollen–stigma interactions or triploid endosperm interactions (reviewed in Turelli and Moyle 2007).

We examine the genetics of reduced hybrid fertility between the two subspecies of A. lyrata: ssp. lyrata and ssp. petraea (Al-Shehbaz and O’Kane 2002; Koch et al. 2008). Arabidopsis lyrata is outcrossing, perennial, and self-incompatible (Schierup 1998), a close relative of A. thaliana, and has become a model species for ecological and evolutionary genetics (Clauss and Koch 2006; Savolainen and Kuittinen 2010). It has an allopatric, disjunct distribution in Central and Northern Europe, Russia, and in North America (Hoffmann 2005). Isozyme, microsatellite and sequence analyses have shown that populations representing North American, most European and European Russian lineages are genetically highly differentiated (Jonsell et al. 1995; van Treuren et al. 1997; Balañá-Alcaide et al. 2006; Muller et al. 2008; Ross-Ibarra et al. 2008). Crosses between these populations produce seeds without any apparent difficulties. An earlier study of one such cross between Swedish and Russian populations revealed a considerable number of transmission ratio distorted loci (Kuittinen et al. 2004). The level of distortion was as high as in many between-species crosses reviewed in Jenczewski et al. (1997). This non-Mendelian inheritance is expected to be caused by selection against heterospecific allele combinations in hybrids. Because in A. lyrata within population segregation distortion appears rare (Leppälä et al. 2008), such high transmission ratio distortion between populations in this species is likely a sign of postzygotic isolation.

Here, we investigate intrinsic postzygotic reproductive isolation that has evolved in allopatry between the subspecies of A. lyrata, using two populations: one North American (Mayodan, North Carolina) representing A. lyrata spp. lyrata, and one European (Spiterstulen, Norway) representing spp. petraea. These populations are genetically highly diverged: 22 microsatellites suggest FST= 0.67 (Muller et al. 2008), and based on sequences from 77 genes FST was estimated to be 0.39 between an American and a Swedish population, closely related to the Norwegian one studied here (Ross-Ibarra et al. 2008). We study whether male and female fertility are reduced in between subspecies hybrids, and use quantitative trait loci (QTL) mapping to find the genomic regions that reduce male fertility (pollen quantity and quality) and female fertility (seed quantity and quality). By comparing QTLs between the reciprocal F2 progenies, we investigate the genetic basis of asymmetric fertility reduction.

Materials and Methods


The present study is based on reciprocal crosses between two populations, representing the subspecies of A. lyrata: a European population of ssp. petraea from Spiterstulen, Norway (abbreviated Sp hereafter, [61°38′N, 8°24′E]) was crossed with North American populations of ssp. lyrata from Mayodan, North Carolina (abbreviated Ma hereafter, [36°25′N, 79°58′W]). It has been shown that A. lyrata populations differ in their morphology (Jonsell et al. 1995) and are locally adapted (Riihimäki et al. 2005; Leinonen et al. 2011). Silent site nucleotide divergence between the two populations is approximately 4.3% (19 genes, E.A. Aalto et al., unpubl. data).

Parents for the cross were grown from field-collected seeds. One Sp plant was crossed as a pollen donor to an Ma pollen recipient to produce F1 hybrids with Ma cytoplasm (See Fig. 1 for the crossing scheme). To be able to obtain a reciprocal F2 progeny, a different Ma plant was crossed as a pollen donor to a second Sp pollen recipient, resulting in F1 hybrids with Sp cytoplasm. Two of the F1 hybrids were then crossed reciprocally to yield two sets of F2 individuals: one set with Ma cytoplasm (hereafter MaSpF2) and one with Sp cytoplasm (hereafter SpMaF2). Note that the same nuclear alleles segregate in both F2 progenies because all F2 individuals are full-siblings. This crossing design was applied because A. lyrata is self-incompatible, and to avoid inbreeding depression.

Figure 1.

Crossing design. The four parents (P) from two populations (Sp, A. lyrata ssp. petraea; Ma, A. l. ssp. lyrata) were crossed to produce two F1 (F1) individuals. In the crosses, pollen recipient is mentioned first. The F1 individuals were crossed reciprocally to produce two F2 (F2) progenies. The gray background indicates cytoplasm from Spiterstulen (Sp).


To assess phenotypic differences between the parental populations and the hybrids, 78 plants from Sp and Ma (both from four seed families), 33 F1 plants (15 SpMaF1 and 18 MaSpF1), and 441 F2 plants (227 SpMaF2 and 214 MaSpF2) were grown in a randomized block design in a greenhouse. To avoid environmental effects, the Sp and Ma plants used for phenotypic measurements were not grown from field-collected seeds but from seeds from laboratory-generated seed families. Seeds were germinated on Petri dishes and grown under 8 h of light in a growth chamber for the first two weeks after sowing. Two weeks after germination, seedlings were individually planted to their pots in the greenhouse of the Botanical Gardens of Oulu, Finland, in April 2005. Plants were grown in a 1:1 mixture of peat and gravel. For the first four weeks after planting, plants were grown without fertilizer in the greenhouse, and then transferred into vernalization, at 4°C and 8 h of light, for eight weeks. At the end of June, after vernalization, plants were taken back to the greenhouse into natural long-day conditions (about 22 h of daylight). The vernalization and long-day conditions were chosen to assure maximal flowering for fertility assessment. Plants were watered when needed and fertilized weekly to avoid reduced fertility caused by drought or lack of nutrients.


To assess male fertility, quality and quantity of pollen was measured. Two unopened buds were collected from each plant, and anthers were dissected. Anthers from one bud were immediately crushed and mixed with 25 μl of lactophenol-aniline blue staining solution (Kearns and Inouye 1993), and samples were stored in cold and dark. The numbers of viable (stained) and inviable (nonstained) pollen grains were counted from a subsample of 4 nL in a hemocytometer under a microscope. From these data, both the total number of pollen grains and the proportion of fertile pollen grains were calculated. As the number of pollen grains, we used the count from the 4-nL subsample, because it is directly proportional to the total number of pollen grains in the flower. Anthers from the other bud were treated similarly, and numbers of pollen grains and percentages of fertile pollen from the two buds were averaged to obtain a single value for each plant for further analyses.

To assess female fertility, we measured seed quality and quantity. Seeds were collected when ripe, counted, and their size measured using image analysis (ImageJ, Seeds were classified as good quality (plump) or low quality (shriveled or green, not having reached maturity). To obtain seeds, plants were crossed with pollen from cloned, tissue-cultured plants: one Sp clone and one Ma clone. The use of pollen from clones eliminates the possibility that differences in seed production are due to genotypic differences between the pollen donors instead of between the pollinated plants. Two flowers from each plant were pollinated with Sp pollen and two flowers with Ma pollen.

The proportion of good quality seeds and the proportion of fertile pollen were not normally distributed. In particular, some SpMaF2 individuals produced only occasionally a few pollen grains (Fig. 2). There was a clear discontinuity in the cumulative distribution of the number of pollen grains at about 30 pollen grains. Therefore, plants that produced less than 30 pollen grains in the counted subsample were considered sterile, and for some analyses male fertility in SpMaF2 was treated as a binary trait: fertile versus infertile.

Figure 2.

Number of pollen grains in the F2 hybrids between subspecies petraea (Sp) and lyrata (Ma) of Arabidopsis lyrata. On the x-axis is pollen number and on the y-axis the number of F2 individuals. Mean pollen counts of parental plants Sp (n= 36), Ma (n= 38), and for each F2 reciprocal are marked with arrows. For Sp cytoplasm, there are two F2 means, one for all SpMaF2 plants marked as “F2” and one excluding male sterile (plants producing less than 30 pollen grains) individuals marked “F2-MS.”

Differences in male and female fertility between parental, F1 and F2 plants were tested for significance with Kruskal–Wallis rank-sum tests. This test is a nonparametric equivalent of one-way analysis of variance (ANOVA), appropriate for nonnormally distributed data. When the Kruskal–Wallis test suggested significant differences, these were identified with pairwise comparisons using Wilcoxon rank-sum tests. In the following, significant differences refer to a P-value smaller than 0.05. Statistical tests were performed with R 2.9.0 (R Development Core Team, 2010).


To determine genotypes of the four parental, two F1, and 441 F2 plants, leaves were collected from the plants at the end of the experiment and DNA was extracted in a plate format either with DNeasy 96 Plant kit (Qiagen, Valencia, CA) or Nucleo Spin Plant kit (Macherey-Nagel, Düren, Germany). The plants were genotyped for altogether 77 marker loci, of which 24 were microsatellites and 53 were SNPs or SNP-based CAPS markers. We attempted to genotype all four parents, two F1, and 441 F2 plants for all 77 marker loci, but due to technical issues we did not obtain all marker genotypes for all individuals. We genotyped 391 of the 441 F2 plants (204 SpMaF2 and 187 MaSpF2). For these 391 F2 individuals, we obtained 98% of the genotypes for the 77 marker loci.

The 24 microsatellites used were a selection of the most informative of 45 loci tested in the F1. Microsatellite primers were either from Bell and Ecker (1994); Clauss et al. (2002); Loudet et al. (2002); Kuittinen et al. (2004), or designed for this study (See Table S1 for primer sequences). Details on microsatellite PCR conditions and fragment analysis are given in Leppälä et al. (2008). The SNP markers were developed by sequencing protein coding genes in F1 parents. Primers were obtained from Hansson et al. (2006), Kivimäki et al. (2007), Kuittinen et al. (2004), from D. Charlesworth and S.I. Wright (from the genes sequenced in Ross-Ibarra et al. [2008]) and from A. Niittyvuopio. Some additional primer sequences were newly designed for this study. See Table S1 for primer sequences and further details on the markers.

For genotyping F2 hybrids with SNPs found from F1 nucleotide sequences, MALDI-TOF mass spectrometry (Sequenom, San Diego, CA) genotyping was used in most cases. We had two sets of SNPs (25 and 20) multiplexed in Sequenom runs at the FIMM Technology Centre (Institute for Molecular Medicine Finland). For some loci, it was feasible to obtain more informative markers by combining information from two SNPs. For some of the sequences, a polymorphism for restriction endonuclease cutting site was available. The PCR product was then cut by a restriction endonuclease and the fragments were visualized on agarose gels. There were 14 CAPS markers of which one was a dCAPS marker. For a dCAPS marker, a new restriction endonuclease recognition site is created by a mismatched PCR primer; otherwise the genotyping is done in the similar way as for CAPS markers. The dCAPS mismatch primer was designed with dCAPS Finder 2.0 (


To construct a genetic linkage map we used JoinMap version 3.0 (Van Ooijen and Voorrips 2001). The markers were grouped with a LOD threshold of 9.0 and Kosambi's mapping function was used to calculate pairwise marker distances.

QTL analyses were performed with R/qtl (Broman et al. 2003), which, apart from mapping normally distributed characters, also allows nonparametric interval mapping as well as interval mapping of binary traits. The numbers of seeds and pollen grains were normally distributed, and the proportion of fertile pollen (excluding male infertile individuals) was normally distributed after logit-transformation. Because we did not find a suitable transformation for the proportion of good quality seeds, we used nonparametric interval mapping for this trait. Because male fertility differed between F2 reciprocal crosses, QTL analyses were conducted separately for the SpMaF2 and MaSpF2. Male sterile individuals in the SpMaF2 were excluded from QTL mapping of the proportion of fertile pollen. Instead, they were included in interval mapping of male sterility as a binary trait.

Genetic incompatibilities affecting fertility between populations, such as Bateson–Dobzhansky–Muller interactions, are typically thought to involve epistasis between two or more loci, so that it is possible that no single-locus effects on fertility are detected. Often, however, effects will be evident at the single QTL level. Therefore, we first performed single-locus QTL scans by interval mapping, followed by two-QTL scans to see whether further QTL appeared, or to confirm the presence of two QTLs on the same chromosome (for which the two-QTL model is more appropriate). Also, the two-QTL model can be used to evaluate whether two QTLs interact epistatically. For each detected QTL, additive, dominance, and parent-of-origin effects were estimated. The genome-wide threshold for significance of a QTL (P < 0.05) was set by permutation (n= 10,000 for single-QTL, n= 800 for two-QTL). The genome-wide thresholds for significance were determined separately for different traits and reciprocal progenies.



The two parental populations were highly male fertile, as measured by the proportions of fertile pollen grains, which were 0.943 (SE = 0.007) and 0.983 (SE = 0.002) in Sp and Ma, respectively. The F1 plants showed significantly reduced fertility when compared to the parental populations (mean = 0.882, SE = 0.019), but the reciprocal F1 hybrids did not differ significantly from each other. The F2 hybrids showed even lower pollen fertility than the F1 hybrids (Fig. 3). Interestingly, male fertility differed also significantly between the reciprocal F2 progenies: in those with Sp cytoplasm (SpMaF2) the proportion of fertile pollen was 0.609 (SE = 0.025), significantly lower than the 0.837 (SE = 0.013) in the F2 with Ma cytoplasm (MaSpF2). On average, pollen fertility was reduced by 37% in SpMaF2 and by 13% in MaSpF2 relative to the mid-parent value.

Figure 3.

Pollen fertility in A. lyrata F2 hybrid flowers. Mean pollen fertility (ratio of fertile pollen grains to total pollen grains) of parental plants (Sp and Ma), F1 (n= 33), and each F2 reciprocal are marked with arrows as in Figure 2.

Seventeen percent of SpMaF2 individuals were male sterile (Fig. 2), producing less than 30 pollen grains in the counted subsample. If these male sterile individuals were removed from the analysis, pollen fertility still remained significantly lower in the SpMaF2 (SpMaF2 without individuals with fewer than 30 pollen grains: 0.736, SE = 0.019) than in MaSpF2.

In addition to the proportion of fertile pollen, we also measured the number of pollen grains per flower. Here the differences between parental populations were substantial (Fig. 2), with Ma plants (197 grains per flower, SE = 10.34) producing almost twice as many pollen grains per flower as Sp individuals (116, SE = 7.49). The SpMaF1 resembled the Sp parent by producing on average 119 (SE = 9.5) pollen grains per flower, but the MaSpF1 was significantly different from its reciprocal cross by producing 159 (SE = 16.47) pollen grains. The number of pollen grains per flower in the MaSpF2 hybrids (153, SE = 4.47) was close to the mid-parent value (156.5), but in the SpMaF2 (115, SE = 5.06; without individuals with fewer than 30 pollen grains 131, SE = 4.26) the number of pollen grains was significantly lower than in the reciprocal cross. The low pollen production of the SpMaF2 was, however, not significantly different from that of the Sp population (Fig. 2). Thus, male fertility was reduced in F1 and F2 individuals as compared to the parental populations if measured as the percentage of fertile pollen. But if measured as the number of pollen grains produced per flower this reduction depended on the cytoplasm: the plants with Sp cytoplasm (both F1 and F2) resembled the Sp population, producing less pollen than those with Ma cytoplasm, which were in between the parental populations in number of pollen grains per flower. Thus, reduced male fertility was altogether more pronounced in the F2 with Sp cytoplasm.


We used R/qtl to locate genomic regions associated with male fertility, separately for the SpMaF2 and MaSpF2 reciprocal crosses. In the SpMaF2 we detected two QTL for the proportion of fertile pollen (individuals with fewer than 30 pollen grains excluded) in the single-locus QTL analysis, and two additional QTLs with the two-QTL additive model (Fig. 4, for LOD scores supporting QTLs and for estimated significance thresholds see Table S2). Two of these QTLs were located at the opposite ends of chromosome AL1: the first one close to marker F20D22 and the other close to marker AT1G36310 (Fig. 4). At both of these QTL, the lowest pollen fertility was found in plants that inherited one allele from Sp and the other from Ma. The direction of the effects of these two QTL was different: near marker F20D22 individuals heterozygous for Sp1Ma2 alleles had the lowest percentage of fertile pollen. In contrast, near marker AT1G36310 Ma1Sp2 heterozygotes had the lowest pollen fertility (Fig. 5).

Figure 4.

QTLs for pollen fertility and seed quality and quantity projected on the linkage map of A. lyrata ssp. petraea Spiterstulen ×A. lyrata ssp. lyrata Mayodan, North Carolina. QTLs for reduced pollen fertility are indicated in gray and for restorer of fertility and seed traits in black. Markers underlined experienced significant transmission ratio distortion.

Figure 5.

Effect plots of QTLs for the proportion of fertile pollen grains. Colors indicate different reciprocal crosses with Sp cytoplasm (black) and Ma cytoplasm (gray). S1S2 is a genotype homozygous for Spiterstulen alleles, M1M2 for Mayodan alleles. M1S2 and S1M2 are heterozygotes, where M1S2 gets its Ma allele from an F1 parent with Sp cytoplasm, and its Sp allele from the parent with Ma cytoplasm, and vice versa for the S1M2 genotype.

The third QTL for the percentage of fertile pollen in SpMaF2 was on chromosome AL4, close to marker TCL1. Also here, the lowest percentage of fertile pollen was found in one of the two heterozygotes (Sp1Ma2) but pollen fertility of plants that inherited both alleles from the Ma population was also low (Fig. 5). The fourth QTL was found with the two-QTL model and was located at AL5. Also at this locus low pollen fertility was observed in plants either carrying the pollen donor's Ma allele and pollen recipient's Sp allele or in plants with both alleles from Ma (Fig. 5). These differences in male fertility between the two heterozygote classes demonstrate that hybrid sterility alleles are polymorphic both within species and populations.

In the reciprocal F2 progeny, with Ma cytoplasm, QTL mapping of pollen fertility gave very different results. No significant QTL were found at the 5% significance level, but at 10% significance level two QTL were found (LOD-score 7.68 against a threshold of 6.34). A scan with two-QTL model lent further support for these same QTL (LOD-score, additive model, 14.94 against threshold 11.2 at 5% significance level). Thus, the evidence for a pair of QTL was clear although it was difficult to distinguish, with a fit of a QTL model, whether they acted epistatically or additively. Comparing different QTL models, the evidence for epistatic versus additive action was close to the 5% significance level with a LOD-score of 7.25 versus 7.81, which was the highest LOD for interaction in this experiment. One of these QTLs was located at about the same location on chromosome AL1 (close to marker F20D22) as one of the QTLs detected in the reciprocal progeny (Fig. 4). The other QTL was located on AL8 close to marker TOC1. The interaction between the QTLs is best visible from a plot showing the percentage of pollen fertility of the different two-locus genotypes (Fig. 6), where it can be seen that having both alleles from Ma at either locus, while having one Ma and one Sp allele at the other locus, leads to low pollen fertility.

Figure 6.

Effect plot of QTLs for the proportion of fertile pollen grains, showing interaction between pollen fertility QTLs on AL1 (F20D22) and AL8 (TOC1), in the reciprocal cross with Mayodan cytoplasm. Genotypes are labeled as in Figure 5.

As mentioned above, 17% of SpMaF2 plants produced less than 30 pollen grains in the counted subsample and were considered male sterile (Fig. 2). Male sterility in the SpMaF2 could thus be regarded as a binary trait (fertile vs. infertile). We mapped male sterility as a binary trait using R/qtl and found a QTL at chromosome AL2 near marker AT1G62520 (Fig. 4). The strong influence of this QTL is clear from the difference in pollen production between genotypes (Fig. 7): All individuals, except one, with fewer than 30 pollen grains had both alleles from Ma at marker AT1G62520. However, not all individuals homozygous for Ma alleles at AT1G62520 were male sterile: 22 such individuals (out of 54) produced more than 30 pollen grains and were thus considered fertile. These plants produced on average 107 (SE = 11.41) pollen grains per flower.

Figure 7.

The number of pollen grains produced by different genotypes at restorer of male fertility QTL near marker AT1G62520 at AL2, in reciprocal crosses of A. lyrata subspecies petraea and lyrata.

We did not find significant QTL for the number of pollen grains per flower when individuals with fewer than 30 pollen grains were removed from the analysis. When male sterile individuals were included in the analysis, a QTL for the number of pollen grains per flower appeared on AL2, in the same region where the male sterility QTL (described in the previous paragraph) was found.


In the crossing design we used, one expects the four parental alleles to be transmitted to the F2 in equal proportions (i.e., each allele 25%). In both F2 reciprocal progeny 34–40% of the markers (underlined in Fig. 4) showed significant deviation from this expectation (as evaluated using χ2 tests), a phenomenon known as transmission ratio distortion (TRD). Details on TRD will be explored in future studies, but we present the location of markers in TRD (Fig. 4) here because three of the five male fertility QTLs described above were located on TRD regions.


Female fertility was measured as the number of seeds produced and as the percentage of good quality seeds. Initially we measured female fertility after pollination with Ma tester pollen, and after pollination with Sp tester pollen. However, pollination with Sp pollen yielded lower numbers of seeds as well as a lower percentage of good quality seeds than pollination with Ma pollen. This difference was present in all plants, that is, also the Sp individuals produced fewer seeds when pollinated with pollen from their own population, and many seeds appeared aborted. Therefore, we assume that the Sp tester pollen was of insufficient quality and we present in the following, only the results on seeds pollinated with Ma tester pollen.

The F2 hybrids and parental populations did not differ significantly in the number of seeds per silique produced. Sp plants produced on average 20.0 seeds per silique (SE = 1.4) and Ma plants 17.5 seeds per silique (SE = 1.0). Also, there were no significant differences in the number of seeds per silique between the MaSpF2 and SpMaF2 reciprocal progenies (combined F2 mean = 16.6, SE = 0.4). The reciprocal F1 progenies differed significantly from each other; mean for the number of seeds per silique in SpMaF1 was 10.2 (SE = 1.3) and for MaSpF1 18.4 (SE = 1.7). The sample sizes were very low for these comparisons, only nine individuals for SpMaF1 and 14 in MaSpF1. Because there were no significant differences between the reciprocal F2 progenies, the data from the F2 progenies were combined for QTL mapping. For the number of seeds produced per silique, we found one QTL at AL6 (Fig. 4), where the highest number of seeds was produced by heterozygous plants with one allele from each parental population (Fig. 8).

Figure 8.

Effect plots of two QTLs (on the left) for the proportion of good seeds and a QTL (on the right) for the number of seeds per silique for the F2 progeny between A. lyrata ssp. petraea and ssp. lyrata.

The parental populations differed significantly in the percentage of good seeds produced (Sp mean = 0.635, SE = 0.06; Ma mean = 0.962, SE = 0.01). The percentage of good seeds produced by the F2 hybrids (combined F2 mean = 0.830, SE = 0.05) was very close to the mid-parent value, and did not differ significantly between the MaSpF2 and SpMaF2 reciprocal progenies. For the proportion of good quality seeds, we detected two QTL; at chromosome AL1 (near marker AT1G31930) and at AL8 (near marker AtCLH2) (Fig. 4). At the QTL in AL1, the lowest proportion of good seeds was produced by plants that inherited both alleles from Sp, and the highest proportion by plants with both alleles from Ma, reflecting the difference in proportion of seeds of good quality in the parental populations. In contrast, at AL8 plants homozygous for Ma alleles (M1M2) produced the lowest proportion of good seeds (Fig. 8), showing that such alleles also are found in this population with generally high proportion of good seeds.


Studies on the genetic background of reproductive isolation in plants have made remarkable progress during the last decade (reviewed in Lexer and Widmer 2008; Lowry et al. 2008). Nevertheless, our understanding of the genetic basis of intrinsic hybrid sterility remains limited to a few well-studied groups, like Mimulus, Helianthus, Solanum (formerly Lycopersicon), and Oryza (Lai et al. 2005; Moyle and Graham 2005; Song et al. 2005; Fishman and Willis 2006; Sweigart et al. 2006; Moyle and Nakazato 2008; 2010). Research has focused on crosses between species, although interest for genetic incompatibilities within species has increased (e.g., Bomblies et al. 2007; Martin and Willis 2010). The number of studies of genetics of postzygotic reproductive isolation in natural populations belonging to the same species remains limited (but for selfing angiosperms see Bomblies et al. 2007; Skrede et al. 2008; Bikard et al. 2009; for outcrossing angiosperm Martin and Willis 2010; or for a bryophyte species McDaniel et al. 2008). The present study on fertility of hybrids between A. lyrata subspecies aims to improve this understanding, investigating the QTLs involved in incipient speciation.


In our experiments, both reciprocal F2 hybrid progenies between subspecies suffered from reduced male fertility. This was clearly mediated partly by cytoplasm, as both the extent of fertility reduction and the underlying QTLs differed between the reciprocal F2 hybrid populations. Only one QTL was common for both reciprocal crosses (upper arm of chromosome AL1) with allelic effects in the same direction. Otherwise the three QTL in SpMaF2 and one in MaSpF2 were different, which demonstrates significant genotype by cytoplasm interactions. This emphasizes the importance of cytonuclear interactions in the formation of reproductive isolation in A. lyrata: the same nuclear gene combinations segregate in both reciprocal crosses, so cytoplasmic or epigenetic interactions must be involved in the reduction of pollen fertility.

Cytoplasm-dependent male sterility (CMS) is common in plants. Most often it is seen in gynodioecious plants, where it controls sex determination within a population, or in crosses of cultivated plants (Hanson and Bentolila 2004). CMS has recently been found in a cross between one population of the hermaphrodite species, Mimulus guttatus and nasutus (Fishman and Willis 2006). Just as in other cases of CMS (reviewed by Hanson and Bentolila 2004), it is caused by altered transcription in mitochondria (Case and Willis 2008). In this system, male fertility is restored by a single dominant nuclear locus that has been mapped to a region with several copies of pentatricopeptide repeat (PPR) genes (Barr and Fishman 2010), which have been identified as male fertility restorers also in other species (Hanson and Bentolila 2004).

In the present cross, the major fertility restorer was found on AL2, near a marker in gene AT1G62520. Interestingly, a cluster of PPR genes, the closest homologs of CMS restorer genes of petunia, radish, and rice, in A. thaliana locates in the region syntenic with our fertility restorer QTL at AL2 (Lurin et al. 2004). Recently, a study by Fujii et al. (2011) identified a cluster of fast evolving PPR genes in A. lyrata, closely related to known Rf genes, between 0.6 and 5Mb at chromosome AL2. This is the same chromosomal region where we mapped our fertility restorer QTL (AT1G62520 is located at 1.7Mb in A. lyrata). Together with the findings in Mimulus and some cultivated plant species (reviewed by Hanson and Bentolila 2004), this suggests a common molecular mechanism of CMS might be universal in plants, and similar in hermaphrodites and gynodioecious species.

In the current study, the major fertility restorer on AL2 is dominant, but there must be at least one other restorer locus (undetected in this study) that restored the male fertility of 22 plants of the 54 plants homozygous for the Ma allele at AL2. The other option is that the severity of F2 hybrid male sterility is affected by factors not related to fertility restorer loci but for example, to the number of mitochondria in the cells or other genetic or nongenetic variance in the hybrids.

It is not fully understood why CMS appears in hermaphroditic plants (McCauley and Olson 2008), and as previous studies were typically conducted on hybrids between hermaphroditic species or within gynodioecious species, standing genetic variation for CMS within hermaphroditic populations has not often been studied. Case and Willis (2008) found a CMS mitotype to be fixed within one hermaphroditic Mimulus guttatus population, whereas it was mostly absent from the other populations studied (see also Martin and Willis 2010). The Sp population harbors male sterile plants in nature (P.H. Leinonen and J. Leppälä, pers. obs.), suggesting that the fertility restorer(s) segregate in the Sp population, perhaps because CMS developed only recently.


In Drosophila hybrid incompatibility loci generally act recessively (Presgraves et al. 2003; Tao and Hartl 2003; Coyne and Orr 2004). In plants both recessive and dominant hybrid incompatibility loci have been observed. In between-species NILs of tomato, the chromosomal insertions induced reduced female fertility only if homozygous (Moyle and Graham 2005; Moyle and Nakazato 2008), suggesting that the genes that affect hybrid female fertility act recessively. Dominant hybrid incompatibility alleles have also been documented (Macnair and Christie 1983; Sweigart et al. 2006). In the present study, (partial) dominance of one of the alleles (either Ma or Sp) could be seen in all the QTL for male fertility except the interacting QTL at AL8 (Fig. 5). Interestingly, pollen fertility differed markedly between different heterozygotes, depending on from which parent they inherited their alleles. For example, at the QTL at AL4, the heterozygotes in the SpMaF2 all have Sp cytoplasm, and all have one allele from Sp and one allele from Ma. Yet, the plants that inherited the Ma allele from the pollen donor (genotype S1M2 in Fig. 5) had much lower pollen fertility than the individuals that inherited the Ma allele from their mother (M1S2 in Fig. 5). Pollen fertility of the Ma homozygote (M1M2) was as low as the fertility of the worse heterozygote.

The above finding that reduced male fertility was caused by one of the Ma alleles (M2), but not by the other (M1), indicates that variation exists within population in hybrid incompatibility alleles. The source of this polymorphism could not be indentified here and could be, for example, due to DNA sequence variation in the sterility causing locus, gene expression differences, epigenetic effects, or interactions between loci. Possibly, ancestral and derived alleles segregate in the populations we studied. Segregating alleles affecting hybrid incompatibility have been observed in populations of other species such as M. guttatus (Christie and Macnair 1987; Sweigart et al. 2007). Polymorphism in hybrid incompatibility genes can provide some insights into their development. If alleles that cause incompatibilities in hybrids experience strong directional selection in their own genetic background, we would expect them to become rapidly fixed within populations. Polymorphisms at hybrid incompatibility loci are therefore expected to be more common if they are selectively (nearly) neutral or under balancing selection in their own genetic background (Shuker et al. 2005). Of course, hybrid incompatibility alleles may also be segregating within a population if they are of recent origin so that they have not had time to go to fixation. Polymorphisms for hybrid incompatibility alleles have been demonstrated for relatively recently diverged species (Wade et al. 1997; Reed and Markow 2004) but also in centrarchid fish that have diverged from their common ancestor 13–16 million years ago (López-Fernández and Bolnick 2007). In our study five of six of the male fertility QTLs appeared polymorphic within populations. The probability of detecting such a number of polymorphic loci under strong selection would seem rather low, suggesting that balancing selection or drift are likely to be involved in hybrid incompatibilities for male fertility. Recent findings suggest that intrinsic postzygotic reproductive barriers may evolve as byproducts of evolution of nearly neutral or selfish genetic changes instead of evolving as byproducts of adaptations to ecological conditions (reviewed by Presgraves 2010). Our results also suggest that these former processes have a role in the development of hybrid incompatibility between A. lyrata subspecies, as polymorphism in some of the male fertility QTL may be due to nearly neutral alleles and CMS is a well-known example of selfish evolution of mitochondrial DNA.

The interaction of the male fertility QTLs at AL1 and AL8 in MaSpF2 was of a peculiar nature: the lowest male fertility was observed in individuals that had both alleles from Ma at either locus, while having one allele from each population at the other locus. Moreover, this interaction also required a third component: cytoplasm from Mayodan (or some epigenetic mechanism), because the same nuclear genotypes did not experience reduced fertility when the cytoplasm was from Spiterstulen. This is neither typical dominant nor recessive interaction. It also does not fit the common expectation of BDMI where the incompatibility would be expected in individuals with both alleles from Sp at one, and both alleles from Ma at the other locus. Co-segregation of alleles causing reduced hybrid fertility has not been studied often, but some studies on Solanum and Drosophila give indications that complex interactions may underlie hybrid incompatibility (Reed et al. 2008; Moyle and Nakazato 2009; Chang and Noor 2010).

Overall, the number of male fertility QTL that we found in A. lyrata was relatively low. This seems to be the trend in plants (Moyle and Graham 2005; Sweigart et al. 2006; Moyle and Nakazato 2008), different from findings in Drosophila where the genetic background of hybrid male fertility reduction is generally complex (Tao et al. 2003). Of course, the probability to detect a QTL depends not only on the number and effect of QTL, but also on the number of individuals studied and the density of the marker map, so one may detect a low number of QTL because of a lack of statistical power. However, the few QTL found in this experiment explain a high proportion of the variance in fertility: about 30% and 50% of male fertility variance in the MaSpF2 and SpMaF2, respectively. The difference in percentage of variance explained may be due to fertility being less reduced in the MaSpF2, and with smaller total variance.


In the present study, hybrids do not appear to suffer reduced female fertility. This is in contrast with earlier studies in plants where both male and female fertility has been found to be reduced in hybrids (Fishman and Willis 2001; Moyle and Graham 2005; Song et al. 2005; Moyle and Nakazato 2008). Whereas in D. melanogaster species group incompatibilities causing hybrid male sterility accumulate faster than those causing hybrid female sterility or hybrid inviability (Presgraves et al. 2003; Tao and Hartl 2003, and references therein). In Drosophila, this is assumed to be due to males being the heterogametic sex, vulnerable to recessive incompatibilities. In addition, according to the faster-male theory, sexual selection is stronger on male reproductive traits, accelerating evolution of male fertility genes. Hermaphrodite plants do not have sex chromosomes, but in A. lyrata sexual selection could well be stronger for male (pollen) related traits, because pollen competition is feasible in this self-incompatible and therefore strictly outcrossing species.

The Ma population produced seeds of better quality than the Sp population, which produced more seeds per silique but also aborted seeds at a higher rate. Seed quality of the F2 hybrids clustered around the mid-parent value, and did not differ between reciprocal crosses. Excluding the data obtained with Sp pollen, we detected two QTL for seed quality. At both of these loci, the poorest seeds were produced by plants that inherited both alleles from the same population (from Sp at AL1 and from Ma at AL8). Therefore, and because seed production did not differ between reciprocal crosses, these QTL likely reflect the differences in seed production between the Sp and Ma populations, and not hybrid incompatibility. The QTL for seed number did not display hybrid incompatibility, but heterotic behavior. Only one of the heterozygotes at this QTL showed hybrid vigor, which illustrates the variability of alleles within populations.


Our study contributes to the growing knowledge of reproductive isolation and its genetic basis in plants. The results presented here stress the importance of cytonuclear interactions in the development of reproductive isolation. In the hybrids between the pair of subspecies we studied, the male function was reduced, whereas seed production remained high. The hybrid incompatibility alleles were observed to be polymorphic within populations. To further understand the generality of these results, additional crosses between different populations of the subspecies will be conducted. Additional crosses will reveal the geographic extent of the mitochondrial variant causing CMS, and whether this mechanism for male sterility is common in between population crosses of A. lyrata. Further crosses will also show if male fertility is always more severely affected compared to female fertility and if fertility reduction has the same genetic background.

Associate Editor: K. Dyer


The authors would like to thank P. H. Leinonen and D. L. Remington for help with R and discussions on the QTL analyses and H. P. Koelewijn for discussions on CMS. We also thank F. Bokma, the members of the Plant Genetics Group in Oulu and the two anonymous reviewers for their valuable comments on this manuscript; L. C. Moyle for encouraging discussion in the early stages of this project; U. Kemi, J. Kiiskilä, P. H. Leinonen, M. Otsukka, S. Remula, R. Syrjänen, T. Teräväinen, and T. Toivainen for their help with the greenhouse experiment or in the laboratory. The staff of Botanical Gardens of the University of Oulu, especially T. Kangas, A. Kestilä, and A. Hämäläinen, are thanked for their help with the plants. This work was supported by the Academy of Finland's Research Council for Biosciences and Environment (grants 107167 and 120809), by the Finnish Population Genetics Graduate School and by Biocenter Oulu.