Identification of diploid males in natural populations of O. bicornis

We morphologically characterized O. bicornis individuals collected from 683 nesting tubes from two locations in Europe (Uppsala, Sweden and Halle, Germany). O. bicornis nests typically contain males closest to the entrance and females at the back, with males being lower in mass (Fig 1C). Potential diploid males were identified based on their occurrence among females in nests or due to elevated mass, or both (see Materials and methods). We also collected females from nests containing potential diploid males, and from randomly selected nests. In total, we performed whole-genome sequencing of 162 individuals (S1–S3 Tables). These comprised 88 potential diploid males and 74 females.

We mapped sequencing reads to the O. bicornis reference assembly iOsmBic2.1 to an average sequencing depth of 13×. We masked heterochromatic regions from the dataset (see Materials and methods, S1A and S1B Fig; S1 Table), resulting in a dataset consisting of 3,573,693 SNPs. We determined the ploidy of our samples by the presence of a significant number of heterozygous sites based on the inbreeding coefficient, F, which is expected to equal one in haploids (when analyzed in the same way as diploids). All females were confirmed as diploid. Out of the 88 potential diploid males, 18 were confirmed to be diploid, whereas 70 were haploid. Our final dataset, therefore, contained 18 diploid males, 70 haploid males, and 74 females (S1C and S1D Fig).

Females from nests containing diploid males had a slight but significantly higher F than those sampled from random nests (mean F = 0.452 compared to 0.230, Mann–Whitney U test, p = 0.0042). However, we found no significant difference in F between diploid males and their sisters (mean F = 0.298 compared to 0.452, p = 0.356). Overall, this indicates that matings that produce diploid males are between individuals that are more closely related than average, but are not close relatives.

After determining the ploidy of all potential diploid males, we were able to estimate the frequency of matched matings in the population in which the mother shares an allele at the CSD locus with the father, leading to diploid male production. We corrected for the probability that a diploid male is not observed in a matched mating based on the number of progeny in each nest. In total, we observed 6 nests containing diploid males from a total of 672 nests found in the nesting tubes. We estimate a matched mating frequency of 1.04% (SD 0.44%; S4 Table). This frequency did not differ significantly between locations or sampling year (Fisher’s exact test, p > 0.05 for all comparisons). The mass of confirmed diploid males was not statistically different from female nestmates, or from all females (S2A Fig; t test, p > 0.05).

Candidate CSD locus orthologous to ANTSR identified on chromosome 1

We next analyzed the genomes of diploid males and females for runs of homozygosity (ROH). The percentage of the genome in ROH blocks >10 kbp ranged from 0.19% to 67.4% between samples (S2B Fig). The distribution of this proportion was similar between diploid males and females. There was a slight but nonsignificant tendency for females from nests with diploid males to have higher homozygosity (average proportion of genome in ROH blocks >10 kbp for females from nests with diploid males = 34.4%, nests without diploid males = 31.6%; S2B Fig).

We then used the genomic distribution of ROH to identify candidate regions for CSD. We divided the ROH into genomic segments of uneven size across all individuals. Using a linear model, we associated the presence or absence of ROH segments across the genome with the sex of the diploid individuals. This analysis identified a single 2 kbp candidate region on chromosome 1 with coordinates 14,404,983–14,406,957 in which the presence of ROH is highly associated with sex (Fig 2A; p < 10⁻¹⁶ after adjustment for multiple testing). All diploid males, and none of the females, are completely homozygous across this region. This is the only region of the genome where strong differentiation is observed in our dataset and is therefore likely to contain the CSD locus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Mapping the CSD locus.

A. Association of presence/absence of runs of homozygosity (ROH) with sex in diploid samples using a linear model (sex ~ ROH; n = 18 diploid males, n = 74 females; correction for genome-wide significance using Benjamin–Hochberg procedure). B. Nucleotide diversity in 10 kbp windows across the genome. C. Close-up of association results on chromosome 1. The red dotted lines between panels B and C indicate the location on Panels A and B. D. Nucleotide diversity across the candidate region estimated in 1 kbp sliding windows. E. Genomic features in and around the candidate region. The red vertical bar indicates the best-scoring ROH segment across all subplots. Data from panels A–E can be found in files Data1–Data3 within https://doi.org/10.5281/zenodo.17424723 and S5 Table.

https://doi.org/10.1371/journal.pbio.3003458.g002

Genomic regions responsible for CSD are expected to evolve under frequency-dependent selection, which results in the presence of a large number of divergent alleles to avoid the occurrence of homozygosity in diploids [15]. This process leads to high levels of nucleotide diversity at CSD loci. We found that the candidate CSD region coincides with a local maximum in nucleotide diversity estimated in 10 kbp windows across the genome (Fig 2B). Nucleotide diversity in the region is 0.017, which is an order of magnitude higher than the chromosome average (mean_CHR = 0.0014, stdev_CHROM = 0.00087). It is the highest peak on chromosome 1, the fourth highest peak genome-wide, and in the top 99.86% of 10 kbp bins genome-wide. This supports the action of frequency-dependent selection in the region.

The candidate CSD region overlaps a single uncharacterized lncRNA in the genome annotation (Fig 2C–2E). The identity and order of genes flanking this lncRNA are identical to those flanking ANTSR, the lncRNA implicated in sex determination in the ant L. humile [10]. These include COPA, CRELD2, miR-315, and THUMPD3 (Fig 2E). This indicates that the CSD locus identified here overlaps the ANTSR ortholog. ANTSR has 2 exons annotated in the O. bicornis genome assembly and the 2 kbp CSD locus overlaps 272 bp of the intron, 354 bp of exon 2, and 1,348 bp of downstream sequence.

High genetic diversity in CSD concentrated in specific subregions

In total, we identified 197 SNPs and 26 indels in the CSD locus. We identified 188 distinct haplotypes in the full sample of 254. A tree showing the relationships between haplotypes reveals high diversity across several clades (Fig 3A). We examined the distribution of heterozygous sites in females across the CSD region to attempt to identify critical regions where heterozygosity is required to activate female development (Fig 3B). Heterozygosity in the ANTSR exons within the CSD locus does not seem to be required to trigger female development as 7/74 females do not have any heterozygous sites in the overlap between the candidate region and exonic regions. The distribution of heterozygous sites in females is variable across the CSD locus. We define three regions where heterozygosity is particularly elevated (S5 Table): in an intron (P1), overlapping the final exon and downstream noncoding region (P2), and a downstream noncoding region (P3). These regions are particularly notable for a high number of indel polymorphisms. These candidate functional regions could potentially affect female development by influencing transcription levels of ANTSR (S6 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Haplotypes across the 2 kbp CSD locus.

A. Maximum-likelihood tree of all unique haplotypes across the mapping population (74 females, 70 haploid males, 18 diploid males, 254 total haplotypes). Circles at branch tips indicate the number of observations of each haplotype. B. Count of differences between the two haplotypes carried by females across the candidate region. The top panel shows differences across all females, with the candidate region highlighted in light red. Three regions with larger differences within the region are labeled P1, P2, and P3, highlighted with blue, green, and orange bars at the bottom of the panel, respectively. The middle panel shows how ANTSR is located within the candidate region. The bottom two panels show a rug plot of heterozygous SNPs across the region in blue, with indel starting positions shown as red bars, and their extent in orange. C. Histogram of the total number of heterozygous sites across the candidate region per individual. D. Histogram of the haplotype frequency spectrum showing the number of times each haplotype is observed in the population samples. Data for panels A–D can be found in files Data4–Data7 within https://doi.org/10.5281/zenodo.17424723.

https://doi.org/10.1371/journal.pbio.3003458.g003

We found that the two haplotypes present in females at this locus differ on average at 24.6 SNPs and 9.5 indels, but there is a large variation in this number (Fig 3C). We examined the importance of heterozygosity in the three regions we defined (S3 Fig). We identified 11 females that were homozygous across one of these regions and one that was invariable across two regions (S6 Table). The minimum number of differences observed between the two haplotypes in a female in our dataset is 3 SNPs, which is found in two individuals. These observations suggest that none of these regions is critical for determining sex on its own, and that heterozygosity at one or more of a large number of critical sites across these regions is required for triggering female development, indicative of genetic heterogeneity.

The majority of haplotypes (n = 152) are seen only once in the sample, with 36 haplotypes observed multiple times (Fig 3D). The frequency of diploid male production in a randomly mating population is expected to be approximately 1/k, where k is the number of sex alleles [4]. In our populations, this proportion is ~0.53% (S4 Table), which is the same as the theoretical expectation of 0.53% assuming that there are 188 haplotypes at the CSD locus, although it is likely that additional haplotypes exist in the population that are not observed in our sample.

Elevated nucleotide diversity implicates ANTSR in sex-determination in other bee and ant species

We identified the location of the ANTSR ortholog in the genomes of nine other bee species (Apis mellifera, Anthophora quadrimaculata, Anthophora retusa, Anthophora plagiata, Bombus monticola, Bombus lapponicus, Bombus balteatus, Lasioglossum albipes, and Lasioglossum pauxillum) and three ant species (Formica polyctena, Formica aquilonia, and Solenopsis invicta; Fig 4A and 4B; S7 Table). In each case, we could identify a candidate ANTSR ortholog including the same set of flanking genes as in O. bicornis, indicating a conserved genomic neighborhood.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Nucleotide diversity around ANTSR in other bee and ant species.

A. Variation in nucleotide diversity (π) is shown in 10 kbp windows across the contig containing ANTSR in each of the following species: Lasioglossum albipes and Lasioglossum pauxillum (both shaded red) Apis mellifera (shaded yellow), Bombus lapponicus, Bombus monticola, Bombus balteatus (all Bombus shaded gray), Anthophora retusa, Anthophora plagiata, Anthophora quadrimaculata (all Anthophora shaded green), Osmia bicornis (shaded white), and three ant species (Formica polyctena, Formica aquilonia, and Solenopsis invicta, all shaded orange). The vertical red line and red arrow above the plots indicate the location of the ANTSR ortholog for each species. B. Phylogenetic tree indicating the relationship between the investigated genera. The tree is based on the distance matrix from BeeTree [25], and Romiguier and colleagues [53]. The data underlying the nucleotide diversity plots can be found in files Data2 and Data8–Data19 within https://doi.org/10.5281/zenodo.17424723.

https://doi.org/10.1371/journal.pbio.3003458.g004

We assayed variation in heterozygosity in 10-kbp windows across the genomes of these species. There is no elevated variation in the orthologous region of Apis mellifera, in which sex determination is known to be controlled at a locus unrelated to ANTSR [8,16]. In all other species of bees and ants that we investigated, we find markedly elevated genetic variation in the vicinity of the ANTSR ortholog (S8 Table). In A. quadrimaculata, A. retusa, B. monticola, B. lapponicus, L. pauxillum, and L. albipes, we observed a peak of elevated genetic variation in the 10 kbp window containing ANTSR (genetic variation in the top 0.3% of windows genome wide). In A. plagiata, B. balteatus, S. invicta, F. polyctena, and F. aquilonia, such a peak was found between 0.5 and 45 kbp (S8 Table), away from ANTSR.

These findings are consistent with the action of frequency-dependent selection in the vicinity of ANTSR in all of the bee and ant species we investigated, with the exception of Apis mellifera. This lncRNA gene is therefore likely to be the sex-determining locus in all of these species in addition to Osmia bicornis. Together, our findings suggest that ANTSR is the predominant sex-determining gene in bees and ants, and likely had this role in their common ancestor ~153 million years ago [13,14].

Sample collection

O. bicornis is a univoltine, protandrous, sexually size-dimorphic, polylectic, and monandrous spring bee. Each individual hibernates as an adult inside a cocoon in the maternal nest. Females construct typically mixed nests of line type containing both sexes. A series of larger cells for daughters is constructed at the back of the tube and a series of smaller cells for sons at the entrance [26] (Fig 1C). Body size of an individual is limited by the provisions in the brood cell [27]. Males developing in a female-dedicated brood cell with more provisions reach a female-like body mass. This situation occurs when the eggs are diploid male (due to homozygosity at the CSD locus) or due to fertilization errors, which result in a haploid egg. The rate of fertilization errors has previously been estimated to be about 6% in female-dedicated eggs [28].

Diploid males occur rarely and are detectable as adults with phenotypic male characteristics in female-dedicated brood cells. We analyzed the contents of 683 Osmia bicornis nests from the botanical gardens Halle (Saale), Germany and from various locations in the vicinity of Uppsala, Sweden during the period 2021–2023 to identify potential diploid males. The nests were contained in Japanese knotweed in Halle (internal diameter 8–10 mm, length 133.4 ± 13.2 mm) or cardboard tubes in Uppsala (internal diameter 7 mm, length 150 mm). They were analyzed during the winter, when adult bees are dormant inside cocoons.

Males and females were distinguished by morphological examination of the contents of each cocoon. Bees with black hairs on the clypeus were considered female, whereas the presence of long white hairs indicated a male. Potential diploid males were identified as morphological males that satisfied these criteria: (1) presence in a female-dedicated brood cell or (2) mass greater than 0.1 g, or both. Ploidy of all potential diploid males was determined by subsequent genome sequencing. We also collected females from nests containing potential diploid males and from random nests for sequencing.

Library preparation and sequencing

Samples were stored in 95% ethanol at −20°C prior to DNA extraction. They were dissected to extract wing muscle from the thorax. DNA was extracted using the Qiagen DNeasy Blood and Tissue Kit. We performed sequencing library preparation using the Illumina Nextera Flex kit. Sequencing was performed using an Illumina NovaSeq 6000 (output per sample shown in S1 Table).

Read mapping, variant calling, and filtering

Reads from all samples were mapped to the Osmia bicornis genome assembly iOsmBic2.1 using bwa v0.7.17 [29]. The resulting files were sorted and indexed using samtools 1.19 [30] before using Picard v2.25.0 [31] to mark duplicates and add read groups. Subsequently, GATK HaplotypeCaller v4.2.0.0 [32] was used to call variants. We considered all samples as diploids for variant calling. The software vcftools v0.1.16 [33] was used to filter the dataset, retaining only biallelic SNPs, MQ > 20, AF ≥ 0.01, as well as missingness <0.1. We also filtered markers for substandard or excess depth (1,000 > DP > 3,000) and quality by depth (QD > 20).

We noted that in certain regions of the genome, genotype calls had a large excess of heterozygosity, read mapping depth was significantly higher than average and quality-by-depth was significantly lower (S1A and S1B Fig). This observation is consistent with the presence of long stretches of highly repetitive heterochromatin in the genome assembly, in which genetic variants cannot be reliably assayed. This has been reported in a wide range of other genomes [34]. We identified regions with suspected heterochromatin based on elevated sequencing depth, reduced quality-by-depth and excess heterozygosity. To define these regions, we ran vcf-tools with the --hardy option, plotting −log(P_{het_excess}). We manually excluded regions of excess heterozygosity that were observed at the distal ends of each of the 16 chromosomes (S1A and S1B Fig; S9 Table). These blocks were excluded for the determination of ploidy and all further genome-wide summary statistics. Additionally, we filtered all markers identified as heterozygous in at least two haploid individuals from the dataset for mapping in the diploid individuals. The regions comprise 23.8% of the total assembled chromosomes. We also excluded 3 samples due to low average sequencing depth (<2 million reads).

Identification of ploidy

We used the --het option in vcftools [33] to calculate the inbreeding coefficient F, the ratio of observed and expected homozygotes per individual. This statistic revealed the ploidy of all samples, and determined whether potential diploid males were haploid or diploid. The male samples clustered into clear groups, with those with (F > 0.95) considered to be haploid males due to lack of heterozygous sites (S1C and S1D Fig).

Estimation of the frequency of matched matings

We used the number and distribution of confirmed diploid males among nests to estimate the frequency of matched matings in the populations. A matched mating is defined as one where a female mates with a male that shares one of its CSD haplotypes, which results in a probability of 0.5 that each fertilized egg will become a diploid male. We estimated the proportion of matched matings by dividing the number of nests containing diploid males with the total number of nests containing females, with a correction based on the likelihood of detecting a matched mating based on independent segregation of alleles at a single CSD locus. This likelihood is defined as d = 1 − 0.5ⁿ, where n is the number of diploid offspring present in a nest. We used this estimate to define the effective number of nests, by summing the values of d across all nests analyzed according to the number of diploid offspring per nest (S4 Table). This was performed using data collected in Halle over two consecutive years, and from 1 year in Uppsala.

Differences in mass

Using the ploidy confirmed by sequencing and the recorded sex, we tested for mass differences between cocoons containing females, haploid males, and diploid males. To compare diploid males to their female nestmates, we used a subset of 15 diploid males with multiple sequenced sisters (n = 37). We used statsmodels in python to make a linear mixed-effect model to test for an association between mass in grams and sex encoded as 0 and 1, with NestID as a random effect. We tested for differences between the mean mass of all confirmed diploid males (n = 18), females (n = 1,745), and haploid males, for which data was available (n = 1,659), using two-sided unequal variance t tests.

SNP density and nucleotide diversity

Nucleotide diversity (π) and SNP density were calculated in 10 kbp windows basis using vcftools [33] with the --window-pi option. For fine mapping nucleotide diversity in the identified putative CSD region, π was calculated per site using vcftools and the --site-pi option, and subsequently evaluated in a 1-kbp window rolling mean using numpy v1.26.0 [35].

Identification of runs of homozygosity associated with sex

ROH were calculated for each individual using plink v1.9 [36] with the ---homozyg group --homozyg-window-snp 50 --homozyg-window-het 3 --homozyg-window-missing 5 --homozyg-window-threshold 0.05 --homozyg-snp 10 --homozyg-kb 5 --allow-extra-chr and otherwise default options. Then, using pandas (2.2.2; [37]) and numpy (1.26.0; [35]) in python 3.10, we segmented the resulting ROH into a binary square matrix across the individuals and the genome, with the borders across the latter being defined by the beginning or end of a ROH within any of the investigated individuals. Subsequently, we tested the association between presence or absence of a ROH segment with sex in diploid males and females with a linear model, using statsmodels (0.14.2; [38]) and controlled for the rate of false positive testing using the Benjamini–Hochberg [39] procedure implemented in statsmodels with α = 0.05.

Haplotype analysis

In order to investigate the haplotype in the candidate region, we extracted the highest-scoring ROH-block (1974 bp), phased it using beagle v5.5 [40], and generated a sequence in fasta format for each haplotype. These were subsequently aligned using MAFFT (v7.520; [41]) and used to generate a tree using RAxML v8.2.12 [42], which was plotted using FigTree v1.4.4 [43].

Comparison with other species

We used blast hits identified with the blastn tool of BLAST+ v. 2.16.0 [44] to locate the ortholog of ANTSR in several other bee species for which genetic variation data were available. We created a query list using the ANTSR sequence of L. humile and its orthologous sequences in A. mellifera (LOC100576403) and Bombus affinis (LOC125386106) reported by Pan and colleagues [10]. We also included the O. bicornis lncRNA sequence studied here. We blasted the four query sequences against two Bombus, three Anthophora, two Lasioglossum, and three ant genome assemblies. These were Bombus sylvicola (BioProject PRJNA646847) [45], Bombus balteatus (PRJNA704506) [46], Anthophora retusa, Anthophora quadrimaculata, and Anthophora plagiata (PRJNA1255204) [47]. Lasioglossum pauxillum (PRJNA13330011), Lasioglossum albipes (PRJNA413432) [48], Formica polyctena, Formica aquilonia (PRJEB51899), [49] and Solenopsis invicta (PRJNA685290) [50]. The region with the highest query cover and identity was identified for each species.

We then examined the annotations of protein-coding and non-coding RNA genes in these regions for each target genome to look for the presence of an annotated lncRNA gene and conservation of the flanking genes. Functional annotations were not available for all of the species. We therefore performed functional annotation for these genome assemblies based on orthology using eggNOG-mapper v.2 [51]. This was performed with eggNOG-mapper default values against the eggNOG 5 database.

We identified the most likely position of ANTSR in each genome based on the annotation of a lncRNA in the vicinity of the best BLAST hit. A candidate gene was present in the annotation of some species, but not available for A. quadrimaculata, L. pauxillum, L. albipes, F. polyctena, F. aquilonia, and S. invicta. For these species, the best BLAST hit was used as the putative location. For A. mellifera, we used the coordinates previously reported by Pan and colleagues [10]. We analyzed genetic diversity in sequenced populations of each of these species. We used 65 and 49 individuals for Bombus lapponicus and Bombus monticola, respectively, mapped to the reference genome of the closely related species B. sylvicola [45]. For B. balteatus, we used 299 bees mapped to the B. balteatus genome [46]. We used 57, 43, and 27 individuals for A. retusa, A. quadrimaculata, and A. plagiata [47]. We used 55 Apis mellifera individuals (BioProject PRJNA357367) [52]. We used 71 individuals for L. pauxillum, 160 for L. albipes, 10 individuals for each of the two Formica species, and 368 for S. invicta. A summary of datasets is presented in S7 Table. We used vcftools v.0.1.16 [33] to calculate genetic diversity based on the pairwise differences (π) in windows of 10 kbp across the genome using vcftools --window-pi. We generated a phylogeny of these species using BeeTree [25], plotted using FigTree 1.4.4 [43].