Sympatric speciation revealed by genome-wide divergence in the blind mole rat Spalax

Significance

Sympatric speciation is still highly controversial. Here we demonstrate, based on genome-wide divergence analysis, that sympatric speciation in the blind subterranean rodent Spalax galili encompasses multiple and widespread genomic adaptive complexes associated with the sharply divergent and abutting basalt and chalk soil populations. Gene ontology enrichment analysis highlights sensory perception, musculature, metabolism, and energetics in basalt against neurogenetics and nutrition in chalk. Population divergence of chemoreceptor genes suggests the operation of mate and habitat choices, substantiating sympatric speciation. Natural selection and natural genetic engineering overrule gene flow, evolving divergent ecological adaptive complexes. Sympatric speciation may be a common speciation mode, as envisaged by Darwin, due to the abundance of sharp divergent geological, edaphic, climatic, and biotic ecologies in nature.

Abstract

Sympatric speciation (SS), i.e., speciation within a freely breeding population or in contiguous populations, was first proposed by Darwin [Darwin C (1859) On the Origins of Species by Means of Natural Selection] and is still controversial despite theoretical support [Gavrilets S (2004) Fitness Landscapes and the Origin of Species (MPB-41)] and mounting empirical evidence. Speciation of subterranean mammals generally, including the genus Spalax, was considered hitherto allopatric, whereby new species arise primarily through geographic isolation. Here we show in Spalax a case of genome-wide divergence analysis in mammals, demonstrating that SS in continuous populations, with gene flow, encompasses multiple widespread genomic adaptive complexes, associated with the sharply divergent ecologies. The two abutting soil populations of S. galili in northern Israel habituate the ancestral Senonian chalk population and abutting derivative Plio-Pleistocene basalt population. Population divergence originated ∼0.2–0.4 Mya based on both nuclear and mitochondrial genome analyses. Population structure analysis displayed two distinctly divergent clusters of chalk and basalt populations. Natural selection has acted on 300+ genes across the genome, diverging Spalax chalk and basalt soil populations. Gene ontology enrichment analysis highlights strong but differential soil population adaptive complexes: in basalt, sensory perception, musculature, metabolism, and energetics, and in chalk, nutrition and neurogenetics are outstanding. Population differentiation of chemoreceptor genes suggests intersoil population's mate and habitat choice substantiating SS. Importantly, distinctions in protein degradation may also contribute to SS. Natural selection and natural genetic engineering [Shapiro JA (2011) Evolution: A View From the 21st Century] overrule gene flow, evolving divergent ecological adaptive complexes. Sharp ecological divergences abound in nature; therefore, SS appears to be an important mode of speciation as first envisaged by Darwin [Darwin C (1859) On the Origins of Species by Means of Natural Selection].

Despite more than a century since first proposed by Darwin (1), the concept of sympatric speciation (SS) as a major mode of speciation, i.e., formation of new species within a freely breeding population with ongoing gene flow, is still highly controversial and evaluated both skeptically and critically (2). Interestingly, recent empirical studies (SI Appendix) and theoretical assessments (3) support SS. Claims of SS must demonstrate species sympatry, sister relationship, reproductive isolation, and that an earlier allopatric phase is highly unlikely (2). We recently described two studies of SS in two evolutionarily divergent mammals, the blind subterranean mole rat Spalax galili (4) (Fig. 1 A–C) and the spiny mice, Acomys, at “Evolution Canyon” (EC), Mount Carmel, Israel (5). Moreover, the Evolution Canyon microsite in Israel has been suggested as a cradle for SS across life, based on incipient SS of five distant taxons: bacteria, wild barley, fruit flies, beetles, and spiny mice (6). Importantly, only a few studies to date have investigated whole genome evolution in an attempt to uncover genome architectural changes during SS (7). Here we show that SS in S. galili encompasses extensive adaptive complexes across the genome associated with the sharply abutting and divergent chalk and basalt ecologies where SS took place, i.e., in sympatry and not in an earlier allopatry (4).

Fig. 1.

Study subject, ecological, and genomic divergence in sympatric speciation. (A) The blind mole rat, S. galili. (B) Vegetation formation exposed with mounds in basalt and covered by dense bushes of Sarcoptherium spinosum in chalk. (C) Geological map. Chalk is in yellow and basalt is in pink, separated by a geological fault. (D) NJ tree based on all of the SNPs. Red, basalt individuals; black, chalk individuals. (E) PCA of two Spalax soil populations. Red triangles, basalt individuals; black circles, chalk individuals. (F) Population genetic structure of two Spalax soil populations when K = 2, 3, and 4.

Results

Population Sequencing and Variation Calling.

Five individuals of the blind mole rat (S. galili) from the chalk rendzina soil and six from the abutting basalt soil (Fig. 1 A–C) were collected for genome sequencing. Of the 11 animals, the generated data for each individual, which had a genome size of ∼3G bp (8), ranged from 19.6 to 30.8 Gb, corresponding to sequencing depths of 6.36×–10× (Table S1). A total of 14,539,199 SNPs were identified, with 3,717,338 and 3,361,317 SNPs unique to the basalt and chalk populations, respectively (Fig. S1). We validated our SNP calling strategy with traditional Sanger sequencing technology and found that the genome-wide false-positive rate is less than 6% and the false-negative rate is less than 13% (SI Materials and Methods).

Fig. S1.

Unique and shared SNPs between the chalk and basalt mole rat populations shown by Venn diagram.

Genetic Diversity and Structure of the Two Populations.

Whole genome genetic diversity, estimated by Watterson’s θ for each individual, was significantly higher in chalk (mean θ = 1.13 × 10⁻³) than in basalt (1.09 × 10⁻³; P < 2.2 × 10⁻¹⁶, Mann–Whitney U test). Two distinct soil population clusters, basalt and chalk, were identified by the neighbor-joining (NJ) method (Fig. 1D) and by principal component analysis (PCA) (Fig. 1E) based on all SNPs. In PCA, the first and second component explained 17.2% and 12.9% of the genetic differences, respectively. A maximum likelihood analysis of population structure was undertaken based on K = 2, 3, and 4 (Fig. 1F). With K = 2, the basalt population is completely separated from the chalk population. In clustering by K = 3, both populations were divided into further subgroups but with the same genetic background (in green), notably, the individual B6 spans both chalk and basalt populations, possibly a recombinant individual. With K = 4, both populations were split into two subgroups, respectively (Fig. 1F).

Linkage Disequilibrium and Population Demography.

Linkage disequilibrium (LD), measured by the correlation coefficient (r²), decreased rapidly below 0.2 within 1,000 bp in both populations (Fig. 2A). Mean population recombination rate per kilobase (ρ = 4Ne × r) in basalt was 1.515, significantly higher than in chalk of 1.125 (P < 2.2 × 10⁻¹⁶, Mann–Whitney U test). Estimated extant effective population size (Ne) for chalk and basalt was 72,380 and 116,800, respectively, and the inferred ancestral Ne was 84,570 (Fig. 2B). The migration rates of chalk to basalt and basalt to chalk were 1.788 and 5.809 per generation (Fig. 2B), respectively. The two Spalax galili soil populations were estimated to split ∼0.2–0.4 Mya based on both whole genome and mitochondrial genome analyses (Fig. 2B and SI Materials and Methods).

Fig. 2.

LD and demographic structure. (A) LD decay of two continuous S. galili populations. x axis stands for physical distances (bp), whereas y axis stands for r². (B) Inferred demographic history for the two abutting soil populations (chalk vs. basalt) of S. galili. The extant and ancestral population sizes (Ne) of the chalk and basalt populations are indicated, and the migration rates between the two populations are provided. The divergence time (T) between two populations was inferred.

Population Genomic Divergence and Functional Enrichment.

Putatively selected genes (PSGs) were identified by screening genomic regions that show low diversity (measured by Tajima’ D) in one population but high divergence [measured by fixation index (F_ST)] between the two populations. A total of 128 genes in the basalt population and 189 genes in the chalk population were identified with the strongest signature of positive selection (Tables S2 and S3). Gene ontology (GO) enrichment analysis revealed functions of musculature, energetics, metabolism, and sensory biology (Fig. 3) distinguished in basalt. In contrast, functions related to neurogenetics and nutrition were enriched in the chalk population (Fig. 3). Our genome-wide scans for positive selection highlight how, where, and why adaptive evolution has shaped genetic variations.

Fig. 3.

GO enrichment analysis of putatively selected genes in chalk and basalt S. galili populations. Gene number is provided next to each GO term.

We sequenced 22 olfactory receptor (OR) genes (i.e., genes are numbered in Table 1) (Fig. S2), 20 bitter taste receptor (Tas2r) genes (Fig. S3), and 18 putatively neutral noncoding regions (NCs). Specifically, ORs refer to a group of olfactory receptor genes; Tas2rs refer to a group of type 2 taste receptor genes; NCs refer to a group of neutral noncoding regions. Pairwise F_ST estimates revealed that 10 ORs, 9 Tas2rs, and 1 neutral region are significantly differentiated between the chalk and basalt populations [P < 0.05 after false discovery rate (FDR) adjustment; Table 1]. The rate of significantly differentiated loci is statistically higher in ORs (10/22 = 45.5%) than in NCs (1/18 = 5.6%) (P = 0.011, two-tailed Fisher exact test), and the same is true for Tas2rs (P = 0.009).

Table 1.

Pairwise F_ST statistics of olfactory receptor genes (OR1-OR22), bitter taste receptor genes (Tas2r1-Tas2r20), and putatively neutral noncoding regions (NC1-NC18) between basalt and chalk populations

Loci	FST	FDR
OR1	0.074	0.093
OR2	0.055	0.029
OR3	0.037	0.051
OR4	0.061	0.011
OR5	0.320	0.000
OR6	0.106	0.029
OR7	0.023	0.465
OR8	0.029	0.093
OR9	0.040	0.220
OR10	0.064	0.188
OR11	−0.003	0.663
OR12	0.000	0.220
OR13	−0.004	0.051
OR14	0.120	0.033
OR15	0.107	0.040
OR16	0.007	0.037
OR17	0.040	0.029
OR18	0.049	0.208
OR19	0.008	0.029
OR20	0.014	0.042
OR21	−0.006	0.155
OR22	0.050	0.169
Tas2r1	−0.025	0.752
Tas2r2	−0.014	0.354
Tas2r3	0.045	0.016
Tas2r4	0.023	0.297
Tas2r5	−0.018	0.604
Tas2r6	−0.001	0.072
Tas2r7	0.000	0.109
Tas2r8	0.000	0.752
Tas2r9	0.015	0.157
Tas2r10	0.166	0.026
Tas2r11	0.104	0.047
Tas2r12	−0.013	0.540
Tas2r13	0.052	0.035
Tas2r14	0.005	0.467
Tas2r15	0.130	0.023
Tas2r16	0.127	0.000
Tas2r17	0.320	0.000
Tas2r18	−0.004	0.516
Tas2r19	0.056	0.000
Tas2r20	0.080	0.016
NC1	0.014	0.458
NC2	−0.008	0.418
NC3	0.037	0.144
NC4	0.056	0.268
NC5	0.064	0.413
NC6	0.061	0.063
NC7	0.042	0.261
NC8	0.005	0.694
NC9	0.050	0.153
NC10	0.002	0.413
NC11	−0.035	0.939
NC12	0.064	0.261
NC13	0.030	0.153
NC14	0.021	0.018
NC15	0.018	0.371
NC16	0.017	0.153
NC17	−0.011	0.413
NC18	−0.009	0.458

Significant P values are underlined. The genes and noncoding regions were numbered in arbitrary order. FDR, false discovery rate; F_ST is the fixation index.

Fig. S2.

Olfactory receptor genes analyses. Neighbor-joining phylogenetic tree of all 1,042 intact and complete olfactory receptor (OR) genes identified from the reference genome of S. galili. Red branches denote the 22 olfactory receptor genes surveyed in 29 blind mole rats from basalt and chalk populations. Nomenclature of the 22 OR genes was arbitrary.

Fig. S3.

Bayesian phylogenetic tree of 32 S. galili and 35 mouse Tas2r genes that are complete and intact. The numbers at the nodes are the Bayesian posterior probabilities shown as percentages. Blue branches indicate the 32 Tas2r genes from S. galili, whereas the remaining genes are from the mouse. Blue branches denote the 20 Tas2r genes surveyed in 29 individuals of S. galili from basalt and chalk populations. Nomenclature of the 20 Tas2r genes was arbitrary.

Differences in Proteostasis of the Two Populations.

Besides genomics, proteomics, particularly mechanisms regulating proteostasis, selectively drive soil population divergence. Mesic basalt population has a threefold significantly greater proteasome chymotrypsin-like activity in muscle and twofold higher trypsin-like and caspase-like activities than the chalk population (Fig. 4A) based on analyses of seven animals from each population. The basalt population displayed significantly higher α7 levels (Fig. 4B) than the chalk population, suggesting higher numbers of proteasomes in these tissues facilitating enhanced degradation of damaged or misfolded proteins through this protein degradation machinery. In contrast, the chalk population showed significantly higher levels of ATG7 and autophagic flux (LC3II/LC3I ratio) (Fig. 4C). This protein degradation profile suggested more of a reliance on autophagy in chalk population.

Fig. 4.

Differences in the proteolytic machinery of Spalax galili basalt and chalk populations. (A) 20S proteasome activity measured by degradation of optimized peptides cleaved by the three active sites, chymotrypsin-like (ChTL), trypsin-like (TL), and postglutamyl, peptide hydrolyzing (PGPH) or caspase-like, show higher levels of activity in the basalt population. (B) Higher levels of the constitutive 20S proteasome subunit α7 support the observation of the increase in activity in basalt population. (C) The chalk population protein degradation profile suggests more of a reliance on autophagy, with significantly higher levels of ATG7 and autophagic flux (LC3II/LC3I ratio).

SI Materials and Methods

Sample Collection.

This study is a continuation of two earlier studies: (i) on possible sympatric speciation, the first report in subterranean mammals (4); and (ii) genome analysis of S. galili (8) (2n = 52). Two abutting but sharply contrasting soil types, white chalky rendzina and brownish volcanic basalt, are typical in two sites (Alma and Dalton) in central eastern Upper Galilee (31), Israel. There are four species of Spalax in Israel: S. galili (2n = 52), S. golani (2n = 54), S. carmeli (2n = 58), and S. judaei (2n = 60) (32). In this study, only S. galili was studied for whole genome analysis from the Alma Plio-Pleistocene basalt plateau vs. the Senonian chalk of Kerem Ben Zimra (11, 31). A total of 11 animals, 5 from chalk and 6 from basalt, were captured alive near Rehaniya, Upper Galilee (33.04E, 35.49N), in January 2014 (Fig. 1 B and C). After injecting Ketaset CIII at 5 mg/kg of body weight, animals were killed. Muscle tissues were isolated and stored in 95% (vol/vol) ethanol until further molecular analysis could be done.

Library Construction and Genome Sequencing.

Genomic DNAs were isolated from the muscle samples of 11 individuals of S. galili using the DNeasy Blood and Tissue Kit (Qiagen) following the manufacturer’s protocol. For each individual, 1 μg of genomic DNA was fragmented into 450–550 bp for libraries, with a 500-bp insert size, using the Covaris S220 system. Genome sequencing libraries were prepared with the TruSeq DNA Sample Preparation kit according to the manufacturer’s protocol (Illumina). Briefly, the cleaved DNA fragments were end-repaired and A-tailed, pair-end adaptors were ligated to the fragments, adapter-ligated products of ∼500 bp were selected and purified on QIAquick spin columns (Qiagen), and purified products were enriched by 10 cycles of PCR amplification with Phusion DNA Polymerase. The purity and concentration of DNA were assessed and quantified using an Agilent 2100 Bioanalyzer (Agilent) and Nanodrop 2000 spectrophotometer (Nanodrop). Massive parallel sequencing was performed on the Illumina HiSeq 2000 platform, and 125-bp paired-end reads were generated.

Genome Mapping and SNP Calling.

In total, 245.69 Gb of raw reads were generated. We removed low-quality reads, including those with >10 nucleotides aligned to the adapter sequences, those of putative PCR duplicates, those with average base quality <15, those with >50% having a base quality score <10, and those with >10% unidentified nucleotides (N). As a result, 241.76 Gb of clean reads were retained. These high-quality paired-end reads were mapped onto the reference genome of S. galili (GenBank accession no. AXCS00000000) using Burrows-Wheeler Aligner (BWA) v0.7.8 (33) with the “mem” option and default parameters. The best alignments were generated in the SAM format by SAMtools v1.1 (34) with the “rmdup” option. Finally, for each individual of the blind mole rats, ∼99.37% of reads were mapped to ∼97.91% (at least 1×) or ∼70.13% (at least 5×) of the reference genome assembly with 7.13-fold average depth.

After genome mapping, we undertook SNP calling for two abutting S. galili populations (chalk population consisting of five individuals and basalt population consisting of six individuals), using a Bayesian approach as implemented in SAMtools v1.1. The “mpileup” command was executed to identify SNPs with the parameters as “-q 20 -Q 20 -C 50 -t DP -m 2 -F 0.002.” The probability of each probable genotype in a given SNP position was calculated with SAMtools, and the genotype with the highest posterior probability was picked. The low-quality SNPs were filtered by the Perl script vcfutils.pl in BCFtools v1.1 (34) package with the “varFilter” option and parameters as “-d 20 -D 140,” and the high-quality SNPs [coverage depth ≥20 and ≤140, root mean square (RMS) of mapping quality ≥10, the distance of adjacent SNPs ≥5 bp] were retained for further analysis. Deviations from Hardy–Weinberg equilibrium (HWE) were tested by using Vcftools (vcftools.sourceforge.net/). SNPs deviating from HWE (P < 0.05) were excluded from subsequent analysis.

Verification by Sanger Sequencing.

Twenty-two ORs, 20 Tas2rs, and 18 putatively neutral noncoding loci from 16 basalt and 13 chalk individuals of the blind mole rats were sequenced by traditional Sanger sequencing technology, aiming to identify genetic differentiation of chemosensory receptor genes between chalk and basalt populations (see details below). We also took advantage of these Sanger sequencing data (SSD) to estimate false-positive rates and false-negative rates in genome-wide SNP calling for our next-generation sequencing data (NSD). We divided SNP sites from SSD into three groups: group 1 contained SNPs that are absent between individuals of the blind mole rats in SSD but present in NSD; group 2 consisted of SNPs that are present between individuals in SSD but absent in NSD; and group 3 included SNPs that are present between individuals in both SSD and NSD. The three groups of SNPs were considered to be false positives (group 1), false negatives (group 2), and true positives (group 3). We identified 291 SNPs from SSD, whereas 273 SNPs were found in the equivalent loci from NSD. The three groups contained 16 (false positives), 34 (false negatives), and 257 (true positives) SNPs, respectively. Thus, our NSD was estimated to have a false-positive rate of 5.86% and a false-negative rate of 12.45%.

Population Structure Analysis.

Population structures were investigated using three approaches. The first approach is the nonparametric PCA (35). PCA is widely used in population structure analysis because it is computationally efficient in handling large numbers of SNP markers (36, 37). The variant calling format was converted to binary ped format using VCFtools and PLINK v1.07 (pngu.mgh.harvard.edu/∼purcell/plink/) (38). PCA was performed using GCTA v1.24 (39) with the parameter “-pca 2.” The second approach is to estimate individual ancestry based on the full maximum likelihood method as implemented in the program frappe v1.1 (40). Because the analysis is computationally intensive, we picked only the first SNP within each nonoverlapping window (window size: 20 kb) for the calculation. We assumed that the probable number of ancestral populations was 2, 3, and 4, respectively, according to the ecological information (4). The third approach is to infer population structures using a neighbor-joining method (41). The neighbor-joining phylogenetic tree was reconstructed using the nucleotide p-distance matrix, and the reliability of the tree was evaluated with 1,000 bootstraps in TreeBeST v1.9.2 (sourceforge.net/projects/treesoft/files/treebest/).

LD Analysis.

To estimate the LD patterns between the two populations, we used the program Beagle v4.0 (42) to phase the genotypes into associated haplotypes with the command “gtgl.” The correlation coefficient (r²) between any two loci was calculated using VCFtools with the “hap-r2” option and default parameters. Average r² was calculated for pairwise SNPs in a 50-kb window with a custom written Perl script and was plotted against physical distance in base pairs with R v3.1.2 (43).

Genetic Diversity and Recombination Rate.

Genetic diversity was estimated by Watterson’s θ (44). For each population, we undertook a sliding window analysis, with a window size of 20 kb and a step size of 5 kb. We calculated θ for each window, and the mean value of θ was considered as whole genome genetic diversity. The significance of difference in θ between two populations was tested with the Mann–Whitney U test.

The interval program in the LDhat v2.2 package (45) was used to infer the population recombination rate (ρ) across 28 scaffolds, each of which is longer than 10 Mb. For each population, the program was run with all of the SNPs from each scaffold. To reduce computational cost, a likelihood look-up table that assumes a population mutation rate (θ) of 0.001 was obtained from LDhat website (ldhat.sourceforge.net), and number of sequences was subsequently specified for each population with an lkgen program in the LDhat package. With a penalty parameter of 5 for changes in recombination rate, the modified look-up tables were applied to the interval program. Ten million iterations were run with the first 10% iterations discarded as burn-in.

Estimation of Demographic Parameters.

We used the software Generalized Phylogenetic Coalescent Sampler (G-PhoCS) (46) to infer demographic parameters for the basalt and chalk populations including effective population size, migration rate, and population divergence time. G-PhoCS models gene flow with a Bayesian manner requiring the input of separate neutral loci and average mutation rate of neutral loci. A total of 3,000 putatively neutral loci with 1 kb in length were randomly picked along the genome, following three criteria: (i) these loci contain at least one SNP; (ii) they are at least 200 kb away from any known gene; and (iii) minimum interlocus distance of these loci is 50 kb. Migrations between two populations were assumed, whereas other parameters were set to the defaults. Sixty million iterations were run with the first 10% iterations discarded as burn-in. The convergence of Markov chain Monte Carlo (MCMC) algorithms was inspected by Tracer v1.6 (47). We repeated this analysis three times independently, all of which yielded similar results.

The demographic parameters generated by G-PhoCS are all scaled by the mutation rate. We estimated the average mutation rate μ = 3.03 × 10⁻⁹ mutations per site per year, among rodents, by the average substitutions per site in neutral loci of 0.28 (48) divided by the average divergence time of 92.3 Mya (49). For each population, the effective population size (Ne) equals $\theta /4g\mu$, where g stands for generation time, and θ refers to population mutation rate. Based on a previous study (50), we assumed an average S. galili generation time of 1-y G-PhoCS estimated the mean population mutation rate of basalt population (θ_basalt = 1.418 × 10⁻³), chalk population (θ_chalk = 8.784 × 10⁻⁴), and ancestral population (θ_ancestor = 1.026 × 10⁻³). Thus, the mean Ne of basalt was estimated to be 116,800, mean Ne of chalk was 72,380, and mean Ne of ancestral population was 84,570.

By specifying a source population (S) and a target population (T), G-PhoCS infers the migration rate from S to T as m_S-T, which can be converted into migrants from S to T per generation (M_S-T) with $M_{S-T}=m_{S-T}\times {\theta }_T$, where θ_T is the population mutation rate of the target population. We estimated the mean migration rate from basalt to chalk (m_b-c = 6.613 × 10³) and migration rate from chalk to basalt (m_c-b = 2.036 × 10³). Thus, the mean number of migrants per generation from basalt to chalk M_b-c was 5.809, whereas the mean number of migrants per generation from chalk to basalt M_c-b was 1.788.

The divergence time between chalk and basalt S. galili populations based on individual genome sequences was inferred using G-PhoCS. Using the average mutation rate μ = 3.03 × 10⁻⁹, the average mutations per site (τ) between the chalk and basalt populations was estimated to be 6.917 × 10⁻⁴. We estimated the divergence time (T) between the two populations by using the ratio of τ to μ. Thus, the mean of T was inferred to be 0.228 Mya, with a 95% confidence interval (CI) of [0.209, 0.250] Mya. For comparison, we also isolated the 13 mitochondrial protein-coding genes from the whole genomes of the 11 individuals of S. galili for dating. The equivalent genes from six outgroup species of mouse (Mus musculus; GenBank accession no.: KP090294), rat (Rattus rattus; GenBank accession no. KM577634), reed vole (Microtus fortis; GenBank accession no. JF261174), bamboo rat (Rhizomys pruinosus; GenBank accession no. KC789518), spiny mouse (Acomys cahirinus; GenBank accession no. JN571144), and a Middle East blind mole rat (Spalax ehrenbergi; GenBank accession no. AJ416891) were downloaded from National Center for Biotechnology Information (NCBI) (www.ncbi.nlm.nih.gov/). The 13 mitochondrial genes were concatenated and aligned. A Bayesian phylogenetic tree was constructed to evaluate the relationships of 11 individuals of S. galili. Divergence times were estimated using BEAST (51) with the Hasegawa–Kishino–Yano (HKY) model and an assumption of constant population size. By using the divergence time (27–34 Mya) (49) between rat and mouse as the calibration point, the divergence time of the basalt and chalk populations was estimated to have a mean of 0.425, and a 95% CI of [0.246, 0.645] Mya.

Putatively Selected Genes During Speciation.

Tajima’s D, calculated by dividing the difference between the mean number of pairwise nucleotide differences and the number of segregating sites by the square root of its SE, was used to test whether a sequence is under selection (52). The fixation index (F_ST) (53) was also chosen as an indicator of population differentiation. Tajima’s D and F_ST values were calculated by Vcftools (vcftools.sourceforge.net/) with a window size of 20 kb and a step size of 5 kb.

Tajima’s D and F_ST values were sorted in descending order separately. Windows that share the highest 5% of F_ST and lowest 5% Tajima’s D estimates of each population were recognized as positively selected regions in a given population, and genes located in these regions are considered putatively selected genes.

Functional Enrichment Analysis.

Functional enrichment analysis of GO terms was conducted using DAVID (Database for Annotation, Visualization, and Integrated Discovery) (54). The gene IDs of the reference genome of S. galili were converted to Uniprot IDs. Putatively selected genes were submitted to DAVID for enrichment analysis of the significant overrepresentation of GO terms, and the whole gene set of the S. galili genome was appointed as the background. We undertook Benjamini-corrected modified Fisher’s exact tests to examine the significance of functional enrichment between various gene sets, and P values [i.e., Expression Analysis Systematic Explorer (EASE) scores] less than 0.05 were considered significant. GO biological processes (GO-BP), molecular function (GO-MF), and cellular component (GO-CC) terminologies were used.

Population Differentiation of Olfactory and Taste Receptor Genes.

We identified olfactory receptor genes (ORs) and bitter taste receptor genes (Tas2rs) from the reference genome of S. galili using TblastN searches, with published mammalian ORs and Tas2rs as query sequences. Other taste receptor genes were not examined, because they evolve much slower than Tas2rs (55). We identified 1,042 ORs and 32 Tas2rs that are complete and intact and putatively functional. Two phylogenetic trees were reconstructed to show phylogenetic relationships of these genes (Figs. S2 and S3). To represent major lineages of the phylogenetic trees, we carefully chose 22 ORs and 20 Tas2rs for Sanger sequencing in 16 basalt and 13 chalk individuals of the blind mole rats. For comparison, we also sequenced 18 putatively neutral noncoding loci in the same 29 mole rats. These noncoding loci were chosen randomly from the genome, with a requirement that they are located at least 200 kb away from any known gene. These putatively neutral regions are all ∼1,000 bp in length, which is similar to the average length of ORs or Tas2rs.

PCRs were conducted with high-fidelity KOD-Plus-Neo DNA polymerase (Toyobo) following the manufacturer’s recommended procedures. PCR products were purified by the QIAquick PCR Purification Kit (Qiagen) and subjected to direct sequencing in both strands with the same primers as used for PCRs. SNPs were identified by visually inspecting sequence chromatogram files. MEGA6 (56) was used to align and edit the sequences. Because mammalian ORs and Tas2rs are intron-less with ∼930 bp in length, we focused exclusively on coding regions for genetic analysis. Population genetic differentiation between the chalk and basalt populations was estimated as the fixation index F_ST using DnaSP v5.10 (57), and the significance level was determined by the Snn test (58). Pairwise F_ST statistics between basalt and chalk populations are provided in Table 1.

Whole Tissue Lysates, Proteasome Activity, and Western Blots.

The muscle tissue from each animal (seven animals from each soil population) was weighed and disrupted in a 2-mL Potter–Elvehjem homogenizer in reticulocyte standard buffer (RSB) (10 mM Hepes, pH 6.2, 10 mM NaCl, and 1.4 mM MgCl₂) at a weight to volume of 1 g of tissue to 5 mL of buffer for muscle. An equal portion of the tissue was placed in a RSB buffer supplemented with the addition of 1 mM ATP, 0.5 mM DTT, and 5 mM MgCl₂ to help maintain intact 26S subassemblies (59) for peptidolytic assays. To the other half, 1 tablet/10 mL protease and phosphatase inhibitor minitablets were added (Thermo Fisher Scientific) for use in Western blots. To clear the lysate of debris, the homogenates were spun at 1,000 × g for 12 min, and the supernatant was then stored as 50-µL aliquots at −80 °C until further use. Protein content was measured using the Pierce BCA Protein Assay (Life Technologies).

In each assay, 20 µg of whole tissue or subfractionated lysates was incubated with 100 µM substrate specific for the type of proteasome activity. A saturating concentration of proteasome inhibitor N-(benzyloxycarbonyl) leucinyl-leucinyl-leucinal (MG132) was added to parallel samples. The difference of the fluorescence released with and without the inhibitor was used as a measure of the net peptidolytic activity of proteasome as previously described using model peptide substrates to represent cleavage after hydrophobic [chymotrypsin-like (ChTL)] residues, basic residues [trypsin-like (TL)], and acidic residues (PGPH) (60).

For Western blots, tissue lysates were separated using a 4–20% SDS/PAGE (Biorad Life Sciences) and transferred to nitrocellulose membranes (Biorad Life Sciences). The membranes were probed with antibodies against the following: α7 (mouse mAb, PW8110, 1:5K), RPT5 (mouse mAb, PW8310, 1:5K), beclin-1 (rabbit mAb, #3495P), LC3A/B (rabbit mAb, #4108P), ATG5 (rabbit mAb, #8540P), and ATG7 (rabbit mAb, #8558P) (Cell Signaling Technology). GAPDH (mouse mAb, G8795, 1:20K; Sigma-Aldrich) antibody (Thermo-Fisher Scientific) was used as a loading control. Primary antibodies were detected using anti-rabbit IRDye 680LT or anti-mouse IR Dye800 CW (Li-Cor) conjugated antibodies. Secondary antibodies were incubated at 1:10,000 for 2 h at room temperature, and images were captured and subsequently quantified using the Odyssey Imaging System (Li-Cor) by quantifying fluorescent signals as Integrated Intensities (I.I. K Counts) using the Odyssey Infrared Imaging System, Application Version 3.0 software. Prism 5 (GraphPad Software) was used to analyze and graph the various datasets generated in this study. Comparisons were analyzed using multiple t tests. Statistical significance was set at the P < 0.05 level with Holm–Sidak corrections to counteract the probability of false positives.

Discussion

Genetic Diversity, Population Structure, Speciation Genes, and Reproductive Isolation.

Genetic diversity at the local chalk-basalt state parallels the regional patterns (9, 10). The basalt and chalk mole rat populations were grouped separately (Fig. 1 D–F), suggesting genome-wide divergence of the two abutting soil populations. Population structure analysis showed similar clustering as revealed by mtDNA (4). The B6 individual showed a genetic mixture of both the chalk and basalt populations, which may indicate some limited gene flow, as was shown earlier by a recombinant individual in the interface of chalk and basalt (4) and a mound row extending from basalt to chalk was also observed in the field. Remarkably, genetic diversity from whole genome analysis is consistent with earlier estimates by mtDNA (4) and AFLP (11), where genetic polymorphism was significantly higher in chalk than in basalt. This finding supports earlier evidence that genetic diversity is possibly associated with a more stressed environment (5, 6, 9, 10). We deduced, across the whole genome, positively selected speciation genes (SGs). We hypothesize that besides highlighting strong selection, additional factors could potentially contribute directly or indirectly to reproductive isolation (RI) in abutting Spalax soil populations associated with SS (Table S4). These factors include habitat selection, chemoreceptor divergence, and preliminary indications of mate choice.

LD and Population Demography.

Rapid LD decay indicates high recombination rates or large effective population sizes (12). LD level is relatively lower in basalt than in the chalk population, possibly because of the larger basalt population size (4) (Fig. 2B) and the higher temperature stress in chalk as was shown experimentally in Drosophila (13). Ne of chalk is smaller than that of basalt (4) (Fig. 2B), probably due to lower food resources (14). The larger Ne of the basalt population (Fig. 2B) is similar to current population estimates (14). The recombination rate is significantly higher in basalt than that in chalk possibly due to adaptation to the novel basalt niche (15). The combined lower LD may suggest that LDs are soil specifically selected for the chalk and basalt separately rather than generally for both. There are more animals migrating from basalt to chalk populations possibly because of the higher population density in basalt.

Functional Adaptive Complexes.

Functional enrichment analysis of GO terms revealed the remarkable divergence of biological differences between the abutting Spalax soil populations, indicating genome-wide adaptive complexes in each soil population resulting from its soil-unique, multiple-specific ecological stresses. Burrow digging, a major activity of mole rats, is harder in compact basalt than in loose chalk (4), evidenced by several GO terms such as musculature, energetics, and sensory biology (Fig. 3). Accordingly, activity patterns (4) and mound density (14) are significantly higher in basalt than in chalk. Remarkably, the clay-rich basalt is harder and sticky, requiring more energy in digging. The enriched metabolic PSGs in basalt reflect the higher resting metabolic rate (RMR) in basalt than in chalk mole rats (4), as is true regionally in northern S. galili compared with southern S. judaei due to low food resources in the latter (16). Olfaction, developed in Spalax (17), is under stronger selection in basalt than in the chalk population, possibly due to higher food diversity and chemical resources in basalt. Likewise, soil moisture is higher in basalt than in chalk (14), increasing odor saturability in the soil and enhancing olfactory reception (18, 19). Finally, sensory perception of chemical stimuli may increase during the colonization of the chalk ancestors into the novel basalt niche (20) due to the higher level of selective pressure for mate choice.

This local pattern of neurogenetics and nutrition parallels the regional one of S. judaei, characterized by high genetic diversity (21), limited food resources, and relatively bigger brain size (22). Both locally (chalk population of S. galili) and regionally (S. judaei), food resources are limited; hence, larger territories and magnetic orientation from the nest to the periphery influence selection for advanced neurogenetics (22) (Fig. 3). Local and regional parallel patterns highlight the adaptive nature of the above traits.

Genetic variations in olfactory and taste receptors may underlie both mate and habitat choice in speciation (23). The divergence of both olfactory and taste receptor genes strongly suggest that olfaction and taste have undergone diversifying selection in population divergence and premating reproductive isolation. Thus, chemosensory habitat and mate preferences may substantiate SS in the blind mole rats.

Greater digging in harder basalt may manifest in greater muscle remodeling, a process dependent on the proteasome (24), leading to higher rates of proteasome mediated protein degradation and higher levels of constitutive proteasome components (α7 levels), as seen in the basalt new derivative species (Fig. 4B). High proteasome activity enhances the removal of damaged proteins and thereby improves the quality of the proteome and the maintenance of proteostasis. This may increase longevity (25, 26), and helps the cell to resist oxidative stress (25, 27). High levels of proteasome activity have also been seen in other underground-burrowing animals, such as the naked mole rat. Alternatively, compromised nutrition from low food resources as seen in chalk Spalax, which relies on lipid metabolism (Fig. 3), may induce autophagy rather than proteasome-mediated degradation for protein recycling (28) (Fig. 4C).

Conclusions and Prospects

During SS, natural selection and natural genetic engineering (29) are nonrandomly diverging 317 soil-biased genes, gene networks, and functional structures involved in the whole genome and proteome, coping with local ecologically, differentially, diverse biological stresses (Fig. 3 and Tables S2 and S3). Because the soil populations are abutting, animals can migrate in both directions (4) (Fig. 2B), yet philopatry or habitat choice restricts gene flow. Natural selection and natural genetic engineering overrule gene flow, as in Evolution Canyon (6), selecting for soil-specific adaptive complexes and generating adaptive ecological incipient SS. Unfolding the regulatory function of the repeatome, and epigenomics could highlight the SS procession in sharply divergent ecologies, which may occur frequently in nature (1) across sharply divergent geological, climatic, edaphic, and biotic abutting contrasts (6). Basalt populations have more SGs, which may involve genes explained by the niche release hypothesis (30). The Spalax example of SS supported by genome-wide adaptive divergence is one of the few cases (SI Appendix) substantiating the occurrence and genetic patterns of SS in nature. Future studies could involve transcriptomes, repeatomes, metabolomics, editomes, and phenomics aspects of the ancestral chalk and derivative basalt populations.

Materials and Methods

All experiments on the blind mole rats were approved by the Ethics Committee of University of Haifa and Wuhan University, and conformed to the rules and guidelines on animal experimentation in Israel and China. Whole genome resequencing of two soil populations of S. galili, the ancestral chalk and derivative basalt populations, was performed. Genome-wide divergence between the two mole rat soil populations was estimated by PCA, neighbor-joining phylogenetic tree, and individual ancestry estimation based on the full maximum likelihood method. Genomic characterization of the two soil populations was revealed by differently enriched GO terms, protein proteostasis, olfactory, and bitter taste receptor gene analyses. Genetic diversity and LD of the two soil populations were compared and contrasted. Population demography, effective population size, and gene flow were estimated to assess the evolutionary divergence of sympatric speciation. Population divergent time was estimated by both mitochondrial and nuclear genome analyses. Full details of the materials and methods are described in SI Materials and Methods.