Samples analyzed in the present study originated from two brown bear populations sampled within the Scandinavian Brown Bear Research Project. The northern population (N) is located near Jokkmokk in Norrbotten County, Northern Sweden, and the southern population (S) consisted of samples collected in Dalarna and Gävleborg counties in Central Sweden and Hedmark County in Southeastern Norway 37 . Details on sampling and genomic DNA (gDNA) extraction can be found in Waits et al. 38 and Bellemain et al. 39 . Altogether samples from 234 individuals were used to characterize polymorphism in MHC and 6 samples from the southern population were used to characterize expression of four MHC genes. Samples were collected under permissions: C7/12 for 2012–2014 and C47/9 for 2009–2012, C59/6 for 2006–2008, C40/3 for 2003–2006 from Uppsala Ethical Committee on Animal Experiments, Uppsala, Sweden.

Comprehensive characterization of variation in highly polymorphic MHC genes requires PCR primers amplifying all alleles. To develop such primers for the second, most variable, exon of several MHC genes, we used two approaches. The first was based on the vectorette PCR and the second employed primers located in conserved portions of exons 1 and 3 or 4 to amplify the intervening fragments from cDNA.

We designed primers in conserved regions of the second exon, identified in the alignment of mammalian MHC sequences downloaded from GenBank (Additional file 1:Table S1, the list of accession numbers in Additional file 2: Supplementary information). Parts of the second exon for all four genes were amplified from several individuals, cloned and sequenced as described in Zagalska-Neubauer et al. 40 . These partial sequences allowed the design of several primers within the second exon, which were used in vectorette PCR, performed as described in Babik et al. 41 , to obtain sequences of 5’ and 3’ ends of the second exon from multiple individuals. In the vectorette PCR approach, total genomic DNA is digested with a restriction enzyme (RE) producing sticky ends; then double-stranded adapters (vectorettes) matching the overhangs but showing some internal mismatch (‘bubble’) are ligated. By using one primer specific to the sequence in question and the other specific to the reverse complement of one of the vectorette strands (in the region of mismatch), it is possible to directionally amplify the genomic fragment between the specific primer and the RE recognition site, i.e. outside of the region of known sequence. The consensus of these sequences and sequences obtained from cDNA (see below) allowed the design of robust primers for all studied genes.

We designed primers in conserved regions of the first and fourth (MHC class I genes, Table 1 – primer pair no. 1) or third (MHC class II genes, Table 1 – primer pairs no.: 2–4) exon using mammalian sequences downloaded from GenBank (Table 1, the list of accession numbers in Additional file 2: Supplementary information). Total RNA was extracted with PAXgene Blood RNA kit (Qiagen) from six blood samples frozen in Qiagen’s PAXgene Blood RNA tubes following the manufacturer’s protocol. Complementary DNA (cDNA) was obtained using Omniscript reverse transcriptase (Qiagen) and Oligo(dT)12-18 primer (Invitrogen). Fragments of MHC genes were amplified from cDNA in 15 μL mixes containing 7.5 μL of HotStarTaq Master Mix (Qiagen), 2 μM of each forward and reverse primers, 5.4 μL of PCR-grade water and 1.5 μL of cDNA template. The polymerase chain reaction (PCR) cycling scheme was as follows: 95°C for 15 min, 28 cycles of 95°C for 30 s, 57°C for 30 s, 72°C for 1 min, and the final elongation step at 72°C for 10 minutes. cDNA amplicons were pooled separately for each gene, cloned and 13 – 28 clones per pool were sequenced. Full second exon sequences were used in combination with sequences obtained with the vectorette PCR technique to design primers used in actual genotyping (Table 1 – primer pairs no.: 5–8). All newly designed primer pairs amplified the previously detected alleles, confirming the successful design of genotyping primers.

xxxxxx indicates the location of a 6-bp tag in the fusion primer sequences.

Investigated MHC genes exhibited varying levels of polymorphism and, consequently, several techniques were used for their genotyping. DQA and DQB genes were slightly or moderately polymorphic so to characterise most variation present in these genes it was sufficient to genotype a relatively small sample by Single Strand Conformational Polymorphism (SSCP) or cloning and sequencing. Genotyping of highly polymorphic class I and class II DRB genes was performed for large samples of individuals by 454 sequencing. For all genes, expression was assessed through genotyping cDNA and gDNA from six individuals.

PCR conditions for both DQA and DQB were: 95°C for 15 min, 28 cycles of 95°C for 30 s, 55°C for 30 s, 72°C for 1 min, and the final elongation step at 72°C for 10 minutes. The SSCP analysis was performed using GMA gels (Elchrom Scientific). We added 4.5 μL of PCR product to 9.5 μL of premix containing formamide and 10 mM sodium hydroxide, denatured it for 5 min at 95°C and immediately cooled it on ice. Electrophoresis was conducted in 1 x TAE buffer at 8°C, 4 V/cm, for 18 hours. Gels were stained with SYBR Gold Nucleic Acid Gel Stain (Invitrogen). Allele sequences were obtained by sequencing the bands excised from gels. Additional screening for variation in these genes was performed by cloning amplicons pooled from 20 individuals and sequencing multiple clones.

454 sequencing was used for genotyping of highly polymorphic MHC class I and MHC class II DRB genes because initial tests with the SSCP technique resulted in complex, uninterpretable patterns. PCR amplification was conducted using fusion primers, which contained the 454 Titanium adapter sequence (A in forward, B in reverse primer) at the 5’ end, followed by a 6-bp tag (barcode), which distinguished amplicons obtained in different PCR reactions, and the gene specific primer. Tag sequences differed from each other in at least three positions, which minimized the chance of misassigning sequencing reads due to errors in tag sequences. Sequences of fusion primers are given in Table 1 (primer pairs no. 9 and 10). Amplification was carried out in 15 μL, as described above. Ten individuals were amplified and sequenced twice to estimate the genotyping error. PCR products were pooled in approximately equimolar quantities, pools were purified with the MinElute PCR Purification Kit (Qiagen) and sequenced at the Functional Genomics Center Uni/ETH in Zurich. Sequencing was performed bidirectionally using the GS FLX Titanium MV emPCR kit for emulsion PCR and the GS FLX Titanium Sequencing Kit XLR70 in combination with GS FLX Titanium PicoTiterPlate Kit 70 x 75 for sequencing (Roche Applied Sciences). Extraction of reads from multifasta files, assignment of reads to individuals and generation of alignments of variants present in each amplicon were performed with jMHC software 42 . The output from jMHC was analysed using BLAST, Excel and Bioedit 43 .

To minimize the occurrence of false alleles that may be the artefacts of PCR or cloning, we followed the guidance of Lenz & Becker 44 . The artefacts that occur in 454 output may be divided into three types: i) substitutions caused by polymerase errors during PCR, ii) PCR-chimeras and iii) insertions, deletions and substitutions due to 454 sequencing errors 45 46 47 . The first two types are not specific to 454 sequencing and their frequency may be reduced at the PCR level 44 . Whereas the point substitutions and insertions – deletions (indels) should be relatively rare, the chimeras may be easily produced during PCR by recombination between true alleles 44 46 . Furthermore, some chimeras may have sequences identical to true alleles, because the latter may originate through historical recombination from other alleles. Distinguishing between the PCR chimeras and the true alleles is based on the rationale that chimeras should always occur with both parental alleles in the amplicon and that the artefacts should be less frequent, as measured by the number of reads per amplicon, than the true alleles.

True alleles were distinguished from the artefacts following the procedure described in Zagalska-Neubauer et al. 40 and Radwan et al. 48 . Briefly, for each sequence variant, we calculated the maximum per amplicon frequency (MPAF) in the whole dataset. Sequences were sorted according to their MPAF. Starting from arbitrary MPAF of 1.5% for MHC class I sequences and 3% for MHC class II DRB sequences, 65 and 28 sequence variants, respectively, were chosen to evaluate whether they represent true alleles or sequence artefacts (see Radwan et al. 48 for details). For MHC class I sequences within the 1.5-2% MPAF interval, 95% (19 of 20) variants were classified as artefacts and within 2-3%, 75% (8 of 12) were classified as artefacts. All of 33 variants with MPAF above 3% were classified as true alleles. MPAF for the least abundant true allele and most abundant artefact (1.53%-2.86%) defined the “grey zone”, which required a decision about whether a sequence was a true allele or an artefact on a case-by-case basis 48 . For MHC class II DRB sequences, all 12 variants in the 3-10% MPAF interval were classified as artefacts. All of 16 variants with MPAF above 10% were classified as true alleles.

Alleles were named according to the nomenclature proposed by Klein et al. 49 and were numbered in ascending order, starting from the most abundant one. We use the term “allele” for unique sequence variants for simplicity, but assigning sequence variants to loci was generally not possible in our study.

Genetic differentiation between populations was measured by calculating pairwise F_ST in ARLEQUIN 3.5 50 . Because assigning of alleles to loci was not possible, each allele was treated as a dominant locus, and binary encoded.

For all alleles, the average pairwise nucleotide distances (Kimura 2-parameter model - K2P), Poisson corrected amino acid distances, as well as the average rates of synonymous (d_S) and nonsynonymous (d_N) substitutions, using the Nei-Gojobori method 51 with the Jukes-Cantor correction for multiple substitutions, were computed in MEGA5 52 . Standard errors were obtained through 1000 bootstrap replicates.

We used two approaches to test whether positive selection shaped the evolution of the second exon of the investigated genes; the one-tailed Z test comparing d_N to d_S and comparison of the likelihoods of codon-based models of sequence evolution. The Z-test, as implemented in MEGA5, compared the rates of synonymous vs. nonsynonymous substitutions at all codons, ABS and non-ABS. The location of putative ABS was inferred from the structure of human HLA genes 6 . For MHC class I, putative ABS location was conservatively inferred from the consensus of ABS common to human HLA-A, B and C genes. The comparison of models of sequence evolution for three genes (excluding MHC class II DQA) was performed in PAML 4.2 53 . Three models were tested: i) M0: a single ω (d_N/d_S) for all codons, ii) M7: nearly neutral (0 < ω < =1), with ω variation approximated by β-distribution, iii) M8: positive selection (a proportion of codons with ω > 1), with ω variation approximated by β-distribution. The best fitting model was chosen on the basis of the lowest value of the Akaike Information Criterion (AIC, 54 ). Positively selected codons under the M8 model were identified through the Bayes empirical Bayes procedure 55 .

Phylogenetic trees visualizing similarities among MHC alleles were constructed under the Bayesian approach with mrBayes 3.1.2 software 56 . The general time-reversible (GTR) model of sequence evolution with the rate-variation (Γ) among sites was used; parameter values were estimated from the data. Priors were set to default values. Two independent Metropolis coupled Markov chain Monte Carlo simulations (four chains each, three of them heated, temp. 0.20) were run for five million generations and sampled every 1000 generations for DQA, DQB and DRB; for the class I longer runs of 20 million generations were necessary to reach convergence, trees were sampled every 2000 generations. The first 1000 (DQA, DQB, DRB) or 2000 (class I) trees were discarded as burn-in. To calculate the posterior probability of each bipartition, the majority-rule consensus tree was computed from the 8000 (DQA, DQB, DRB) or 16000 (class I) sampled trees.

We genotyped 228 bp fragment of MHC class I 2^nd exon in 224 bears, 100 from the northern and 124 from the southern population, and identified 37 alleles (GenBank accession numbers JX469853-89). The nucleotide sequences translated into 33 unique amino-acid sequences (Figure 1a). Two alleles had premature stop codons (PSC). Because assignment of alleles to loci was impossible, instead of true allele frequencies, frequencies of individuals possessing particular alleles are given in Table 2. Frequencies of MHC class I alleles differed significantly between the two Scandinavian brown bear populations (F_ST = 0.216, P < 10^-5)

Alleles are sorted according to their total frequency.

A comparison of genotypes obtained from cDNA and gDNA of 6 individuals confirmed the expression of 10 alleles (Urar-U*01, *03, *04, *07, *10, *11, *13, *17, *26, *27), whereas 6 alleles were found only in gDNA (Urar-U*02, *05, *06, *08, *09, *12). The non-expressed alleles form two distinct clusters, consistent with the presence of two non-expressed MHC class I loci (Figure 2). We found 5–11 alleles per individual in gDNA. The allele Urar-U*01 was present in all individuals. The 6 bears assayed for MHC class I expression had 6–9 alleles in genomic DNA, but only 3–5 were detected in cDNA (Additional file 3: Table S2). This indicates the expression of at least 3 MHC class I loci in the brown bear, and the presence of a minimum of two non-expressed loci; at least one of these numbers is certainly an underestimate, because a minimum of six loci were present in gDNA when the entire population sample is considered. There was complete concordance between genotypes of both replicates in all cases.

Alleles Urar-U*35 and *36 group with Aime-1906 locus of giant panda (Figure 2). Similar to the panda, alleles of this locus contain amino acid substitution at position 59, where Y is replaced by F, indicating non-classical nature of these alleles 57 58 . A similar substitution was found in Urar-U*02, *09 and *18 alleles, which form a separate, non-expressed cluster. Allele Urar-U*01 (present in all individuals) and the *24 group with Aime-152 locus. This locus, monomorphic in giant panda, does not bear the landmarks of a non-classical gene 58 .

Ninety of 228 (39.47%) nucleotide positions and 42 of 76 (55.26%) amino-acid positions were variable. Pairwise differences between alleles varied from 0.44% to 31.74% and amino-acid translations showed between 0 and 61.72% pairwise differences. Average nucleotide and amino-acid distances are listed in Table 3. Across all sites, d_N and d_S were similar and consequently d_N/d_S did not differ significantly from 1 (Table 4). For ABS, however, d_N exceeded d_S by a factor of two, although the excess of non-synonymous substitutions was marginally non-significant. The model of codon evolution allowing for positive selection (M8) fitted the data better than models without positive selection (Table 5). The Bayes Empirical Bayes procedure identified eight codons under positive selection (positively selected sites, PSS; Figure 1), five of which were located at ABS, which is more than random expectation (Fisher’s exact p = 0.001).

Distances and standard error values are given as percentages per site; standard errors, obtained through 1000 bootstrap replicates, are in parentheses. K2P, Kimura 2-parameter models.

Distances and standard error values are given as percentages per site; standard errors obtained through 1000 bootstrap replicates in parentheses. Significant results are highlighted in bold.

Parameter description : ω – d_N/d_S per all sites; ω1 – estimated value of ω for sites under positive selection, p0 – proportion of sites with ω ≤ 1 ; p1 – proportion of sites with ω > 1.

Because in pseudogens the signal of positive selection may erode over time, we also carried out tests for positive selection after excluding sequences of pseudogens and non-classical MHC class I genes, but this did not change the results qualitatively (Additional file 4: Table S3).

We assayed a 192 bp fragment of the DRB second exon. Sixteen DRB alleles were identified among 234 individuals (100 from the north and 134 from the south) (Additional file 5: DRB sequences, these sequences did not reach the minimum of 200 bp required currently for GenBank submission). Four of these were identical to alleles reported by Goda et al. 36 (Urar-DRB*11, *13, *16, *17). The sequences did not contain indels or PSC. The 16 nucleotide sequences translated into 15 unique amino-acid sequences (Figure 1b). Frequencies of individuals possessing particular alleles are given in Table 2. F_ST between the north and south was 0.304 (P < 10^-5).

Seven expressed alleles were found in six bears assayed for expression (Figure 3). Two to four alleles per individual were found in gDNA, all expressed, which implies the presence of at least two expressed loci. No discrepancies between replicates were found in 10 replicated individuals (maximum genotyping error 2.6%).

DRB alleles grouped with DRB of other ursids, and separately from DQB alleles (Figure 3). The phylogenetic tree did not reveal clusters corresponding to two loci inferred above: there were 3 well supported clusters, and relationships among the remaining alleles were poorly resolved (Figure 3).

Thirty-seven of 192 (19.27%) nucleotide and 21 of 64 (32.81%) amino-acid positions were variable. Pairwise nucleotide differences ranged from 0.52% to 16.87%, and amino-acid sequence differences ranged between 0 and 33.03%. Nucleotide and amino-acid distances are reported in Table 3. d_N significantly exceed d_S for all codons, and especially for ABS codons (by a factor of about 5); also at non-ABS sites d_N was higher than d_S (by a factor of nearly 3), but the difference was not significant (Table 4). PAML analysis showed the best fit of the positive selection model M8 (Table 5). The Bayes Empirical Bayes procedure identified 18 PSS (Figure 1), of which 12 were in ABS. The excess of PSS at ABS was significant (Fisher’s exact p < 0.0001).

Four alleles were found in the 224 bp fragment of the DQB 2^nd exon in 26 genotyped individuals (GenBank accession numbers JX469892-5), each translating into a unique amino-acid sequence (Figure 1c). The sequences did not contain indels or PSC. One to 3 alleles per individual were present and all of them appear expressed.

Urar-DQB 2^nd exon sequences formed a cluster separate from Urar-DRB sequences, but grouped with giant panda DQB (Figure 3). However, brown bear sequences were more similar to each other than to any of the giant panda DQB sequences.

Twenty-five of 224 (11.16%) nucleotide and 14 of 74 (18.92%) amino-acid positions were variable. Pairwise differences ranged between 2.27% and 11.04% for nucleotide sequences, and pairwise amino-acid differences ranged between 2.74% and 20.97%. Nucleotide and amino-acid distances are presented in Table 3. Across all sites, d_N was nearly equal to d_S, but for ABS d_N was significantly higher (by a factor of about 3; Table 4). However, PAML analysis did not provide evidence for positive selection, as M7 model fitted the data best.

Two alleles were found in the 202 bp fragment of DQA 2^nd exon in 26 genotyped individuals (Genbank accession numbers JX469890-1). The sequences did not contain indels or PSC. Each allele translated into a unique amino-acid sequence (Figure 1d). All six individuals assayed for expression had only allele Urar-DQA*05 in gDNA and cDNA. Allele Urar-DQA*06 is more similar to one of the giant panda’s alleles than to Urar-DQA*05 (Figure 4). The two nucleotide sequences differed by 2.97%, and the difference at the amino-acid level was 5.97%. d_N was not significantly different from d_S (Table 4).