Genetic variation levels of Aime-MHC class I genes
In this study, we identified 24 exon 2–3 haplotypes for the 4 classical Aime-MHC class I genes in 218 wild individuals, averaging 6 haplotypes per locus. In our previous study , we detected 13 exon 2 and 16 exon 3 sequences, which formed 17 haplotypes in the Chengdu captive population, revealing that most diversity from wild populations was conserved in captive populations. Compared with the brown bear, the giant panda has similar or fewer MHC class I alleles. A total of 37 alleles (2 pseudo-alleles) were observed from at least 5 loci in 234 brown bear individuals, averaging 7 alleles per locus. However, compared with other endangered felids, Aime-MHC class I genes maintain a relatively high level of genetic diversity. For example, a total of 10 alleles (9 functional alleles and 1 pseudo-allele) were detected from 4 putative MHC class I loci in 108 Namibian cheetahs, averaging 2.5 alleles per locus . While 13 putatively functional alleles and one pseudo-allele were found from at least 4 MHC class I loci in 16 highly endangered India Bengal tigers . Furthermore, Aime-MHC class II genes also showed higher polymorphism relative to other endangered species . These findings suggested that the giant panda had relative higher genetic variation at their MHC genes, which is necessary for them to cope with changing environmental conditions (e.g., pathogens).
Genetic variation within populations
According to a survey conducted by the State Forestry Administration of China , XXL occupies the smallest habitat area and includes only 32 giant pandas. Interestingly, XXL represented more haplotypes, higher AR, and higher expected heterozygosity at MHC class I genes than those in the larger mountain populations, i.e., MSH, QLA, and QLI (Table 1). Our microsatellite data further revealed that XXL had the highest genetic variation among all of the populations in terms of AR, expected heterozygosity, and number of alleles. Furthermore, a recent MHC II study revealed that XXL has the greatest number of alleles within wild giant panda populations . These results, regardless of adaptive or neutral markers, suggested that the XXL population may have arisen from an ancestral population that had a higher level of genetic diversity, which was also supported by the results of MHC class II study . Although the MSH population covers the largest habitat area and contained 708 individuals as of the last survey round, it did not show the highest level of genetic variation, as was reflected by Ne estimates. Ne estimates based on microsatellites at 6 populations indicated that MSH had an Ne of 90.5, which was smaller than that of the majority of giant panda populations (see Additional file 3: Table S2).
ESUs, MUs, and AUs in giant panda populations
Population genetics data are useful to identify ESUs, MUs, and AUs in some endangered species [9,36]. In this study, we first defined ESUs in giant pandas in order to protect evolutionarily important groups. Second, we identified MUs in each ESU for management purposes. Finally, we looked for possible AUs to help the government make management decisions. MHC and microsatellite variations in this study revealed that the 6 giant panda populations formed 3 distinct groups. Based on these data, we recommended that the 3 groups be 3 AUs, but partitioned into 2 ESUs, and that one of the ESUs consists of 2 MUs.
The QLI population should be viewed as a separate ESU. Funk et al. defined ESU as “a population or group of populations that warrant separate management or priority for conservation because of high genetic and ecological distinctiveness,” and they recommended using neutral and adaptive markers to define ESUs, since neutral and adaptive processes both shape ESUs. Therefore, our recommendation is based on our present genetic data and previous ecological and molecular genetics studies [22, 23, 37, 38]. Our NJ trees based on microsatellite and MHC class I genes revealed that QLI formed a distinct cluster from other populations, which is consistent with our STRUCTURE analysis and previously reported genomic, microsatellite, and DNA fingerprinting data [22, 23, 37]. The QLI population is currently isolated from other populations by the Hanjiang and Jialingjiang rivers. Additionally, QLI giant pandas live in the south-central range of the QLI Mountain at elevations between 1300 and 2600 m, where the bamboo Bashania fargesii (E. G. Camus) Keng f. et Yi grows. In contract, other populations of giant pandas live at elevations of 2100 to 3400 m throughout the year and mainly eat bamboo of the genus Fargesia . Additionally, Wan et al.  revealed that QLI giant pandas have smaller skulls, larger molars, and different pelage color as compared to other populations’ individuals; these differences may be due to different habitat characteristics in QLI and other mountains. Based on DNA fingerprint and morphological data, Wan et al.  suggested that the QLI should represent a separate subspecies. However, whether this population represents a subspecies or a distinct ESU is still controversial . Because our evidence indicated that there is significant genetic and ecological distinctiveness between QLI and the other 5 southern populations, we propose that QLI should be a separate ESU and should be monitored and managed separately. Moreover, given that the QLI population has lower genetic diversity at MHC genes and microsatellites and fewer offspring in the captive population compared to the other 5 southern populations, captive breeding of Qinling giant pandas should be encouraged.
The other ESU contains 2 MUs, represented by MSH-QLA and DXL-XXL-LSH. MUs are usually defined as demographically independent populations . If the dispersal rate (m) is smaller than 10%, populations become demographically isolated . Dispersal rate or gene flow is shaped by neutral processes; therefore, neutral markers should be used to define MUs . Our Bayesian clustering analysis using microsatellites showed that 3 clusters existed within giant panda populations. Our results are different from those of a previous study based on microsatellites , where they detected 4 clusters (QLI, MSH, QLA, and XXL-LSH). In the present study, MSH and QLA formed 1 cluster, which was confirmed by an NJ tree and was consistent with the data from previously reported DNA fingerprinting and mtDNA analyses [22, 40], but was inconsistent with the results of Zhang et al.’s study . These inconsistencies could be the result of difference in samples used in the different studies. Three populations, i.e., DXL, XXL, and LSH, formed another cluster, which may not have conflicted with Zhang et al.’s study. Because there was only 1 sample collected from the DXL population in the previous study, this sample was considered part of the QLA population for the analysis . The Ne values for MSH-QLA and DXL-XXL-LSH were 200 and 300, respectively (see Additional file 3: Table S2). Given that the threshold dispersal rate is 10%, this corresponded to an FST of ~0.0125 (FST = 1 / [1 + 4Nem]). The FST between MSH-QLA and DXL-XXL-LSH was 0.038 (Table 2), which was greater than the threshold of 0.0125; therefore, we can conclude that these 2 clusters should be separate MUs. Moreover, QLI also deserved a separate MU given the greater pairwise FST between QLI and the other 2 clusters (Table 2). Since MSH and QLA showed no genetic structure among wild populations, we suggest that green corridors should be constructed between these 2 similar populations in order to preserve its existing genetic diversity and evolutionary potential of the populations. In addition, intrapopulation habitat fragmentation is a serious problem for the giant panda , so it is essential that we reconnect the patches inhabited by each population in order to enhance contemporary gene flow (individual dispersal) and ensure the long-term survival of the giant panda.
When discussing AUs, adaptive loci should be used . We determined 3 possible AUs (QLI, MSH-QLA, and DXL-XXL-LSH) based on patterns of variation at MHC loci that reflected the ability to adapt to various pathogens. These analyses suggested that QLI should be a separate AU, which was supported by our NJ tree and structure analyses and genomic structure data .Our NJ trees revealed that MSH and QLA were most similar (Figure 2A; bootstrap value = 78%); this was supported by our structure analysis and the FST value between these 2 populations (FST = 0.003), but was inconsistent with the results of Zhao et al. . They detected 3 distinct populations (QLI, MSH, and QLA-DXL-XXL-LSH) based on genomic data. The discrepancy lies in whether MSH and QLA should be together considered as a single AU and could be due to differential sensitivity of these 2 groups of markers. However, given that it is better to use adaptive loci to delineate AUs, it is hard to say whether MSH and QLA should be viewed as separate AUs, though genomic data is much more sensitive than specific genes of known function (i.e., MHC loci) . The genomic structure results reported by Zhao et al. were based on all loci . Furthermore, we do not have any data on different types of pathogens within giant panda populations that could directly reflect the different characteristics among possible adaptive groups. Therefore, we can only recommended 3 possible AUs given the above limitations to our data.