Mitochondrial DNA Variation of the Striped Hyena (Hyaena hyaena) in Algeria and Further Insights into the Species’ Evolutionary History

Abstract

Background: The striped hyena (Hyaena hyaena) occurs in a wide range from north and east Africa, through southwest Asia to India, but its distribution is increasingly patchy and many of its populations are in decline due to intense human pressure. Its genetic diversity and structure, phylogeography, and evolutionary history, remain poorly understood. Methods: In this study, we investigated mitochondrial DNA variation in Algerian striped hyenas. Moreover, with the aim of contributing to our understanding of the evolutionary history of the species, we also examined samples from other geographic regions and compared our results with those of the only previous study in which individuals from across the range of the species were analyzed. In particular, we performed a wide range of analyses of demographic history and estimation of the age of the extant mitochondrial DNA variation. Results and Conclusions: The Algerian population sample was monomorphic. Overall, the global patterns of genetic diversity and the results of some demographic history analyses support a scenario of population growth in the species, estimated to have occurred in the Late Pleistocene, but many of the analyses did not detect a significant signal of growth, most likely a result of the limited power provided by a small number of segregating sites. The estimates, from three different methods, for the time to the most recent common ancestor (TMRCA) of the mitochondrial DNA variation hovered around 400 ka, coinciding with one of the longest and warmest interglacials of the last 800,000 years, with environmental conditions similar to the Holocene.

1. Introduction

The striped hyena Hyaena hyaena (Linnaeus, 1758), the only extant representative of its genus [1,2], is a large mammalian carnivore distributed from north and east Africa, through southwest Asia to India [3,4,5,6]. Although once abundant and widespread across this range, its distribution is now mostly patchy, with the species extinct in many areas and the remaining populations generally in decline [4,6,7,8,9,10,11,12,13]. The global population is estimated at less than 10,000 individuals and, given the intense human pressure (persecution, poaching, poisoning, conflict with pastoralists, target of negative feelings and superstitious fears, prey base depletion, use as folk medicine, road mortality) to which it is subject, it is listed as ‘Near Threatened’ on the International Union for the Conservation of Nature (IUCN) Red List of Threatened Species [6].

The striped hyena is one of the four living species in the family Hyaenidae, the other three being the brown hyena (Parahyaena brunnea), the spotted hyena (Crocuta crocuta), and the aardwolf (Proteles cristata). The latter represents the oldest lineage of extant hyaenids, being quite distinct from the other three [1,2,14,15]. It has been estimated that the lineage ancestral to Crocuta diverged from the lineage ancestral to Hyaena and Parahyaena 9–11 Ma [2,15,16], while the latter two lineages diverged from each other about 4–5 Ma [2,17,18,19]. Traditionally, based on differences in body size and skull and pelage characteristics [20], the following five subspecies were accepted [3,4,8,21]: barbara in Northwest Africa (type locality: Oran, Algeria); dubbah in East Africa (type locality: Atbara, Sudan); syriaca in Syria and Anatolia (type locality: Antakya, Turkey); sultana in southern Arabia (type locality: Mt. Qara, Oman); and hyaena from Iran to India (type locality: Benna Mountains, Larestan, Iran). However, the distributions and limits of these classically recognized subspecies are not precisely known [3], and attempts to delineate their geographic ranges and boundaries should be seen as tentative [4]. Jenks and Werdelin [14] noted that molecular data could be very useful for assessing subspecies designations in the striped hyena. In fact, a study that sequenced a 340 base pairs (bp) fragment of the mitochondrial Cytochrome b (Cyt b) gene in 13 striped hyena samples from scattered locations throughout the historical and current distribution of the species, found low genetic diversity (four haplotypes with two variable sites) and a general lack of obvious phylogeographic structure, with three of the four haplotypes shared between Africa and Eurasia [17]. Thus, although the species exhibits marked geographic variation in size and coat color patterns, the currently accepted view is that there is no evidence for the recognition of subspecies [5,22].

In Algeria, the striped hyena had the same history of decline over the past few centuries as observed in many other regions of its distribution, and the species remains vulnerable and threatened by anthropogenic causes of mortality and in need of conservation efforts, but it has managed to persist throughout much of the country [23,24].

In this study, we sequenced a large sample of striped hyenas from Algeria (i.e., representing the traditional subspecies barbara) for a Cyt b fragment containing the shorter region analyzed by [17] to allow for comparisons. We also examined samples from geographic regions and traditional subspecies not included in that study, such as southern Arabia (subspecies sultana) and southern Iran (subspecies hyaena). Our objectives were to assess mitochondrial DNA (mtDNA) diversity of striped hyenas in Algeria and, expanding on the pioneering and seminal study of [17], contribute to increase our knowledge of the evolutionary history of the species.

2. Materials and Methods

2.1. Sampling and Laboratory Procedures

To investigate the mtDNA diversity of the striped hyena in Algeria, we collected samples from 25 specimens found dead on roads or in the field. We also analyzed samples of striped hyenas from Jordan (2), Saudi Arabia (3), Oman (5), and Iran (2) (Figure 1 and Table S1). DNA was extracted from muscle tissue and blood samples using the EZNA Tissue DNA kit (Omega Bio-Tek, Norcross, GA, USA), while that from dry hair samples was extracted following a simple polymerase chain reaction (PCR) buffer-based extraction protocol [25,26]. For every batch of samples, we used DNA extraction blanks to monitor exogenous contamination. We amplified a fragment containing the first 753 bp of the Cyt b gene with the primers L14724 (5′-TGATATGAAAAACCATCGTTG-3′ [27]) and H15791 (5′-AATGTAGTTGTCTGGGTC-3′ [28]). This fragment corresponds to positions 14952—15704 in the reference mitochondrial genome for the striped hyena (GenBank accession number NC_020669 [29]). For four hair samples (one from Algeria, one from Saudi Arabia, and the two from Jordan), we were unsuccessful in amplifying this fragment, which we interpreted as a result of DNA degradation. Thus, for those samples, we amplified a 363 bp part of the fragment (corresponding to positions 16—378 of Cyt b) containing the 340 bp region analyzed by [17], in two overlapping pieces using respectively the two following primer pairs: Hyena Cytb F1 (5′-CCAATGACCAACATTCGA-3′)/Hyena Cytb R1 (5′-TCAGCCATAGTTGACGTC-3′) (for a fragment of 198 bp) and Hyena Cytb F2 (5′-AACCGCCTTTTCATCAGT-3′)/Hyena Cytb R2 (5′-GACGTAACCTATGAATGC-3′) (for a fragment of 181 bp). PCRs were carried out in volumes of 15 μL with 1x PCR Buffer (NZYTech, Lisbon, Portugal), 2 mM MgCl₂, 0.2 mM of each dNTP (Meridian Bioscience, London, UK), 0.5 μM of each primer, 0.75 U of Supreme NZYTaq DNA Polymerase (NZYTech, Lisbon, Portugal), and 3 μL of DNA extract. DNA extraction and PCR blanks were included to check for contamination. Thermal cycling conditions consisted of an initial denaturation at 95 °C for 5 min, followed by 45 cycles of 30 s at 94 °C, 30 s at 50 °C, 30 s at 72 °C, and a final extension of 7 min at 72 °C. The results of the PCR amplifications were visualized on 2% agarose gels to verify PCR quality and absence of contamination, and the PCR products were purified with an Exo-SAP protocol [30,31] and sequenced by the Sanger method at Macrogen Spain (Macrogen Inc., Madrid, Spain).

2.2. Data Analyses

Sequences were edited, assembled, aligned, and checked for the absence of indels and stop codons using Sequencher 4.7 (Gene Codes Corporation, Ann Arbor, MI, USA) and the Translate program of the Sequence Manipulation Suite 2 [32]. We downloaded striped hyena Cyt b sequences from GenBank (mainly the sequences from [17] and 39 sequences of Iranian individuals from [33]) and included these in our alignments (Table S1). Trimming our sequences to 648 bp allowed us to include those Iranian sequences from GenBank, but in order to incorporate other shorter sequences from diverse geographic origins, in particular those from [17], we created a smaller alignment of 340 bp (Table S1). Sequence alignments were analyzed with FaBox 1.5 [34] to identify samples with identical sequences. File format conversions of sequence alignments for use in different computer programs were carried out using ALTER [35].

To characterize patterns of mitochondrial genetic diversity, and what they suggest in terms of demographic history of the striped hyena, we computed the number of polymorphic or segregating sites (S), the number of haplotypes (n_H), haplotype diversity (h), nucleotide diversity (π), the mismatch distribution [36,37], Tajima’s D [38], Fu’s F_S [39], and the R₂ statistic [40] in Arlequin 3.5.2.2 [41] and DnaSP 5.10.1 [42]. We also computed Fu and Li’s D* and F* tests [43], as they are particularly powerful for detecting background selection [39]. If only these tests are significant, this suggests the action of background selection; if, on the contrary, they are not significant, but, for example, the Fu’s F_S test is, then this is more compatible with a history of population growth [39].

To visualize and examine the genealogical and geographical relationships among haplotypes, we estimated haplotype networks using the integer neighbor-joining method (IntNJ) in PopART 1.7 [44]. The IntNJ method was run with the reticulation tolerance parameter α set to zero. We used outgroup rooting [45,46,47] to estimate the ancestral node of the ingroup. The outgroups used here, and in the phylogenetic tree reconstructions described next, were the brown hyena (P. brunnea) and the spotted hyena (C. crocuta).

To infer evolutionary relationships among haplotypes and assess the presence of clades within H. hyaena, we constructed phylogenetic trees. The tree reconstructions were conducted with the data partitioned by codon position, as this was the best-fit partitioning scheme according to the corrected Akaike information criterion (AICc) [48,49,50] in PartitionFinder 2.1.1 [51] using PhyML [52]. Phylogenetic analyses were performed using Bayesian inference (BI) and maximum likelihood (ML) as implemented in MrBayes 3.2.6 [53] and IQ-TREE 1.6.12 [54], respectively. Analyses in MrBayes were conducted with two parallel Markov Chain Monte Carlo (MCMC) runs, each with four Markov chains (one cold and three heated), default heating parameter (t = 0.1), and 20 million generations. The first five million generations were discarded as burn-in and, thereafter, chains were sampled every 500 generations. The entire general time-reversible (GTR [55]) substitution model space was sampled within the analyses [56]. Convergence was indicated by an average standard deviation of split frequencies between parallel runs of less than 0.01. For all model parameters, the effective sample size (ESS) was greater than 700 and the potential scale reduction factor (PSRF) was 1.0. Support for tree nodes was determined according to the values of Bayesian posterior probability (BPP) obtained in a majority-rule consensus tree [57,58]. In IQ-TREE, we used the nucleotide substitution models estimated as best fit for each codon position, according to the Bayesian information criterion (BIC; [59]), by ModelFinder [60], a model-selection method implemented in IQ-TREE (648 bp dataset, according to codon position: the Kimura two-parameter model (K2P [61]); Hasegawa–Kishino–Yano model ([62]) with a proportion of invariable sites (HKY + I); HKY) (340 bp dataset, according to codon position: K2P + I; HKY; Tamura–Nei model (TN93 [63])). Tree search runs were performed using IQ-TREE default settings, with the exception of the ‘-allnni’ option which turns on a more thorough Nearest-Neighbor Interchange (NNI) tree search, and support for each node was evaluated by 1000 non-parametric bootstrap replicates [64]. Majority-rule consensus trees [57,65] were computed with SumTrees 4.5.2 of the DendroPy library version 4.5.2 [66] and visualized and edited with FigTree 1.4.4 (available online: https://github.com/rambaut/figtree/releases (accessed on 1 October 2022)).

To try to learn more about the recent evolutionary history of the species, we performed several analyses specifically on demographic history. In these analyses, we used only the 648 bp dataset (76 striped hyenas; 13 segregating sites), as the 340 bp dataset (50 individuals) only contained five segregating sites. For these analyses, we used several different methods: MCMC sampling-based skyline plots, single tree-based skyline plots, the likelihood approach of Weiss and von Haeseler [67] as implemented in IPHULA 1.16 [68], the MCMC coalescent genealogy sampler LAMARC 2.1.10 [69,70], and the Markov chain simulation technique for ancestral inference implemented in Genetree 9.01 [71,72,73,74]. Methodological details of the demographic history analyses are given in the Supplementary Materials.

To estimate the time to the most recent common ancestor (TMRCA) of the observed mitochondrial genetic variation, we obtained an estimate of the substitution rate for the analyzed Cyt b fragment as follows. Given our observed Cyt b divergence between striped hyena and its closest relative, the brown hyena, and the divergence between the two lineages that the fossil record suggests may have occurred 4–5.2 Ma ago [1,16,19,75,76,77], an estimate of the substitution rate can be obtained using equation d_xy = 2 μT [78,79], where d_xy is the mean sequence divergence per site, μ is the mean substitution rate per site, and T is the time since divergence. Genetic estimates of divergence time between Hyaena and Parahyaena also fall at about 4–5 Ma [2,17,18,19]. We confirmed the assumption of nucleotide substitution rate homogeneity among the striped hyena haplotypes and between these and Parahyaena, using Crocuta as the outgroup, via generalized relative rate tests with topological weighting [80] in RRTree 1.1.11 [81], and with Tajima’s relative rate test [82] as implemented in Mega 11.0.9 [83]. In RRTree, statistical significance (p < 0.05) of rate heterogeneity among lineages was assessed using both synonymous substitution rates (Ks) and nonsynonymous substitution rates (Ka). The d_xy between striped hyenas and Parahyaena was computed in Mega using the substitution model that best fits the data, according to the BIC in both IQ-TREE and Mega, among the models available in Mega for distance estimation (TN93 for the 648 bp alignment; K2P for the 340 bp alignment), and the standard error was estimated using 1000 bootstrap replicates. As a point value for the divergence time between Hyaena and Parahyaena, we used 4.6 Ma, the midpoint in the 4–5.2 Ma interval suggested by the fossil record. Similar point values have been used or estimated for the divergence of the two lineages in recent genetic studies of hyaenids [16,18,19]. The d_xy estimate for the 648 bp alignment was 0.091 ± 0.013 (mean ± standard error), and for the 340 bp alignment it was 0.086 ± 0.017 (mean ± standard error). Approximating 95% confidence intervals (CIs) based on 1.96 standard errors yielded 95% CIs of [0.066, 0.116] and [0.053, 0.119], respectively. Using the above equation d_xy = 2 μT with T = 4.6 Ma, the d_xy estimates yielded substitution rate estimates of 0.99% per Ma (95% CI: 0.72–1.26%) and 0.93% per Ma (95% CI: 0.58–1.29%), respectively; i.e., ≈1% per Ma, the traditional average rate for mtDNA [84,85]. Also, integrating the uncertainty in the fossil record about the divergence time between Hyaena and Parahyaena results, respectively, in the following CIs: 0.63–1.45% and 0.51–1.49%.

We estimated the TMRCA of the extant striped hyena mitochondrial genetic variation, based on both 648 bp and 340 bp datasets and using different methods implemented in Thomson’s estimator [86], Genetree, and BEAST. An attractive feature of Thomson’s estimator is that it does not rely on a specific population genetic model [86]. On the other hand, the estimator requires data consistent with the infinite sites model. Accordingly, in the 648 bp dataset, we removed a polymorphic site (site 546), which caused the haplotypes H1 and H3, as well as H7 and H10, to no longer differ from each other. In the 340 bp dataset, it was also necessary to remove one polymorphic site incompatible with the said model (site 317), which caused the haplotypes H1 and H7, as well as H4 and H5, to no longer differ from each other. We computed Thomson’s estimator and its variance (the latter according to Equation (4) in [87]) using NumPy code kindly provided by Dr. Helmut Simon (https://github.com/helmutsimon), and 95% CIs (as in Equation (5) in [87]). The required estimate of the unfolded site frequency spectrum was obtained using the site.spectrum function of the R package pegas 1.3 [88] and an estimate of the ancestral sequence of the ingroup. We estimated this ancestral sequence using the two outgroups (P. brunnea and C. crocuta) in two different computer programs, FastML (http://fastml.tau.ac.il/ (accessed on 1 October 2022) [89]) and IQ-TREE, for comparison purposes, and both provided the same estimate of the ingroup’s ancestral sequence for each of the two datasets. In Genetree, we used both the constant population model and the exponential growth model, with the ML estimates of θ and growth rate (obtained as described in the Supplementary Materials) as generating parameters. Both the 648 bp and 340 bp datasets were made compatible with the infinite sites model by removing one inconsistent site in each dataset, as described above. The ingroup’s ancestral sequences used, one for each of the two datasets, were those above-mentioned, estimated using FastML and IQ-TREE. We ran 100 million simulations to obtain TMRCA estimates and likelihood surfaces with 10,000 points. In BEAST, we ran analyses with both outgroups and only with P. brunnea. The outgroups corresponded to the molecular clock calibrations used: (i) the divergence between the Crocuta lineage and the lineage ancestral to Hyaena and Parahyaena, for which we used a normally distributed calibration prior with mean 9.5 Ma and standard deviation 2 Ma [14,15,16,17,18]; and (ii) the split between Hyaena and Parahyaena, with a normal calibration prior with mean 4.6 Ma and standard deviation 0.3 Ma [2,16,17,18,19]. The datasets were partitioned by codon position. We used the piecewise-constant skyline tree prior [90], which has been shown to perform well in molecular dating analyses involving a mixture of inter- and intraspecific data [91,92], and the approximate continuous time Markov chain (CTMC) rate reference prior [93]. In the analyses of the 648 bp dataset with two outgroups, we used relaxed clock models (uncorrelated lognormal model, UCLN, [94]; random local clocks model, RLC, [95]) because an LRT (performed in Mega) of the strict molecular clock hypothesis rejected it (p < 0.001). In contrast, this hypothesis was not rejected (p = 0.060) for the dataset with only P. brunnea as the outgroup, in agreement with the results from RRTree and Tajima’s relative rate test, and therefore we assumed a strict molecular clock in the analysis of this dataset. Similarly, the strict clock hypothesis was rejected for the 340 bp dataset with two outgroups (p = 0.007), but not for the one with only P. brunnea as the outgroup (p = 0.226). For each BEAST analysis, we ran four independent replicate MCMC simulations, each with 50 million generations and sampled every 5000 generations following a pre-burnin of 10%. We also ran two replicate simulations without data to obtain estimates from the prior distribution and test the influence of priors on posterior distributions [94,96]. For each analysis, we used Tracer with the default burn-in to assess convergence of the chain to the stationary distribution, obtain estimates and ESS of parameters, and plot marginal posterior densities. After verifying convergence, and confirming that the posterior distributions of estimates were markedly different from the prior distributions, the tree files from the four runs with data were combined using SumTrees, with the first 25% (2500) trees from each run discarded as burn-in, into a maximum clade credibility tree (MCCT) with median node ages. This method has a good overall performance, in terms of accuracy in estimating ages and model fit, compared to other tree summary approaches [97,98,99,100]. After analyzing the results of the runs with the relaxed clock models, we noticed that in general, the TMRCA estimates for the striped hyena mitochondrial variation seemed too high (>800 ka) when compared with those obtained with Thomson’s estimator (range of mean estimates: ~410–420 ka), Genetree (range of mean estimates: ~240–330 ka), and the only previously published estimate that we are aware of (340 ka; [17]). Thus, we considered these high estimates likely to be erroneous. The apparent poor performance of the UCLN clock model does not seem to be due to the presence of significant rate autocorrelation among lineages, since, for both datasets with two outgroups, the 95% highest posterior density (HPD) interval of the covariance between parent and child branch rates contained zero [94]. The CorrTest method [101], implemented in Mega, also did not reject the hypothesis of independent evolutionary rates among lineages (p > 0.05). However, we also noted that the standard deviation of the UCLN clock (ucld.stdev parameter) was low in the analyses of both datasets (648 bp dataset: mean = 0.049, median = 0.010, 95% HPD interval = [3.7 × 10⁻⁷, 0.243]; 340 bp dataset: mean = 0.138, median = 0.055, 95% HPD interval = [9.9 × 10⁻⁶, 0.557]), which suggests that the data are quite clock-like. The same indication was given by the results for the posterior distribution of the number of rate changes (K) (rateChangeCount parameter) in the RLC analyses, as a greater posterior than prior probability for K = 0 supports the global clock hypothesis, while small or negligible posterior probability for K = 0 strongly rejects the hypothesis [95]. In the analysis with the 648 bp dataset, the posterior distribution for K (mean: 0.619; median and mode: 0; 95% HPD interval: [0,2]) showed a higher probability for K = 0 than the prior distribution. This was not so with the 340 bp dataset (mean: 0.769; median and mode: 1; 95% HPD interval: [0,2]), for which 43% of the probability mass of the posterior distribution was for K = 1, but still 42% of the probability was for K = 0. In such cases, a strict clock (SC) may perform better in dating estimation [102,103]. Marginal likelihood estimates, obtained using stepping-stone sampling in BEAST with 100 path steps and a chain length of 500,000, also supported the use of an SC (648 bp dataset: log marginal likelihood = −1344.5; 340 bp dataset: log marginal likelihood = −711.2) compared with the RLC (648 bp dataset: log marginal likelihood = −1344.5; 340 bp dataset: log marginal likelihood = −711.1) and the UCLN (648 bp dataset: log marginal likelihood = −1347.4; 340 bp dataset: log marginal likelihood = −713.2). Therefore, we decided to also use the SC in the analyses with two outgroups.

3. Results

3.1. Genetic Variation, Haplotype Relationships and Phylogenetic Analyses

The new striped hyena DNA sequences discovered in this study were deposited in GenBank under accession numbers PP328974-PP328975 (Table S1). Several observations indicate that the generated sequences are mitochondrial and not nuclear-integrated copies of mtDNA [104,105,106]. First, PCRs consistently yielded single products of the expected size. Second, independent replicate PCRs produced identical sequences. Third, the sequences were unambiguous and did not contain indels or stop codons [107,108,109].

Genetic diversity measures and results of mutation-drift equilibrium tests, for both the 648 bp and 340 bp alignments of striped hyenas, are presented in

Table 1. Estimates of mitochondrial genetic diversity and mutation-drift equilibrium tests in striped hyenas, based on the two Cyt b alignments with different sequences and lengths.

	S	nH	h	π	D	FS	R2
648 bp dataset (n = 76)	13	11	0.794 ± 0.022	0.0026 ± 0.0017	−0.993 (p = 0.166)	−2.662 (p = 0.126)	0.064 (p = 0.181)
340 bp dataset (n = 50)	5	7	0.498 ± 0.082	0.0020 ± 0.0017	−0.922 (p = 0.198)	−3.345 (p = 0.020)	0.071 (p = 0.149)

n, number of samples; S, number of segregating sites; n_H, number of haplotypes; h, haplotype diversity (mean and standard deviation); π, nucleotide diversity (mean and standard deviation); D, Tajima’s D statistic [38]; F_S, Fu’s F_S statistic [39]; R₂, Ramos–Onsins and Rozas’ R₂ statistic [40]; p, p-value of the test.

. In the longer alignment, the number of polymorphic sites was 13, with nine being parsimony-informative and four singletons, and in the shorter alignment, the number of polymorphic sites was five, four of them parsimony-informative and one a singleton. In both datasets, none of the mutation-drift equilibrium tests used rejected at the 95% confidence level the null hypothesis of a constant size population under the neutral model; note that in the F_S statistic test, an α = 0.02 corresponds to a 5% significance level [39]. The D* and F* statistics were respectively −0.739 (p = 0.233) and −0.983 (p = 0.180) for the 648 bp alignment, and 0.134 (p = 0.529) and −0.228 (p = 0.396) for the 340 bp alignment. Analysis of the mismatch distribution for the 648 bp dataset under a model of sudden demographic expansion [37], implemented in Arlequin, did not reject the expansion hypothesis (p = 0.678; 20,000 bootstrap replicates). In agreement, the Harpending’s raggedness index [110] was low (0.032; p = 0.552). The model parameter for the expansion time was estimated at 1.877 (in mutational time units) (95% CI: 0.715–3.736), which, assuming a nucleotide substitution rate of 1% per Ma, translates into 144,830 years (95% CI: 55,170–288,272). For the 340 bp dataset, the analysis of the mismatch distribution under the sudden expansion model failed to converge, possibly due to the low number of polymorphic sites in this dataset. However, analyses in DnaSP showed that the observed mismatch distribution fit better with the expected distribution under a sudden expansion model [36] than under a model of constant population size (Figure S1). We also attempted in Arlequin to test the mismatch distribution under a model of spatial expansion [111,112], but the least-square procedure to fit the expected and observed mismatch distributions did not converge.

Both the haplotype networks (Figure 2) and the BI and ML phylogenetic trees (Figures S2–S4) for both the 648 bp and 340 bp datasets indicate a shallow genetic divergence among the striped hyena haplotypes, and essentially an overall lack of phylogeographic structure.

3.2. Demographic Simulations

The different MCMC sampling-based skyline plot analyses conducted in BEAST did not identify signs of significant population size change in the Late Quaternary demographic history of the striped hyena (Figures S5–S7). Details of the results of these analyses are given in the Supplementary Materials. The results of the single-tree skyline plot analyses performed in APE suggested population growth during the Late Pleistocene (Figures S8 and S9). The fit of the different demographic models tested in genieR was compared using the Akaike information criterion (AIC). The constant population size model was rejected (ΔAIC = 90), while the exponential growth (AIC = −1575.919) and expansion growth (AIC = −1573.919) models were equally supported (ΔAIC ≤ 2).

The implementation of Weiss and Von Haeseler’s method [67] in IPHULA estimated a maximum likelihood model parameter combination (ρ = 10, θ = 2.8, τ = 0.3) that suggests slow exponential growth since the Late Pleistocene. Specifically, assuming a nucleotide substitution rate of 1% per Ma and a generation time for the striped hyena of 9.2 years [113], a τ of 0.3 translates to about 23 ka. In turn, a θ of 2.8 corresponds to a female N_e of 23,484 at that time. However, the 95% confidence set of models, based on the likelihood ratio of each combination of demographic parameters tested and the maximum likelihood model, included scenarios of constant population size.

In LAMARC, analyses using the exponential growth or shrinkage model suggested a slow exponential growth scenario. The highest likelihood estimate of g was 910 (95% HPD interval: [−83, 4704]; ESS > 700 in all individual replicate runs). Therefore, the 95% HPD interval crossed zero, not ruling out scenarios of constant population size or even contraction. Moreover, LAMARC estimates of g tend to be biased upwards when only a few loci are analyzed [114]. The highest likelihood estimate of current θ per site was 0.008 (95% HPD interval: [0.002, 0.020]; ESS > 350 in all individual replicate runs), which corresponds to a θ of 5.2 for the 648 bp fragment and, again assuming a nucleotide substitution rate of 1% per Ma and a generation time of 9.2 years, can be translated into a female N_e (N_ef) of 43,478. However, LAMARC estimates of θ in the varying population size model may also be slightly biased upward due to correlation with g when only a few loci are analyzed [114]. In the analyses with the constant population size model, the highest likelihood estimate of θ per site was 0.004 (95% HPD interval: [0.002, 0.006]; ESS > 850 in all individual replicate runs), which corresponded to a θ of 2.6 for the 648 bp fragment and, assuming a nucleotide substitution rate of 1% per Ma and a generation time of 9.2 years, can be translated into an N_ef of 21,739. This θ estimate was very similar to that obtained using Watterson’s estimator [115] based on the number of segregating sites: 2.7.

In Genetree, the analyses using the constant population size model also yielded an ML θ estimate of 2.7 (thus, a N_ef of 22,645), while in the exponential growth model, the ML θ estimate was 3.1 (i.e., N_ef = 26,000); the latter model suggested a slow exponential growth rate. However, a LRT did not detect a significant difference in the fit to the data between the simpler constant population size model and the exponential growth model (p = 0.460).

3.3. Time to Most Recent Common Ancestor (TMRCA)

Thomson’s method estimates for the TMRCA of the observed genetic variation in H. hyaena were 419 ± 112 ka (mean ± standard error; 95% CI: 225–672 ka) based on the 648 bp dataset and a nucleotide substitution rate of 1% per Ma, and 410 ± 160 ka (mean ± standard error; 95% CI: 153–793 ka) based on the 340 bp dataset and a nucleotide substitution rate of 0.93% per Ma (

Table 2. TMRCA estimates for the striped hyena and between striped hyena and the outgroups used for time calibrations in the phylogenetic dating analyses.

	Hyaena hyaena	H. hyaena/P. brunnea	(H. hyaena + P. brunnea)/C. crocuta
Thomson’s estimator
648 bp dataset	418.938 ka a (95% CI: 225.273–672.201)
340 bp dataset	410.256 ka b (95% CI: 152.578–792.834)
Genetree
648 bp dataset
Constant population size model	333.334 ka a (95% CI: 125.000–541.668)
Exponential growth model	303.784 ka a (95% CI: 136.344–471.224)
340 bp dataset
Constant population size model	279.418 ka b (95% CI: 44.118–514.718)
Exponential growth model	240.043 ka b (95% CI: 83.493–396.593)
Beast c
648 bp dataset; calibration with two outgroups d
Strict clock (estimated clock rate: mean = 1.3% median = 1.3% 95% HPD = [0.9%, 1.8%])	377.8 ka (mean: 395.9 ka) (95% HPD: 172.7–657.2)	4.598 Ma (mean: 4.599 Ma) (95% HPD: 4.015–5.164)	9.323 Ma (mean: 9.393 Ma) (95% HPD: 6.808–12.118)
Random local clock (estimated clock rate: mean = 1.3% median = 1.3% 95% HPD = [0.9%, 1.8%])	387.3 ka (mean: 422 ka) (95% HPD: 161.5–711.9)	4.600 Ma (mean: 4.602 Ma) (95% HPD: 4.010–5.163)	9.304 Ma (mean: 9.369 Ma) (95% HPD: 6.705–12.070)
648 bp dataset; calibration with one outgroup e
Strict clock (estimated clock rate: mean = 1.6% median = 1.5% 95% HPD = [0.8%, 2.5%])	333.8 ka (mean: 353.1 ka) (95% HPD: 135.0–613.3)	4.605 Ma (mean: 4.605 Ma) (95% HPD: 4.028–5.199)
340 bp dataset; calibration with two outgroups d
Strict clock (estimated clock rate: mean = 1.3% median = 1.2% 95% HPD = [0.8%, 1.8%])	440.7 ka (mean: 472.1 ka) (95% HPD: 143.4–872.3)	4.567 Ma (mean: 4.569 Ma) (95% HPD: 3.973–5.136)	9.874 Ma (mean: 9.940 Ma) (95% HPD: 6.908–13.058)
340 bp dataset; calibration with one outgroup e
Strict clock (estimated clock rate: mean = 1.1% median = 1.1% 95% HPD = [0.6%, 1.7%])	483.8 ka (mean: 522.3 ka) (95% HPD: 150.8–974.3)	4.603 Ma (mean: 4.604 Ma) (95% HPD: 4.010–5.183)

^a Estimate using a substitution rate of 1% per Ma. ^b Estimate using a substitution rate of 0.93% per Ma. ^c For each of the analyses carried out in BEAST, we also present the respective estimated clock rate in % per Ma. ^d Two calibration points: Hyaena hyaena/Parahyaena brunnea (4.6 Ma; 95% interquantile range: 4.012–5.188 Ma) and (H. hyaena + P. brunnea)/Crocuta crocuta (9.5 Ma; 95% interquantile range: 5.58–13.42 Ma) (normally distributed calibration densities). ^e One calibration point: Hyaena hyaena/Parahyaena brunnea (4.6 Ma; 95% interquantile range: 4.012–5.188 Ma) (normally distributed calibration density).

). Adding the uncertainty in the fossil record and in the d_xy estimate for the divergence between Hyaena and Parahyaena into the nucleotide substitution rate estimate (i.e., 95% CIs of 0.63–1.45% and 0.51–1.49% for the 648 bp and 340 bp datasets, respectively) yielded the following respective 95% CIs for the TMRCA: [155,1066] and [95,1447]. In Genetree, analyses of the 648 bp dataset with the constant population size model and the ML estimate of θ of 2.7 yielded a TMRCA estimate, in coalescent time units, of 1.6 ± 0.5 (mean ± standard deviation), which, using a nucleotide substitution rate of 1% per Ma and a generation time of 9.2 years, corresponds to 333 ± 104 ka; approximating the 95% CI by ±2 standard deviations [116] gives 125–542 ka (86–860 ka including uncertainty in the fossil record and in the d_xy estimate for the divergence between Hyaena and Parahyaena in the nucleotide substitution rate estimation). With the exponential growth model and the respective ML estimate of θ of 3.1, the TMRCA estimate in coalescent time units was 1.27 ± 0.35 (mean ± standard deviation), i.e., 304 ± 84 ka (95% CI: 136–471 ka) (94–748 ka incorporating the aforementioned uncertainty in the nucleotide substitution rate estimate). Analyses of the 340 bp dataset with the constant population size model and an ML estimate of θ of 0.9 yielded a TMRCA estimate in coalescent time units of 1.9 ± 0.8 (mean ± standard deviation), which, using a nucleotide substitution rate of 0.93% per Ma and a generation time of 9.2 years, translates into 279 ± 118 ka (95% CI: 44–515 ka) (28–939 ka considering the uncertainty in the nucleotide substitution rate estimate). Analyses of the same dataset with the exponential growth model and an ML estimate of θ of 1.1 resulted in a TMRCA estimate in coalescent time units of 1.38 ± 0.45 (mean ± standard deviation), i.e., 240 ± 78 ka (95% CI: 83–397 ka) (52–723 ka considering the uncertainty in the nucleotide substitution rate estimate) (Table 2). The fact that Genetree TMRCA estimates using two very different demographic models were quite similar suggests a minor effect of the underlying demographic model on the estimates. TMRCA estimates from the BEAST analyses for different combinations of dataset, calibrations, and clock model, are given in Table 2; ESS values were > 3000 for all parameters in all individual runs under the strict clock, and > 200 under a random local clock. Across the different analyses, the TMRCA point estimate for the observed mtDNA variation in striped hyenas ranged from 330–480 ka, with the clock rate point estimate between 1.1% and 1.6% per Ma, hovering around 1.3% per Ma (Table 2), which is quite close to the rate estimates of ≈1% that we inferred based on the d_xy distance estimate between Hyaena and Parahyaena.

4. Discussion and Conclusions

Topics such as genetic diversity and structure, phylogeography, and evolutionary history of the striped hyena remain poorly investigated [117]. To date, our knowledge of the mitochondrial genetic diversity and phylogeographic structure of the striped hyena essentially comes from the analysis of a 340 bp Cyt b fragment in 13 individuals from across the historical and present distribution of the species, which was part of a ground-breaking study on the evolutionary history of extant hyenas [17]. That analysis found low genetic diversity and no evident geographic structure of this diversity. In our study, we wanted to investigate the mitochondrial genetic variation of the striped hyena in Algeria, using for this purpose a Cyt b fragment encompassing the one analyzed by [17]. It was also our objective, with the addition of samples from the Arabian Peninsula and Iran (regions not covered in [17] and where classic morphological subspecies have been described) and using a diverse set of demographic history and coalescence time inference analyses, to conduct a more comprehensive and better-sampled global analysis of the species than in [17]. The finding that a sample of 24 Algerian striped hyenas was monomorphic for a 753 bp Cyt b fragment was striking, and agrees with the notion of low mitochondrial diversity in the species. The apparent absence of phylogeographic structure in the species seemingly also extends to southern Arabia and Iran (Figure 2 and Figures S2–S4), with, for example, haplotype sharing between Oman and Iran for the 648 bp fragment (Figure 2a), which therefore also calls into question the validity of the sultana and hyaena subspecies.

4.1. Demographic History

The observed patterns of moderate haplotype diversity and low nucleotide diversity (Table 1) are compatible with a historical population bottleneck in the species, followed by demographic growth and the appearance/accumulation of new haplotypes [118]. Moreover, the mismatch distributions for both 648 bp and 340 bp datasets showed a significant fit with expected distributions under a sudden expansion model. Nevertheless, it should be taken into account that tests based on the mismatch distribution have been noted as conservative in rejecting the null hypothesis of population expansion [40,119]. On the other hand, none of the mutation-drift equilibrium tests used rejected the hypothesis of constant population size (Table 1). A reduced number of segregating sites, as is the case here, can limit the power of these tests [40]. In particular, the values of the F_S statistic were clearly negative, a result consistent with expectations in a scenario of population growth, which may suggest a lack of power in the datasets due to a small number of segregating sites. However, a limited number of segregating sites does not necessarily prevent significant results in those tests [120,121,122,123,124]. These tests also have less power if the demographic expansion is historically very recent (e.g., Holocene), especially if it was not very intense [39,40]. Rohland et al. [17] also did not detect a statistical signal of demographic expansion in their data (the authors did not report which analyses and tests of demographic history were performed), but the observed phylogeographic pattern was inferred to be the result of a recent and rapid range expansion, probably within the last 100 ka and possibly as recently as the Holocene, from a small refuge population, most likely located in Africa as, according to [1], no Late Pleistocene fossils of striped hyenas are known outside of Africa. The results of the single-tree skyline plot analyses and demographic model fit comparisons all support the population growth hypothesis. However, these single-tree methods have the limitation of not taking into account uncertainty in the tree topology [125,126]; we attempted to mitigate this problem by obtaining the tree to be used by Bayesian analysis and summarizing the posterior into an MCCT, the tree that maximizes the product of the posterior clade probabilities. All the most sophisticated demographic inference methods used (MCMC sampling-based skyline plots, IPHULA, LAMARC, Genetree) could not reject with high confidence (>95%) the hypothesis of constant population size. MCMC sampling-based skyline methods may allow us to correctly identify demographic trajectories based on single-locus data [127], but such datasets may also lack the power to reject the hypothesis of constant demography [128]. For example, although the sample size we analyzed cannot be considered small (n = 76), it has been noted in the literature that the BSP method may lack the power to correctly detect demographic expansions using single locus mitochondrial data when the sample size is <100 [129]. Again, also for those more complex methods, it is obviously preferable that the number of segregating sites is not small [130], but they have been applied to datasets with numbers of segregating sites similar to our data and produced significant results (e.g., [72,121,122,123,131,132,133,134]). A small number of segregating sites may not be the only challenge for the methods implemented in IPHULA, LAMARC, and Genetree. Another factor is the assumed population growth model, which in the case of these three methods is the exponential model. It is conceivable that the hypothesis of constant population size may not be rejected with confidence if an exponential growth model is very far from the true demographic history, which could be, for example, one of slow, more or less linear growth, or a complex history alternating recurrent episodes of expansion, stability, and/or decline (e.g., [125,135]). For example, in the BEAST analysis with the exponential growth coalescent model as tree prior, the marginal posterior distribution of the exponential growth rate parameter included zero, implying that a contrasting model of constant population size could not be rejected. Overall, the observed patterns of haplotype and nucleotide diversity, the mismatch distributions, and the single-tree skyline plot analyses and demographic model fit comparisons support the hypothesis of population growth, confirming the inference by [17]. This hypothesis was not statistically significant in the more sophisticated and exacting methods used, probably largely as a result of the limited power provided by a small number of segregating sites.

It was surprising to observe only one haplotype in 24 individuals from Algeria, whereas, for example, 41 animals from Iran showed eight haplotypes (648 bp fragment; Table S1, Figure 2a). It is therefore possible that the post-bottleneck demographic growth within Africa was not as rapid and intense as that associated with the colonization of Eurasia. If this was the case, then signatures of demographic expansion will perhaps be easier to detect using large Asian population samples. In the same way as in the human species, signatures of demographic expansion are easier to detect in non-African populations than in African ones [136,137] or in pooled or global datasets (e.g., [138]). Given the typical substitution rates in mitochondrial DNA, it is expected that recent demographic expansions (e.g., postglacial), particularly in organisms whose generation time is not short, may be difficult to detect with data from relatively short mtDNA fragments (e.g., <1000 bp), even using efficient methods in extracting the information present in the data [139]. Data from mitogenomes can be very useful in such cases [140,141]. Future mitogenomic studies will likely help to clarify open questions regarding the nature and age of the Late Quaternary striped hyena demographic and geographic expansion(s) suggested by the currently available mitochondrial data.

4.2. Age of the Extant Mitochondrial DNA Variation

For the estimation of the TMRCA of the sampled genetic variation, we also used very different methods, always based on the same rationale that if different analytical approaches yield congruent results, this increases our confidence in them. The Thomson’s estimator is simple and does not require the assumption of a model for the demographic history of the population. Genetree computes TMRCA likelihood distributions conditional on the gene tree mutation structure, a given value of θ, and the assumed model of population size history. For both approaches, the TMRCA estimates were translated into absolute time using the estimated substitution rates for each fragment (340 bp and 648 bp). In contrast, in BEAST, the time information to calculate the TMRCA in years was derived from fossil calibrations for the divergence between the striped hyena lineage and the brown hyena and spotted hyena lineages. The TMRCA point estimates of the observed genetic variation ranged between about 300 and 420 ka based on the 648 bp dataset, and between about 240 and 480 ka based on the 340 bp dataset (Table 2); hence there was relative agreement between the estimates from the two datasets. The variability of the estimates was lower for the dataset with longer sequence length and more segregating sites, as would perhaps be expected. For both datasets, estimates were higher with Thomson’s method (648 bp: 419 ka; 340 bp: 410 ka) and with BEAST (648 bp: 334–387 ka; 340 bp: 441–484 ka) than with Genetree (648 bp: 304–333 ka; 340 bp: 240–279 ka). This pattern is in line with arguments that Genetree results should be interpreted as estimated lower bounds for the true values [142,143]. In fact, most estimates from Thomson’s method and BEAST tended to hover around 400 ka, coinciding with one of the longest and warmest interglacials of the last 800,000 years (within Marine Isotope Stage 11, MIS 11: 374–424 ka [144]), with environmental conditions similar to the Holocene [145,146,147]. Rohland et al. [17], using the phylogenetic program r8s [148] and a point estimate of 10 Ma for the divergence between spotted and striped/brown hyenas as a calibration date, estimated the TMRCA of striped hyenas at about 340 ka, which is quite congruent with our estimates. One of our goals in this study was precisely to provide additional estimates, using different methods, for comparison with that one obtained with r8s. For instance, it has been suggested that BEAST may be more accurate than r8s for clade age estimation, particularly with single locus mitochondrial data [149].

4.3. Nuclear DNA Perspective

A recent whole nuclear genome sequencing study on the evolutionary histories of extant hyenas [19], in which the striped hyena was represented by a single captive individual of unknown geographic origin, indicated a very low genetic diversity in the nuclear genome of the species, and estimated a demographic growth in the lineage starting at ~500 ka that was eventually followed by a sharp decline since ~100 ka. A priority in the near future should be a genome-wide study with a broad geographic sampling throughout Africa and Asia to allow for a better characterization of genetic diversity and the impact on it of the anthropogenic reduction of the species in the last centuries as well as a more powerful and informative assessment of population structure and demographic history, as already attempted for its closest extant species, the brown hyena [150].