Different Microsatellite Mutation Models May Lead to Contrasting Demographic Inferences through Genealogy-Based Approaches: A Case Study of the Finless Porpoise off the East Asian Coast

Abstract

Understanding the population history of wide-ranging species, especially those ranging over varying landscapes, helps in deciphering the evolutionary force (s) that shaped the present genetic diversity/structure of regional fauna. In the shelf region, evolution of coastal morphology through glacial oscillations played an important role in shaping the contemporary genetic structure of coastal marine organisms, although the type and extent of such influence may differ between ecologically dissimilar species, such as marine mammals vs. other marine vertebrates. We reconstructed the demographic trajectories of four populations of the finless porpoise (Neophocaena spp.), covering a wide latitudinal range in the western Pacific and using coalescent-based techniques. Subsequently, we compare the findings with the evolution of suitable ecological niche by reconstructing historic sea level fluctuations with a maximum entropy method. Our results indicate that the finless porpoise was distributed along the continental slope during the low stand of sea level, while the post-glacial marine transgression enabled the porpoise to re-colonize a vast region of the shelf, leading to the most recent expansion of the genus in east Asia. We underscore that inferences of past demographic events are sensitive to the evolutionary model of microsatellite loci and the proportion of multi-step mutation. For coastal cetaceans inhabiting complex coastal habitats, caution has to be exercised when examining demographic parameters to prevent biased inferences due to historic gene flow during marine transgression. Systematic sampling scheme should be encouraged for rigorous quantification of demographic parameters, which may be further applied to more adaptable methods such as approximate Bayesian computation.

1. Introduction

It has been long recognized that the evolutionary history of populations has left imprints on their genetic material [1]. Early studies used to detect the demographic signals by examining the distribution of genetic diversity [2,3], while more recent developments in coalescent-based techniques examine the distribution of allele genealogies, which allows for population parameters such as the timing of demographic events and extent of demographic change to be rigorously estimated [4,5,6,7]. Examination of reconstructed population history against known records of environmental change offers a possibility to investigate the evolutionary forces that have likely shaped the contemporary occurrence and genetic diversity and structure of species and populations [8,9].

Microsatellites consist of a series of tandemly repeated (STR) sequence motifs across the genome. Though recent revolutionary progress in genome-sequencing technologies has enabled researchers to analyze genome-wide single nucleotide polymorphisms (SNPs) at lower costs, STRs present a much higher mutation rate than the point mutations in most sequence markers [10], and therefore provide genetic signals of more recent demographic events (e.g., dozens of generations) [8]. Furthermore, considerably more tools have been developed to reconstruct the past demographic history based on STR (e.g., MSVAR, VarEff, MIGRAINE, BEAST, etc.), which can make demographic inferences ranging from a few dozen generations to several thousands of generations conditional on the severity and timing of demographic change [4,7]. The robustness of the genealogy-based methods has been examined against multiple factors such as the demographic model (e.g., linear change vs. exponential change), the presence of gene flow and the mutation model of STR (e.g., the single-step mutation model/SMM, the two-phase mutation model/TPM and the generalized stepwise mutation model/GSM). Violation of these model assumptions may lead to a biased estimate of parameters or even a spurious conclusion [1,11]. As an example, in a mathematical framework based on the SMM, a departure from the SMM with large proportion of large step mutation [11] may lead to an upward biased estimate of the ancestral population size, and therefore a false detection of bottlenecks [1]. Though some model assumptions (such as the gene flow) can be easily assessed, many others are difficult or even impractical to validate (such as the demographic change model and mutation model of STR) for most species. As a result, the discrepancy of simultaneous use of multiple techniques has been rarely voiced and the potential bias was mostly ignored.

During glacial–interglacial transitions, the fluctuation of sea level has reshaped the seascape of the continental shelf region across the world, which is frequently used to explain the recent population history of coastal marine organisms [12,13,14]. In general, the glacial sea level changes had two major biological impacts: (1). altering the size of continental shelf areas and thus the suitable niche size of coastal fauna [13]; and (2). reshaping the topologic structure of coastline and therefore influencing the gene flow between marine realms [15]. Meanwhile, intrinsic features of marine organisms, such as habitat specialization or mobility of species, may have also influenced their response to glacial oscillations [16]. For instance, benthic fishes, which generally exhibit limited mobility as compared to their pelagic counterparts, experienced a more pronounced decline in the Antarctic shelf region during the formation of ice sheet [12]. Therefore, the demographic population histories of species, especially those displaying ecological dissimilarities, even if sympatric, should be examined independently.

Off the coast of China, a considerable divergence has been observed between marine organisms from the East and South China Sea [17,18,19], largely due to a prior land bridge that connected Taiwan Island with the continental mainland, which interrupted the water flow (and thus the larval dispersal) during the sea level low stands [20]. In both these marine basins, the topological development of coastline was greatly affected by the sea level change through glacial oscillation, with profound impacts on shelf-dwelling fish and much of invertebrate fauna [21]. However, the impacts on marine mammals remain less known as far fewer studies have been conducted to date [8,14].

The finless porpoise (Neophocaena spp.) is a small coastal toothed cetacean that inhabits coastal waters of the western Pacific and Indian Oceans [22]. The latest taxonomic classification categorizes the finless porpoise into two species: the narrow-ridged finless porpoise (N. asiaeorientalis) in waters north of Taiwan Strait (TWS), and the Indo-Pacific finless porpoise (N. phocaenoides), also known as (and hereafter referred to as) the wide-ridged finless porpoise, with a wide distribution from the Taiwan Strait in the east to the Persian Gulf in the west [22]. The genus is mostly allopatric throughout its range, with the only sympatric zone in the Taiwan Strait [23]. As reported for many other coastal marine organisms, the gene flow between wide- and narrow-ridged forms across the Taiwan Strait was interrupted during the last glacial maxima (LGM) [24], which has been frequently used as the genetic evidence to support the current taxonomic classification of the genus [22,25]. Genetic differentiation was also observed within the porpoise populations of each marine basin. For instance, Li et al. [26] found a moderate differentiation between the narrow-ridged porpoise in the Bohai Sea and Yangtze River mouth, while Lee et al. [27] reported a genetic differentiation between the narrow-ridged porpoises off the east- and west coast of Yellow Sea. Another recent study suggested that the wide-ridged finless porpoise from the Pearl River Delta (PRD) region and Taiwan Strait represent two genetic populations [28], and even deeper divergence was recently found between the porpoises in northern South China Sea and the Gulf of Thailand [29]. Although much of the east Asian coastal region has not yet been covered with a sufficient sampling effort, these recent findings suggest a limited gene flow between putative geographic populations [30].

Given the above, the current taxonomic classification of the finless porpoise may be over-conservative, as it cannot address the range-wide genetic pattern. For example, the divergence between wide- and narrow-ridged finless porpoise across the Taiwan Strait was found to be unnaturally low as compared to species-level divergence averaged over cetacean species [31], which is especially striking given the life expectancy and generational length, both of which are notably short for the finless porpoise [32,33,34,35]. Therefore, we should expect a relatively fast rate of substitutions (calculated in generations) compared to other cetacean species. Moreover, evidence of genetic structure was commonly noted within the range of currently recognized species [26,27,29,30], and some of the “intra-species” divergence was considerably higher than the “inter-species” divergence [28,29,36,37,38]. Instead of being an exceptional case [31], it was proposed that the interplay between glacial barrier, geographic isolation, and unknown vicariance may have shaped a multi-level divergence of the genus [28,36], which, however, remains to be tested with a more comprehensive sampling scheme [28].

Before such more comprehensive research is conducted, a study of the demographic history of populations may provide valuable insights into the evolutionary process of the extant species [39]. In the present study, demographic histories of four finless porpoise populations (including both wide- and narrow-ridged forms), covering a wide latitudinal range across various seascapes, were reconstructed using three coalescent-based methods. Subsequently, the current and ancestral niches at different sea levels were modeled for the finless porpoise along the coast of east Asia, to see whether the hypothetical population history as inferred by the genetic tools is supported by the evolution and development of the continental shelf since the LGM.

2. Materials and Methods

2.1. Sampling Locations and Sample Information

We collected the STR data of four finless porpoise populations from three sites in Chinese coastal waters (Figure 1), including:

1. The wide-ridged finless porpoises from the Pearl River Delta region (PRDw): the PRD is located in a transition zone between subtropical and temperate climatological zones in south China; it connects the Pearl River to the South China Sea (Figure 1). Following the end of LGM (c. 17–19 ka) [40,41,42], the sea level in the PRD reached its maximum around 8 ka [43]; thereafter, the coastline evolution was predominately shaped by sedimentation as result of the formation of monsoons [44,45] and in the recent two millennia by anthropogenic factors such as human migration, population explosion, and deforestation associated with the development of agriculture in south China [43]. The wide-ridged finless porpoise occurs in the PRD year-round [46]. This population differs from the conspecific population in the Taiwan Strait [37], but its distribution (and genetic relationship) in waters further west remains unknown.

2. The Taiwan Strait (TWS): two currently recognized species, the narrow-ridged and the wide-ridged finless porpoise inhabit waters of TWS and they appear to have been reproductively isolated since the LGM [24]. Occasional sally of species were seen, e.g., Lee et al. [27] reported two wide-ridged finless porpoises stranded on Jeju Island, Korea, while narrow-ridged finless porpoises have been found stranded, albeit very rarely, in Hong Kong [29]. Whether these distant migrations can lead to efficient gene flow remains unknown. For simplicity, we thereafter refer to the wide- and narrow-ridged finless porpoise from the TWS as TWSw and TWSn, respectively (Figure 1).

3. The narrow-ridged finless porpoise from the Bohai Sea (BSn): as the only inner sea of China’s coast, the BS is demarcated from the Yellow Sea by the Shandong Peninsula in the south and the Liaodong Peninsula in the north (Figure 1). The BS, along with much of Yellow Sea and the shelf region of the East China Sea, were exposed above the sea level during the sea level low stands, forcing coastal marine organisms to the narrow rim of Okinawa Trough [47,48]. Meanwhile, the Yangtze River was connected to the East China Sea through the paleo-river mouths, strengthening the connection between the epicontinental areas of the northern East China Sea and southern Yellow Sea [48]. The finless porpoise found in the BS is exclusively the narrow-ridged form, and it differs genetically from the porpoises seen around the Yangtze River mouth (the mid-East China Sea) and off the Korean coast in the Yellow Sea [27].

Figure 1. The distribution of samples obtained from finless porpoises in China’s coastal waters that were used in the present study, including both the narrow-ridged finless porpoise (N. a. sunameri) from Bohai Sea and wide-ridged finless porpoise (N. phocaenoides) from South China Sea (Pearl River Delta region). Data were also reconstructed for the two species occurring sympatrically in Taiwan Strait [24,49].

Figure 1. The distribution of samples obtained from finless porpoises in China’s coastal waters that were used in the present study, including both the narrow-ridged finless porpoise (N. a. sunameri) from Bohai Sea and wide-ridged finless porpoise (N. phocaenoides) from South China Sea (Pearl River Delta region). Data were also reconstructed for the two species occurring sympatrically in Taiwan Strait [24,49].

Muscle/skin tissues were collected from 45 wide-ridged and 16 narrow-ridged finless porpoise carcasses stranded in the PRD and BS, respectively (Figure 1). The tissues were preserved at −20 °C before genome extraction. Twenty-two STR loci [50,51,52,53] were cloned and genotyped as described further. STR data of 120 wide-ridged finless porpoises with known stranding locations were recently reported in [29] which were collected, processed, and genotyped together with the data reported by the present study, and are also included here. Ten STR data were reported for the TWSw and TWSn by Ku [49], and another 11 STRs were reported by Wang et al. [24]. As most STR loci did not overlap between these two studies, no calibration was required. For TexVet2, which was reported by both studies, we kept the records with larger sample size only to prevent errors associated with sequencing bias. We further discarded the data with unidentified allele size (see detail in the Supplementary File), which is essential for coalescent analysis. Moreover, while Wang et al. [24] combined samples of the wide-ridged finless porpoise from Hong Kong waters (equivalent to the PRDw in the present study) with those from TWS, assuming that finless porpoises from these two areas represented one genetic population, this assumption has been recently challenged [37]. As sampling structured populations may lead to false detection of bottleneck events, we omitted the data of the wide-ridged finless porpoise from Wang et al.’s study. The remaining data were then combined for the following analysis, which consisted of 16 loci (dinucleotide: EV1PM, EV37MN, EV94MN, FCB11, GT310, NP321, PPHO130, PPHO131, RW34, TexVet19, TexVet2 and TexVet5; tetra-nucleotide: NP309 and NP431; and di-/tetra-nucleotide: NP454 and GT271) and 10 loci (dinucleotide: GT310, TexVet2, EV37MN, PPHO130, EV94MN, PPHO131 and NP321; tetra-nucleotide: NP431 and NP309; di-/tetra-nucleotide: NP454 and GT271) for TWSn and TWSw, respectively.

2.2. DNA Preparation and STR Amplification

For the samples collected in the PRD and BS, the genomic DNA was extracted using the phenol/chloroform method. Twenty-two STR loci were amplified using fluorescently labeled primers according to the source papers [50,51,52,53]. All the PCR products were sequenced by the Invitrogen (Guangzhou, China), and the size of sequencing results were determined using GeneMarker v2.2.3. The presence of allelic dropout, null alleles, and stuttering of STR genotypes was tested with Micro-Checker 2.2.359, and the linkage disequilibrium of pair-wise loci was calculated using PopGene v1.3 (http://www.ualberta.ca/~fyeh, accessed on 25 February 2020).

2.3. Demographic Reconstruction Using MSVAR

As significant genetic differentiations were previously reported amongst the four finless porpoise populations [24,26,37], demographic inferences were run for each population under one population model.

The recent development of coalescent-based genealogical modeling methods enables quantitative inferences of effective population size before and after the demographic change, and also of the time of demographic event. We firstly detect the demographic event using MSVAR v1.3 [54], which assumes an isolated population experiencing demographic change from N₁ (effective population size before change) to N₀ (current effective population size) at generation-scaled time t, with an additional parameter of μ (mutation rate). This approach allows the mutation rate to vary between loci and simulate the genealogy separately for individual loci, then provides a posterior distribution of demographic parameters over all tested loci to preclude the bias associated with aberrant markers.

Exploratory simulations were first run with a wide range of prior distribution and hyperprior values for N₀, N₁, t, and μ for each population, to evaluate the robustness of simulation to prior parameterization. For the final analyses, the prior distribution of N₁ and N₀ were both set to the same range to avoid bias towards the hypothetical model. The variance of log-transform N was set as 2, to allow the effective population size to change from a few individuals to over a few million, which were expected to cover all the possible range of parameters. Log(T) was set to 5 with variance as 3, to allow a wide search range from a few hundred years to 10 million years before the present, which has been suggested to be the likely origination of the genus Neophocaena [28,55]. For each population, we ran three replicates of simulation under the exponential population model, with different starting points of estimated parameters randomly assigned for each locus. Each chain was run with 50,000 iterations and records were drawn every 20,000 steps. Fifty percent of the total chain was discarded as burn-in. Convergence between replicates was further evaluated using Brooks, Gelman, and Rubin Convergence Diagnostic (BGR Convergence Diagnostic). An estimate close to 1 and with the 97.5% quantile ≤ 1.2 was taken as a sign of good convergence [56]. All convergence tests were carried out with R package BOA [57]. The combined simulations of multiple replicates were used for the final inference of parameters using R package locfit.

2.4. Demographic Reconstruction Using VarEff

MSVAR assumes a strict SMM for the STR [54,58], which, however, is rarely met, especially for dinucleotide loci. When considerable departures from SMM model occur, the accuracy and precision of certain parameters might be compromised [4,8].

To evaluate the robustness of our data and simulation to possible deviation from SMM model, we further reconstructed the demographic trajectory of finless porpoises using the R package VarEff [7], which includes the SMM model, TPM and the more realistic GSM. The current and ancestral effective population size were both set as 5000 with a variance of log-transform N_e (VARP1) as 3; the time span of simulation was set as 10,000, and the variance of log distribution of searching time was 3. The maximal distance between alleles (DMAX) was adopted according to the observed value in each population to guarantee its representation of over 95% of data. The number of demographic events (JMAX) was set as 3; as the number of demographic events detected was exclusively less than JMAX, we did not rerun the simulation with further JMAX value. The coefficient of correlation between N_e (RHOCORN) was set as 0 to allow abrupt change of population size under some circumstances. The smoothing parameter (D) of 0.5 was adopted as indicated by Nikolic and Chevalet [7]. Three independent chains were run for 20,000 steps and sampled every 100 steps. Convergence of simulation of the individual chain was evaluated using the Heidelberger and Welch stationarity test [59], which tests the hypothesis that the Markov chain is generated through a covariance stationary process. The Heidelberger and Welch stationarity test was started by discarding the first 10% of iterations, with the procedure repeated until 50% of iterations were discarded when there was evidence of non-stationarity. Since VarEff did not include the inference of mutation rate in its model, we projected the result of VarEff following the μ according to MSVAR simulations and the generation length of the taxon to better illustrate the demographic trajectory in calendar year. Recent life table studies suggested the generation length of finless porpoise to range from 7.9 to 10.3 [32,60,61]. However, as our studied populations are under considerable anthropogenic stress [32,60,62,63], the observed values may be lower than the natural unaffected life expectancy of the species. Therefore, we set the generation length as the average age of reproductively mature females (BSn: 15 years; TWSn and TWSw: 16 years; and PRDw: 15 years) [61] as proposed by Lin et al. [36], to better scale the demographic history of the genus.

2.5. Demographic Reconstruction Using MIGRAINE

We inferred the demographic history of the finless porpoise using the maximum-likelihood framework via a OnePopVarSize model embedded in the program MIGRAINE [11]. OnePopVarSize gives inference of the current (2Nμ) and ancestral (2N_ancμ) effective population size (scaled in mutation rate), the duration of demographic change (scaled in generation, Dg), and the geometric distribution of allele size change (p_GSM). Analyses were run assuming the SMM for tetra-nucleotide loci and GSM for di- and di/tetra-nucleotide loci [11,64]. A wide boundary was set for the initial search of both N_anc and N (0.001–100) to preclude the hypothetical bias. The searching range was set as 0.1–0.9 for p_GSM and 0.0001–10 for Tg, which was expected to cover the wide time frame since the Pleistocene. The initial run was performed with 2000 parameter points with 2 iterations, and each point was run 1000 times. In the second run, the upper/lower boundaries of parameters were modified according to the result of the previous run, and a final run was repeated with three iterations, each with 50,000 data points run 2000 times, to make a final length of iteration up to 60,000,000.

2.6. Ecological Niche Modeling

The ecological niche modeling (ENM) was built on the basis of present geomorphological characteristics. The occurrence of finless porpoises was extracted from two periodic line transect studies in Korean waters [62,65], Bohai Sea [66], and Hong Kong waters [67], and georeferenced in ArcGIS. Although regular monitoring surveys have been performed in Hong Kong waters since the late 1990s, the porpoises in Hong Kong comprise only a small fraction of the entire population in the coastal PRD region [32]. As a high proportion of incompletely covered occurrence amongst the overall training data may lead to an inshore-biased prediction, we randomly choose one-year sighting data from the 2009–2010 monitoring program in Hong Kong [67]. Two geomorphological features, depth and slope, were extracted from the elevation map at a spatial resolution as 1 arc minute (accessed on 1 February 2018 at https://www.ngdc.noaa.gov/mgg/global/), which, together with the distance-to-shore (DS) and distance-to-the-closest-river-mouth (DRM), were taken as predictors for the porpoises’ niche. Pearson correlation coefficients between selected environmental variables were calculated to evaluate the correlation between variables [68,69]. Although other environmental features, such as currents, the volume of runoff, marine productivity, etc., may also shape the distribution of porpoises, these features were not available in most of our cases, especially for the paleo-coastline.

The ENMs were then projected for China’s coastal region at different sea levels using maximum entropy modeling embedded in the program MaxEnt [70]. In short, the program generates the probability distribution of species over the pixels of the study area by ensuring that the expected value of an individual environmental variable matches the empirical value over the sample area [70,71]. MaxEnt analysis starts with a uniform distribution of species (0 gain) and repeats until the difference between model iterations is below the convergence threshold. The regularization multiplier was set to 5 to smooth the output model. A threshold value of minimum training presence was followed to generate the absence data. We then used the threshold-independent measures, which calculate the area under the curve (AUC) metric of the receiving operator characteristic (ROC) curve, to evaluate the model fitness. The ROC curve plots the sensitivity values (true positives) against the specificity (false positive). AUC value of 0.5 indicates random prediction, while the value of 1 indicates perfect discrimination.

The relative importance of variables was measured with the percent contribution and permutation importance. The percent contribution calculates the increase in gains with each step and ranks the variables accordingly, with the higher increase in gains as a sign of greater importance of a variable. Alternatively, the permutation importance measures the decrease in AUC after randomly permuting a single variable of training data, where the larger decrease in training AUC, the more important a variable. The percent contribution is dependent on the particular iteration of MaxEnt algorithm and is sensitive to the correlation between variables.

3. Results

3.1. Data Summary

In the present study, muscle/skin tissues were collected from 45 wide-ridged and 16 narrow-ridged finless porpoises stranded in the PRD and BS (Figure 1), respectively, which generated unambiguous genetic results for the 22 microsatellite (or short tandem repeat, STR) loci, including eleven dinucleotide loci (YFP59, YFP8, YFP20, YFP1R, NP391R, YFP66R, NP399, GT271, PPHO130, Texvet2, and GT310), eight tetra-nucleotide loci (NP428, NP426, NP403, NP431, NP427R, NP309, NP445x, and NP432) and three combinations of both di-/tetra-nucleotide loci (NP321, NP464, and NP454). No sign of allelic dropout and linkage-disequilibrium between loci was detected, but null alleles were apparent at NP432, NP445x, and NP464; a signal for stuttering was also detected for locus NP432. Thus, we discarded these three loci in all subsequent analysis. PCR amplification failed for most of the BSn samples at PPHO130, which was further excluded for the BSn population. The remaining dataset was then combined with the data from 120 wide-ridged finless porpoises reported by Lin et al. [29]. The gene copies ranged from 22 to 36 (28.6 on average) for the BSn population and from 170 to 220 (197.0 on average) for the PRDw population.

After excluding data with inaccurate or incomplete allele size information (see details in the Methods), unambiguous allele frequency of five STRs (EV1Pm, TexVet5, RW34, FCB11, and IGF-1) was reconstructed from Wang et al.’s study [24] for 16 narrow-ridged finless porpoises stranded in the Taiwan Strait (TWSn), along with ten STR data reported for the TWSw and TWSn by Ku [49]. As mentioned above, since no overlap was found between the STRs of these two independent studies, no calibration was required. A final combined dataset consisted of 16 STRs allele frequency over 17 TWSn individuals and 11 loci over 29 TWSw individuals (as summarized in Table S2).

3.2. Microsatellite Genotype and Diversity

As only the allele frequencies were available for the two TWS populations, we could not assign alleles to specific individuals for heterozygosity analysis. Therefore, allele richness and expected/observed heterogeneity of these two populations were adopted from the source papers [49]. As shown in

Table 1. Molecular diversity of four finless porpoise populations from China’s coastal waters.

Population	N	# of Loci	A	He	Ho
BSn	16	19	6.6	0.73	0.60
BSn [26]	147	9	9.6	0.79	0.79
TWSn [49]	17	11	7.0	0.70	0.70
TWSn [29]	13	18	3.9	0.69	0.64
TWSw [49]	29	10	6.1	0.50	0.50
TWSw [29]	12	18	3.3	0.57	056
PRDw	112	19	10.5	0.65	0.50

BSn: the narrow-ridged finless porpoise population from Bohai Sea; TWSn and TWSw: the narrow-ridged and wide-ridged finless porpoise population from Taiwan Strait; PRDw: the wide-ridged porpoise population from the Pearl River Delta region; N: number of individuals; A: allele richness; He/Ho: expected/observed heterogeneity.

, allele richness was highest for the two datasets with the largest sample size (PRDw and BSn from [26,29]), suggesting that increasing sampling effort will raise the probability of discovering new alleles. When the diversity was measured in heterogeneity, the narrow-ridged finless porpoise displayed a higher level of diversity than the two populations of wide-ridged finless porpoise (Table 1).

3.3. Demographic History Based on MSVAR

The BGR convergence diagnostic indicated a good convergence of MSVAR simulations, except for the inference of current effective population size (N₀) and timing (T) of the population contraction for the BSn and PRDw populations (

Table 2. Convergence test using Brooks, Gelman, and Rubin diagnostics, the mode value and 95% Highest Posterior Density (HPD) distribution of demographic parameters, including the mutation rate (μ), current effective population size (N₀), effective population size before the demographic change (N₁), and time (T) of the demographic event (years before present) estimated with MSVAR simulations. Parameters with signs of non-convergence are indicated in bold. Narrow-ridged forms: N. a. sunameri; wide-ridged form: N. phocaenoides.

Population	Parameters	Convergence Test		Estimate
		Estimate	0.975 Quantile	Mode	95% HPD
Pearl River Delta (N. phocaenoides)	μ	1.003968	1.016194	0.00027	0.00011–0.00069
	N0	1.014079	1.066325	313	96–1560
	N1	1.000077	1.000437	245,625	91,368–660,314
	T	1.021028	1.090848	6987	2266–22,653
Taiwan Strait (N. phocaenoides)	μ	1.000183	1.001017	0.00026	0.00010–0.0006
	N0	1.068236	1.271684	19	5–75
	N1	1.001211	1.005221	252,187	25,2187–784,675
	T	1.049098	1.205557	1722	493–6901
Taiwan Strait (N. asiaeorientalis)	μ	1.000591	1.002447	0.00030	0.00012–0.00077
	N0	1.026537	1.109023	11	4–34
	N1	1.003475	1.016309	101,903	31,184–324,717
	T	1.040346	1.171372	722	250–2209
Bohai Sea (N. asiaeorientalis)	μ	1.000466	1.002282	0.00033	0.00013–0.00082
	N0	1.076863	1.293392	56	20–160
	N1	1.000356	1.001944	183,364	67,306–489,324
	T	1.057863	1.234439	2789	952–8089

). We further doubled the replicates to make a final length of simulations up to 10,000 × 2000 steps for these two populations, which, however, achieved little improvement. Thus, the lack of convergence was most likely due to the data structure instead of over-dispersion. We then discarded the burn-in steps and combined the latter half of the simulated data from different replicates for the subsequent analyses.

A clear separation of log(N₀) and log(N₁) indicated a recent bottleneck in all four populations (Figure 2). While the ancestral effective population size seems to be comparable in all four locations, estimates of the contemporary sizes differ considerably. Inconsistent with the genetic diversity, the largest effective population size was detected for the BSn and PRDw populations, with notably smaller values for the two TWS populations. The onset of population reduction was found to be after the end of LGM, except for the PRDw population which covered a wide timescale for T estimate. When the posterior distribution of T was rescaled into calendar year, it suggested that the population reduction most likely occurred during the marine transgression (Figure 2), with the two TWS populations displaying a most recent timing of a few thousand years.

3.4. Demographic History Based on VarEff

All VarEff simulations passed the Heidelberger and Welch’s stationarity test. When only a small proportion of multi-step mutations was allowed (C = 0.2), the coalescent simulations resulted in a recent bottleneck event regardless of the STR mutation models (Figure S1). The inferences of the contemporary and historic effective population size appeared relatively constant under different scenarios, although using the TPM or GSM resulted in more ancestral timing of the demographic decline than those following the SMM. When the C value was raised to 0.6, simulations under TPM still displayed population decline but with the timing of bottleneck varying between populations. Contrarily however, simulations under GSM indicated that present datasets were better explained by population expansion within a comparable timeframe (Figure 3). When the demographic trajectories were rescaled into calendar years, the expansions under GSM+0.6 C fell on or around the starting time of the post-glacial sea level rise (Figure 4).

3.5. Demographic History Based on MIGRAINE

As MIGRAINE requires the STR alleles to be precisely assigned to specific individuals, MIGRAINE simulations were conducted for the PRDw and BSn populations under the “OnePopVarSize” demographic model (see details in the Supplementary Material and Table S1), but were not applicable to the TWSn and TWSw populations. MIGRAINE simulations propose a high proportion of large step mutation for both PRDw (p_GSM = 0.571, 95% CI: 0.508–0.632, Figure S2) and BSn populations (p_GSM = 0.437, 95% CI: 0.314–0.579). For the PRDw population, the θ (3.95, 95% CI: 2.90–5.15) was substantially larger than θ_anc (0.35, 95% CI: NA–1.41), with the 95% confidence interval of N_ratio (42.3, 95% CI: 2.52–8486) larger than 1, indicating a significant expansion of the PRDw population. Using a population-specific substitution rate (0.00025, Table 2) and the generation length as 15 years, the point estimate for the onset of the PRDw population expansion (18.7 ka), albeit with a wide 95% CI range from 0.9 ka to 92.4 ka, was slightly earlier than, though comparable with, the estimate of VarEff under GSM + 0.6 C. However, for the BSn population, estimates of the parameter θ, θ_anc, and Dg were wide-ranging, with N_ratio covering a wide range both below and above 1, which provides little biological information to understand the demographic history of this population.

3.6. Ecological Niche Modeling

Of the four selected environmental variables, a moderate correlation was detected between DRM (distance-to-river-mouth) and DS (distance-to-shore, r: 0.621,

Table 3. Correlation between the four geographic characteristics of the present study area, including the depth, slope, distance-to-shore (DS), and distance-to-the-river-mouth (DRM). Value > 0.75 was taken as a sign of strong correlation, while those between 0.5 and 0.75 were taken as evidence of moderate correlation.

	Slope	DS	DRM
Slope
DS	0.271
DRM	0.094	0.621
Depth	0.619	0.386	0.190

). Although these two variables can often be expected to provide similar information, exceptions are found in the Shandong Peninsula and Hainan Island where the river mouths are not distributed continuously along the coastline as in most other regions. Therefore, we kept both DRM and DS as predicting variables. Correlation was also detected between the slope and depth (r: 0.619), likely because the slope was generated based on water depth. These two parameters, however, may provide different ecological information. As such, we kept both of them for the MaxEnt analyses.

The mean AUC (0.901, S.D.: 0.013) indicated a good fit of the present data (Figure 5). The DRM (percent contribution: 51.8%) outperformed the Depth (percent contribution: 24.2%) in describing the occurrence of the finless porpoise, which was consistent with the results of permutation importance (DRM: 43.4%; Depth: 21.9%). As expected, the slope contributed the least in predicting the porpoise presence probability (Figures S3 and S4). While generating predictions of present-day porpoise distribution, the score lower than the minimum training presence threshold (0.064) was set as not-suitable. Our projection indicates a continuous range of the genus, with a possible barrier between the Beibu Gulf and the rest of South China Sea (Figure 5); and lower densities off Shandong and Liaodong Peninsula and off the south coast of Hainan Island, probably due to the lack of a large estuarine system.

In the interglacial period, when the sea surface level was 60 m below the present level, the Taiwan Strait and Qiongzhou Strait were exposed above the sea surface, which prevented migration through the straits (Figure 5). The Beibu Gulf and Bohai/Yellow Sea experienced a severe marine regression, with over 40% loss of most suitable niches (suitability score > 0.50) in these areas. Moreover, the porpoises previously living around the rim of Beibu Gulf and Bohai/Yellow Seas had to retreat to small refuges, which likely facilitated the gene flows within these regions. Habitat loss continued in the paleo-Yellow Sea along with the sea level receding to its lowest level around 17–19 ka; while in the remaining part of what currently constitutes the coast of China, the coastal morphology remained relatively constant, though the porpoises had to shift their range from the inner continental shelf to the narrower outer shelf region that borders the continental slope.

4. Discussion

4.1. Robustness of Demographic Inferences

Coalescent simulations generated contrasting demographic inferences when different STR mutation models were applied. Under the SMM, both MSVAR and VarEff detected a population decline. Under the TPM, population decline was detected by VarEff with a wide range of large step mutations’ proportion (0.2 ≤ C ≤ 0.6). Under the GSM, population decline was detected with a small proportion of large step mutation (C = 0.2), but expansion was detected instead when the C value was raised up to 0.6. Taken together, the inconsistency of demographic inference mainly arose from the selection of STR mutation models rather than the mathematical framework used.

Though the mutation model of STR remains poorly examined for most of the non-model species, it is nowadays widely accepted that most STRs, especially dinucleotide STRs, do not evolve in a strict SMM manner [42,73]. Girod et al. [4] evaluated the influence of mutation model in MSVAR simulation, which detected a false signal of demographic decline when STRs were generated under strong GSM (p = 0.74). Similarly, the signal of population expansion may be overlooked by the TPM, as it will generate a similar allele composition of the deme by simply increasing the proportion value of large step mutation. As an example, early studies suggested a relatively small proportion of multi-step mutation in humans (13%) [11,74], while Di Rienzo et al. [75] simulated STR data under the TPM for human populations with well-known expansion history and found that only the simulations with a proportion of multi-step mutations higher than 80% were sufficient to describe the observed data.

Alternatively, the GSM is thought to be a more realistic model to describe the evolutionary process of STR [11], as it incorporates the confounding effect on large step mutation [76] as indicated by empirical data [77]. However, VarEff resulted in contrasting demographic inferences with different C values under GSM, indicating that the proportion of large step mutation may also influence the simulation process. Peery et al. [74] reviewed the STR of a wide range of non-human vertebrates and found that 60% of them displayed a large proportion of multi-step mutations (22% < pGSM < 68%), covering most of the range that we set for the VarEff simulation (0.2 ≤ C ≤ 0.6). Of the three programs used in the present study, MIGRIANE allows the coefficients of geometric distribution (p) approximately estimated together with the other demographic parameters, which resulted in a point estimate of p = 0.571. This value is highly comparable to the one computed for the Indo-Pacific humpback dolphin (59.1%) [8], suggesting that multi-step mutations may also occur frequently in marine mammals. Taken together, simulation under GSM with a high p value may provide a more accurate estimate of the demographic parameters [78].

4.2. Niche Suitability Modeling

The development and evolution of habitat is generally considered a major driving force of population history, especially during the pre-Holocene period [79,80]. As such, we reconstructed the niche suitability along the East Asian coast under different sea level scenarios, which we further compared with the demographic histories of the porpoise reconstructed under different genetic models.

MaxEnt modeling results suggested that the occurrence of finless porpoise was restricted within the inshore waters ca. > −100 m, which is supported by field observation in Bohai Sea [66] but is much wider than the range recorded in the Inland Sea (>−50 m, [81]) and along the Pacific coast of east Japan (−10 m to −35 m, [82]). It is noteworthy that the porpoise habitats in the remote Japanese waters may not resemble those along the continental shelf. Nevertheless, field studies are scarce in the continental coastal region, hindering us from evaluating the robustness of the niche modeling. Given the relatively small numbers of predicting variables used in the present study, the current results could be over-conservative and should be treated with caution.

Our results have implications for the population connectivity on a broad geographic scale. As an example, the low niche suitability recorded in the Yellow Sea basin and along the coast of Shandong Peninsula may account for the low gene flow between the east and west Yellow Sea [27], and between the Bohai Sea and west Yellow Sea [26]. This would then suggest that deep-water regions (e.g., the Taiwan Strait and Beibu Gulf basin) and coastal waters with distant river mouths may possibly function as geographic barriers. This hypothesis should be given further attention in future studies and sampling design to assess the population connectivity.

4.3. Change in Habitat Size vs. Demographic Change

Lowering the sea level a few dozen meters reduces most of the present shelf habitat around the world [13]. Although the glacial maxima generally lasted a relatively short period of time, the sea level remained at an intermediate level for a major part of the time since the Pleistocene [40]. Therefore, for coastal marine organisms, any contraction in population size associated with the shrinking habitat was likely initiated around 100 ka during the transition from the early to mid-Pleistocene [83], which predates the bottleneck events detected under SMM, TPM, or GSM with small C. Alternatively, the post-glacial expansion detected under the GSM with large C could be well explained by the marine transgression following the LGM ~18 ka. As such, we suggest that amongst the models generated in our study, the simulations under the GSM with large C reflect most accurately the demographic history of the finless porpoise.

During the LGM, the exposure of Bohai Sea, Yellow Sea, and the Taiwan Strait had led to a habitat size shrinkage of ca. 95% in the regions of Bohai Sea (including Yellow Sea) and Taiwan Strait, compared to the contemporary seascape. In the PRD region, due to its more linear coastline, the extent of change in habitat size (~70%) was relatively less. In line with the habitat size change, more severe demographic changes appear amongst the BSn (θ_anc/θ = 0.24) and TWSn (θ_anc/θ = 0.11) population, followed by that of the wide-ridged finless porpoise in the PRD region (θ_anc/θ = 0.31).

Unexpectedly, the wide-ridged finless porpoises from the Taiwan Strait, which was also subject to the glacial exposure of the Taiwan Strait, presented the least demographic change (θ_anc/θ = 0.60). A likely explanation is that the core area of the TWSw is not within the Taiwan Strait where the samples were collected, but is located in the northern apex of the South China Sea where the coastline remained relatively unchanged and with continuous freshwater input through the periods of sea level fluctuation. Approximately 70% of finless porpoises stranded ashore on the west coast of the Taiwan Strait were identified as the narrow-ridged form (L.Y. Zhao, personal communication), which supports the above hypothesis. If proven right, the TWSw population was likely less susceptible to the glacial exposure of the Taiwan Strait, and its habitat reduction during the low stand of sea level may be overestimated by the present study. This hypothesis, however, remains to be tested with more sampling and concerted survey effort in this region.

4.4. Change in Habitat Patchiness through the Sea Level Fluctuations

The coalescent framework of the present study has adopted the assumptions of one isolated population model, in which the extant genetic diversity of populations is not impacted by the influx of external genetic material [4,7,11]. When violated, the alleles within a population will give inference of the effective size of the target population, while the distant alleles brought by immigrants will coalesce in the larger external population, leading to a bimodal posterior distribution for the estimated effective population size or a “vanishing” population of the VarEff results [7]. Vanishing population was detected for the PRDw, TWSn, and BSn populations. Though the gene flow between these populations appeared low [24,26,37], genetic exchange between these populations and neighboring unsampled populations cannot be ruled out. This is further supported by our ecological niche modeling which did not identify geographic barriers along most of the east Asian coastline. As such, the assumption of no gene flow of the one isolated population model was very likely violated for the finless porpoise.

It is well recognized that geographic barriers caused by receding sea level, such as land bridges connecting offshore islands with the mainland, the closure of straits, etc., may and often do interrupt the gene flow between marine refugia [84]. Within the glacial refugia, the sea level fluctuation may influence the population connectivity or gene flow of coastal cetaceans through two major mechanisms. Firstly, the contours of inner seas, gulfs, bays, and straits shrink along with the retreating sea level. For example, the present-day coastline of Bohai/Yellow Sea of over 3000 km contracted to less than 300 km during the glacial epoch, forcing the distant populations previously inhabiting this region to restrict their range to the northern apex of the Okinawa Trough [47,48]. The reduced inter-population distance was likely to facilitate the gene flow, or might even homogenize previously differentiated populations during the low stand of sea level [36]. Such a scenario seems likely for marine mammal populations from the Beibu Gulf, the shallow waters in the Sunda Shelf. For example, the dugong (Dugong dugon) inhabiting the Gulf of Thailand, an area that completely dried up during the glacial maxima, exhibit little evidence of genetic structure, while those from the Andaman Sea form at least two genetic populations [85].

Furthermore, the gene flow within refugia may be reshaped by changing physiographic dynamics of estuaries, and this represents the second major mechanism that may affect population connectivity and/or gene flow. During the low stand of sea level, the organic-rich sediments were transported more directly into the deep sea through submarine canyons and raised the marine productivity in these areas [86]. Consequently, large predators such as marine mammals may have had to intensify their activities around the paleo-river mouths. Under such a scenario, population connectivity and its history have to be carefully evaluated in the context of the evolution of the local/regional coastal and oceanographic/hydrologic systems.

In our case here, the river mouths in the northern South China Sea and the southern East China Sea extended steadily seaward during the marine regression, which may have facilitated a continuous accumulation of the divergence associated with geographic distance [36]. In contrast, the complex estuarine system of the Yangtze River was considerably more spread out in the northern East China Sea during the LGM. As depicted in Figure 5, in its lower reaches, the Yangtze River was split into several branches, with a major branch located between the mid- and northern Okinawa Trough and likely functioned as a glacial corridor for the finless porpoise. This would well explain the lack of genetic differentiation between the finless porpoise from the East China Sea and those from the extant Yellow and Bohai Sea [28], though their present geographic distance is much larger than that of genetically differing populations in the present-day northern South China Sea [28].

4.5. Limitations of the Present Study

There were several limitations to the present study. Firstly, genetic study of marine mammals in Chinese waters are primarily based on sampling opportunistically beached carcasses or bycatch individuals [8,26,87,88], which makes it extremely difficult to control the sampling scheme to address specific research questions. For example, genetic differentiation was detected between all the examined putative population of the finless porpoise [22,26,37,89,90], but the present sampling effort covers only a small fraction of the entire genus range [28], with the only exception of the porpoise in Japan waters [30]. Without a better understanding of the porpoise dispersing biology, the genetic structure as observed by previous research may arise from multiple factors including vicariance, isolation-by-distance, or breeding isolation [24,28].

VarEff simulation provided an indirect evidence of gene flow. Given the stranding records of wide-ridged finless porpoise in Korean waters [27], an area previously thought to be allopatric to the narrow-ridged porpoises, the probability of ongoing gene exchange must not be overlooked. However, large unsampled areas between the three sampling sites of the present study hinder an incorporation of gene flow into the population modeling process. Furthermore, the disproportional sampling effort may also raise some concern. Of the four examined populations, the two narrow-ridged porpoise populations (BSn and TWSn) presented only marginally acceptable sample sizes (15 < n < 20) [78]. Therefore, the wide range of N-ratio of BSn population may not be translated into a lack of demographic change, but could be due to the lack of a sufficient genetic signal to depict the demographic change. To sum up, a more comprehensive and systematic sampling effort should be encouraged to assess the dispersing biology of the finless porpoise, which is vital for a more accurate inference of its population history.

Finally, some important environmental features, such as marine productivity and temperature [91], were not included in the present niche suitability modeling. The effect of marine productivity may be partially accounted for by the DRM, as the marine productivity in the coastal shallow waters is largely shaped by the terrigenous matter flux carried by the freshwater runoff [92]. The temperature, on the other hand, which seemingly plays a vital role in determining cetacean ranging [24], is not available for ancestral niche modeling. Exclusion of these factors may lead to an over-conservative prediction of suitable niche (e.g., overestimating the niche size). Nevertheless, as we have modeled the suitable niche within a region with well-known presence of the porpoise, the present study is likely to be representative of the change in habitat size and patchiness across different sea level conditions.

5. Closing Remarks

Our work presents a comparative study of the evolutionary history of the finless porpoise and its coastal habitat in what is a present-day coastal region of China. We suggest that the genus has undergone a post-glacial expansion along with the rising sea level, while the shortening of coastline and changing estuarine morphology during the low stand of sea level might have facilitated a gene flow within otherwise increasingly restricted (insular) environments. Post-glacial expansion may be symptomatic of the biogeographic history of cetacean species inhabiting the extant shelf region, yet this scenario remains speculative due to the uncertainty of demographic parameter inference such as the time-since-the-last-demographic-event. This limitation of the present study, however, cannot be addressed without a more comprehensive sampling scheme covering the clinal distribution of the genus. In a broader perspective, the use of coalescent-based methods to infer the population history of extant species is on the rise [4], but the inference of population parameters may be biased if the model assumptions (either the mutation model of genetic marker or the population model) are violated. We present a case study where the coalescent inference of the population parameters is not only sensitive to the misspecification of mutation model but also to the size-dependent mutation rate of STR. Similar cases have rarely been evaluated with current mathematical frameworks or have not been fully examined in previous studies that applied coalescent-based methods.