Centrality to the metapopulation is more important for population genetic diversity than habitat area or fragmentation

Abstract

Drift and gene flow affect genetic diversity. Given that the strength of genetic drift increases as population size decreases, management activities have focused on increasing population size through preserving habitats to preserve genetic diversity. Few studies have empirically evaluated the impacts of drift and gene flow on genetic diversity. Kryptolebias marmoratus, henceforth ‘rivulus’, is a small killifish restricted to fragmented New World mangrove forests with gene flow primarily associated with ocean currents. Rivulus form distinct populations across patches, making them a well-suited system to test the extent to which habitat area, fragmentation and connectivity are associated with genetic diversity. Using over 1000 individuals genotyped at 32 microsatellite loci, high-resolution landcover data and oceanographic simulations with graph theory, we demonstrate that centrality (connectivity) to the metapopulation is more strongly associated with genetic diversity than habitat area or fragmentation. By comparing models with and without centrality standardized by the source population’s genetic diversity, our results suggest that metapopulation centrality is critical to genetic diversity regardless of the diversity of adjacent populations. While we find evidence that habitat area and fragmentation are related to genetic diversity, centrality is always a significant predictor with a larger effect than any measure of habitat configuration.

1. Introduction

Genetic diversity is one of the three internationally recognized levels of biological diversity by the United Nations Convention on Biological Diversity and it can impact the evolutionary trajectory of species [1,2]. Given that natural selection acts on heritable phenotypic variation, genetic diversity dictates the ability of populations to evolve in response to environmental conditions [3,4]. Preserving genetically distinct populations (i.e. preserving species-level genetic diversity) has been championed to maintain the potential for peripheral populations to rescue declining populations with low genetic diversity through natural (i.e. evolutionary rescue) or augmented (i.e. genetic rescue) gene flow [5,6]. Decreases in genetic diversity are associated with increased extinction risk [7,8]. While all genetic diversity ultimately originates from mutation, observed levels of genetic diversity result from previous episodes of gene flow, genetic drift and selection [9]. Thus, genetic diversity is the product of past evolutionary forces while also dictating future evolutionary responses.

Genetic diversity often is attributed to gene flow and drift. Gene flow—the exchange of genetic material between populations—can either increase or decrease genetic diversity, depending on whether the focus is on the population or species level. Gene flow can increase population genetic diversity through the introduction of alleles from adjacent populations [10,11]; however, gene flow can reduce species-level genetic diversity by homogenizing allele frequencies and driving the loss of private alleles [12]. The pattern of gene flow across a group of populations is the result of immigration and emigration within a group of populations followed by successful reproduction (i.e. metapopulation structure) [13,14]. This genetic connectivity (i.e. gene flow) is impacted by the distance [15,16] and environmental conditions between populations [17]. Hence, environmental spatial heterogeneity between populations and the distribution of populations impact patterns of gene flow and, subsequently, patterns of genetic diversity. However, population genetic diversity is likely not the product of incoming gene flow from one population, but it is the sum of all incoming connections.

Genetic drift refers to stochastic changes in allele frequencies unrelated to fitness and it opposes local genetic diversity. However, independent bouts of drift between populations can maintain species-level genetic diversity through the retention of private alleles in isolated populations [18]. Because drift is stronger in smaller populations [19,20], demographic declines decrease genetic diversity [21,22] and limit the population’s ability to recover from or respond to environmental change [23,24]. Environmental changes that impact population size, such as habitat loss [25,26] and fragmentation [27,28], increase drift and decrease genetic diversity [29,30]. Gene flow and drift can operate simultaneously [31]; therefore, identifying environmental drivers of genetic diversity requires concurrently evaluating the abiotic conditions that influence both patterns of gene flow and the strength of drift.

Kryptolebias marmoratus [32], hereafter rivulus, is a cryptic self-fertilizing androdiecious killifish [33] that inhabits the highly threatened and fragmented mangrove forests in North America, Central America, the Caribbean and the Bahamas [34–36]. As anthropogenic activities such as greenhouse gas emission and land development continue, rivulus will face reduced habitat availability [37] and may be ill-prepared to evolve in response to these novel environmental conditions because populations often have low genetic diversity [38,39]. Rivulus dispersal is likely passive through eggs attached, via adhesive filaments, to flotsam or adults rafting within debris [39,40], thus limiting rivulus’ ability to leave unsuitable habitats. Gene flow between rivulus populations is generally low and asymmetric, with asymmetries associated with ocean currents [40], resulting in complex patterns of gene flow that may limit the introduction of adaptive alleles and genetic rescue.

Given rivulus’ limited genetic diversity, restricted gene flow, increased habitat loss and the future impact of climate change on their distribution [37], preserving existing genetic diversity is essential for the persistence of rivulus populations. By quantifying abiotic factors that impact the strength of drift (i.e. habitat area, fragmentation) and patterns of gene flow (i.e. oceanic connectivity), we can explicitly evaluate their independent contributions to genetic diversity while comparing their relative impacts. Because population genetic diversity is likely associated with gene flow patterns across the range, we use a network approach to quantify each population’s centrality to the metapopulation. Centrality refers to statistics that characterize how populations are connected within a directed network (see electronic supplementary material —Network Centrality Measures). We hypothesized that centrality to the metapopulation via ocean currents and patch qualities (e.g. habitat area) would influence population-level genetic diversity (H1). We predicted that increased centrality to the metapopulation would increase genetic diversity, while decreases in habitat area and increased fragmentation would decrease genetic diversity.

2. Methods

(a). Genetic and environmental data

We used 1245 published genetic samples [38–41] genotyped at 32 microsatellite markers [42] previously collected from 56 sites across Central America, the Bahamas, the Caribbean and North America between 1994 and 2014 (for details see electronic supplementary material,—Genetic Data). Using R v. 4.3.1 [43], we grouped sites into larger populations, filtered samples and estimated oceanographic connectivity between each pair of populations following Snead et al. [40] (for details see electronic supplementary material, —Biophysical Modeling), resulting in 17 populations and 1120 individuals (figure 1; electronic supplementary material, table S1). Using the biophysical model (Connectivity Modeling System - CMS) output files and the Global Land Cover and Land Use 2019 dataset [44], measures of oceanographic connectivity and habitat configuration were calculated. Oceanographic connectivity was calculated following [40] (for details see electronic supplementary material, Biophysical Modeling). The Global Land Cover and Land Use 2019 dataset [44] was downloaded at ~30 m² resolution and reclassified into either suitable habitat—wetlands (except salt pans) and water or unsuitable habitat—all remaining classifications resulting in a binary measure of potentially suitable habitat. Total habitat area, cohesion, edge density and the number of patches were calculated for suitable habitat within each population cluster buffer using Landscapemetrics [45]. Total habitat area is the measure of all suitable habitats within the population buffer, while the number of patches is a count of the number of disconnected clusters of suitable habitats. Cohesion is a measure of aggregation between 0 and 100 that characterizes the continuity of habitat within the buffer [46], and edge density represents the configuration of the landscape by calculating the number of edges (i.e. where suitable habitat meets unsuitable habitat) and standardizing it by the total area of within the buffer [47].

Figure 1.

A map of all the sampling locations of Kryptolebias marmoratus in Florida and the Caribbean. The sampling locations were grouped in populations as shown on the map and described in the text. The number of samples in each population is shown in the table.

(b). Genetic analysis

Snead et al. [40] use the same microsatellite data to explore local and regional population structure with an Analysis of MOlecular Variance [48], a Discriminant Analysis of Principal Components [49], TESS3 [50], sNMF [51], STRUCTURE [52] and InStruct [53], while patterns of gene flow were investigated using G _ST [54], G′_ST[55], Jost’s D [56], R _ST [57] and BayesAss [58]. Similarly to Snead et al. [40], deviations from Hardy–Weinberg equilibrium (HWE) were tested for each microsatellite locus at the population level and for the entire dataset with the package Pegas [59]. With the entire dataset, all loci deviated significantly from HWE, as expected owing to nonrandom mating; however, no locus deviated from HWE in all populations. Therefore, all loci were retained. While there is a large temporal range across the samples, previous work found low genetic differentiation (F _ST = 0.023) between samples collected over 10 years apart in Twin Cayes, Belize (a population with more males and higher genetic diversity) and even lower patterns of isolation by time in three populations across the Florida Keys (F _ST = 0.002), which predominantly self-fertilize with few males [39,41]. Therefore, previous results suggest little change in genetic diversity across the sampling period.

Unique to this experiment, the rarefied number of multilocus genotypes (eMLG; the average number of unique multilocus genotypes after randomly subsampling ten individuals across 1000 iterations), Zahl’s unbiased estimator (Z) [60,61], rarefied Stoddart and Taylor’s index (G) [62], expected heterozygosity (H _exp) [63], observed heterozygosity (H _obs) and the mean rarefied allelic richness (Ar) were calculated for each population with the packages poppr [64], PopGenReport [65] and adegenet [66]. To account for uneven sampling across populations, both Ar and G were calculated with rarefaction. There were no significant correlations between sample size and genetic diversity (eMLG[r = 0.13, p = 0.61], Z[r = 0.37, p = 0.15], G[r = 0.1, p = 0.71], H _exp[r = 0.38, p = 0.14], H _obs[r = 0.12, p = 0.64], Ar[r = 0.37, p = 0.14]). This combination of metrics was chosen to facilitate comparisons between typical population genetic diversity metrics that lack strong assumptions (i.e. Z, H _obs, Ar) with H _exp, which assumes random mating, and a measure of genotypic diversity specifically developed for mixed mating systems (G) [62]. While H _obs, H _exp and Ar are common metrics of genetic diversity in other mixed mating systems such as plant [67,68], comparing results with Z and G enables us to evaluate whether our inference is robust to metric choice and mating systems by comparing across metrics with different assumptions.

(c). Statistics

Snead et al. [40] used measures of genetic differentiation and oceanic connectivity to demonstrate that patterns of gene flow were primarily associated with ocean currents. Novel to this experiment, oceanic connectivity values were used to calculate two measurements of network centrality (closeness and strength). Closeness is the inverse average distance from any node or vertex in the network to the target node, while strength is the sum of all oceanographic connectivity estimates to a given vertex [69,70]. Models were constructed with centrality calculated in two ways: with or without standardization of oceanographic connectivity by source population genetic diversity. Comparing these models enabled us to determine whether source genetic diversity modulates the impact of ocean connectivity on sink population genetic diversity (for details see electronic supplementary material, Network Centrality Measures). All variables were scaled and centred prior to variable reduction and modelling. The number of variables was reduced using a Variance Inflation Factor (VIF) threshold of 5 before being further reduced to retain a metric of area (total area), fragmentation (number of patches) and the two centrality measures (closeness and strength). The VIF variable reduction was an iterative process where the variable with the highest VIF was removed before the VIF for all variables was recalculated until no variables had a VIF greater than 5. In fact, no variables had a VIF greater than 2.1, with the maximum absolute correlation coefficient being between closeness and total area (−0.53), while the minimum absolute correlation coefficient was between strength and total area (0.013).0.53.

Linear models were run separately using genetic diversity metrics as response variables and every combination of landscape metrics, centrality measures and all two-way interactions between landscape metrics and centrality measures using the package MuMIn [71]. Models were run once with centrality measures standardized by source population genetic diversity and once without. To meet normality assumptions Z, G and H _exp were raised to the 2^nd, 3^rd and 3^rd power, respectively, while Ar and H _obs were left untransformed. Models were compared via [ 72].

3. Results

(a). Genetic diversity

The rarefied number of multilocus genotypes (eMLG), Zahl’s estimator (Z), Stoddart and Taylor’s index (G), expected heterozygosity (H _exp), observed heterozygosity (H _obs) and allelic richness (Ar) varied considerably across populations. These metrics are segregated largely on a regional basis, with few exceptions. Populations in North Florida (CC, IR, NS, SL, TB) and the Bahamas (EI, LB, SS) had fewer eMLGs, lower G and lower genetic diversity (H _exp, H _obs, Ar) than populations in South Florida (EG, FL, LK, UK) and Central America (NC, LC, TA, TC, UH). Notable exceptions included that Honduran populations (UH) were less diverse than Belizean populations (NC, LC, TA, TC), low diversity in the southeastern-most population in peninsular Florida (FL) was more like Bahamas and Northern Florida populations than the other south Florida populations—Keys (LK, UK) and Everglades (EG), and two of the most genetically diverse populations, one from south Florida (LK) and another from Belize (LC), showed fewer eMLGs relative to other populations from the same regions (table 1, electronic supplementary material, table S2).

Table 1.

The rarefied number of multilocus genotypes (eMLG), Stoddart and Taylor’s index (G), Simpson’s index (λ), expected heterozygosity (H _exp), observed heterozygosity (H _obs), mean allelic richness (Ar), total habitat area (A), number of patches (NP), network closeness (C) and network strength (S) for each population along with the population abbreviation.

population	N	eMLG	Z	G	H exp	H obs	Ar	A (m2)	NP	C	S
Charlotte County (CC)	17	8.56	0.42	7.81	0.25	0.020	1.92	5907.72	677	0.10	0.03
Everglades (EG)	28	11.65	0.97	24.50	0.45	0.002	3.86	20 605.72	142	0.05	6.28e−7
Exuma Island (EI)	12	9.00	0.21	8.00	0.14	0	1.37	806.16	289	0.08	0.05
Fort Lauderdale (FL)	13	12.00	0.78	13.00	0.44	0	2.79	645.06	515	0.10	6.01e−3
Indian River (IR)	14	2.70	0.14	1.34	0.07	0	1.45	5156.91	905	0.08	2.12e−3
Lower Bogue (LB)	14	5.43	0.60	3.50	0.36	0.020	2.22	7278.16	474	0.04	1.44e−7
Long Caye (LC)	272	10.95	1.25	38.69	0.60	0.160	4.59	244.26	14	0.15	0.01
Lower Keys (LK)	143	11.31	1.19	53.96	0.54	0.010	4.59	10 959.34	711	0.14	0.21
Northern Caye (NC)	67	12.00	1.12	65.06	0.55	0.200	4.10	272.70	8	0.14	8e−3
New Smyrna (NS)	92	9.85	0.34	20.35	0.18	0.001	1.93	112 717.53	1413	0.06	9.42e−5
Saint Lucie (SL)	29	11.51	0.47	24.03	0.29	0.030	2.00	4061.90	704	0.08	0.01
San Salvador (SS)	81	9.27	0.77	14.10	0.39	0.010	3.05	5906.90	265	0.06	5.2e3
Turneffe Atoll (TA)	30	12.00	1.29	30.00	0.59	0.280	4.82	3026.49	51	0.18	0.03
Tampa Bay (TB)	130	5.49	0.45	4.12	0.28	0.001	1.89	3682.99	1069	0.09	7.26e−3
Twin Cayes (TC)	59	12.00	1.61	59.00	0.69	0.520	6.13	287.14	40	0.21	0.04
Utila, Honduras (UH)	20	11.65	0.99	18.18	0.50	0.004	3.66	3435.69	154	0.01	1.43e−4
Upper Keys (UK)	99	11.92	1.28	88.30	0.56	0.050	5.15	3733.57	793	0.14	0.35

(b). Habitat metrics

Total suitable habitat area ranged from 244.26 to 20 605.72 m² and number of patches from 8 to 1413 across rivulus populations. The general trend was for Belizean populations (LC, NC, TA, TC) to have much less area but more contiguous area than most populations from Florida and the Bahamas. Exceptions included the southeastern-most population on the Florida peninsula (FL) and one Bahamas population (EI) having low area and the Everglades (EG) population having fewer patches compared with other non-Central American populations (table 1, electronic supplementary material, table S2).

(c). Network variables

Network closeness ranged from 0.04 to 0.21, and network strength from 1.40 × 10⁻⁷ to 0.35. The Florida Keys (LK, UK) and larger islands off the coast of Belize (LC, TA, TC) had the highest closeness values and showed some of the highest values for strength as well. Two populations with the highest area—Everglades (EG) in south Florida and New Smyrna (NS) in north Florida—had relatively low centrality. Populations on the southern fringe of island systems in Central America (UH) and the northern fringe of island systems in the Bahamas (LB) had some of the lowest measures of centrality. The Exuma Island (EI) population was the only one to show considerable disagreement in the two measures of centrality, closeness and strength; this population showed moderate-to-low closeness but high strength, indicating that the population receives a large number of immigrants from a few adjacent populations but is not well connected to the entire metapopulation (table 1, electronic supplementary material, table S2).

(d). Statistical models

The model rankings, coefficient estimates and R ² were similar between models that used centrality measures calculated with or without standardizing oceanic connectivity by the genetic diversity of the source population (table 2, electronic supplementary material, table S3); therefore, models without centrality standardized are reported and discussed. Regardless of the diversity measure used as the response variable (Z, G, H _exp, H _obs, Ar), closeness was always included within the best model and was significant (p < 0.05). In the set of models within two AICc units of the best model, habitat area was included in at least one of the best fit models for H _exp and Ar. The number of patches was included in the set of best fit models for all but diversity metrics. Strength was included only in the Ar set and was not significant. Habitat area, number of patches and closeness were all significant (0.05 < p < 0.1) for at least one model in each set except for G (H _exp, H _obs, Ar), with closeness being the only significant predictor of G (table 2).

Table 2.

A table with the formula, coefficient values, standard errors, significance, AICc, AICc weight (AICc_w) and adjusted R-squared (R ²) for all models using unstandardized centrality measures within two AICc units of the best model for all the diversity metrics (Stoddart and Taylor’s Index = G, expected heterozygosity = H _exp, observed heterozygosity = H _obs, allelic richness = Ar). Covariates are symbolized by their abbreviations (total area = A, number of patches = NP, closeness = C, strength = S), with interactions between variables indicated with an x between the two covariates; the intercept is reported for all models. Estimates shown in italics are significant at 0.01 ≤ p < 0.05, and those shown in bold are significant at p < 0.01. If a cell is blank, this indicates that the covariate was not included in the best fit model(s).

		habitat		centrality
response	intercept	A	NP	C	S	A x C	AICc	AICcw	R 2
Z ~	0.85 ± 0.16 p < 0.0001	0.27 ± 0.19 p = 0.01	−0.24 ± 0.18 p = 0.01	0.67 ± 0.2 p < 0.0001			18.26	0.41	0.84
G ~	1014.5 ± 285.578 p < 0.0001			404.8 ± 0294.4 p = 0.01			268.66	0.37	0.32
G ~	1014.5 ± 279.889 p < 0.0001		−196.2 ± 312.65 p = 0.2	329.2 ± 312.68 p = 0.04			270.06	0.16	0.36
H exp ~	0.1 ± 0.02 p < 0.0001	0.02 ± 0.02 p = 0.04	−0.03 ± 0.02 p = 0.006	0.08 ± 0.02 p < 0.0001			−54.67	0.37	0.85
H exp ~	0.1 ± 0.02 p < 0.0001		−0.03 ± 0.02 p = 0.02	0.07 ± 0.02 p < 0.0001			−53.04	0.17	0.81
H obs ~	0.05 ± 0.03 p = 0.005		−0.05 ± 0.03 p = 0.004	0.06 ± 0.04 p = 0.005		−0.07 ± 0.04 p = 0.001	−39.00	0.49	0.84
Ar ~	3.27 ± 0.35 p < 0.0001	0.68 ± 0.43 p = 0.005	−0.5 ± 0.4 p = 0.02	1.31 ± 0.45 p < 0.0001			45.61	0.38	0.79
Ar ~	3.27 ± 0.33 p < 0.0001	0.6 ± 0.41 p = 0.008	−0.63 ± 0.4 p = 0.005	1.09 ± 0.49 p = 0.0004	0.34 ± 0.41 p = 0.09		46.34	0.26	0.82

4. Discussion

The spatial distribution of genetic variation is the product of drift, gene flow, natural selection and mutation [73,74]. Because decreases in habitat area [25,26] and increases in fragmentation often decrease population size [27,28] and because the strength of drift increases as population sizes decline [15,19], habitat area and configuration are frequently prioritized when attempting to maintain genetic diversity [75,76]. However, comparing the relative importance of habitat measures against connectivity is uncommon. In this study, we combined over a thousand genetic samples from across rivulus’ range, ocean current simulations and land classification data within a network framework to test the role of habitat area, fragmentation and connectivity in maintaining genetic variation. While our models show that both habitat configuration and connectivity dictate genetic variation, connectivity was repeatedly identified as the most important determinant with the largest effect size.

Considering that mating system impacts genetic diversity [77], rivulus’ status as a self-fertilizing vertebrate may spark warranted apprehension regarding the applicability of this study to other species, while variation in outcrossing and selfing rates across rivulus populations may raise concern regarding the determinants of genetic diversity. However, mixed mating systems are extremely common in plant studies using the same genetic diversity metrics [67,68]. Research suggests that mixed-mating systems can maintain genetic diversity at similar levels to purely outcrossing populations [78,79]. Within this study, there are examples of populations that primarily self and have low genetic diversity (North Florida) along with populations that primarily self and have high genetic diversity (South Florida). Populations with high genetic diversity and in which self-fertilization is the predominant mode of reproduction [38,39,41] also have high centrality to the metapopulation (table 1, electronic supplementary material, table S2). Studies suggest that the genetic diversity metrics applied within this study and the comparison across populations with different outcrossing rates are robust and can be applied to other systems. However, mating systems should still be considered when designing management plans and interpreting patterns of genetic variation because mating systems have large impacts on genetic diversity.

Habitat area and fragmentation are often significantly associated with decreased genetic diversity, a finding that has inspired many management decisions [80]. While we found evidence for habitat area or fragmentation impacting the distribution of genetic variation for rivulus (table 2), these variables were not always within the best model nor did they have the largest effect size. When habitat area and fragmentation were included within the model, habitat area was positively associated with genetic diversity, while fragmentation was negatively associated with genetic diversity, supporting previous studies in plants and mammals [29,30]. When testing our genotypic measure of diversity (G), neither habitat area nor fragmentation were important determinants. Hence, we find support for habitat configuration dictating genetic diversity but not genotypic diversity (H1).

Drift and gene flow are regularly described as antagonistic, with drift decreasing and gene flow increasing population-level genetic diversity [10,29]. We find that closeness (i.e. the number and magnitude of incoming connections) was a significant predictor for all measures of genetic diversity (i.e. Z, G, H _exp, H _obs, Ar) (H1; table 2). We ran the analysis with and without scaling measures of connectivity (used to calculate closeness and strength) by the source populations’ genetic pool (i.e. rarefied number of multilocus genotypes). Given that the results of the two analyses were similar (table 2; electronic supplementary material, table S3), genetic diversity may be more impacted by centrality to the metapopulation than by the specific genetic source pools of immigrants. While there has been recent interest in preserving populations with high emigration that harbour genetic diversity to facilitate natural genetic rescue [81], our results indicate that, for rivulus, genetic diversity is linked more tightly with metapopulation structure than the level of genetic diversity within connected populations or local habitat configuration.

While this research uses connectivity and measures of habitat configuration as proxies for gene flow and drift, gene flow and drift are complex evolutionary forces that cannot be reduced to any single environmental measure. Patterns of genetic variation are the product of historical changes such as demography [73,74] that may not necessarily be represented in current environmental conditions. Hence, the use of habitat configuration and connectivity as proxies for drift and gene flow, respectively, should not be misconstrued as proposing equivalency because current environmental patterns may not represent past patterns of evolutionary forces. Furthermore, this study does not include all populations of rivulus across the range, meaning that some aspects of connectivity may have been missed. The sampling does represent populations from all major areas across the range (i.e. Caribbean, Central America, South Florida, East Florida and West Florida), which suggests that our estimated patterns of oceanic connectivity are representative even without some of the unsampled populations.

Anthropogenic activities are increasing fragmentation, decreasing habitat area and exposing species to novel environmental stressors [82]. Hence, understanding the determinants of genetic variation, which is essential for the evolvability of populations [23,24], is critical to mitigate population extirpation. Using a network approach, we calculated connectivity with respect to the entire metapopulation and compared inferences with and without standardizing connectivity by source genetic diversity. While previous research emphasized associations between habitat configuration and genetic diversity, we found that patterns of connectivity—the population’s location within the metapopulation network—are more important for genetic variation than the amount of habitat area or fragmentation, suggesting that range-wide connectivity assessments are essential for designing effective management plans that not only protect populations in the present but preserve the evolvability of populations under future environmental change.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

While the raw biophysical modelling data files were too large to provide, the R data file and an additional copy of the code for the manuscript are available at in a Figshare repository [83]. All code and microsatellite data for the manuscript are provided at [84].

Supplementary material is available online [85].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

A.A.S.: conceptualization, data curation, formal analysis, investigation, methodology, software, visualization, writing—original draft, writing—review and editing; A.T.: investigation, methodology, resources, writing—review and editing; D.S.T.: investigation, resources, writing—review and editing; K.M.: investigation, resources, writing—review and editing; R.L.E.: investigation, resources, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

Sampling and genotyping were funded by an EPSCor GRSP and an FSBI research grant to R.L.E., as well as a Howard Hughes Medical Institute Undergraduate Science Education grant. A.A.S. is supported by the NSF Postdoctoral Research Fellowships in Biology Program under Grant No. 2305939.