Cave beetles of the eastern USA are one of many poorly studied groups of insects and nearly all previous work delimiting species is based solely on morphology. This study assesses genetic diversity in the monotypic cave carabid beetle genus Darlingtonea Valentine 1952, to test the relationship between putative geographical barriers to subterranean dispersal and the boundaries of genetically distinct groups. Approximately 400bp of the mitochondrial cytochrome oxidase I (COI) gene was sequenced from up to four individuals from each of 27 populations, sampled from caves along the escarpments of the Mississippian and Cumberland plateaus in eastern Kentucky, USA. The 81 individuals sequenced yielded 28 unique haplotypes. Hierarchical analyses of molecular variance (AMOVA) within and among geographically defined groups tested two a priori hypotheses of structure based on major and minor river drainages, as well as genetic distance clusters defined a posteriori from an unrooted analysis. High genetic differentiation (F_ST) between populations was found across analyses. The influence of isolation by distance could potentially account for much but not all of the variation found among geographically defined groups at both levels. High variability among the three northernmost genetic clusters (F_CT), low variability among populations within clusters (F_SC), and low within-cluster Mantel correlations indicate the importance of unidentified likely intra-karst barriers to gene flow separating closely grouped cave populations. Overall phylogeographic patterns are consistent with previous evidence of population isolation among cave systems in the region, revealing geographically structured cryptic diversity in Darlingtonea over its distribution. The landscape features considered a priori in this study were not predictive of the genetic breaks among the three northern clusters, which are genetically distinct despite their close geographic proximity.

mitochondrial DNA, Mississippian Plateau, Pennyroyal, phylogeography, Southern Appalachians, troglobites, troglobionts

Variation within a species is usually not random, but structured in some way and typically forms a metapopulation with various levels of deviation from panmixia (Hanski 1999). Landscape features that correlate with intraspecific variation may represent boundaries reducing gene flow among discrete groups of populations. Alternatively, differences between populations of a species may increase linearly with physical distance, especially for less vagile organisms (e.g., Lee and Mitchell-Olds 2011, Goudarzi et al. 2019). The limestone karst regions of the Eastern United States support a remarkable diversity of cave-specialized animals (Barr 1985, Peck 1998, Hobbs 2012, and see White et al. 2019). Troglobionts, i.e., obligate and permanent cave inhabitants, can be predicted to demonstrate high levels of population genetic structure owing to a lack of gene flow between caves. Even long-term population isolation, however, may not yield diagnosable morphological differentiation due to phenotypic convergence in similar cave environments (Wiens et al. 2003, Derkarabetian et al. 2010, Hedin and Thomas 2010). Therefore, many troglobiotic taxa may harbor cryptic variation (Niemiller et al. 2012), and the biodiversity of cave-dwelling organisms may currently be underestimated.

Patterns of gene flow among caves in karst areas vary mostly in accordance with the geographical distribution of subterranean limestone (e.g., Caccone 1985, Katz et al. 2018). In limestone-rich parts of the Eastern United States (Fig. 1) where karst exposure is patchy, structurally fragmented, and discontinuous, caves are generally smaller and more isolated from one another (e.g., Currens 2002, Christman et al. 2005). One such region is the Appalachian Valley (AV), located primarily in eastern Tennessee and Virginia, which supports a high diversity of endemic cave beetles and other troglobites per unit area, many of which are limited in range to one or a few caves (Barr 1967, 1981, 1985, Christman et al. 2005, Niemiller and Zigler 2013). Conversely, troglobiotic invertebrates that inhabit large and highly interconnected cave systems which have permeated the large and uninterrupted exposures of limestone in the Mississippian Plateau (MP) region have comparatively broader ranges and less predictable distributional boundaries (e.g., Barr 1979). Species numbers and abundances differ among cave communities in the interior low (“Mississippian”) plateau (referred to below as “MP”) and Appalachians (Appalachian valley and ridge, referred to below as “AV”) regions; MP cave systems support larger and richer communities of troglobionts compared with those in the AV to the east (Barr and Holsinger 1985). With fewer endemics per unit area, cave species in the MP have been suggested as more likely to occur in sympatry than those inhabiting AV caves (Barr 1967, 1985, 2004). More recently though, Christman et al. (2016) presented contrasting evidence that despite the greater dissection of karst in the AV, cave species actually have lower rates of endemicity in the AV than in the MP.

The cave-rich limestone of the MP is bisected by the Cumberland Saddle, a low point in the Cincinnati Arch formation, which separates the MP into two regions: the MP-I to the west and the MP-II to the east (Fig. 1). Within both bands of the MP, cave interconnectivity has helped establish and maintain diversity by facilitating subterranean dispersal, leading to extensive range overlap and sympatry of species that were previously isolated, and linking populations together through gene flow which likely has reduced stochastic extinction events (Barr 1985, Barr and Holsinger 1985).

Isolating barriers between cave systems restrict gene flow and promote divergence among populations of cave organisms, effectively dividing parts of cave systems into subterranean islands (Culver 1970). Major waterways like the Cumberland and Ohio rivers serve as important fluvial barriers to dispersal of terrestrial troglobionts (Barr and Holsinger 1985, Barr 1985) and even some stygobionts (Niemiller et al. 2013), but smaller streams and rivers may actually promote their dispersal; Barr (1985) compared the “meander frequencies” of rivers dividing the distributions of cave beetle species, finding support for his hypothesis that the more turns a river takes over a given distance, the more often beetles washed out of caves will survive to encounter limestone outcrop karst refugia leading to an increase in distribution range via colonization of new cave systems.

Figure 1. 

(Adapted from Barr 1985, Figure 3) Map showing the major geologic features important for cave development in the southeastern United States: MP-I and MP-II (green) are western and eastern bands of the Mississippian Plateau. Dots indicate collecting records (see Figure 3).

Study species

Darlingtonea Valentine, 1952 is a monotypic genus of cave carabid beetle found in a narrow distributional band from north-central Tennessee (known from a single cave near the Kentucky border) extending northeastward into east-central Kentucky (mainly the northern part of “MP-II” in Fig. 1 and see Fig. 3). Like many of the other cave-specialized carabids of the subtribe Trechina, Darlingtonea are true troglobionts, with adaptations for subterranean life: they lack eyes and wings, possess enlarged mouthparts, lengthened appendages, and specialized sensory setae, and are depigmented compared with their epigean relatives (Fig. 2). Darlingtonea kentuckensis Valentine is usually abundant in caves within its range compared to many species of closely related Pseudanophthalmus (Valentine 1952). Molecular phylogenetic evidence from a 2012 study including representatives of all five eastern North American cave genera shows the genus shares common ancestry with a lineage of Pseudanophthalmus and is essentially derived from within the latter (Philips and Valkanas, unpublished). The close relationship of those genera together with Ameroduvalius Valentine, Nelsonites Valentine, and Neaphaenops Jeannel within the Trechoblemus series and within the Trechina is also strongly supported by Maddison et al. (2019).

Figure 2. 

Gravid female Darlingtonea kentuckensis photographed in Fletcher Spring Cave, Rockcastle County, Kentucky. Photo courtesy of Dr. Matthew Niemiller, University of Alabama, Huntsville.

Figure 3. 

Cave localities of currently known sites for Darlingtonea kentuckensis. White dots were the caves sampled for this study while black dots represent caves unsampled.

Regarding the origin and diversity of North American cave trechines, most authors have favored some version of a “Pleistocene-effect” model (Holsinger 1988). In contrast, Faille et al. (2015) puts the divergence times between two European trechine Aphaenops cave species around 9 my (with a credibility range of 4–17 my). Regardless of age, the proposed evolutionary scenario can be summarized as follows: As climate cycles associated with glacial advance and recession led to fluctuation of surface conditions, ancestral trechines followed cool, moist microhabitats from the deep soil which was abundant during glacial maxima to subterranean or montane refugia during warmer, drier glacial minima (Barr 1969, 1971, 1973, 1985). Periods of isolation in caves during warm intervals were punctuated by periods of introgression during cool intervals until a warm, stable post-Pleistocene climate restricted surface dispersal and promoted subterranean allopatric speciation (and see Jeannel 1948, 1949 for further details on the effects of glaciation).

Other authors have found isolation and divergence in allopatry to be an unsatisfactory model for cave colonization in other taxa, which may be better viewed as a parapatric ecological transition or ‘adaptive shift’ occurring in the presence of gene flow via diversifying selection (Niemiller et al. 2008). Further, surface characteristics of the Earth, such as latitude, percent karst, and landscape rugosity (Topographic Position Index) may have significant effects on the evolution of a cave-adapted fauna (Christman et al. 2016)

It is currently unclear what factors have led to the evolution of any morphological or genetic diversity within Darlingtonea kentuckensis. Darlingtonea kentuckensis has a broader than average distribution compared to most terrestrial Eastern North American troglobionts based on our review (Philips et al. unpublished). Both Valentine (1952) and Barr (1985) noted some morphological diversity among populations of D. kentuckensis. For example, Valentine noted subtle differences including a slightly more convex body form, slightly wider elytra, and more rounded elytral humeral angles (in populations on either side of the Cumberland River), but concluded there was not enough support for subspecific designation. In contrast, the population from Big Saltpeter Cave in Rockcastle County by the Rockcastle River was thought to be distinct enough to warrant the subspecific name D. k. lexingtoni Valentine. Morphologically, this taxon diagnosis was based on a slightly paler body color, very slightly narrower pronotum, flatter elytral disc, and claimed differences in the male genitalia that included subtle differences in the apex of the median lobe and one lobe of the internal sac (see Valentine 1952, Plate IV).

Barr (1985) speculated that D. kentuckensis includes at least seven subspecies or races isolated by landscape barriers. Kane et al. (1992) sampled ten D. kentuckensis populations from across the MP-II for a study of allozyme diversity. Polymorphism in nine of the eleven electrophoretic markers examined combined with the lack of variation within populations and high F_ST across loci suggested long-term isolation.

The exceptional species diversity in North American cave trechines (Peck 1998) makes this lineage valuable to understanding the speciation processes in troglobitic insects and other terrestrial cave organisms. Since populations of Darlingtonea occur across a broad geographic range relative to other troglobiotic taxa while belonging to a single morphologically, geographically, and genetically distinct lineage, D. kentuckensis is a convenient model for comparing observed patterns of genetic variation against those predicted by a climate-mediated process of cave colonization.

Successful PCR amplification was found to be less reliable for older samples (some as old as five years), despite storage at -80 °C in 95% or stronger ethanol. Despite careful optimization of thermal cycling conditions, agarose gel purification of PCR products was found to considerably improve sequence read quality and was performed for most samples included in the final data set.

The distribution of cave collection sites and proportions of haplotypes from 27 populations are shown in Figs 3 and 4 respectively. Frequencies of COI haplotypes and their proportions in each major faunal region (I) or minor watershed (II) are shown in Fig. 5. A minimum spanning network of COI haplotypes is color coded for each structure hypothesis in Fig. 6A–C). A network of the 81 COI sequences (Fig. 6D) reveals five genetically distinct clusters of structure hypothesis III.

The analysis of molecular variance (AMOVA), from which F-statistics (F_ST, F_CT, and F_SC) were calculated to describe nucleotide sequence diversity at hierarchical levels, within and among groups from each hypothesis of structure are summarized in Table 2. The first two hypotheses were based upon a priori geographical hypotheses: (I) the location of caves sampled relative to two zoogeographic barriers proposed by Barr (1985) to be biologically important in MP-II, and (II) the ten minor watersheds to which sampled caves were classified based on assumptions about hydrology gathered from topographic maps (see Table 1).

AMOVA for the a posteriori structure hypothesis III, based on five distinct genetic clusters from a neighbor-joining network of COI sequences produced the greatest difference between F_CT and F_SC among all three analyses. In other words, when nucleotide diversity is partitioned among hierarchical levels, variance in nucleotide diversity is maximized among groups and minimized within groups. The northernmost 15 sampled populations make up three genetic clusters within an approximately ten-kilometer physical radius of one another. In this arrangement, no haplotypes are shared between the three groups, and the clusters contradict both a priori hypotheses about the locations of important major and minor water barriers to gene flow, especially in the northern part of the MP-II. Mantel tests of group submatrices found population pairwise F_ST to be independent of geographic distance within each cluster. Among all 15 of these populations, only a maximum of 14% of the observed variation can be explained by geographic distance.

Examination of male genitalia generally showed only slight differences among cave localities examined (Fig. 7). The median lobes were of a consistent shape as were the parameres and internal sac morphology with one notable exception. In specimen 14, the paramere expansion is absent and the internal sac appears to have a different shape within the median lobe. This morphology was found only in Hicksey Cave (abbreviated HIC in all Figs) located in the northern part of the distribution. One should note that paramere expansion is more visible in those specimens that have darker cuticle and hence individuals can appear more different than they actually are due to superficial color differences. No support for a distinct genitalic morphology of D. kentuckensis lexingtoni was observed and the three caves sampled with this subspecies (Great Saltpeter, Teamers, and Mullins Spring Caves) are no more distinct than some of the populations from sets of caves or even single caves such as individuals from Pourover Cave.

Figure 4. 

Distribution of cave collection sites and proportions of haplotypes from 27 populations of Darlingtonea kentuckensis in eastern Kentucky, USA. Circle area corresponds to number of individuals sampled per locality. Different colors indicate different haplotypes; similarity in hue qualitatively indicates sequence similarity. KR: Kentucky River; RR: Rockcastle River; CR: Cumberland River; MVF: Mount Vernon Fault; DD = drainage divide between Kentucky and Rockcastle rivers.

Figure 5. 

Frequencies of COI haplotypes and their proportions, color coded for each hypothesis of structure; circle area corresponds to number of individuals assigned to each group. Overlain transparent dots show collecting localities. A Four faunal regions of hypothesis I (fifth region unsampled in this study: see discussion and Barr 1985, Kane et al. 1992) B ten minor watersheds of hypothesis II C five genetic clusters of hypothesis III.

Figure 6. 

A–C Minimum spanning networks of COI haplotypes, color-coded for each hypothesis of structure. A Four faunal regions of hypothesis I B ten watersheds of hypothesis II C five genetic clusters of hypothesis III D A split network of 85 COI sequences revealing the five genetically distinct clusters of hypothesis III.

Figure 7. 

Representative male genitalia from 17 of the sampled caves: 1 Wells Cave; 2 Pine Hill Cave; 3–5 Wind Cave; 6 Richardson’s Cave; 7, 8 Lainhart #1 Cave; 9, 10 and 15, 16 Pourover Cave; 11, 12 John Griffin Cave; 13 Climax Cave; 14 Hicksey Cave; 17, 18 Stab Cave; 19, 20 Piney Grove Cave; 21, 22 Dykes Bridge Cave; 23 Great Saltpeter Cave; 24 Teamers Cave; 25 Mullins Spring Cave; 26 Jesse Cave; 27 Steel Hollow Cave. Note that Wells and Dykes Bridge Caves were not included in the genetic study.

F_ST measures allelic identity within populations, or among-population variation. Across partitioning schemes, F_ST values close to one indicate that individuals within populations are more similar to each other than to individuals in other populations, corroborating the idea that in general, cave populations in this study are isolated from one another. Structure hypotheses I and II were developed based on a priori information about the locations of cave collection sites relative to (I) two hypothesized major geographic barriers to gene flow or (II) ten watersheds of higher-order streams. Results of AMOVA for evaluating structure hypotheses I and II indicated that for both hypotheses, the majority of total variation (56–61%) is accounted for by variation among the groups defined under each hypothesis. These results support both structure hypotheses I and II over a null hypothesis of panmixia. Due to the similarity of results for both structure hypotheses I and II and because they are not mutually exclusive, neither can be concluded to better represent geographic structure of genetic diversity among the populations sampled. Hence both the major rivers and even some of the smaller watersheds may be geographic barriers to gene flow. High estimates of F_SC relative to F_CT (Table 2), as well as shared haplotypes among groups in the northern MP-II indicates that neither hypothesis provides the most optimal scheme for partitioning the observed genetic diversity. The lack of robust support for a partitioning scheme based on small watersheds is not necessarily evidence against the influence of climate cycles on the process of lineage diversification. Many caves do not “belong” to a single watershed, but rather may connect or fall between two or more. This factor, along with the uncertainty surrounding cave connectivity via small passages accessible only by small taxa like these beetles, can make it difficult to truly know the possible connectivity of some caves to one watershed over another. Additionally, it is possible that the shape, size, and the pathway of the watersheds in this area changed throughout the recent Pleistocene and earlier. Hence the separation of populations by hypothesized barriers between caves assigned to different watersheds may have resulted from actual watershed barriers, intra-karst heterogeneity, and or climate cycles at various times that in turn helped drive or prevent cave colonization.

Structure hypothesis III was developed based on the five genetic clusters resulting from a split network. The boundaries for the five population clusters in this hypothesis were determined solely by clustering based on genetic distances among sequences, independently of any a priori geographic information. AMOVA statistics for structure hypothesis III (Table 2) indicate that for each hypothesis of structure, among-group variation accounts for a higher percentage of the total variation than within-group variation. These results support all three hypotheses as better models for structured diversity compared with a null model of panmixia. However, variation among groups (genetic clusters) in hypothesis III accounts for much more of the total variation (82%) than either hypothesis I or II (56% and 61%, respectively). Further, only in structure hypothesis III does diversity among groups (F_CT = 0.82) exceed diversity within groups (F_SC = 0.65). Unlike hypotheses I and II, no haplotypes are shared between the five clusters. Lastly, these five genetic clusters form natural, geographically proximate groupings. Hence the evidence supports hypothesis III as the most representative model for the geographic structure of genetic diversity among sampled populations, and especially for those in the northern MP-II part of the distribution.

If geographic distance is strongly positively correlated with genetic distance, gaps in sampling (rather than specific geographic features acting as barriers to gene flow) could be responsible for at least some of the observed clustering of populations. Results of partial Mantel tests (Table 3) indicate up to 18% of the total observed genetic variation across all 27 populations can be attributed solely to the influence of geographic distance. Across the 15 northern populations (three of the five genetic clusters), IBD could explain up to 14% of the total variation. However, low Mantel correlations for population subsets corresponding to each of these three clusters suggests that the genetic structure observed in this region (Rockcastle, Jackson, and Estill counties) is most likely due to actual barriers to gene flow and not simply isolation by distance.

Barr (1985) recognized that the fragmented geology of Rockcastle County, Kentucky may account for the morphological (and genetic) variability in the region, which is topographically complex and dissected with many rivers and streams. The five clusters (including two completely outside Rockcastle County) could represent distinct lineages important in considering the ecology and evolution of Darlingtonea, but divergence times and particular geographic or geologic features consistent with the apparent locations of most putative isolating barriers have not yet been investigated systematically; only the Mount Vernon fault has been well studied.

The Mount Vernon fault (Fig. 4) runs through a cave-rich area of Rockcastle County, Kentucky. Based on its position in the otherwise relatively less faulted MP-II compared to other karst formations (KGS 2017), it may serve as a stratigraphic barrier isolating one of the three northern clusters (D. kentuckensis lexingtoni populations) from the other two (Fig. 5C red colored pie #1 and Fig. 6C). The relatively cave-poor divide between the Kentucky and Rockcastle river drainages (KGS 2017), hypothesized by Barr (1985) to represent an important stratigraphic barrier, is not supported in this study given that populations of the northernmost genetic cluster fall on both sides of the barrier. The influence of the three-way fluvial barrier proposed by Barr (1985), formed by the confluence of the Cumberland River and its Big South Fork, is not explicitly supported but cannot be ruled out due to lack of breadth and spatial resolution in population sampling. Examination of geographically proximate populations in each sector of this “river triangle” (Barr 1985, see fig.1b in Kane et al. 1992) would help to clarify its role as an isolating barrier. Though our study did not explicitly test the effect of meander frequency (Barr 1985) on terrestrial troglobiont dispersal potential, the distributional patterns observed (Fig. 4) do not conflict with the hypothesis that smaller, meandering waterways are less likely or even unlikely to act as dispersal barriers compared to large rivers.

The sampling scheme of our study makes it difficult to extricate signal due to population structure from that due to IBD for the two genetic clusters on either side of the Cumberland River, which are strongly clustered spatially (Fig. 4). An ideal scheme would evenly sample many population pairs on either side of and at increasing distances from each proposed barrier. Under this sampling regime, results of partial Mantel tests within and among groups separated by each proposed barrier could be used to detect population structure amid underlying “noise” from IBD. Isolation between groups across fluvial barriers with different calculated meander frequencies could also be formally compared. Such a systematic sampling method would be challenging for this group of organisms however, as caves are unevenly distributed across the landscape and access restrictions further reduce the number of available cave sampling localities.

Overall, the limits of neither major nor minor watersheds alone adequately model the observed distribution of genetic diversity across sampled populations of D. kentuckensis. Geographic distance and landscape features, both stratigraphic and fluvial, appear to have each contributed to this distribution. Determination of the boundaries of cryptic species or subspecies, inference of their pattern of relatedness, and identification of predictive characteristics of isolating barriers will require further sampling of additional populations and more complete and/or additional molecular loci.

Based on CO1 data alone, there is a wide range of divergence values between taxa that can be defined as separate species on their own evolutionary trajectory from other lineages (Hebert et al. 2003). No formal taxonomic changes are proposed herein as a result of this study, as full or nascent species could be represented by all, some, or none of the five genetic clusters discovered among twenty-seven sampled populations of D. kentuckensis, depending upon the species definition favored. Both genetic and some morphological evidence supports the hypothesis that D. kentuckensis consists of isolated populations that could be considered as separate cryptic species or perhaps subspecies. Hebert et al. (2003) gives an average sequence divergence of 11.2% between species of beetles within the same genus, but divergence ranges from below 1% to 16–32% depending upon the paired taxa examined. Genetic divergence between each of the five populations studied herein differ by ~1.3%, a percentage that is within the range of CO1 sequence divergence between species, although it is certainly on the low side. Regardless, populations within the range of the subspecies D. kentuckensis lexingtoni do form a genetically distinct cluster that is especially supported by this study; additionally all three northernmost clusters are geographically proximate but genetically distinct, with little evidence that isolation by distance is an influence on the pattern of genetic structure. The observed strong correlation between pairwise F_ST and geographic distance among the two southern populations may either be an artifact of sampling deficiency that overlooks intermediate haplotypes or a reflection of a real historical sequence of colonization events. Therefore, these results can be viewed as a starting point for continued investigation, using additional molecular markers and denser sampling, of the historical phylogeography and species limits in this group and other related taxa.

This research was supported by funds from TKP and JRJ, in addition to internal funding from Western Kentucky University through a Graduate Student Research Grant and a Research and Creative Activities Program (RCAP) grant. Naomi Rowland provided valuable advice and assistance with molecular work. We thank Dr. Matthew Niemiller for his general guidance and valuable comments on the manuscript, Dr. Karen Ober for her contribution of specimens, data, and advice, and Elise Valkanas for her phylogenetic studies on eastern North American cave beetles (supported by an NSF- REU grant) which inspired and contributed to this project. We thank Den and Sheila Roenfeldt for supporting a travel fellowship which allowed OFB to present this research and gather feedback from colleagues at the International Society for Subterranean Biology’s 2016 conference. Our appreciation to Arnaud Faille and one anonymous reviewer whose insightful comments improved the manuscript. We thank the many landowners who granted permission for us to sample caves on private property, as well as Jim Currens, Julian Lewis, Ben Miller, Jason Polk, Lee Florea, Kurt Helf, The Rockcastle Karst Conservancy, the Green River Grotto, John Andersland, Rob Neidlinger, and the Kentucky Speleological Society.