A multi-method approach for assessing the distribution of a rare, burrowing North American crayfish species

Introduction

Primary burrowing crayfishes are crayfish that spend the majority of their lives in burrows and only leave occasionally to mate and find food (Hobbs Jr, 1942). Burrowing crayfish play an important role as ecosystem engineers by disturbing the soil in which they burrow and by creating habitat for a wide range of other taxa (Pintor & Soluk, 2006; Welch et al., 2008). Crayfish are understudied relative to other aquatic taxa (Reid et al., 2019), and even among crayfishes, primary burrowing crayfish are particularly understudied (Bloomer, DiStefano & Taylor, 2021; Moore, DiStefano & Larson, 2013). About one third of all crayfish species globally are threatened with extinction (Richman et al., 2015), and about 22% of North American crayfish species of conservation concern are primary burrowing species (Taylor et al., 2007). The high rate of endemism and narrow ranges among crayfish species exacerbates their imperilment rate (Dudgeon et al., 2006; Taylor et al., 1996). Additionally, primary burrowing crayfish are known to inhabit a wide diversity of habitats such as open-canopied grasslands (Rhoden, Taylor & Peterman, 2016), pitcher plant bogs (Welch et al., 2008), and upland forested seeps (Loughman, 2010), and can be habitat specialists or generalists (Loughman, Welsh & Simon, 2012). To effectively conserve a species, resource managers must understand its distribution and habitat requirements in order to make informed decisions (Richter et al., 1997; Taylor et al., 2007).

Species distribution modeling (SDM) is one tool for analyzing the habitat variables driving the range of rare and endemic taxa (Fois et al., 2018; Peterman, Crawford & Kuhns, 2013; Rhoden, Peterman & Taylor, 2017). Species distribution models use species locality and environmental data to make correlative models depicting relative habitat suitability for focal organisms (Warren & Seifert, 2011). A common and effective approach for creating SDMs is maximum entropy modeling or MaxEnt (Elith et al., 2011; Phillips, Anderson & Schapire, 2006). MaxEnt is useful for rare and understudied species because it performs well with small sample sizes and is capable of using presence-only data (Elith et al., 2011; Hernandez et al., 2006; Rhoden, Peterman & Taylor, 2017).

While SDMs can be used for analyzing the drivers of a species’ distribution, ground-truthing or validation should be included as an important part of the process (Egly & Larson, 2018; Peterman, Crawford & Kuhns, 2013; Rhoden, Peterman & Taylor, 2017; Stirling et al., 2016). Habitat modeling using traditional field sampling can also incorporate variables influencing a species’ distribution that are not easily accounted for in large-scale spatial data (Hirzel et al., 2006). This is particularly true for sessile organisms or organisms with a small home range, which benefit from finer grain analyses (Guisan et al., 2007). Species distribution models can poorly account for the influence of biotic interactions, lack fine-scale habitat information, or be limited by biased or unreliable historical data (Fourcade et al., 2014; Guisan & Thuiller, 2005; Peterman, Crawford & Kuhns, 2013). These potential limitations of SDMs illustrate the need for incorporating ground-truthing and utilizing traditional habitat sampling when studying a species’ habitat needs and distribution.

Another emerging method of detecting rare, endemic taxa is employing molecular approaches collectively termed “environmental DNA” (eDNA). Environmental DNA refers to the capture and identification of DNA from an environmental sample (air, soil, or water), generally without the collection or identification of entire organisms (Rees et al., 2014). Environmental DNA may detect some cryptic or low-abundance organisms better than conventional sampling methods (Barnes & Turner, 2016). As a recent example, Rice, Larson & Taylor (2018) found eDNA to successfully detect a rare, large-river crayfish in the Missouri Ozarks. However, eDNA has not yet been used to detect primary burrowing crayfish, which are often more difficult to collect than their stream dwelling counterparts (Taylor, Engelbert & DiStefano, 2015). Environmental DNA offers a unique opportunity to detect primary burrowing crayfish, while avoiding the high amounts of effort and low detection rates associated with traditional methods, such as hand excavating burrows.

The Boston Mountains Crayfish (Cambarus causeyi Reimer, 1966) is an understudied and rarely observed burrowing crayfish endemic to the Boston Mountains of Arkansas, United States of America (USA). Cambarus causeyi spends the majority of its life in generally shallow, yet complex, burrows that are often found underneath large rocks (Robison, Crandall & Wagner, 2009; Robison & Leeds, 1996). Occasionally C. causeyi occurs in “small springs and tiny creeks under rocks” when reproducing, and was noted to be associated with intermittent mountain seeps and streams (Robison, Crandall & Wagner, 2009; Robison & Leeds, 1996). Between 1941 and 1986, only nine C. causeyi individuals were captured over six surveys. These specimens were collected from four localities in Pope County and one locality in Stone County (Robison & Leeds, 1996). Subsequently, Robison & Leeds (1996) reported the capture of 87 individuals from 39 separate localities (erroneously reported as 40). Robison & Leeds (1996) expanded the known range of C. causeyi to include Madison, Johnson, Franklin, Newton, and Searcy counties. However, they reported that 67 of the individuals were collected from Johnson County. Robison, Crandall & Wagner (2009) conducted a follow-up study that located C. causeyi at four new localities, but otherwise could not recover the species from the vast majority of sites sampled by Robison & Leeds (1996). The extreme decrease in the number of individuals caught in the 2009 study caused C. causeyi to be updated to the status of Vulnerable on the 2009 IUCN assessment (Robison, Crandall & Wagner, 2009). Only two additional records of C. causeyi have been made by Arkansas Game and Fish Commission (AGFC) since 2009; they are from Newton and Van Buren counties in 2017 and 2019, respectively (B Wagner, pers. comm., 2019).

We applied SDMs, fine-scale modeling, and eDNA to improve understanding of the distribution and habitat associations for C. causeyi. Combined, these methods were intended to help us determine if C. causeyi declined across its range as suggested in Robison, Crandall & Wagner (2009) and identify habitat needed to protect the species. We also evaluated whether SDM approaches like MaxEnt are an effective means of predicting occurrences of rare, difficult to detect burrowing crayfish. Additionally, we tested the efficacy of utilizing eDNA as a surveillance tool for C. causeyi and other primary burrowing crayfish species. The results of this study will inform managers looking to monitor C. causeyi and primary burrowing crayfish as a whole.

Materials & Methods

Results

Species distribution modeling

Our MaxEnt model’s test AUC was 0.911, indicating good fit, with a standard deviation of 0.0342 (Fig. 1). Average annual precipitation was the best predictor of C. causeyi presence estimated from a jacknife analysis (Fig. 2). All other variables performed poorly. Percent contribution and permutation importance mirrored results of the jackknife analysis. Precipitation had a contribution of 65.0%, Euclidean distance from a stream had a contribution of 13.4%, elevation had a contribution of 12.0%, and all remaining variables had a contribution under 5%. We found a unimodal relationship between average annual precipitation and C. causeyi presence, which peaked at approximately 140 cm/yr (Fig. 3). Cambarus causeyi habitat suitability declined below this value, plateauing around 150 cm/yr.

Fine-scale habitat modeling

We detected C. causeyi at nine of 51 sites during our habitat sampling (Fig. 4; Table S4). We failed to find C. causeyi at three sites where the species had been found during our March 2019 preliminary survey, and these preliminary detections were omitted from models (Table S3). We found male and female pairs sharing burrows during the spring preliminary sampling, and found one ovigerous female and many young instars during the summer habitat sampling. No models were particularly well-supported, and our most-supported model (proportion of sand and presence of P. liberorum) was only marginally better than the null model (Table 2). Both the proportion of sand and the presence of P. liberorum had negative associations with C. causeyi relative abundance (Fig. 5). Proportion of sand and presence of P. liberorum were also the only single variable models to perform better than the null model (Table 2). Interpolated MaxEnt habitat suitability generally performed poorly in predicting C. causeyi relative abundance from conventional sampling. Surprisingly, MaxEnt habitat suitability had a negative effect on C. causeyi relative abundance in all models including this variable (Table 2). Similarly, AUC for the field validation of our MaxEnt model was only 0.548, indicating that this model only performed marginally better than random.

Table 2:

Candidate models for our fine-scale habitat analysis.

Candidate models relating Cambarus causeyi relative abundance to predictor variables (Sand, P. liberorum, Large Rock, Interpolated Maxent, Stream Dwelling Crayfish, Canopy Cover) with the number of parameters (K), delta corrected Akaike Information Criteria (AICc), and log likelihood (LL) values representing model performance. Estimates of parameter coefficients with standard errors in parentheses are given for each predictor variable when it is included in a model, whereas predictor variables omitted from a model are represented by a dash (-).

Sand	P. liberorum	Large Rock	Interpolated Maxent	Stream Dwelling Crayfish	Canopy Cover	K	ΔAICc	LL
−7.331 (5.001)	−20.401 (16810.675)	–	–	–	–	4	0.00	−27.107
–	−20.860 (19040.000)	−2.196 (1.189)	–	–	–	3	0.89	−28.729
−6.290 (3.634)	–	−2.084 (1.285)	–	–	–	4	0.95	−27.582
−7.252 (4.977)	–	–	–	–	–	4	1.02	−27.616
−7.832 (5.224)	−20.904 (19922.950)	–	−0.827 (1.430)	–	–	3	1.71	−29.144
–	−20.443 (19897.237)	–	–	–	–	5	2.12	−26.938
–	–	–	–	–	–	2	2.40	−30.618
−7.728 (5.260)	–	–	−0.568 (1.468)	–	–	3	2.83	−29.701
–	–	−1.631 (1.174)	–	–	–	4	3.09	−28.653
–	−20.457 (19168.973)	–	−0.551(1.357)	–	–	4	3.91	−29.060
–	–	–	–	−0.271 (0.916)	–	3	4.57	−30.574
–	–	–	−0.244 (1.373)	–	–	3	4.63	−30.602
–	–	–	–	–	−0.128 (1.158)	3	4.65	−30.612

DOI: 10.7717/peerj.14748/table-2

Discussion

We assessed the distribution and habitat preferences of a rare, understudied crayfish species using SDMs, fine-scale modeling, and eDNA. Our MaxEnt model for this species appeared to perform well with a high test AUC and identified average annual precipitation as a factor in defining the distribution of C. causeyi. However, ground-truthing revealed poor performance of the MaxEnt model, as predicted habitat suitability was not included in any of our subsequent models explaining C. causeyi relative abundance. Instead, we found site-specific measures of soil texture and presence of another burrowing crayfish (P. liberorum) best explained C. causeyi relative abundance, albeit in relatively poor-performing GLM models. Large-scale analyses like MaxEnt and other SDMs often lack fine-resolution data and biotic interactions that can affect distributions (Anderson, Peterson & Gómez-Laverde, 2002; Guisan & Thuiller, 2005; Peterman, Crawford & Kuhns, 2013). Lastly, our eDNA assay detected C. causeyi from the field at a greater rate than, and in moderate agreement with, conventional sampling. Although SDMs and fine-scale habitat modeling from conventional surveys had moderate or weak performance explaining C. causeyi habitat associations, eDNA appears to be a promising approach to sampling for this and similar burrowing crayfish in the future.

We encountered several limitations during the course of our study. Our most supported generalized linear model from conventional field sampling performed better than the null model but had a mediocre fit (Fig. 5). We initially designed our field sampling data to be analyzed with occupancy models using replicated quadrats or timed searches for estimating detection probabilities (Durso, Willson & Winne, 2011; MacKenzie et al., 2003). However, we did not detect C. causeyi at enough sites for our models to converge. Instead, we employed zero-inflated Poisson generalized linear models on relative abundance of C. causeyi, but this approach was still limited by our infrequent crayfish detections. We had planned to improve detection of C. causeyi by sampling intensively in spring of 2020, but COVID-19 travel restrictions reduced our sampling in this season to only one week of eDNA surveys and no conventional surveys. We suggest that future sampling for C. causeyi be conducted in the spring when this species is most actively exhibiting mating behaviors (Robison, Crandall & Wagner, 2009), supported by our finding of male and female crayfish sharing burrows in March 2019 preliminary sampling but not in subsequent summer sampling. Rhoden, Peterman & Taylor (2017) had better success than our study in ground-truthing SDM models for burrowing crayfish species in Arkansas, but had sampled for these species in spring rather than summer. We expect this seasonal sampling recommendation also applies to eDNA sampling for C. causeyi and other burrowing crayfishes, as timing of eDNA sampling affects detection probabilities relative to life history and organism activity (e.g., de Souza et al., 2016; Troth et al., 2021). For example, Dunn et al. (2017) found that ovigerous crayfish produce more eDNA than other individuals in a laboratory experiment. We suspect that C. causeyi would be most likely to be ovigerous in the middle to late spring, as we found many juveniles during our summer sampling but only one female with eggs.

Our MaxEnt model had a high testing AUC when validated on historic C. causeyi occurrences, but did not predict detections of this species well during ground-truthing in 2019 and 2020. MaxEnt identified average annual precipitation as the best predictor of C. causeyi presence, and this climate factor may limit the distribution of C. causeyi to wetter locations at a broad scale. However, our MaxEnt model seemingly missed factors affecting C. causeyi presence and relative abundance at the site or local scale. For example, in our fine-scale modeling, proportion of sand in soil was negatively associated with C. causeyi relative abundance, but water storage capacity of soil was not identified as an important variable in our MaxEnt model. Given these disparate results, we subsequently compared water storage capacity of soil from gSSURGO to our field-observed sand proportion at sites. Sandier soils should have low water holding capacity, but we found a very weak, noisy relationship between proportion of sand in soil and estimated water holding capacity from gSSURGO (Fig. S2). Soil conditions are important to burrowing crayfish (Dorn & Volin, 2009; Grow, 1982), but our MaxEnt model seemingly lacked high resolution or accurate soil data for our study sites. Improved spatial data on soil attributes (i.e., Vergopolan et al., 2022) may lead to better SDM models for burrowing crayfish in the future.

Biotic interactions also appear to affect C. causeyi presence and relative abundance. Cambarus causeyi was never found at the same site as the primary burrowing crayfish P. liberorum (Fig. 5), which may be excluding C. causeyi by competition. Egly & Larson (2018) found poor ground-truthing performance of an SDM for crayfish to be partially explained by biotic interactions, in which native species were absent from suitable habitat occupied by invasive species. Biotic interactions can be incorporated into SDMs (Anderson, Peterson & Gómez-Laverde, 2002; Guisan & Thuiller, 2005; Peterman, Crawford & Kuhns, 2013), but we lacked a priori data to know that P. liberorum may be an effective competitor excluding C. causeyi from otherwise suitable sites. Our study emphasizes the importance of having familiarity with a study species’ biology and ecological interactions when selecting variables for SDMs.

We designed and validated a robust qPCR assay for C. causeyi that does not amplify the DNA of non-target crayfishes at realistic, field concentrations. Five of the six eDNA detections for C. causeyi were at locations where the species was detected by conventional sampling, whereas only one eDNA detection was at a location where C. causeyi was not directly collected in our study. In this one instance, eDNA might be interpreted as out-performing conventional sampling for C. causeyi, particularly as we documented no evidence of field or laboratory contamination in our study commonly associated with false positives for eDNA studies. However, eDNA may have been advantaged relative to conventional sampling because eDNA samples were taken in the spring during the presumed active, reproductive season of C. causeyi (Robison & Leeds, 1996). Future comparisons of eDNA and conventional sampling for burrowing crayfish would ideally be done during the same season and year. Alternatively, we failed to detect C. causeyi eDNA from six sites where the species was detected by conventional sampling in either 2019 or 2020. These apparent non-detections or false negatives could be the consequence of mismatches between years of sampling eDNA and conventional surveys, especially because only one of the potential false negatives was conventionally sampled in 2020 (Table S3). For example, C. causeyi may have experienced population declines at some sites between conventional sampling in 2019 and eDNA sampling in 2020 (Fournier, White & Heard, 2019). Our C. causeyi eDNA samples had low copy numbers, near the LOD, when the species was detected, which may have been caused by low local abundance or biomass at most sites. Therefore, the performance of our eDNA assay for C. causeyi might also be improved by increasing either the volume of water filtered or number of field replicates taken (Dougherty et al., 2016; Sepulveda et al., 2019). Additionally, future burrowing crayfish monitoring could be done by testing burrow water or soil for eDNA (Baker et al., 2020), although sampling C. causeyi associated surface water (ephemeral streams) during the reproductive season appears to be a good approach for this particular species. Given the difficulty of detecting C. causeyi by conventional sampling, eDNA appears to be an encouraging approach for monitoring the presence of this species and other primary burrowing crayfish.

Conclusions

Our study was motivated by the possibility that C. causeyi had experienced severe range and population declines between Robison & Leeds (1996) and Robison, Crandall & Wagner (2009), which resulted in elevated IUCN conservation status for this species. Our results suggest that C. causeyi still occurs throughout its historic range (Fig. 4), but the species is difficult to detect. Further, detection probabilities for C. causeyi are likely higher in spring than in summer, and may be higher in wetter than drier years. For example, MaxEnt revealed a positive association between C. causeyi presence and wetter locations as represented by average annual precipitation. Robison, Crandall & Wagner’s (2009) C. causeyi survey was preceded by and coincided with the 2006–2007 extreme drought in Arkansas (National Drought Mitigation Center, U.S. Department of Agriculture & National Oceanic and Atmospheric Administration, 2022). Although C. causeyi is associated with higher elevation ephemeral and intermittent streams, which dry during a typical year, the species may still be negatively affected by severe drought. Robison, Crandall & Wagner (2009) may have detected widespread, drought-associated population declines of C. causeyi that the species has since recovered from. Alternatively, C. causeyi may also have been difficult to detect during Robison, Crandall & Wagner (2009) due to drought conditions. Cambarus causeyi uses shallow burrows, often under large rocks, but may have burrowed deeper to follow the receding water table during the 2006–2007 drought (Novakowski & Gillham, 1988; Robison, Crandall & Wagner, 2009). Future work could investigate seasonal and annual effects of precipitation on both C. causeyi occupancy and detection probability.

Although we did not find a decrease in the range of C. causeyi, there are still reasons this species may be vulnerable to population declines or range contraction. Future work might investigate biotic or competitive interactions with other burrowing crayfishes like P. liberorum, which could displace C. causeyi if they spread due to factors like climate or land use change. Cambarus causeyi could also be directly vulnerable to climate change if severe weather events like droughts become more common in Arkansas over time (Mukherjee, Mishra & Trenberth, 2018), particularly given the apparent preference of this species for locations with high average annual precipitation. Despite our success finding C. causeyi at more sites than Robison, Crandall & Wagner (2009), we suggest the species still deserves its vulnerable conservation status and increased management attention due to its narrow range, low local abundances (Table S3), and the threats described above. Furthermore, eDNA was effective at detecting C. causeyi, and future monitoring for this and other burrowing crayfishes might benefit from incorporating eDNA into sampling programs, although this emerging methodology should continue to be tested against conventional field sampling and distributional predictions from models like SDMs.

Supplemental Information