The application of stable hydrogen isotope (δD) techniques has swiftly advanced our understanding of animal movements, but this progression is dominated by studies of birds and relatively long-distance, north–south migrants. This dominance reflects the challenge of incorporating multiple sources of error into geographic assignments and the nature of spatially explicit δD models, which possess greater latitudinal than longitudinal resolution. However, recent progress in likelihood-based assignments that incorporate multiple sources of isotopic error and Bayesian approaches that include additional sources of information may advance finer-scale understanding of animal movements. We develop a stable-isotope method for determining probable origins of bats within hibernacula and show that this method produces spatially explicit, continuous assignments with regional resolution. We outline how these assignments can be used to infer hibernacula connectivity, an application that could inform spatial modeling of white-nose syndrome. Additionally, estimates of seasonal and annual flight distances for many cave-dwelling bat species can be derived from this approach. We also discuss how this application can be used in general to provide insights into variable migratory and foraging strategies within bat populations.
Movement patterns, especially long-distance dispersal and migration, are of central importance to the ecology and conservation of most species, but their study is frequently difficult (Webster et al. 2002). As a result, the movements of many animals, including cave-dwelling bats, are poorly understood (Popa-Lisseanu and Voight 2009). Cave-dwelling bats primarily escape seasonal fluxes in resource availability through hibernation, but the natural rarity of suitable hibernacula may prompt annual migrations between seasonal roosts (Kurta 1996). Banding studies have documented Myotis species traveling more than 455 km between summer roosts and winter hibernacula (reviewed by Fleming and Eby 2003). The emerging, catastrophic impact of white-nose syndrome in North America (Blehert et al. 2009, Frick et al. 2010, Foley et al. 2011) emphasizes how the conservation of bats depends on better understanding their movements.
White-nose syndrome (WNS), associated with the psychrophilic fungus Geomyces destructans (Blehert et al. 2009), causes the sudden death of hibernating bats. WNS has spread rapidly through eastern North America since its detection in 2006, and bat populations in affected hibernacula frequently experience 75–100% mortality (Frick et al. 2010). WNS threatens all cave-dwelling bat species in North America (Foley et al. 2011). Human activity may translocate G. destructans (Linder et al. 2011), but seasonal bat movements and interactions are thought to be the driving force behind the spread of the disease (Frick et al. 2010, Hallam and McCracken 2010, Foley et al. 2011). Quantifying the connectivity between summer roosts and winter hibernacula is essential to predict the spread of WNS.
Banding initiatives have modestly increased our understanding of bat movements, but such studies mostly ended in North America after bands were linked to significant bat population declines (Ellison 2008). Widespread use of other external markers is limited by size and expense, but recent techniques using endogenous stable-isotope markers have greatly advanced our knowledge of terrestrial animal movements (Rubenstein and Hobson 2004). Stable-isotope approaches to animal tracking rely on the predictable spatial variation of isotope abundances due to biogeochemical processes. In particular, stable hydrogen isotope ratios (δD) of growing-season precipitation (hereafter δDp) show predictable continental variation and are incorporated into consumer tissues through diet and drinking water (Rubenstein and Hobson 2004). For organisms traveling through isotopically distinct food webs, the stable-isotope composition of a tissue can be analyzed and related to the geographic region in which it was grown using spatially explicit δDp models (e.g., Hobson et al. 2009, Van Wilgenburg and Hobson 2011).
Stable hydrogen isotope techniques are already widely used to study animal movement (Rubenstein and Hobson 2004) but have only been applied to bats in a few instances and with mixed success (Cryan et al. 2004, Britzke et al. 2009). Cryan et al. (2004) found strong correlation between the stable hydrogen isotope ratios of bat hair (hereafter δDh) and δDp. Conversely, Britzke et al. (2009) found the correlation between δDh and δDp explained little of the variance observed in three of the four studied species. However, stable-isotope approaches require tissues of known origin, which for bats requires sampling within the molt period (Cryan et al. 2004), a condition not met by Brtizke et al. (2009). Here we describe a method using stable-isotope techniques and a combination of new and synthesized molt-period data to determine the geographic region from which a hibernating bat population originates, i.e., a cave catchment area. We use this method to derive continuous, spatially-explicit probability surfaces for little brown bat populations in three mines in Western Upper Peninsula, Michigan (WUP). To our knowledge, no attempts have been made to apply stable-isotope techniques to determine the origins of hibernating bats. This study provides a framework that can be used to study the spatial dynamics of migratory animals in general and, specifically, provides novel ecological data that can be incorporated into white-nose risk assessment models. Such data help address a major research need in efforts to combat WNS (Frick et al. 2010, Foley et al. 2011).
To develop this application we (1) estimated the seasonal molt period of the little brown bat, Myotis lucifugus, (2) determined the relationship between model-predicted δDp (stable hydrogen isotope ratios of growing-season precipitation) and those of little brown bat hair (δDh), and (3) determined the provenance of little brown bats hibernating in three mines in the Western Upper Peninsula (WUP) of Michigan, USA.
To determine the molt period and quantify the δDh–δDp relationship, we captured little brown bats from 11 June through 23 August 2003 at three U.S. National Park Service sites: Pictured Rocks National Lakeshore in Michigan, Apostle Islands National Lake Shore in Wisconsin, and Grand Portage National Monument in Minnesota. Additionally, we used previously published data from little brown bat hair collected from 15 May to 1 August 2001 and 2005 throughout the eastern United States (Britzke et al. 2009). A total of 78 female and 46 male bats from 34 summer sites were sampled (Appendix: Fig. A1).
To determine the probable summer origins of hibernating bats, hair was collected from 20 hibernating bats (September, October 2003; March 2004) at each of three WUP mines: Quincy Mine (47°8′ N, 88°34′ W), Caledonia Mine (46°45′ N, 89°7′ W), and Norway Mine (45°47′ N, 87°54′ W) (Appendix: Fig. A1). An estimated tens of thousands of little brown bats overwinter in Quincy Mine, as do at least 23 000 bats in Norway Mine, and a quarter of a million bats hibernate in Caledonia Mine (Kurta 1996). Only bats roosting singly or in small clusters (≤5 bats) were sampled. Hair samples (1–2 mg) were taken from the interscapular region on each bat and frozen until analysis. Handling time was minimized and no more than three people entered a mine.
Hair samples were cleaned using a 2:1 chloroform : methanol solution and prepared for analysis at the National Water Research Institute, Saskatoon, Canada. Stable hydrogen isotope measurements were performed on H2 derived from high-temperature flash pyrolysis of hair subsamples (350 μg [to within 10 μg]) and keratin standards using continuous-flow isotope-ratio mass spectrometry (CF-IRMS). Because a portion of the hydrogen in keratin freely exchanges with ambient moisture and affects comparability of results, we report nonexchangeable δD values of bat hair determined using the comparative equilibration method described by Wassenaar and Hobson (2003; ∼13% of the H in M. lucifugus hair is exchangeable with the environment). Results are reported in the delta notation in units of per mil (‰) and normalized on the Vienna Standard Mean Ocean Water-Standard Light Antarctic Precipitation (VSMOW-SLAP) standard. Repeated analysis of the reference material IAEA-CH-7 (−100‰) yielded a repeatability better than ±1.5‰, based on a six-month running average.
We first estimated molt period by plotting the difference between δDh and estimates of collection-site δDp (ΔδDhp) as a function of ordinal collection date (Cryan et al. 2004), and determining for which dates ΔδDhp most closely approached the generalized offset value for keratinous tissues (−25‰; Wassenaar and Hobson 2001). The expected result was that ΔδDhp values would clearly converge on the offset value during mid-summer. However, results using this method (Appendix: Fig. A2) were inconclusive and ΔδDhp values were randomly scattered around a −5‰ offset with no clear temporal trend, possibly because M. lucifugus does not regularly undergo seasonal movements large enough to produce the distinct temporal trend in ΔδDhp observed by Cryan et al. (2004). We then estimated molt period based on the prediction that a single annual molt occurs in late summer, prior to fall migration and after parturition and lactation (Constantine 1957, Cryan et al. 2004). While parturition and weaning dates for M. lucifugus vary among and within maternity colonies (e.g., Humphrey and Cope 1976, Henry et al. 2002), available evidence indicates most females throughout their range give birth by late June in most years (Cagle and Cockrum 1943, Davis and Hitchcock 1965, Henry et al. 2002, Krochmal and Sparks 2007), and weaned juveniles are commonly present by early July (Cagle and Cockrum 1943, Davis and Hitchcock 1956). Because females molt later than males, a molt estimate based on female life history should be a conservative estimate for males (Cryan et al. 2004). For the purposes of this study, we define the molt period as occurring between 1 July and 23 August but note further molt observations are warranted.
Assigning probable geographic origins
We used ordinary least-squares regression to determine the relationship between δDh from adult bats collected within the defined molt period and collection-site growing-season δDp derived from the Bowen and Ravenaugh (2003) model. This relationship was used to convert the GIS-based δDp model (Bowen and Ravenaugh 2003) to a spatially explicit model depicting mean expected δDh values across North America. Despite the strength of the observed δDp–δDh relationship, biological, ecological, and biogeochemical factors, such as inter-individual physiological, behavioral, and dietary differences and annual weather variation, influence the distribution and incorporation of stable hydrogen isotopes and cause deviations from model expectations (Rubenstein and Hobson 2004, Wunder 2009). Additionally, measurement error associated with isotope mass spectrometry is a small but known source of variance (e.g., Wunder and Norris 2008). The δDh value expected from a given location is therefore more accurately characterized as a distribution of potential values (Wunder and Norris 2008). To propagate the sum of these variances into geographic assignments, we used maximum-likelihood estimation to fit a gamma probability distribution to within-site variance δDh estimates obtained from summer sampling locations at which ≥3 bats were captured (Appendix: Table A1), which yielded the parameters shape k = 3.05 and scale θ = 2.45. We simulated 1000 new δDh values for each sampled hibernating bat, where the mean of the simulations was equal to the observed δDh value and the standard deviation was drawn at random from the fitted distribution of δDh bulk variance estimates (Wunder 2009). Kernel density estimates with a Gaussian kernel and “solve-the-equation” bandwidth estimates (Sheather and Jones 1991) were fit to the simulated data set for each mine. For ease of display and interpretation, probabilities were relativized to the maximum value (Hobson et al. 2009, Van Wilgenburg and Hobson 2011). To determine the probability that a given geographic coordinate on the GIS-based δDh model represented the origins of a hibernating bat, we reclassified the model according to the derived probability density function. This yielded a map depicting the probable isotopic origins of bats hibernating within each mine. Maps were then constrained to the known range of the little brown bat (Patterson et al. 2007).
Probable maximum migration distances
We derived the probable maximum migration distance for hibernating bats within each mine from the 95th percentile of the simulated δDh data sets. For each mine we determined the 95th percentile δDh values and then measured the geographic distance from the mine to the nearest pixel on the δDh base map with the same value. Spatial manipulations were done using ArcGIS Spatial Analyst, version 10.0 (ESRI 2010). All statistical analyses were conducted using R, version 2.6.2 (R Development Core Team 2008).
Regression analysis indicated that 63% of the variance in the stable hydrogen isotope ratios of bat hair, δDh, could be explained as a function of modeled δDp, the similar ratios of growing-season precipitation, at the site of sample collection. Examination of regression residuals for constant variance across ordinal collection dates to assess the appropriateness of our molt period estimate resulted in no indication of heteroscedasticity (Fig. 1). Origins based solely on isotopic assignments exhibited unrealistic longitudinal range, i.e., bats almost certainly did not originate from the Atlantic or Pacific coasts (Fig. 2). This result is due to the nature of precipitation δD models, which possess greater latitudinal than longitudinal resolution in North America. Realistic origins are likely within the intersection of regional isotopic assignments and estimates of maximum migration distance (Fig. 3). Hence, the continental extent of assignments (Fig. 2) is an overestimate as derived maximum distances from which bats were entering the mines in Western Upper Peninsula, Michigan, USA, ranged from 300 to 565 km (365 ± 154 km [mean ±SD]). While Quincy and Caledonia mines had similar predicted maximum migration distances of 350 and 300 km, respectively, the furthest migrants in Norway Mine originated from a locality at least 565 km away from the mine (Fig. 3). Post hoc tests for differences in migration-distance distributions between sexes were significant for all mines (Kolmogorov-Smirnov; all pairs P < 0.0001, α = 0.05) and indicated that females are more likely to originate in more distant summer habitats than males.
The probability density surface for hibernating bats in Norway Mine suggested a bimodal distribution of summer origins: the most probable isotopic origin centered on −70‰ δDh, consistent with the area north of the mine, but a second, less probable mode at −15‰ δDh indicated possible origins in the central and southeastern United States (panel inset in Appendix: Fig. A2). When intersected with the maximum migration distance with this mine, the most probable origins for this second, more migratory population lie in southern Michigan, Wisconsin, or northern Illinois (Fig. 3A). The probability densities for little brown bats in both Caledonia and Quincy mines also centered on −70‰ δDh, isotopically consistent with the region containing the mines themselves (Fig. 2B and C, inset). Given the maximum migration distance for these mines, most bats hibernating in Quincy and Caledonia mines likely originated from summer habitats near these hibernacula (Fig. 3B and C). The probability density of Quincy Mine indicated a slightly broader catchment area than the other two mines (Fig. 2C, inset).
Stable hydrogen isotopes offer a means to study the cryptic movement patterns of cave-dwelling bat species. Our new framework incorporates an improved molt estimate, sources of inter-individual isotopic variance, and a precipitation model using stable hydrogen isotope ratios of growing-season precipitation, δDp (Bowen and Ravenaugh 2003) with GIS resources and isotopically derived migration distances to advance our understanding of connectivity among hibernating bat populations. This advancement provides the opportunity to inform the spatial modeling of white-nose syndrome (WNS) disease transmission (Foley et al. 2011) without the unacceptable level of mortality associated with traditional banding techniques (Ellison 2008). Inferring connectivity between seasonal habitats, identifying areas of spatial and temporal interaction among populations, and deriving migration distances are critical prerequisites to understanding the spatial dynamics of WNS spread (Foley et al. 2011).
Although assignments theoretically spanned North America, this result can be improved by eliminating biologically implausible origins (e.g., Van Wilgenburg and Hobson 2011). Little brown bats are not continental migrants and likely originate from the most proximal geographic location corresponding to their isotopic assignment. The best estimate of a catchment area involves the intersection between isotopic assignments and dispersal distances. Adding a spatially explicit representation of the maximum migration distances modeled in this analysis and of the maximum migration distance reported from banding studies (455 km; Humphrey and Cope 1976) to the probability surfaces allows a realistic interpretation of origins on a regional, rather than continental, scale (Fig. 3). Geographic origins of animals determined using this approach should be interpreted cautiously because upper limits of migration distances of cave-dwelling bats are still unknown. Incorporating other stable isotopes (e.g., sulfur, strontium) and additional prior information, such as species abundance and habitat use, may yield more certain geographic assignments (Royle and Rubenstein 2004, Van Wilgenburg and Hobson 2011).
For some life-history questions the proportion of migrating individuals in a population is more important than absolute geographic origins of individuals. For example, migration in cave-dwelling bats is highly plastic, may be sex biased, and migratory and non-migratory phenotypes potentially coexist in the same hibernacula (Fleming and Eby 2003). Multimodal probability densities of isotopic assignments may indicate migratory relative to sedentary phenotypes and our study provides some evidence of partial migration in the little brown bat. The probability density for Norway Mine (Western Upper Peninsula, Michigan, USA) possesses two modes (Fig. 3A), with the mostly likely δDh (stable hydrogen isotope ratios of bat hair) value consistent with an origin 100 km north of the mine, but the δDh values of the second mode are isotopically consistent with summer origins in southern Wisconsin and northern Illinois. A sedentary, local population and a migratory population originating from the naturally hibernacula-depauperate regions of southern Wisconsin or northern Illinois both occupying the Norway Mine could explain this result, but analysis of a larger number of individuals is warranted to support this explanation. Migration is likely a driving factor behind the spread of WNS (Frick et al. 2010), and migratory phenotypes in a hibernaculum may increase the risk of infection.
Conclusions and future work
Incorporating informative priors in a Bayesian approach with quantified, partitioned estimates of stable-isotope variances in a multi-isotope framework will yield more accurate geographic assignments. While informative priors can be developed from existing data and multi-isotope frameworks are already used with mixed successes (e.g., Kelly et al. 2008), the sources of variance in isotopic assignments are still poorly characterized. Aquatic foraging is a source of isotopic variance of particular concern for little brown bats (Britzke et al. 2009) and other organisms utilizing aquatic food webs because such foraging may decouple the relationship between tissue deuterium, δD, and δDp (Coulton et al. 2000). While our study demonstrated strong, positive correlation between δDh and δDp despite presumed aquatic foraging (Anthony and Kunz 1977), a preliminary test demonstrated that the proportion of aquatic habitat within the 50-ha foraging area (Henry et al. 2002, Broders et al. 2006) around each bat's capture site explained 31% of the variation in δDh–δDp regression residuals (P < 0.001, n = 50 sites). Further investigation of the effect of aquatic foraging on the relationship between tissue δD and δDp may allow this source of variance to be explicitly incorporated a priori into assignment models.
Our delineation of hibernacula catchment areas illustrates an analytical approach with both general application to understanding subpopulation connectivity for congregating species and specific application to modeling WNS transmission. By incorporating multiple sources of isotopic variance, this application produced probabilistic, spatially explicit mine catchment areas, which yielded insights into likely connectivity among regional seasonal habitats. Additionally, our analysis revealed differences in the geographic breadth of the three mines' catchment areas, while indicating that most bats in all three mines originate from the geographic area containing the Quincy and Caledonia mines. Predicting connectivity among regional bat hibernacula is important to resources managers considering hibernacula closures, culling select populations (Hallam and McCracken 2010) and for understanding meta-population dynamics among regions.
We thank Eric Britzke for generously providing data, Bill Route, Paul M. Cryan, Allen Kurta, and Keith Hobson for helpful discussion, and Jessie Barber, Oliver Gailing, and John Vucetich for review of earlier drafts. Two anonymous reviewers provided helpful comments. Ed Yarborough, Paul Brandes, Bill Scullon, and Richard Whiteman assisted in mine hibernacula access. Len Wassenaar assisted in sample analysis. Support was provided by the National Park Service Great Lakes Network, the Ecosystems Science Center and the School of Forest Resources and Environmental Science at Michigan Tech, and support to R. O. Peterson from the Robbins Chair in Sustainable Management of the Environment at Michigan Tech.
Map, geographic coordinates, and predicted stable hydrogen isotope ratios of growing-season precipitation at collection localities (Ecological Archives A022-074-A1).