The imprint of Quaternary glaciers on the present-day distribution of the obligate groundwater amphipod Niphargus virei (Niphargidae)
*Arnaud Foulquier, UMR CNRS 5023, Ecologie des Hydrosystèmes Fluviaux, Université Claude Bernard Lyon 1, Bâtiment Forel (403), 43 Bd 11 Novembre 1918, F-69622 Villeurbanne Cedex, France. E-mail: email@example.com
*Arnaud Foulquier, UMR CNRS 5023, Ecologie des Hydrosystèmes Fluviaux, Université Claude Bernard Lyon 1, Bâtiment Forel (403), 43 Bd 11 Novembre 1918, F-69622 Villeurbanne Cedex, France. E-mail: firstname.lastname@example.org
Aim A plausible yet untested biogeographical scenario suggests that Quaternary glaciers shaped the present-day distribution of the groundwater amphipod Niphargus virei. This study was designed to test two hypotheses pertaining to this scenario: (1) the probability of occurrence of N. virei in ice-free areas decreases in the vicinity of the Würm glacier; and (2) dispersal is sufficiently low for the historical record of glacial effects to persist over time.
Location The study area was located in the southern Jura Mountains, France.
Methods A total of 497 sites were sampled to ascertain the distribution of N. virei in the southern Jura. Amplified fragment length polymorphism was analysed from a subset of 24 sites. The relationships between the probability of occurrence of N. virei and distance to the Würm glacier or elevation were investigated using a logistic regression. Spatial autocorrelation analyses were performed on both the residuals of the logistic regression and genetic distance to test the significance of dispersal and barriers to post-glacial recolonization. The influence of catchment boundaries as barriers to dispersal was examined using different neighbouring relationships between sites. We tested the statistical significance of the reduction in deviance and gain in precision of an autologistic regression that took into consideration the influence of dispersal constraints on the distribution of N. virei.
Results Niphargus virei rarely occurred in formerly glaciated areas, and its probability of occurrence in ice-free areas decreased in the vicinity of the Würm glacier. Combined autocorrelation analyses of spatial distribution and spatial genetic structure showed that: (1) the distance at which spatial autocorrelation was no longer significantly positive did not exceed 16 km; (2) genetic differentiation fitted a model of isolation by distance; and (3) catchment boundaries acted as barriers to dispersal. The autologistic regression with dispersal constraints significantly increased our capacity to predict the distribution of N. virei. Maps of probabilities of occurrence suggested that post-glacial recolonization was impeded by the extension of glacial outwash.
Main conclusions The present distribution of N. virei in southern Jura is probably the result of a historical range reduction driven by glaciation coupled with restricted dispersal and isolation by distance.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Ecologists are increasingly aware of the prominent role of historical factors in shaping the present-day patterns of biodiversity at local, regional and continental scales (Ricklefs, 2004). Several studies demonstrate that the spatial distribution and genetic diversity of species, as well as the richness and composition of communities of animals and plants in North America and Europe, retain the imprint of cyclical changes in climate and glacier extent that occurred during the Pleistocene (Taberlet et al., 1998; Hewitt, 2004; Svenning & Skov, 2004, 2005). The historical imprint on the distribution of plants and animals can persist for longer among species showing low dispersal ability, because it is less likely to be overwritten by subsequent dispersal phases. Historically caused patterns of distribution are presumably widespread among obligate groundwater species because of the supposedly poor dispersal ability of organisms and the highly fragmented nature of groundwater systems (Gibert et al., 1994; Humphreys, 2000). The distribution patterns of several groundwater species have been interpreted to reflect the past distribution of their marine ancestors during colonization events induced by eustatic cycles (Ginet, 1971; Notenboom, 1991; Boutin & Coineau, 2000). Coineau (1994) argued that the distribution of several Microcharon species (microparasellid isopods) in the Mediterranean basin coincided with distinct expansion phases of the Tethys from the Turonian (upper Cretaceous) up to the Pliocene. More recent historical events that potentially shaped groundwater biodiversity patterns include glacial expansion–contraction cycles and changes in river catchment connections during the Quaternary (Ward et al., 2000). However, most inferences regarding the influence of historical factors on the current geographical distribution of groundwater species have been derived from descriptive studies that overlaid the distribution of a limited number of species occurrences with the spatial extent of palaeogeographical events (Lefébure et al., 2006, 2007; Zaksek et al., 2007). Distribution maps alone are not sufficient to infer the effect of historical events on the present-day occurrence of species. In the present study, we developed a multifaceted approach involving presence–absence data from multiple sites, DNA polymorphism data from a subset of sites, generalized linear models, and spatial autocorrelation analyses, to test for the historical influence of Quaternary glaciers on the present-day distribution of a groundwater amphipod.
Niphargus virei Chevreux, 1896 is an obligate groundwater amphipod that completes its whole life cycle exclusively in groundwater (Fig. 1a). The biology, ecology and physiology of this species are known in exceptional detail (Ginet, 1960, 1971, 1996; Mathieu & Turquin, 1992; Hervant et al., 1999). Niphargus virei colonizes the vadose and saturated zone of karst aquifers, but has never been found in interstitial aquifers or the hyporheic zone of rivers. Its distribution in France and Benelux spans 1000 km of latitude, but recent phylogenetic analyses have demonstrated that this widely distributed amphipod comprises three highly divergent clades (1–3 in Fig. 1b; Lefébure et al., 2006). Ginet (1960, 1971) suggested that the distribution of N. virei in the Jura (clade 3 in Fig. 1b) had been shaped by the Würm glaciers (80,000–10,000 years bp). He proposed that the expansion of ice would have devastated populations in the eastern Jura, and that post-glacial recolonization would have been limited by dispersal constraints. This palaeogeographical scenario implies a number of hypotheses to be tested: (1) the geographical distribution of N. virei was much larger prior to the expansion of Quaternary glaciers; (2) its rarity in formerly glaciated areas is not attributable to habitat features of the current environment or interspecific competition; (3) the probability of occurrence of N. virei in ice-free areas decreases in the vicinity of the last glacial maximum (LGM); and (4) dispersal was sufficiently low for the historical record to persist over the past 10,000 years. The first hypothesis is not amenable to testing because of the lack of fossil records. The second hypothesis was somewhat supported by a transplant experiment, introducing individuals of N. virei into a formerly glaciated cave system located outside the actual species range (Ginet, 1965). Barthélémy (1982) showed that individuals of N. virei not only persisted in the cave 20 years after their transplantation, but also outcompeted individuals of the resident groundwater amphipod Niphargus rhenorhodanensis Schellenberg, 1937. The present study was designed to test the third and fourth hypotheses. We sampled about 500 sites in southern Jura and used a logistic regression model to test for variation in the probability of occurrence of N. virei with distance from the terminus of the Würm glacier. The significance of dispersal and the occurrence of barriers to post-glacial recolonization were examined using spatial autocorrelation analyses with different neighbouring relationships on presence–absence data and genetic distance.
The study area extended over 2300 km2 in the southern part of the Jura Mountains, France (Figs 1 & 2). The area comprises six distinct river catchments (Upper Ain, Valouse, Suran, Oignin, Albarine, Lower Ain), which belong to the catchment of the Ain River. In addition, sampling was extended west (Bresse) and east (Burbanche-Séran) of the catchment of the Ain River (Fig. 2a). Karst aquifers formed in Jurassic and Tertiary limestone, which outcrop in 57% of the study area (Fig. 2b). Interstitial aquifers lie in a variety of unconsolidated sediments including glacial outwash, glacial till, modern alluvium and mountain scree. Elevation increases from 220 to 1510 m eastward (Fig. 2c). The Würm glacier, which covered the eastern part of the study area, retreated about 10,000 years ago (Fig. 2d). Grid coverage of aquifer type, elevation and distance to the glacier was generated in a geographic information system (GIS, ArcView) using commercially available maps (elevation: Institut Géographique National (IGN), Base de Données Altimétriques (BD ALTI), spatial resolution 50 m; aquifer type: Bureau de Recherches Géologiques et Minières (BRGM) geological map, 1 : 50,000; glacier: contour maps of Quaternary glaciers). Distance to the glacier was obtained by delineating 100-m equidistant areas from the LGM (Würm glacier). Negative and positive values corresponded to areas located beneath and outside the glacier, respectively.
A total of 242 karstic sites and 255 interstitial sites were sampled from July 2002 to June 2004 (Fig. 2d). The 497 sites comprised 61 caves, 181 karstic springs, 43 interstitial springs, 56 hyporheic sites and 156 interstitial wells. Several methods were used to sample amphipods in caves, including collecting by sight, filtering of water in streams and gours, and trapping with baits (Camacho, 1992). Drifting amphipods at the outlets of springs were sampled by filtering water with nets (mesh size 100 μm) for at least 1 day. Individuals that found refuge in the benthic layer near springs were captured by disturbing the sediment and filtering water immediately downstream of the disturbed area with a Surber sampler (Malard et al., 2002). Amphipods were collected from the hyporheic zone of rivers by pumping four replicate samples of 5 L water and sediment with a Bou-Rouch pump (Boulton et al., 2003) at each site. Phreatic wells were sampled by pumping at least 50 L water and sediment with a manual piston pump. In large-diameter wells, sediment and associated amphipods were also collected using a phreatobiological net (Dumas & Fontanini, 2001). More details of the sampling methods are given in Malard et al. (2002). Samples were preserved in 70% alcohol and sorted in the laboratory under a dissecting microscope. Obligate groundwater amphipods were counted and identified according to Ginet (1996). A subset of 24 sites distributed throughout the study area was selected for genetic analyses (Fig. 2d).
Amplified fragment length polymorphism (AFLP) analyses
Genomic DNA (from three to nine individuals per site) was extracted using the DNAeasy kit (Qiagen, Hilden, Germany) or Wizard SV 96 Genomic DNA Purification kit (Promega, Madison, WI, USA) on c. 20 mg of the whole animals. AFLP reactions were adapted from Vos et al. (1995). Restriction reactions were performed with 200 ng DNA with 5 U MseI, 5 U EcoRI and 0.15 μg BSA with supplied MseI buffer (NewEngland Biolabs, Ipswish, MA, USA) for 2 h at 37°C in 15 μL. Two Weiss units T4 DNA ligase, 1 μmMseI and EcoRI adaptors, and supplied ligase buffer (NewEngland Biolabs) were then added to the restriction reaction and run for 2 h at 16°C. Samples were subsequently diluted 1 : 5 in water. Primary amplification was carried out using 5 μL diluted restriction/ligation products and 0.4 μm preselective primers in a standard 25-μL polymerase chain reaction (PCR) containing 2.5 mm MgCl2, 1 × Mgfree Taq buffer (Invitrogen, Carlsbad, CA, USA), 250 μm dNTPs and 1.5 U Taq polymerase. Reactions were run with the following cycling parameters: 120 s at 72°C, 25 cycles of 30 s at 94°C, 30 s at 56°C, 120 s at 72°C and 10 min at 72°C. Products were finally diluted 1 : 20 in water. Selective amplifications were conducted using 5 μL diluted primary amplification, 0.25 μm of unlabelled (MseI) primer, and 0.15 μm fluorescently labelled EcoRI primer in a 20-μL PCR reaction. PCR conditions were as described above, except for the hot-start polymerase (Hot Master Taq DNA Polymerase; Eppendorf, Hamburg, Germany). After denaturation at 94°C for 2 min, reactions were run through 10 cycles, each cycle beginning with 30 s at 94°C, followed by a 30-s annealing step at temperatures decreasing from 66 to 57°C with each cycle, followed by a 120-s extension at 72°C. Reactions were subjected to an additional 20 cycles of 30 s at 94°C, 30 s at 56°C, and 120 s at 72°C, followed by a final 10-min extension at 72°C. Primer pairs EcoRIACA/MseICGA and EcoRIACT/MseICCA with EcoRI primers labelled with FAM were used to generate AFLP profiles. Samples were run on a capillary sequencer (Megabace 1000; Amersham, Uppsala, Sweden) with ROX size standard following 75 s of injection at 4 kV. Several blanks were added at each step and the whole procedure was carried out twice for five individuals. All individuals run twice showed identical AFLP profiles. Profiles were analysed with genetic profiler (Amersham), which returns bands by their length (in bp) and fluorescent intensity. During a pilot analysis of 50 individuals, we selected a set of polymorphic fragments whose presence or absence was unambiguous. This procedure led to the selection of 103 polymorphic loci (59 loci for EcoRIACT/MseICCA and 44 for EcoRIACA/MseICGA) for which we defined a length window (minimum and maximum length in bp) and an intensity cut-off below which a fragment is considered absent. For each individual and locus, data were coded as binary characters (1 = presence; 0 = absence), yielding a data matrix of 130 individuals × 103 loci.
The relationships between the probability of occurrence of N. virei and elevation or distance to the glacier in ice-free areas were investigated using a generalized linear model with logit link and binomial error (logistic regression) (McCullagh & Nelder, 1989). Regression models were performed on presence/absence data rather than number of individuals, because sampling methods differed among sites. The two variables were tested for inclusion in the model by considering linear, quadratic and cubic terms. The reduction in deviance associated with each term was tested for significance at α = 0.05 using a χ2 test (Venables & Ripley, 2002). The most parsimonious model was selected using Akaike’s information criterion (Akaike, 1974).
Spatial autocorrelation under different neighbouring relationships
We performed spatial autocorrelation analyses both on the residuals of the logistic regression and on genetic distances to examine the significance of dispersal and hydrological barriers to post-glacial recolonization. Spatial autocorrelation in the residuals of a logistic regression may reflect dispersal processes (the occurrence of a species at a site increases its probability of occurrence at neighbouring sites), but it may also be due to the omission of environmental variables that are spatially structured (Keitt et al., 2002). Spatial autocorrelation analysis using genetic distance is another indirect approach estimating the strength of dispersal in aquatic systems (Primmer et al., 2006). Under restricted dispersal and isolation by distance, spatial genetic autocorrelation is expected to decline monotonically with increasing distance between sites (Arnaud et al., 2001).
Spatial autocorrelation in the residuals of the logistic regression was quantified using Moran’s I statistic (Moran, 1948). Under the null hypothesis of a random spatial distribution, the expected value for Moran’s I is –(n – 1)−1, higher or lower values indicating positive and negative autocorrelation, respectively. The normalized Mantel statistic (rM) was used to analyse spatial autocorrelation in the genetic data (Mantel, 1967). This statistic tested the correlation between the genetic distance matrix computed from Nei–Li distance (Nei & Li, 1979) between individuals, and the geographical distance matrix computed from spatial coordinates between sites, under the null hypothesis of no relationships between these two matrices. The distance decay of the autocorrelation was examined using correlograms (Sokal & Wartenberg, 1983; Diniz-Filho & De Campos Telles, 2002; Torres et al., 2003). The distance at which the correlogram first intercepted the x axis, or when coefficients were no longer significant for successive values, defined the mean autocorrelation distance between sites or patch size (Sokal & Wartenberg, 1983). Distance classes were used to allow a sufficient number of pairwise comparisons between sites to be considered in each coefficient calculation (Aubry, 2000). Correlograms were computed for 10 distance classes that were defined so as to contain a similar number of pairwise comparisons between sites (Legendre & Legendre, 1998). The significance of both Moran’s I and rM in each distance class was tested by a randomization test (10,000 iterations) under the null hypothesis of a spatially random distribution. The global significance of each correlogram was tested using Bonferroni’s criterion: at least one of the Moran’s I or rM must be significant at a 0.05/k significance level, where k represents the number of distance classes defined (Oden, 1984).
The influence of hydrological barriers to dispersal was examined by considering different neighbouring relationships between sites (Thioulouse et al., 1995). In particular, we examined the role of river catchment boundaries as potential barriers to dispersal. Three cases were tested using the following neighbouring relationships.
• Case 1: dispersal occurs within and between river catchments. Two sites were neighbours if their separation distance fell within the distance class considered.
• Case 2: dispersal was solely possible within catchments because of the weakness of hydrological connections between catchments. Two sites were neighbours if they belonged to the same catchment and their separation distance fell within the distance class considered.
• Case 3: designed to examine spatial autocorrelation exclusively due to dispersal between catchments. Two sites were neighbours if they belonged to distinct catchments and their separation distance fell within the distance class considered.
These different neighbouring relationships, and the constraint to maintain a similar number of pairwise comparisons between sites, prevented the use of similar distance classes among correlograms. In the first two cases, spatial autocorrelation was investigated for distances up to half the maximum distance between two sites in the study area to prevent edge effects (Fortin et al., 2002). Mantel tests in distance class 0 km were carried out to test inter-individual distances within populations (Stehlik et al., 2001). The result was included in the Mantel correlograms. In the third case, spatial autocorrelation was investigated up to distance classes showing significant autocorrelation in the two preceding correlograms.
Under the assumption that the spatial autocorrelation in the residuals of the logistic regression reflected dispersal, we examined the reduction in deviance and the gain in prediction associated with the inclusion of an autocovariable. This autocovariable took into account the probability of occurrence at neighbouring sites. It was defined as (Augustin et al., 1996):
where wij= 1/hij (hij is the linear distance between sites), ki is the number of points neighbouring site j, and yj is the probability of occurrence at site j.
Based on the results of correlograms, the autocovariable at site i was calculated using neighbouring sites, located at distances up to 6000 m, that belonged to the same catchment as site i. The statistical significance of the deviance reduction due to the inclusion of the autocovariable in the logistic regression, yielding to an autologistic regression, was tested with a chi-square test. The predictive performances of the logistic and autologistic regressions were evaluated by a cross-validation using receiver-operating characteristic (ROC) analyses (Fielding & Bell, 1997). Sensitivity values (true presence) and specificity values (true absence) from model predictions were calculated for threshold values between 0 and 1. A ROC curve is a plot of sensitivity values against (1 – specificity) values. The area under the ROC curve (AUC) is a measure of the predictive performance of the model (Fielding & Bell, 1997). AUC values vary between 0.5 for a model with predictions no better than random, to 1 for a model with predictions exactly reproducing the observations. The data set was randomly divided into two parts: two-thirds of the data set were used for calibration and the remaining one-third was used for evaluation of predictive performances using ROC analyses. This step was repeated 1000 times for the logistic and autologistic models, yielding 1000 AUC values for each model. Student’s t-test for pairwise observations was performed to test the statistical significance of the increase in mean AUC due to the inclusion of the autocovariable. A Moran correlogram was calculated on the residuals of the autologistic regression.
Mapping probabilities of occurrence
Logistic and autologistic regressions were used to draw prediction maps of the probability of occurrence of N. virei in the catchment of the Ain River. Maps were constructed according to Augustin et al. (1996). The logistic regression calibrated on the presence/absence data set was used to generate a first prediction map for the catchment of the Ain River (cell size 300 m). The autocovariable was then calculated for all the cells of this first prediction map and the autologistic regression was adjusted using the presence/absence data set. The autologistic regression was used to generate a second prediction map that took into consideration dispersal constraints.
Of a total of 497 sites sampled, only 81 sites harboured N. virei, 209 contained amphipods other than N. virei, and 207 had no amphipods. Niphargus virei was never collected from interstitial groundwater and occurred in only 33% of the 242 karstic sites. Moreover, the geographical distribution of this species fell almost entirely within the limits of the non-glaciated area (Fig. 3). Niphargus virei was collected at 70 sites out of 120 in the non-glaciated area, and at only 11 sites out of 122 in the glaciated area. The 11 formerly glaciated sites containing N. virei were located at an average distance of 2765 ± 1655 m from the glacial boundary. Consequently, the regression models and spatial autocorrelation analyses were conducted on a subset of 136 karstic sites belonging to the area (Bresse, Suran, Valouse and Lower Ain) that was not covered by ice during the LGM. Of these 136 karstic sites, 75 sites harboured N. virei.
The probability of occurrence of N. virei decreased with increasing elevation (Fig. 4a). The relationship between the probability of occurrence of N. virei and distance to the glacier was best described using a unimodal response curve (Fig. 4b). Probabilities of occurrence were maximal at distances ranging from 5 to 10 km and were minimal for distance classes < 0 km and > 15 km. Elevation and distance to the glacier, expressed respectively in their linear and quadratic forms, produced a significant reduction in deviance in the logistic regression (Table 1). Distance to LGM was included in second position in the regression, but it accounted for the same reduction in deviance as elevation.
Table 1. Results of the logistic and autologistic regressions (inclusion of the autocovariable) for testing variables for effects on the probability of occurrence of Niphargus virei.
Distance to LGM
Moran and Mantel correlograms were significant (Bonferroni’s level) when dispersal was allowed to occur within and between catchments (case 1, Fig. 5a,b). Moran’s I was significantly positive for the first distance class 0–5.5 km and became significantly negative for the second class. Moran’s I was not significant for all other distance classes, except class 5. Mantel correlograms decreased with increasing distance; the last significant positive value was observed for distance class 12.5–14.5 km.
Both correlograms were significant (Bonferroni’s level) when dispersal was allowed to occur solely within catchments (case 2, Fig. 5c,d). Moran’s I was significantly positive up to distance class 5–6.7 km. Values were not significant for all other distance classes, except for classes 6.7–8.5 km (significantly negative) and 16–19 km (significantly positive). Mantel coefficients were continuously and significantly positive up to distance class 13–16 km.
Both correlograms were significant (Bonferroni’s level) when we examined spatial autocorrelation that was exclusively due to dispersal between catchments (case 3, Fig. 5e,f). Moran’s I was significantly positive for the first distance class 0–4 km and was not significant for the following distance classes. Mantel coefficients were significantly positive up to distance class 12–15 km and significantly negative in distance class 15–17 km.
The inclusion of the autocovariable in the logistic regression resulted in a significant reduction in deviance (P < 0.0001; Table 1). The autocovariable accounted for the same reduction in deviance as the elevation and distance to the glacier. The AUC values of the autologistic regression (mean AUC = 0.803, SD = 0.054) were significantly higher than those of the logistic regression (mean AUC = 0.749, SD = 0.06) (Student’t-test for paired observations, P < 0.001), indicating that the consideration of spatial autocorrelation increased the predictive performance of the model. Moran’s correlogram performed on the residuals of the autologistic regression was not globally significant because none of the autocorrelation coefficients was significant at a 0.05/10 significance level (Fig. 6).
Map of probabilities of occurrence
Probability maps generated with the logistic and autologistic regressions reliably portrayed the observed distribution of N. virei and provided predictions at the scale of the catchment of the Ain River (Fig. 7). In both maps, probabilities of occurrence were distinctly higher in the catchment of the Suran River. However, the logistic regression predicted high probabilities of occurrence in the non-glaciated part of the Albarine catchment, whereas N. virei was never collected in this area. This erroneous prediction was partly corrected by the autologistic regression, which took into consideration the spatial pattern of occurrences and boundaries between catchments.
The first hypothesis of a decrease in the probability of occurrence of N. virei with decreasing distance to the LGM was clearly supported by the spatial pattern of occurrences. Niphargus virei rarely occurred in formerly glaciated areas, and its probability of occurrence in ice-free areas increased westward, up to distances ranging from 5 to 10 km to the LGM. The decrease in the probability of occurrence for distances > 10 km probably reflected sampling inefficiency due to the presence of Quaternary sandy and clayey deposits covering karst aquifers located along the western Jura margins. Distance to the LGM was included after elevation in the logistic regression, but it accounted for a significant reduction in deviance (8%). Part of the reduction in deviance accounted for by elevation could also reflect the influence of Quaternary glaciers. Indeed, elevation and distance to the LGM were negatively correlated because Quaternary glaciers covered the most elevated areas of the Jura Mountains (elevation = −0.020 × distance to LGM + 481; r2 = 0.44, P < 0.001, n = 497) (see also Castellarini et al., 2007). Elevation directly influences groundwater temperature which, in turn, may affect the physiology of N. virei [mean annual spring temperature (°C) = −0.0061 × elevation (m) + 12.62, r2 = 0.92, P < 0.001, n = 16 springs, F.M., unpublished data]. However, the mean annual temperatures in the study area (5.8–11.3°C) are close to the thermal preferences of N. virei and typically fall outside lethal temperatures measured in the laboratory (< 3 and > 17°C) (Issartel et al., 2005a,b). Thus the decrease in the probability of occurrence of N. virei with increasing elevation might partly reflect the correlation between elevation and distance to the LGM, rather than the biological effect of groundwater temperature or other current environmental variables related to elevation (e.g. food resources).
The influence of Quaternary glaciers in shaping the present-day distribution of groundwater organisms has long been debated (Vandel, 1964; Holsinger, 1980). Most studies, but not all (see Culver et al., 2003), suggest that the groundwater fauna of northern Europe and America was impoverished as a result of the last glaciation (Strayer, 1994; Proudlove et al., 2003; Gibert & Culver, 2005). Several hypotheses are proposed to explain the presence of groundwater species in formerly glaciated areas, including extensive post-glacial recolonization from distant southern refugia via hyporheic corridors (Proudlove et al., 2003) and the persistence of cold-adapted populations in local ice-free mountain tops (nunataks) or subglacial refugia (unfrozen groundwater beneath the ice) (Holsinger, 1980). Lefébure et al. (2007) recently provided strong DNA evidence that several lineages of the groundwater amphipod N. rhenorhodanensis survived the expansion of French alpine valley glaciers in nunataks. Paleogeographical reconstructions of the LGM in the Jura indicated that this middle-elevation mountain range was entirely covered by a thick ice cap, leaving no nunatak for survival (Buoncristiani & Campy, 2004). Our study clearly shows that the distribution of N. virei paralleled the terminus of the Würm glacier, suggesting that any pre-existing populations east of that glacier were extirpated, but that populations continued living to the west of it. The eastward decrease in the probability of occurrence of N. virei in ice-free areas might reflect the worsening periglacial environmental conditions with decreasing distance to the glacier terminus.
Dispersal and post-glacial recolonization
Our second hypothesis to account for the persistence of a glacial imprint was that of limited dispersal by N. virei. This hypothesis was supported by the autocorrelation analyses of distributions of occurrences and spatial genetic structure. For both approaches, the distance at which spatial autocorrelation was no longer significantly positive did not exceed 16 km. Moran correlograms indicated that N. virei exhibited a non-random distribution pattern characterized by small clusters (< 7 km) of higher density. Mantel correlograms, which were characterized by positive spatial genetic autocorrelations over short distance classes, subsequently declining through zero and becoming negative, indicated a model of isolation by distance (Slatkin, 1993; Peakall et al., 2003). Results of the combined analyses were consistent with the expectedly low dispersal capacity of N. virei due to its phenology, including a low reproduction rate (one clutch of 20–30 eggs per year), and ecological characteristics, such as its inability to disperse across non-karstic environment (Ginet, 1960). Obligate subterranean taxa have generally been considered to be poor dispersers because of their short distribution range and low gene flow, resulting in high levels of genetic differentiation among local populations (Caccone, 1985; Strayer, 1994; Gibert & Deharveng, 2002). Strong genetic divergences among nearby groundwater populations of crustaceans were reported for Amphipoda belonging to the genera Niphargus (Lefébure et al., 2006, 2007), and Isopoda belonging to the genera Monolistra (Caccone et al., 1986) and Asellus (Sbordoni et al., 1980; Verovnik et al., 2004). Lefébure (2005) calculated that the genetic structure of the 24 sites of N. virei used in the present study was Θ = 0.297 (CI 0.266–0.326) (Weir & Cockerham, 1984), suggesting an important genetic differentiation between locations separated by only a few kilometres.
The congruence between the results of the distributional and genetic approaches suggests that spatial autocorrelation in the residuals of the logistic regression reflects limited dispersal, rather than the potential effect of spatially structured environmental variables not included in the model. The consideration of dispersal constraints in the form of an autocovariable resulted in the same reduction in deviance as distance to the LGM, and also significantly increased our capacity to predict the distribution of N. virei. As observed in other studies (Svenning & Skov, 2004, 2005), dispersal limitation of the post-glacial recolonization from ice-age refugia probably caused N. virei to occupy only a limited portion of its potential distribution range. Both the aggregated distribution of N. virei and the genetic pattern of isolation by distance suggested that long-distance dispersal was not sufficiently common to enable the recolonization of formerly glaciated areas.
Barriers to dispersal
Karst landscape in the meridional Jura is a mosaic of small aquifers, the groundwater of which feeds multiple springs. The study area (2300 km2) includes over 180 springs, and the average area of surface basins feeding karstic aquifers is only about 15 km2 (F.M., unpublished data). Although surface streams, hyporheic corridors and interstitial aquifers are potential dispersal routes for aquatic invertebrates between isolated karstic aquifers, small pore size in unconsolidated sediments severely restricts movement of the large-sized amphipod N. virei (body length > 20 mm) (Fig. 1a). Moreover, dispersal as juveniles is probably limited by predation by epigean invertebrates. None of the 43 interstitial springs, 56 hyporheic sites and 156 interstitial wells sampled in this study yielded even a single individual of N. virei. The predictions of the autologistic regression also suggested that the extension of glacial outwash along the Ain River might have impeded the colonization of the Albarine catchment from glacial refugia in the Suran catchment. Thus N. virei dispersed via subsurface karstic hydrological connections between sites (Lefébure et al., 2006). Owing to the hierarchical nature of karst landscapes, we expected that dispersal would be more likely to occur within catchments than between sites belonging to different catchments. Results of spatial autocorrelation analyses with different neighbouring relationships supported the view that watersheds acted as barriers to dispersal. Moran and Mantel autocorrelation coefficients were significantly positive over higher distance classes when dispersal was allowed to occur solely within catchments (Fig. 5c,d) rather than between any sites (Fig. 5a,b). The between-catchment Moran correlogram (Fig. 5e) provided significant positive autocorrelation for lower distance classes than the within-catchment correlogram (Fig. 5c). The Mantel coefficient between sites belonging to distinct catchments declined markedly for distances higher than 7 km, although it remained significantly positive up to distances of 15 km (Fig. 5f).
Lefébure (2005) analysed the same AFLP data set using a Bayesian model-based clustering method to delineate populations of N. virei in the southern Jura, and concluded that the spatial genetic structure within N. virei coincided with hydrological landscape features, including catchment boundaries. Gooch & Hetrick (1979) and Culver et al. (1995) also found that subsurface hydrological connectivity, as reflected by the presence of surface and/or subsurface drainage divides, was the most significant variable for understanding the genetic structure of spring and cave populations of Gammarus minus (Amphipoda) in karst landscapes of Virginia, USA. Genetic structuring of groundwater populations of the water louse Asellus aquaticus and the European cave salamander Proteus anguinus was also found to follow the present-day hydrological drainages (Verovnik et al., 2004; Goricki & Trontelj, 2006). We suggest that post-glacial colonization of karst groundwater in the Suran catchment could have been facilitated by its continuous synclinal structure and the occurrence of larger karstic aquifers (surface basins c. 100 km2). Caccone & Sbordoni (1987) found that populations of North American cave crickets were genetically less differentiated in continuous limestone outcrops, showing higher opportunities for dispersal. A detailed understanding of dispersal pathways in fragmented karst landscapes of the Meridional Jura would require a more extensive genetic data set and a precise knowledge of aquifer boundaries.
Understanding the role of history and dispersal constraints in shaping the present-day distribution of groundwater species is not only an important theoretical issue, but is also a critical step for predicting the potential effects of rapid climatic changes and selecting appropriate conservation strategies. According to combined autocorrelation analyses of spatial distribution and spatial genetic structure, the present-day distribution of N. virei in southern Jura is probably the result of a historical range reduction driven by glaciation coupled with restricted dispersal, and isolation by distance since glacial retreat. The long-lasting effect of historical large-scale events severely limits the relevance of current environmental variables for modelling the present-day distribution of groundwater species. Our findings suggest that fast future climatic changes are unlikely to result in a rapid displacement of species distribution ranges because of the importance of dispersal constraints in groundwater.
This work was supported by the European project PASCALIS (Protocols for the Assessment and Conservation of Aquatic Life in the Subsurface, NEVK2-CT-2001-00121). We thank the large number of people who contributed to the collection and sorting of samples: G. Bouger, T. Datry, M.-J. Dole-Olivier, D. Ferreira, C. Malard, D. Martin, J. Mathieu, V. Noune and V. Règue. We are particularly grateful to V. Règue for his help in generating the geological coverage of southern Jura and to B. Hugueny for his insightful statistical advice. We thank François Pompanon and the DTAMB, and especially Laure Granger for their technical assistance with AFLP genotyping. We are indebted to R. Ginet and M.-J. Dole-Olivier for their valuable help in the identification of amphipods. We thank Andrew Boulton for reviewing an earlier draft of this manuscript and two anonymous referees for their valuable comments.
Arnaud Foulquier is a doctoral student interested in statistical methods for modelling the effects of history and human disturbance on biodiversity patterns and biogeochemical functioning of groundwater ecosystems. The research for this paper was conducted during the completion of Arnaud’s Master’s degree within the framework of the European Project PASCALIS (N°EVK2-CT-2001-00121).
Florian Malard is a groundwater ecologist with research interests in landscape ecology and the factors influencing the assemblage of ecological communities.
Tristan Lefébure is an evolutionary biologist interested in diversification and adaptation of subterranean animals and pathogenic bacteria.
Christophe Douady is an expert in phylogenetic, tokogenetic and phylogeographical reconstructions from both methodological and biological points of view.
Janine Gibert leads the groundwater Hydrobiology and Ecology Laboratory at the University of Lyon 1 (http://groundwater-ecology.univ-lyon1.fr). She is an expert in groundwater ecology and biodiversity.