Habitat requirements and ecological niche of two cryptic amphipod species at landscape and local scales

Cryptic species are phylogenetically diverged taxa that are morphologically indistinguishable and may differ in their ecological and behavioral requirements. This may have important implications for ecosystem services and conservation of biodiversity. We investigated whether two ecologically important cryptic species of the freshwater amphipod Gammarus fossarum (types A and B) are associated with different habitats. We collected data on their occurrence at both the landscape scale (large watersheds) and at the local scale (river reach) to compare macroand microscale environmental parameters associated with their presence. Analysis of the landscape scale data showed that occurrence of types A and B differ with respect to watershed and river size and, interestingly, human impact on river ecomorphology. Whereas type B was mainly found in less forested areas with higher human impact, type A showed the opposite occurrence pattern. Analyses of the local scale data suggested that habitats occupied by type A were characterized by larger gravel, larger stones and less macrophytes than habitats occupied by type B. The landscape and local data set showed contradicting patterns with regard to stream size. Overall, the observed differences between the two types of G. fossarum most likely reflect ecological differences between them, but alternative explanations (e.g., historical colonization processes) cannot be completely ruled out. Our study underlines that common cryptic species can differ in their ecology and response to anthropogenic influence. Such differences in habitat requirements among difficulttoidentify taxa present a challenge for biodiversity and ecosystem management. Our results emphasize the importance of conservative and precautionary approaches in maintenance of habitat diversity and environmental heterogeneity.

INTRODUCTION v www.esajournals.org EISENRING ET AL. similarity does not necessarily imply ecological similarity, assessment of ecological niche diferences among cryptic species is crucial for many research ields such as biological control and use of biological indicator taxa (Feckler et al. 2012(Feckler et al. , 2014. Cryptic species may difer with respect to their ecological niches (e.g., Bidochka et al. 2001, Davidson-Wats et al. 2006) and may interact differently with their environment (Bickford et al. 2007). Newly discovered cryptic species may therefore have important implications for conservation of biodiversity where the goal oten is to deine and protect endemic species richness (Bickford et al. 2007).
The arthropod order of Amphipoda lends itself to the study of ecological diferences between cryptic species, as its members play several important roles in aquatic ecosystems (MacNeil et al. 1997, 1999, Dufy and Hay 2000 and oten contain complexes of cryptic species (e.g., Wellborn and Cothran 2004, Wit et al. 2006, Lefébure et al. 2007, Fišer et al. 2015. Of speciic signiicance is the freshwater keystone amphipod Gammarus fossarum, which consists of several cryptic species, some of them probably millions of years old (Scheepmaker and Van Dalfsen 1989, Müller 2000, Weiss et al. 2014. Reproductive isolation of the cryptic G. fossarum species is indicated by the absence of intermediate genotypes in mixed populations (Müller 2000). G. fossarum has a central position in food chains of prealpine streams, processing dead organic material and representing a major source of ish nutrition (MacNeil et al. 1997, Dangles et al. 2004. Furthermore, G. fossarum is used as an indicator organism for habitat quality and ecotoxicological studies (e.g., Lukancic et al. 2010, Bundschuh et al. 2011, 2013. In headwaters of our study area (Switzerland) G. fossarum is frequently the dominant benthic macroinvertebrate (Altermat et al. 2014), and loss of these populations may have pronounced efects on the ecosystem (MacNeil et al. 1997(MacNeil et al. , 1999. Despite the ecological importance of this species complex, relatively litle information regarding ecological requirements of G. fossarum is available (but see Mejering 1991, Peeters 1998, Stürzbecher et al. 1998. This is especially true for potential diferences between the cryptic G. fossarum species. Given their age and the potential for evolution in distinct refugia during Pleistocene glaciations (Müller 2000), diferences in various biological traits due to divergent selection (and drit) are conceivable. Indeed, a few studies have found diferences in biological characteristics including sensitivity to toxic chemicals (Feckler et al. 2012), infection with parasites (Westram et al. 2011a), and timing of reproduction (Stürzbecher et al. 1998).
The most common cryptic species within the G. fossarum complex, namely types A and B according to the terminology of Müller (1998), have geographically distinct distribution patterns in Europe. In Switzerland, their distribution ranges overlap and sympatric populations occur. Whereas type A is common in the eastern parts of Switzerland, type B is more frequent in the western parts (Altermat et al. 2014). Several streams in the Rhine drainage are known where both species occur in sympatry (Westram et al. 2011b, Altermat et al. 2014. Preliminary evidence suggests that types A and B might have diferent preference of microhabitats (e.g., Stürzbecher et al. 1998. Müller et al. (2000) classiied streams in Germany into two categories: streams with plant-rich substrates and mud versus streams characterized by gravel and the presence of leaf liter. Type B mostly occurred in lower altitude streams of the former category, whereas type A did not show a clear tendency. These diferences in ecological niche were detected albeit the study restricted to relatively crude habitat categorizations, and very litle information was available on microhabitat associations within streams. However, a study on a single mixed population indicated that type A was mostly associated with stones, whereas type B was more common in areas with macrophytes (Stürzbecher et al. 1998). Our motivation was to expand these studies to more detail on microhabitat occupation across a large number of streams and ask if the reported diferences in ecological niche are general enough to conclude that the species are ecologically diferent.
We analyzed environmental factors to characterize habitat occurrence patern of the two cryptic species, G. fossarum types A and B, in Switzerland both at the landscape scale (watershed) and at the local scale (river reach). The landscape data include a large number of stream characteristics and parameters describing anthropogenic efects. On the local scale, we quantiied various v www.esajournals.org EISENRING ET AL. microhabitat types within each stream and associated them with G. fossarum occurrence. Our study allows for a general test of environmental correlates of distribution of G. fossarum species, and evaluation of potential ecological niche differences between the two cryptic species. A detailed description of the distribution paterns of cryptic G. fossarum species is of high relevance for adjusting its future role as a biological indicator species and will add to our understanding of fundamental ecological processes in prealpine headwater streams.

Field survey
Landscape scale.-We used data on the occurrence (presence/absence) of G. fossarum within the River Rhine drainage in Switzerland. The data are based on standardized and representative sampling program conducted within the project "Bio diversity Monitoring in Switzerland" (BDM, 2009, 2014. In this project, all macroinvertebrates are monitored on a systematic sampling grid covering the whole of Switzerland. At each site, specifically trained ield biologists collected and identiied macroinvertebrates to species level, using well-established and highly standardized methods. The sampling sites are randomly chosen across Switzerland, and the data sampled therein thus relect a representative depiction of both biological (e.g., occurrence of species) and abiotic (e.g., altitudinal distribution) variables at the landscape scale (see also Stucki 2010. Observed diferences can thus be directly linked to biological diferences in the two species or colonization events. We analyzed all 222 sampling localities within the Rhine drainage (36,500 km 2 ) that were sampled once between 2010 and 2012 (Fig. 1a). The BDM sampling scheme includes watercourses higher than 1st order streams (Stucki 2010). Standing waterbodies, irst-order streams, and watercourses completely inaccessible by wading are not included in the BDM program for methodological reasons (Stucki 2010, BDM 2014. The sampling scheme speciically considers the representative inclusion of headwaters, which are naturally much more numerous (Altermat 2013). Sampling sites are located between 280 and 2718 m a.s.l. All sites together relect the Swiss watercourses and amphipod species diversity therein. The seasonal timing of the sampling was optimized for the local phenology of macroinvertebrates (Stucki 2010) and took place between March and July. For each site, we collected information on the presence of G. fossarum A and B (Altermat et al. 2014)  and on environmental variables (Stucki 2010). The environmental variables that were evaluated for each sampling site are presented in Table 1, for details of the sampling methods, see Stucki (2010).
We identiied amphipods based on morphology using standard identiication literature (Eggers and Martens 2001). We further used previously established microsatellite markers for separating G. fossarum A from G. fossarum B. We extracted DNA from 5 to 50 individuals of the G. fossarum complex per site (either extracting DNA from whole individuals or from pereopods) and analyzed 10 microsatellite markers using the method described in Westram et al. (2010). Speciic allelic combinations in these 10 microsatellite markers have been described as a diagnostic tool to tell types A and B apart (for details see Westram et al. 2010Westram et al. , 2011bWestram et al. , 2013. The microsatellite markers diagnostic for type A is gf27 polymorphic with alleles >200 bp (but ≠205), whereas for type B the marker is monomorphic at 205 bp. The BDM method is not optimized for a quantitative sampling of macroinvertebrates, and we thus only used presence/absence data.
Local scale.-Gammarus fossarum specimens were collected from 17 diferent streams in Switzerland during August and October 2010.
( Fig. 1b). The streams were inhabited either only by types A or B, as known from previous studies based on molecular markers (Westram et al. 2013, Altermat et al. 2014. To distinguish microhabitat characteristics of A and B streams and to investigate potential microhabitat preferences of the two G. fossarum species, each stream was examined on a total length of 30 m, subdivided into 15 sections of 2 m length each. Exceptions were three stream sites where only 10, 11, and 14 sections were deined due to spatial restrictions. For each section the proportions of six microhabitat types were estimated by eye and channel width and low velocity were measured (Table 2). A quantitative sample of benthic invertebrates was taken from a random position within each section using a HESS-sampler (sample surface area = 452.4 cm 2 , mesh size 0.5 mm). As an exception from random sampling, rare microhabitats were sampled at least once per location, without paying atention to randomization. Gammarids of a sample were washed out in a sieve (mesh size 1.5 mm) and presence/absence was recorded. Because of the mesh size, only individuals with a body size of 5 mm or more were included in the study.

Data analysis
Landscape scale.-We grouped the 222 biodiversity monitoring sites into four classes according to presence-absence of G. fossarum types A or B (no G. fossarum, types A and B, or both). Next we tested how these categories associated to environmental and biological parameters that were measured at the sites. We irst reduced the dimensionality in the continuousscale environmental variables using a principal component analysis (PCA) ( Table 1). We then used multinomial logistic regression where the dependent variable was the G. fossarum classiication and independent covariates (continuousscale) were the two statistically signiicant PCA axes. As additional categorical independent factors we used the ecomorphology-based evaluation of the human impact at the sites, sotsediment abundance class, and classiication of algal coverage (Table 1). We evaluated the model using a forward-stepwise evaluation of independent variables excluding interactions between factors.
Local scale.-Our local scale data set consists of environmental parameters from a total of 245 sections from 17 study streams. We also included sections that had no G. fossarum to get a reference of the environment that was neither preferred by G. fossarum A nor B.
We irst reduced the dimensionality in the environmental parameters with a categorical PCA (Linting et al. 2007) to three dimensions, using each section as an independent sample. We then used the object scores of the three dimensions to ask if type A of B were associated with any speciic habitat dimensions. To test for diferences in habitat characteristics between the two G. fossarum types we used a multivariate response proile analysis (MANOVA) where object score proiles of types A and B presence were compared to each other and to samples with no G. fossarum. We did not include stream identity in the statistical analyses because in each stream we found only one type of G. fossarum, therefore stream identity as a factor would have been confounded with G. fossarum type. For completeness we ran the same statistical models using stream as a factor instead of G. fossarum type and present the results graphically for comparison.
All statistical analyses were conducted with SPSS version 22.

Environmental correlates of Gammarus fossarum types A and B distribution in the landscape
We had data on the G. fossarum presenceabsence and all environmental variables from 222 BDM sites. At 75 of these sites we found G. fossarum A, at 17 sites we found G. fossarum B. The two types of G. fossarum were co-occurring at 27 sites. No G. fossarum was found at 102 sites.
The three irst PCA axes captured 70% of variation in the data. Factor loadings (Table 1) show that the irst axis (31%) loaded positively to channel and catchment size and negatively to proportion of forest in the area. The second axis (24%) loaded positively to altitude and negatively to proportion of agricultural land in the area, whereas the third axis (15%) loaded positively on proportion of forest in the area. Overall, G. fossarum A was found between 280 and 1620 m a.s.l., whereas G. fossarum B was found between 280 and 867 m a.s.l.
The multinomial logistic regression model gave us several interesting insights on how G. fossarum presence-absence relates to environmental parameters and how the predicted occurrence paterns difered between the G. fossarum types. Two of the three PCA axes had a signiicant association with the presence of G. fossarum in the inal multinomial logistic regression model. PCA2, which captured largely variation in altitude, entered the model irst, being the most signiicant (P < 0.001) explanatory variable, whereas PCA1, which captured variation in size of the watershed and river size, entered second (P = 0.010). Interestingly, human impact on ecomorphology (MSK) remained as the only signiicant factorial variable in the inal model. Efect of human impact on morphology was weaker than that of the two principal components, but still statistically signiicant (P = 0.035).
More detailed interpretation of the model brings three main results. First, high elevation sites (which are also characterized by less agricultural use) are less likely to support G. fossarum populations (Fig. 2). This is shown by very low predicted occurrence probabilities for sites where PCA2 scores are high. Predicted probabilities for both G. fossarum types declined at a similar rate as a function of PCA2, indicating that species responded similarly to variation captured by PCA2.
The second main result concerns diferences among G. fossarum types in response to PCA1, which mainly corresponds to size of the watershed and size of the river. Occurrence probability of type B increases with PCA1 score, suggesting higher likelihood to ind type B populations in larger and less forested rivers (Fig. 2b). On the contrary, type A has much higher occurrence probabilities for low values of PCA1 (forested, smaller streams) (Fig. 2a).
The third interesting result is that species response to human impact seems to be opposing for the two types. Type A occurrence probabilities are highest for near natural sites, whereas they were lowest for type B (Fig. 2).

Environmental correlates at the local scale
The irst three dimensions of the categorical PCA analysis captured 61% of total variance among the environmental parameters. Highest object scores for the irst dimension (27% of variance) were associated with samples that had high proportion of large gravel substrate and wide channels, characterizing larger streams, with less large stones and dead organic material (Fig. 3). The second dimension (19% of variance) was associated positively to macrophytes and slow low, separating samples with macrophyte habitat. The third dimension (15% of variance) associated positively on large stones and gravel, characterizing sections with heterogenous habitat. Distribution of samples along these habitat dimensions is shown in Fig. 4a,c and e.
Proile analysis suggests diferences in all three habitat dimensions between presence of the two G. fossarum types (Figs. 4 and 5) (Wilks' Lambda = 0.56, F 4,482 = 40.1, P < 0.001). Stream sections where type A was found were characterized by wider channels, more large gravel and large stones and less macrophytes than sections with type B (Fig. 4). Sites where no G. fossarum were found were fastest lowing sections with largest stones and gravel (Fig. 4).

DISCUSSION
We investigated environmental parameters associated with presence of G. fossarum populations. In the following, we discuss potential ecological and historical factors driving G. fossarum distribution and especially focus on differences between the two cryptic G. fossarum species A and B in this respect. Our data contribute to the growing body of research showing that cryptic species, although morphologically indistinguishable, can differ in ecological characteristics (Narins 1983, Henry 1994, Feulner et al. 2006, Westram et al. 2011a, Cothran et al. 2015. Since G. fossarum is one of the most common macroinvertebrate species in prealpine headwater streams of central Europe, the understanding of the ecological differences between these cryptic species are of high importance and have implications for the conservation of freshwater ecosystems.

Environmental parameters associated with occurrence of the Gammarus fossarum complex
We found several key parameters that are associated with the presence of G. fossarum. Some of them afect both species similarly, but others indicate diferences between the two species.
Our analyses show that altitude is a key factor determining G. fossarum distribution (Fig. 2). Sites above about 1600 m a.s.l. seem not habitable for both G. fossarum types. This can probably be atributed to harsher climatic conditions and lower food availability (i.e., fewer decaying leaves) in high-altitude compared to lowland v www.esajournals.org EISENRING ET AL. streams. Furthermore, G. fossarum generally avoided fastest lowing stream sections with large stones and gravel. Both species occurred with high probabilities in stream sections in agricultural land or setlement. This can potentially be explained with the comparatively high nutrient levels of these stream sections and therefore a high amount of submerged aquatic vegetation, which mainly positively correlates with aquatic invertebrate abundance (Krull 1970, Anteau et al. 2011. Interestingly, however, human impact, characterized via the MSK classes, had clearly diferent efects on the distribution paterns of the two cryptic G. fossarum species. Whereas type A mostly occurred at near natural sites, the opposite patern was found for type B. As the MSK classiication can correlate with water pollution (or human inluence in general), these diferences may be linked to diferences in physiochemical tolerances of the two species. Indeed it has been demonstrated that type A shows a higher overall sensitivity toward speciic insecticides and fungicides compared to type B (Feckler et al. 2012). However, since type A was not negatively afected by the proportion of setlement and agricultural land, diferences in physiochemical tolerances can probably be ruled out as an explaining factor. Based on MSK classiications, stream sections that are under high human impact are oten characterized by a high degree of artiicial ground modiications and compaction (Stucki 2010). This may afect the abundance of aquatic vegetation of certain ish and macroinvertebrate predators as well as intra-and interguild cannibalism and competition with multiple direct and indirect efects for G. fossarum population dynamics (MacNeil et al. 1999, Anteau et al. 2011 andtherein). Type A might react diferently to these changes in vegetation, predator and competitor abundance than type B, and therefore show diferent paterns regarding MSK characterization. This hypothesis is partially contradicted by the indings of Seymour et al. (2016) who found no covariation between macroinvertebrate diversity and the genetic diversity within cooccurring G. fossarum.
Our landscape scale data show that type B occurs with a higher probability in large streams and less forested areas, whereas type A tends to the opposite patern but is showing a more generalist behavior regarding these environmental factors (Fig. 2). Similar results were found on the local scale. Type A mainly occurred in streams that were comparatively rich in gravel and stones but poorer in macrophytes, whereas type B was mainly found in streams with typical grassland characteristics (high proportion of macrophytes but less dead organic material or stony substrate). Our results are in line with the study of Müller et al. (1998) in which type B was more oten associated with grassland streams than woodland streams. Diferences in the degree of feeding plasticity between the two types could explain some of the found distribution paterns. Friberg and Jacobsen (1994) state that species of the genus Gammarus, which are generally characterized as typical detrivore-shredders, are oten able to exploit additional food sources such as fresh aquatic plant material. This could explain why type B was mainly found in streams with less dead plant material but with a higher proportion of macrophytes compared to streams inhabited by type A. Furthermore, parasitism and predation may have a strong impact on G. fossa- rum population dynamics (MacNeil et al. 1999 and therein). Diferent predation or parasite pressure (Westram et al. 2011a), or diferences in defensive behavior of the two types could therefore potentially lead to a selection for diferent environmental conditions in which the respective predation or parasite avoidance strategies are optimized. However, in order to get a beter un-derstanding of the microhabitat diferentiations between G. fossarum types A and B, clearly more detailed local scale studies are needed, especially at sites were types A and B occur sympatrically.
Stream size was the only parameter that showed an opposite outcome on the local compared to the landscape scale. Based on our landscape scale model, type A was mainly found in smaller streams than type B, whereas the local scale data suggest the opposite pattern. However, due to extreme diferences in the range of river widths included in the landscape data (streams of all sizes were included) compared to the local data (only smaller streams were selected) comparisons of the two data sets regarding stream size have to be treated with caution.
The landscape and the local data set were not sampled at the same time. This raises the question regarding phenological efects biasing our results. Stürzbecher et al. (1998) conducted a study were they focused on seasonal diferences in abundance and reproduction of G. fossarum A and B. The local data set was sampled during August and October, a time when according to Stürzbecher et al. 1998 the relative abundance between A:B is roughly 50:50. We did not have time-series data, and each site at the landscape scale data was sampled once between March and July. Sampling time was optimized with respect to the elevation (for details see Altermat et al. 2013, Stucki 2010. While Stürzbecher et al. (1998) shows that numbers of A and B can luctuate diferently during that period of time, we do not think this is afecting our interpretations for two reasons. First, the sampling time was optimized for local phenology for all sites, such that similar phenological (but diferent Julian) data was compared. Second, our analysis is mostly focusing on presence/ absence and not abundance data. Thus, even if there were changes in abundances, they were likely not strong enough to override the pres-ence-absence paterns, and would also only be relevant for the relatively small number of sympatric populations.

Possible evolutionary and historical factors driving Gammarus fossarum distribution
The central inding of this study is that the two cryptic G. fossarum species are associated with diferent environmental parameters. In the following we discuss potential evolutionary and historical mechanisms that could alternatively explain the found distribution paterns.
Historical recolonization processes, rather than distinct adaptations, might partly explain the differences between the two species. A clear latitudinal distribution patern of the two cryptic species was found for Switzerland ( Fig. 1). Type A is more common in the eastern parts of the country; type B is more frequent in the western parts (e.g., Westram et al. 2011b, Altermat et al. 2014. Other studies (e.g., Müller et al. 2000) imply that a similar distribution patern is also true for Germany. Although both cryptic G. fossarum species diverged before or during the last glacial period in the Pleistocene and then probably came into contact due to the withdrawing of the glaciers (Scheepmaker 1990), a lack of time for a complete overlap of the two cryptic species could explain current distribution paterns. Alternatively, competitive exclusion between A and B species could have prevented the spread of the two species across the whole studied area and therefore be responsible for the separation we ind nowadays. For example, intraguild predation is wide spread in freshwater gammarids (MacNeil et al. 1997, Fig. 5. Proile plots of mean object scores of the Categorical PCA dimensions for (a) each Gammarus fossarum type and (b) for each stream. See Fig. 3  1999 and therein) and can result in rapid species exclusion and replacement (Dick et al. 1993). In both cases, if the eastern and western parts of our study area difer with respect to environmental characteristics, the distinct species distributions could generate species-environment associations even if the two species are not diferently adapted. We argue that this is unlikely to be the sole explanation. First, no obvious geological pattern that matches the distribution of the cryptic species of the G. fossarum complex in Switzerland can be found. The only environmental factor that changed with latitude and that was associated with G. fossarum distribution was altitude. Although it had a strong impact on G. fossarum distribution in general, we found that both G. fossarum types were similarly afected by altitudinal changes. Second, in the area where the distributions of the two species overlap, it has been found that species composition is clearly distinct even between geographically very close populations (few km) (Alp et al. 2012). A similar patern was found for the area of overlap in this study (Fig. 1a). In these cases, local environmental differences seem more likely to explain distribution paterns than historical reasons.
We suggest that, rather than being explained by historical factors alone, the observed diferences are partially explained by diferent adaptations of the two cryptic species. They existed in diferent refugia during the last glaciation for extended periods of time (Müller 2000). These refugia are in geographically very diferent regions, with potentially very diferent selection pressures. Therefore, it is quite plausible that distinct adaptations have evolved, with the two species specializing on diferent habitats. Future work should further investigate this, for example, using lab experiments to test for diferential adaptation.

Implications for conservation management
The knowledge of cryptic species and the potential ecological and behavioral diferences within such species complexes necessarily lead to new questions regarding conservation management strategies.
In this study, we found good evidence for ecological diferences between the two cryptic G. fossarum species, A and B, which are oten associated with diferent environmental parameters and difer in many habitat requirements.
Gammarus species play a fundamental role in many freshwater ecosystems and drastic decreases of their populations can have severe consequences for other trophic levels. Besides the two studied species types A and B, several additional cryptic species, which also may difer in their ecology, exist within the G. fossarum complex (Müller 1998, Weiss et al. 2014.
In order to preserve G. fossarum populations it is therefore necessary not to preserve a single type of freshwater stream but to maintain a highly diverse set of heterogeneous stream types, as the distribution of cryptic G. fossarum species oten can depend on diferent environmental factors.