Extreme selective environments are commonly believed to funnel evolution toward a few predictable outcomes. Caves are well-known extreme environments with characteristically adapted faunas that are similar in appearance, physiology, and behavior all over the world, even if not closely related. Morphological diversity between closely related cave species has been explained by difference in time since colonization and different ecological influence from the surface. Here, we tested a more classical hypothesis: morphological diversity is niche-based, and different morphologies reflect properties of microhabitats within caves. We analyzed seven communities with altogether 30 species of the subterranean amphipod (crustacean) genus Niphargus using multivariate morphometrics, multinomial logit models cross-validation, and phylogenetic reconstruction. Species clustered into four distinct ecomorph classes—small pore, cave stream, cave lake, and lake giants—associated with specific cave microhabitats and of multiple independent phylogenetic origins. Traits commonly regarded as adaptations to caves, such as antenna length, were shown to be related to microhabitat parameters, such as flow velocity. These results demonstrate that under the selection pressure of extreme environment, the ecomorphological structure of communities can converge. Thus, morphological diversity does not result from adaptive response to temporal and ecological gradients, but from fine-level niche partitioning.

The paradigm of convergent evolution states that the stronger the selection pressure, the more similar distantly related organisms will appear. Extreme environments exercise strong directional selection upon their inhabitants, leading to convergent and predictable outcomes (e.g., Chen et al. 1997; Melville et al. 2006; Conway Morris 2010; Langerhans 2010). Caves, along with deep seas, polar regions, deserts, and many other habitats, are examples of extreme environments (Poulson and White 1969; Howarth 1993; Lefébure et al. 2006; Culver and Pipan 2009). Species that have successfully colonized such extreme environments undergo substantial and predictable adaptive changes (Hoffmann and Parsons 1997). The transition to subterranean lifestyle is usually thought of as a slow, gradual process, associated with a suite of morphological changes known as troglomorphies, typically including the reduction of eyes and pigment, elongation of appendages, augmentation of extra optical sense organs, increase in body size, and change in body shape (Culver et al. 1995; Christiansen 2012).

Darwin (1859) already noticed the striking similarity of these changes among several groups of taxonomically unrelated animals and used it as argument in favor of the parallel action of natural selection. Troglomorphies have retained their appeal as evidence for repeated, independent, and linear evolution under strong directional selection (Poulson and White 1969; Jones et al. 1992; Schluter and Nagel 1995; Hüppop 2000; Langecker 2000; Aden 2005; Arnedo et al. 2007; Hedin and Thomas 2010). They have made many cave species look so similar that molecular techniques are necessary to distinguish them (Lefébure et al. 2006; Guzik et al. 2008; Trontelj et al. 2009; Zakšek et al. 2009).

Nevertheless, considerable morphological diversity is frequently encountered among subterranean animals. Some cave species appear to be more, others less troglomorphic, and some even lack troglomorphic characters altogether (Stoch 1995; Sket 2008; Romero 2009; Christiansen 2012). A traditional explanation is that the degree of troglomorphy in a lineage corresponds to the time passed since its invasion of the subterranean realm (Poulson 1963; Barr 1968; Wilkens 1986; Derkarabetian et al. 2010). This, however, turned out not to be the case in troglomorphic planthoppers (Wessel et al. 2007). Moreover, large-scale molecular phylogenies suggest that species that must have dwelled in the underground for the same amount of time because they descended from a common subterranean ancestor, can display different degrees of troglomorphy (Fišer et al. 2008; Faille et al. 2010; Ribera et al. 2010). Recently, Culver et al. (2010) tried to explain the diversity among troglomorphic traits of 56 species of the North American amphipod genus Stygobromus by ecological gradients among different subterranean habitats. As food availability decreases when the subterranean environment becomes more extreme, antennal length was expected to increase in response to the stronger selection pressure. Unexpectedly, they found no support for this prediction.

Because models based on gradients, like time since colonization and severity of the subterranean environment, cannot explain the diversity among cave species, other ecological factors might determine the evolution of subterranean diversity. According to classical ecological theory, spatial and temporal coexistence implies niche division (Tilman 1982; Chase and Leibold 2003). One way species can achieve this is by divergent morphological adaptations that facilitate the use of different microhabitats, or niches (Ricklefs and Miles 1994; Schluter 2000). The aquatic cave environment is not entirely homogeneous and might provide microhabitats to which species adapt thus minimizing competitive interactions (Fišer et al. 2012).

We therefore decided to test the hypothesis that morphological diversity within cave communities is niche-based. Specifically, we tested three predictions resulting from this hypothesis: (1) species should group into discrete morphological clusters across communities; (2) these clusters should be consistently related to a certain type of microhabitat; (3) niche-specific selection should drive similar niche-related traits to evolve several times independently (Wainwright and Reilly 1994; Schluter 2000; Harmon et al. 2005; Losos, 2009; Losos and Ricklefs 2009). These predictions are central to the ecomorph concept (Williams, 1972; Losos 2009), which states that species with the same niche, even if not closely related, tend to resemble each other in morphology and behavior. We use the ecomorph concept as framework for our test. Our study taxon is the subterranean amphipod (crustacean) genus Niphargus, known for its high diversity of species and its capacity to establish local communities.

Material and methods


In the present study, we looked exclusively at communities from karstic aquifers, which are typically connected to or found inside of caves in carbonate rocks. By the term community we refer to the spatial and temporal coexistence of three or more Niphargus species. Communities living in aquatic subterranean habitats not associated with karst caves, like the hypotelminorheic and interstitial alluvial groundwaters (Juberthie 2000; Culver and Pipan 2009), were not within the focus of our research. This is an important distinction compared to the study by Culver et al. (2010) that encompassed the whole range of aquatic subterranean habitats. We use the term microhabitats to refer to the variation of physical properties of the aquatic environment within single caves.

The standard hydrogeological scheme for caves and karstic groundwaters (Gibert et al. 1994; Ford and Williams 2007) implies four types of aquatic microhabitats associated with caves (Fig. 1, Table 1): (1) deep stagnant and slowly flowing groundwater of the saturated or phreatic zone of the karstic aquifer—for simplicity called cave lakes—characterized by stable hydrological conditions and an overabundance of space; (2) cave streams of the unsaturated, or vadose, zone of the karstic aquifer characterized by variable flow regimes with high to very high water flow velocity and an unlimited to slightly reduced (e.g., under rocks) availability of space; (3) epikarst with its associated system of small conduits, typically forming the cave ceiling, characterized by highly variable hydrological conditions (Williams 2008) and a small diameter of available spaces, technically termed pore size; (4) interstitial water inside the cave, that is, water in unconsolidated cave sediments, with a small diameter of available spaces and often hydrologically associated with large cave streams. During periods of low water table, these microhabitats are either separated from each other or connected directionally by water flowing from the epikarst through the vadose zone and via cave streams to the phreatic zone. During periodical floods, however, the karst water table often rises up into the vadose zone, bringing into contact animals from all four types of microhabitats. Essentially, all parts of a karstic aquifer are hydrologically and ecologically interconnected (Perrin et al. 2003; Bonacci et al. 2009).

Figure 1.

Simplified view of aquatic microhabitats found in caves, along with their typical amphipod inhabitants. Niphargus subtypicus with long legs and antennae represents the cave lake ecomorph; N. stygius with short legs and antennae belongs to the cave stream ecomorph; Niphargobates orophobata and N. dobati are representatives of the small-pore ecomorph, characterized mainly by small body size. Animals are not drawn to scale; size is indicated by bars. Uropods (hind legs) are strongly sexually dimorphic.

Table 1.  Physical properties of aquatic cave microhabitats analyzed in this study.
MicrohabitatPore sizeFlow velocity
  1. 1The terms limiting and nonlimiting denote the potential of physical properties to restrict the occurrence of species. See text for further explanation.

Cave lake (phreatic)Large, nonlimiting1Slow, nonlimiting
Cave streams Large, nonlimiting High, limiting
EpikarstSmall, limitingPeriodically high, limiting
Cave interstitial Small, limiting Slow, nonlimiting

Another kind of water bodies commonly found in caves are isolated pools of stagnant water. They are fed by percolating water from the epikarst, by cave streams during periods of high flow rates, by the floods from the phreatic zone, or by the three sources combined. Although they often harbor Niphargus species, they are not their primary reproductive environment (e.g., Pipan et al. 2010) and were therefore not treated as an independent type of microhabitat.

The four cave microhabitats listed above can be described using a general two-dimensional template where temporal heterogeneity of the habitat represents the first and spatial heterogeneity the second component (Townsend and Hildrew 1994). To describe the microhabitat properties for individual Niphargus species, we simplified the template and used water flow velocity as a surrogate for temporal heterogeneity, and the variation of available space, or pore size, as measure of spatial heterogeneity. In the simplest possible subdivision, space and flow velocity are considered as either limiting or nonlimiting factors. The rational is that a species adapted to small spaces can—all other things being equal—survive in a large-space environment, but not vice versa, and a species adapted to resist strong water currents should be able to survive in slow flow conditions as well. The resulting simple matrix of two by two attributes stands for the four types of aquatic microhabitats in caves (Table 1).


The genus Niphargus (family Niphargidae) is the most diverse genus of subterranean organisms in the world. It inhabits several kinds of subterranean waters throughout the Western Palearctic (Karaman and Ruffo 1986; Sket 1999; Väinölä et al. 2008; Fišer et al. 2009a,b). Niphargus is a particularly suitable taxon for this kind of studies because (1) it constitutes a monophylum of over 300 closely related species thus offering ample possibilities for within-group comparisons where the expected impact of phylogeny is low (Losos and Miles 1994; Fišer et al. 2008). (2) Up to nine species can coexist in one place, which is more than in any other subterranean genus (e.g., Sket 2003). (3) Most species are poor dispersers and have small ranges; geographically separated subterranean communities are therefore likely to be of independent origins (Trontelj et al. 2009).

Information about the structure of seven Niphargus cave communities was retrieved from publications and from collections (Table S1). Six of the caves are located in the Dinaric Karst in Slovenia and Bosnia and Herzegovina, one in the Central Apennines in Italy (Fig. 2). The enigmatic epikarst specialist Niphargobates orophobata has been found to be congeneric with other Niphargus species (this study) and consequently included in the analyses. We always tried to determine a species’ main microhabitat, relying on our own field observations and using any other available published and unpublished data for confirmation (Table S1).

Figure 2.

Caves harboring the studied Niphargus communities. Slovenia: 1—Postojna Planina Cave System, six species; 2—Podpeška jama, four species; 3—Luknja, five species, 4—Jama pod Krogom, three species. Bosnia and Herzegovina: 5—Vjetrenica, nine species; 6—Dejanova pećina, three species. Italy: 7—Frasassi Cave System, three species.

Morphological analyses were conducted on 147 specimens belonging to 30 species. Three species were found in two communities each, but belonged to separate populations. Difficult access to the studied communities is a common problem researchers of extreme habitats have to cope with. In our case, up to 15 individuals per species were examined, but for some rare species only few or even single individuals were available. Some samples were partly damaged. We studied adult individuals of both sexes. Traits known to display strong sexual dimorphism were not used. Missing data were taken from taxonomic descriptions.


Troglomorphic traits are classically divided into constructive, or progressive, troglomorphies, which include elaboration of existing structures and the evolution of new ones, and regressive traits such as loss of eyes and body pigment (Hüppop 2000; Langecker 2000; Aden 2005; Christiansen 2012). All Niphargus species lack eyes and body pigment, so we analyzed the variation of their constructive troglomorphic traits. In addition to those, we measured traits involved in the hydrodynamics of an amphipod. Dahl (1977) elegantly showed how gammaridean amphipods depend on self-generated water currents (Fig. 3). Paired, ventrally extended coxal and epimeral plates form a longitudinal ventral channel. In the posterior part of the body, pleopod action generates water currents that enter the ventral channel in the anterior part and exit at the rear. The sum of water currents equals zero when the animal rests on its side in a flexed pose, whereas in the stretched animal they generate jet propulsion. In addition, currents passing the antennae before entering the ventral channel provide chemical cues. Oxygenated water passes the gills situated in the channel, and some species even feed by filtering organic particles.

Figure 3.

Functional model of an amphipod. Arrows indicate water currents at rest; arrow size roughly corresponds to strength of current. Below is a cross-section of an amphipod mid-body segment showing the ventral channel used for gill ventilation and, in some species, microfeeding (modified from Dahl 1977).

The 11 traits included in the analysis under the above considerations were as follows. (1) body length, (2, 3) total lengths of antennae I and II, (4, 5, 6) lengths of pereopods V–VII, (7, 8) depth (proximo-distal length) of coxal plates II and III, (9, 10, 11) width of bases of pereopods V–VII. Among these, body size has been generally considered as an adaptive trait (e.g., Simberloff and Boecklen 1981; Kozak et al. 2009), but also as a troglomorphy (Jones et al. 1992; Christiansen 2012). Antennal length and leg length are classical troglomorphic traits. The lengths of the appendages selected for analysis have been demonstrated to evolve independently (Fišer et al. 2006, 2008). The depths of coxal plates and pereopod bases describe the efficiency of the ventral channel. They also contribute substantially to the diameter of the animal, which is possibly more restraining than body length in habitats with narrow pores.

Specimens were prepared for analysis and measured using morphometric landmarks and procedures as described by Fišer et al. (2009a). Measurements were conducted using an Olympus SZX9 stereomicroscope (Olympus Optical Co. [Europa] GMBH; Germany, Hamburg) and Olypmus software (DP soft and Cell B) with a precision of 0.01 mm. Measurements of the body, antennae, and pereopods, where ambiguities were possible, were repeated at least three times, and the mean value was used as the best estimate of the trait value (Fišer et al. 2009a). Species averages used for all subsequent statistical analyses are listed in Table S2.


Population means of traits were used for all subsequent statistical procedures. Nontransformed data passed the Kolmogorov–Smirnoff test for normality of distribution at P > 0.2, so all subsequent analyses were performed on raw data. The effect of body size was accounted for by either using body length as covariable, or by substituting original trait values by residuals obtained from regressing against body length.

The first prediction, that species form discrete morphological clusters across communities, was addressed by using principal component analysis (PCA) and hierarchical clustering. PCA was performed on a covariance matrix using body length and residual values for all other traits. For hierarchical clustering, we used Ward's method that minimizes within-cluster variance by aggregating species so as to keep the sum of squared Euclidean distances at minimum. We further tested morphological clusters found within the cave lake (phreatic) microhabitat by analysis of variance for differences in body length and analysis of covariance for other traits, controlled by body length as covariable. Analyses were run under PASW Statistics 18 (SPSS Inc., Chicago, IL).

Associations between morphological clusters and microhabitat were first explored by mapping species habitat affiliation on the PCA plot. In addition, we asked how well morphological data predict microhabitat preferences. We used the multinomial logit model (MLM), a regression model that generalizes logistic regression by allowing more than two discrete outcomes. Morphological traits were used as independent explanatory variables and microhabitats as four dependent categories (outcomes). The models were built using the multinomial function from the R package nnet (R Development Core Team, 2009; Venables and Ripley, 2002). Misclassification rate of MLM predictions and the Akaike information criterion (AIC) were used to assess the importance of morphological traits for assignment to ecomorphological classes.

For the actual test of the proposed ecomorph classes, we employed an MLM cross-validation procedure in which we compared the quality of assignment under observed species–microhabitat associations against the quality of assignment under the null model of random microhabitat associations. The total pool of species was randomly divided into two groups, each containing about one-half of the species. The first (training) group was analyzed by MLM. The estimated model was then used to predict the microhabitat for species in the other (validation) group based on their morphological traits. The mismatch of predicted and actual microhabitat was used as a measure of prediction quality. This cross-validation procedure was repeated 100 times, giving the sample of misclassification rates for actual microhabitat membership of species. To test if such misclassification rates can occur purely by chance, we used a permutation test for which we allocated species at random to microhabitats and performed the same cross-validation procedure as for the original data. Each of 100 random microhabitat allocations was cross-validated 100 times. In this way, we got a distribution of 100 mean misclassification rates under the assumption of no relation between morphological traits and microhabitats that formed our null model. If the coexistence of species is deterministic, that is, governed by the same ecomorphological rules in independent subterranean communities, the prediction quality of observed microhabitat–morphology associations should be significantly better than that of randomly allocated microhabitats. One species, Niphargus balcanicus, a PCA outlier that did not cluster with any of the other ecomorphs (see Results), was omitted from the MLM tests.

Finally, we were interested in the functional aspects of individual traits and their possible adaptive value. Using general linear model (GLM), we asked whether variation of morphological traits can be explained by habitat properties and/or their interactions. In the first GLM, body size was treated as dependent variable and pore size and flow velocity as fixed factors. All other traits are size-dependent. Therefore, all other GLMs were controlled for body size by including body length as an extra covariate in the model. We used PASW Statistics 18.


Phylogenetic relationships among the studied species were assessed based on two variable segments of the 28S rRNA gene and the histone H3 gene. Niphargus zavalanus was not included because we were not able to collect fresh specimens for DNA isolation. The first 28S fragment measuring about 830–870 bp was amplified using primers 5′-CAAGTACCGTGAGGGAAAGTT-3′ and 5′-AGGGAAACTTCGGAGGGAACC-3′, the second 28S fragment measuring about 1240–1600 bp was amplified using primers 5′-GCCCTTAAAATGGATGGCGCT-3′ and 5′-CCGCCGTTTACCCGCGCTT-3v, and the 330-bp H3 gene using primers 5′-ATRTCCTTGGGCATGATTGTTAC-3′ and 5′-ATGGCTCGGTACCAAGCAGAC-3′. DNA isolation, PCR, and sequencing procedures were as described in Fišer et al. (2008). Some sequences were taken from previous studies (Fišer et al. 2008; Trontelj et al. 2009; Flot et al 2010; GenBank accession numbers and further information in Table S3).

28S rDNA sequences were aligned using MAFFT (Katoh et al. 2005) under the E-INS-I allowing for large indels separating conserved blocks. Gap-rich regions of the alignment were removed prior to phylogenetic analysis with the help of Gblocks (Talavera and Castresana 2007). Computer alignment was not necessary for H3 sequences as they were all of equal length. The sequence matrix used for phylogenetic analysis was deposited in TreeBASE (Study ID S12919).

Bayesian MCMC tree search was performed in MrBayes 3.2.1. (Ronquist and Huelsenbeck 2003) under six different substitution rate categories and gamma-distributed rate heterogeneity with a proportion of invariable sites. All model parameters except topology were unlinked between 28S and H3. Two independent runs with four chains each were sampled every 1000 generation until the SD of split frequencies stabilized at about 0.002, which happened at roughly 7 million generations. The first 4500 trees from both runs were rejected and from the remaining 5000 trees a 50% majority rule consensus was calculated.


Ecomorph classes as revealed through PCA and cluster analyses were mapped onto the molecular phylogenetic tree to establish whether they had a single or multiple origins. The single-origin hypothesis was tested for each ecomorph class separately by conducting a new Bayesian search with the given ecomorph class constrained to monophyly and same parameter settings as in the unconstrained search. We used Bayes factors (Kass and Raftery 1995) calculated in Tracer 1.5 (Rambaut and Drummond 2009) to evaluate the evidence in favor of the unconstrained tree.

To estimate the number of independent origins of ecomorphs, we employed ancestral state reconstruction as implemented in Mesquite version 2.74 (Maddison and Maddison 2010). We used maximum likelihood ancestral state reconstruction under a Markov k-state one-parameter model (Mk1) and compared it to the results of parsimony-based reconstruction. To account for topological uncertainty, we used the “trace character over trees” option of Mesquite to summarize the reconstruction over the same sample of 5000 trees that was used to calculate the Bayesian consensus tree. As cave animals constitute a system with high inherent rate of evolutionary transitions and low dispersal, we treated the reconstructed number of independent origins as a tentative minimal value.



PCA performed on all Niphargus communities simultaneously partitioned the common morphospace into at least four groups (Fig. 4). They correspond to a group of species living in small-pore microhabitats (epikarst and cave interstitial), a group of species from cave streams, and two groups found in cave lakes (phreatic). The latter two groups from the same microhabitat were separated along the first principal component (PC), which chiefly explains variation in body size (Table 2). The second PC accounted for variation in appendage length and body shape. Both PCs together explain 97.5% of the total variation. A single outlier of the analysis, N. balcanicus, is a very large species with extremely long antennae and legs.

Figure 4.

Principal component analysis on morphometric traits (mean values) of 33 Niphargus species and populations from seven cave communities. The first two axes (PC1 and PC2) together explain 97.5% of the total variation. Cave microhabitats (symbols) and proposed ecomorphs (colors) are only partly in agreement. There is no morphological distinction between inhabitants of the epikarst and the cave interstitial, and there are three distinct morphological groups within the phreatic habitat. Ellipses encircle ecomorph clusters as they were revealed by Ward's hierarchical clustering (Fig. 5).

Table 2.  Loadings by individual traits on the first three principal components.
TraitPC1 (72.4% of total variance)PC2 (25.1% of total variance)PC3 (0.99% of total variance)
  1. *We used residual values for individual traits, calculated from interspecific regressions in which traits were regressed onto body length.

Body length8.2 0.0 0.0
Length of antenna I* 0.0 −3.8  0.5
Length of antenna II*0.0−0.9−0.7
Length of pereopod V* 0.0 −1.6 −0.2
Length of pereopod VI*0.0−1.8−0.3
Length of pereopod VII* 0.0 −1.6 −0.1
Depth of coxa II*0.0−0.1≈0
Depth of coxa III* 0.0 −0.2 ≈0
Width of basis V*0.0≈0≈0
Width of basis VI* 0.0 ≈0 ≈0
Width of basis VII*0.0−0.1≈0

Four discrete ecomorphological classes were revealed also by Ward's hierarchical clustering (Fig. 5). Again, species from two different small-pore habitats showed no distinction in morphology, whereas the division into a large and a very large bodied group in cave lakes was even clearer. Analyzed independently, the two cave lake groups differ significantly in body length (t-test, P < 0.001), but not in other traits when controlled for body length.

Figure 5.

Phenogram of overall morphological similarity between 33 Niphargus species and populations, based on squared Euclidean distances and Ward's clustering method. Four distinct clusters correspond to the five proposed ecomorphs. Typical representatives of each ecomorph are drawn approximately to scale: N. stygius of the cave stream ecomorph, N. dabarensis (not included in the tree but closely resembling N. vjetrenicensis) of the lake giants, N. subtypicus of the cave lake ecomorph, and N. dobati (slender) and Niphargobates orophobata (stout) of the small-pore ecomorph in which species from the epikarst and the cave interstitial are lumped together. Niphargus balcanicus (marked by star) is the only analyzed species of its ecomorph, the “daddy-longleg.” Numbers indicate the caves in which a species or population was found and are as per Figure 1. Species found in more than one cave community were analyzed as separate samples and are marked by the initial of the respective cave name.

Based on these results, we assigned each species to one of four tentative ecomorphs: the small pore, the cave stream, the cave lake, and the lake giants ecomorph. We erected a separate tentative ecomorph for N. balcanicus that, with its oddly long appendages, did not fit with any of the four groups. (See Discussion for a functional explanation of ecomorphs.)

We then built a multinomial logit regression model (MLM) for all species except N. balcanicus. The model correctly predicted the ecomorph for all species, requiring minimally two characters for correct classification. The most important character was body length, which correctly classified 85% of all species if used alone. In combination with any other character, 100% of the classifications were correct. The combination of body length and length of the fifth thoracic leg (pereopod V) received the best AIC support, indicating that appendage length is the second most important ecomorphological trait.

The validity of the ecomorph classes obtained by ordination and clustering was put to test in an MLM cross-validation procedure. Now, only one-half of the species was used to build the model, whereas the other half was used to validate it (see Statistical analyses for details). By repeating this 100 times, we obtained a measure of microhabitat predictability for ecomorph classes. This measure was compared against the null-model predictability obtained by the same procedure but assuming random species–habitat associations. The mean misclassification rate for data with actual microhabitats (12%) was significantly lower than the null-model misclassification rate for randomly permutated microhabitats (38%, P < 0.0001), suggesting that deterministic factors shape ecomorphological convergence of communities (Fig. 6).

Figure 6.

Average misclassification rates for cross-validation MLM models built on observed species–microhabitat associations (left) and on 100 sets of randomly assigned microhabitats (right). Each cross-validation was performed by randomly selecting 50% of the species to build the model, and validating it on the other 50%. The mean misclassification rate of models built on observed species–microhabitat associations was significantly lower than for null models (P < 0.0001).

In some caves, more than one species was associated with a particular type of microhabitat (Table 3). Also, not all ecomorphs were found in all caves. Cave lakes were the most species-rich microhabitat harboring more than twice as many species as small-pore microhabitats and cave streams.

Table 3.  Number of species per ecomorph and cave.
  1. 1Caves are numbered as per Figure 1.

  2. 2When from interstitial and epikarst microhabitats, the respective number of species are shown as summation; in all other caves only epikarst species occur.

Small pore22+1113+17+2
Cave stream 2 1 1 1    1 6
Cave lake11211219
Lake giant   2 1   3 1 1 8


Strong correlation between functional morphological traits and microhabitat properties indicated that variation of these traits is likely to be adaptive. GLMs revealed three kinds of effects (Fig. 7). First, body length was strongly dependent on pore size, but was not affected by flow velocity, with smaller species living in the epikarst and cave interstitial, and larger ones in the phreatic and cave streams. In addition, body size had a significant effect on all other traits in the analyses, where it was used as covariable. Second, the length of sensory and ambulatory appendages (antennae I, II, pereopods V–VII) was affected by flow velocity, but not by the available space. Species with longer appendages live in stagnant and slowly flowing water, whereas species with shorter appendages live in cave streams. Third, traits defining body diameter and the ventral channel capacity (depths of coxal plates and pereopod bases) were affected by both, flow velocity and pore size. Species with deep ventral channel and proportionally large body diameter tended to associate with stagnant or slowly flowing water (cave lakes), and slender, elongated species with shallow ventral channel with cave streams. Somewhat less pronounced but in part still significant was the association between body diameter and pore size, with species from small-pore habitats tending to be more slender. A functional explanation of these morphologies in relation to specific microhabitats is given in the Discussion.

Figure 7.

Variation of three groups of morphological traits in relation to microhabitat properties across 33 Niphargus species and populations. Appendage lengths were measured for antenna one (I) and two (II) as well as legs (pereopods) of the fifth (V), sixth (VI), and seventh (VII) thoracic segment. Body shape was described by width of the second (II) and third (III) pereopod coxa as well as the depth of the fifth (V), sixth (VI), and seventh (VII) pereopod basis. Mean values are plotted together with 95% confidence intervals. Significance of effects of pore size and flow velocity according to general linear models is indicated for each trait: n.s., not significant, *P < 0.05, **P < 0.001. Significance of interactions of both ecological parameters is as follows: body length n.s.; appendage length I*, II n.s., V*, VI*, VII**; body shape II*, III*, V n.s., VI n.s., VII n.s.


The coexistence of morphologically different yet closely related species is by itself an indicator of niche-based mechanisms. An additional pointer toward natural selection acting at the level of microhabitat adaptation comes from the phylogenetic reconstruction with two or more independent origins of the same ecomorph (Fig. 8). Likelihood-based reconstruction led to largely inconclusive ancestral states, whereas parsimony suggested an overall ancestral small-pore ecomorph (Fig. S4). With an ancestral small-pore ecomorph, two to three independent gains of the stream and the lake ecomorphs, as well as three independent gains of the lake giant ecomorph can be postulated. However, other scenarios with an early transition to the cave stream and up to two additional gains for the small-pore ecomorph seem probable, too. Given the inconclusiveness of the likelihood-based reconstruction and the known shortcomings of the parsimony method, the exact sequence of evolution of ecomorphs cannot be reproduced. What we can conclude is that each nonancestral ecomorph evolved twice or more than twice.

Figure 8.

Phylogenetic relationships among the studied Niphargus species based on combined 28S rDNA and histone H3 sequences obtained by Bayesian inference. Posterior probabilities > 0.50 are shown. The tree was rooted using topological information from a genus-wide molecular phylogeny (Fišer et al., 2008). Ecomorphs do not form coherent monophyletic groups or paraphyletic series but, rather, seem to have originated several times each (see also ancestral state reconstructions in Fig. S4). All phylogenetic scenarios with monophyletic ecomorphs are decisively rejected by Bayes factors. Numbers indicate the caves in which a species was found and are as per Figure 1. Most cave communities are not monophyletic by origin, with the exception of cave Vjetrenica (number 5), where most of the species seem to have evolved by local adaptive radiation.

Most importantly, all phylogenetic scenarios with monophyletic ecomorphs received decisively less support than the original phylogenetic hypothesis, as implied by log10 Bayes factors between 67 (monophyletic small pore) and 135 (monophyletic lake giants) in favor of the original tree. This makes any evolutionary scenario with even a single ecomorph originating only once highly improbable.



With this study, we demonstrate that classical, niche-based adaptation is still a valid hypothesis to explain morphological variation among closely related cave species. We show that four distinct Niphargus ecomorph classes—small pore, cave stream, cave lake, and lake giants—consistently occur across geographically and phylogenetically independent communities in the same type of microhabitat. These microhabitats are really facets of the aquatic environment of a single cave, not separate subterranean compartments. Competitive interactions between species are therefore likely to be driving morphological differentiation and controlling community assembly. Accordingly, morphologies of subterranean animals are not only general subterranean adaptations, for example, to darkness and low resources, but also adaptations to the particular microhabitat. Although we here equate niches with cave microhabitats (cf. Culver 1970, 1973), other niche dimensions, for example related to feeding behavior or life history, are just as likely. In our case, niche separation within cave lakes might have proceeded by size.

As expected and commonplace as these results may seem, they are but a first challenge to the paradigm of time-correlated troglomorphies that has persisted to the present date: “If, on the other hand, the selective environment is the same in the four subterranean habitats, then differences in antennal length or flagellar number would be the result of different levels of adaptation, possibly the result of different lengths of time of isolation in caves and other subterranean habitats” (Culver et al. 2010), and Derkarabetian et al. (2010): “A logical expectation for cave-dwelling organisms is that the degree of troglomorphism is correlated with time, i.e., more highly modified taxa are expected to be relatively old.”

Therefore, the question arises whether the observed pattern of ecomorphological differentiation in Niphargus communities necessarily contradicts the scenario of directional selection toward an increasingly adapted troglomorphic animal. If troglomorphies do progress with time, different ecomorphs might represent evolutionary stages in a repeated pattern of radiation (Losos 2009). Culver and Pipan (2008) suggested that superficial subterranean habitats could be a gateway used to colonize deeper subterranean space. Accordingly, the first stage in the process of troglomorphic adaptation would resemble the small-pore ecomorph. One could think of cave streams as possible second type of aquatic habitat, finally to be followed by a shift to the deep phreatic cave waters. Long-legged animals from cave lakes with their enormous antennae could in a sense truly be considered as a final stage of ecomorphological evolution in caves. This scenario is in considerable agreement with the distribution of morphotypes on the phylogenetic tree (Figs. 8 and S4). However, it does not imply that extant small-pore and cave stream species are transitional stages that will eventually evolve to lake giants. Also, it is difficult to explain why not all ecomorphs were found in all caves. The reasons could be ecological (a particular type of microhabitat is missing or not abundant enough to support a species). Cave Vjetrenica (number 5), for example, has no permanent, sizable cave stream, which might explain why there is no cave stream ecomorph. They could also be related to unequal sampling effort and bias, as small epikarst species are least likely to be discovered. Finally, a wider historic and faunistic context might reveal that there are simply no species of the missing ecomorph sufficiently close to colonize the cave.


The four ecomorphs differ in three groups of traits: body size, structures contributing to body diameter, and appendage length. In the following, we attempt to functionally explain their variation among different types of cave microenvironment. We propose that the morphologies related to different cave microhabitats reflect trade-offs between traits generally advantageous in the subterranean environment and the selective constraints of specific microhabitats.

Body size is a complex trait, related to available space, but evidently also in close relation with body shape. Changes can go both directions as reports have been made of both, miniaturization in interstitial groundwater (Coineau 2000; Galassi et al. 2009) as well as gigantism in cave lakes (Koenemann and Holsinger 1999). Similarly, our data imply that species specialized for life in small-pore habitats are small, and that species living in phreatic water are, on average, large. Small size, however, is not a prerequisite for a species to colonize a small-pore habitat. Interstitial and epikarst amphipods can be in fact quite long, as also noted in the North American amphipod Stygobromus (Culver et al. 2010). Niphargid examples are N. likanus and N. dolenianensis collected from alluvial interstitial groundwater of the rivers Sava (Slovenia) and Torre (Italy), respectively. Both of them exceed 10 mm in length. The cost they pay is a reduction of coxal plates and pereopod bases, and thereby a reduction of the ventral channel. By this reduction, the body diameter becomes small enough to permit movement through tiny voids. This makes long-bodied amphipods living in small-pore habitats look slender. A deeper ventral channel creates stronger currents enabling faster swimming, as exemplified by N. ictus, a rapid swimmer moving between the bottom and upper layers of water (Dattagupta et al. 2009; Flot et al. 2010). Furthermore, a weaker water current passing the ventral channel means less oxygen transported to the gills and to the broods in female marsupia (cf. Dahl 1977). A large body, on the other hand, makes it easier for an amphipod to prey upon other animals and less likely become prey itself (MacNeil et al. 2008; Luštrik et al. 2011). Other reported advantages of a larger body in caves include lower energetic demand per unit mass (Hüppop 2000), higher fecundity (Culver et al. 1995), and larger eggs (Roff 1992). Body size of the small-pore ecomorph—while undoubtedly restricted by the strongly limited space—may thus also reflect a trade-off imposed by the capacity of the amphipod ventral channel.

Length of antennae and legs has been one of the least disputed troglomorphies throughout the history of biospeleology. From Racovitza (1907) till the present time (Mejía-Ortíz et al. 2006; Culver et al. 2010; Derkarabetian et al. 2010), there has not been the slightest doubt that among a bunch of similar animals the one with the longest antennae is best adapted to cave life. Conversely, our results suggest that appendage length can no longer be considered as a general troglomorphy in amphipod crustaceans. Instead, they strongly suggest that variation in appendage length is determined by fine-scale habitat affinity of a species. At one extreme are cave lake species with strongly elongated antennae (up to 1.3 times body length) and pereopods, and at the other extreme species from cave streams that tend to have the shortest appendages (down to 0.35 of body length). For comparison, in the generalist freshwater amphipod genus Gammarus, the second antenna measures about 0.5–0.6 of the body length. The finding that appendage length is affected by flow velocity, and not by the availability of space, seems to contradict another traditional concept of subterranean biology. Namely, interstitial crustaceans are described as having their appendages, except the posterior ones, shortened or even reduced in number (Danielopol et al. 1994; Coineau 2000). We can only speculate about the reasons for this stark reversal in selection pressure in cave microhabitats with high velocity of water flow. The advantages of long tactile sense organs in the absence of optical perception seem straightforward. Elongated appendages extending far from body can improve perception of water movements or chemical cues by carrying more sensillae, and a more precise location of the source of stimuli. Extremely elongated pereopods are associated with behavioral changes. The long-legged N. balcanicus no longer lies on its side in the typical amphipod fashion, but rest and moves in an upright posture (own unpublished field and laboratory observations). Together with two morphologically and behaviorally similar species not included in this study (N. croaticus and N. dolichopus), we tentatively assign them to an ecomorph class of its own, the daddy-longlegs. In fast flowing streams, however, long and fragile arthropod appendages are exposed to injury, and an upright posture can be dangerous in high drift conditions (Wilzbach et al. 1988). In summary, increased antennal and leg length can only be considered as an adaptation to low flow velocity subterranean waters. The trait seems to be subject to a trade-off between increased sensory and ambulatory function in the lightless environment and the mechanical hindrance it poses in fast flowing cave stream. At least for niphargid amphipods, small-pore size does not represent as severe a restriction to appendage length as flow velocity.


The main intention of this work was to find an explanation for the high morphological diversity among closely related species living in the same kind of ecologically constraining aquatic environment. Most of the measured morphological variation can be explained by environmental predictors that are a combination of flow velocity and available space. Nonetheless, the morphospace of the tested traits showed no response to variation of flow velocity in the two small-pore microhabitats, the cave interstitial and the epikarst. These two microhabitats are ecologically very different. Physicochemical conditions are constant in the former, whereas in the epikarst they fluctuate drastically (Pipan et al. 2006). It seems unlikely that small spaces and darkness are sufficient for a uniform ecomorph to evolve, even though data from North American subterranean amphipods seem to imply so (Culver et al. 2010).

In addition to the lack of ecomorphological distinction of epikarst and interstitial species, we encountered a similar enigma posed by the repeated sympatric occurrence of several cave lake and lake giant species within single communities. Often these species are grouped closely together by PCA, and constitute sister species or entire clades. Data on diet and behavior is needed to underpin the distinction between the cave lake and the cave giant ecomorph. Fine-scale morphological differences in gnathopod shape and size could be informative of feeding behavior. The known variation of these structures (Fišer et al. 2009a) suggests dietary difference among cave lake and lake giant Niphargus, and observations indicate that some species are more potent predators than others (Fišer et al. 2010). To a lesser extent, a similar question comes up for occasionally co-occurring species of other ecomorphs. These are ecologically less challenging, because fewer species coexist, and their microhabitats are spatially more heterogeneous, potentially permitting them to avoid direct competitive interactions. However, as models do predict the possibility of functionally equivalent co-occurring species in condition similar to those in caves (Hubbell 2006), this possibility should be considered, too.

Finally, the particular factors driving Niphargus species into divergent niches need to be determined. Competition for scarce resources comes to one's mind in caves, but predation among species might be just as important.

Associate Editor: J. Wares


We are grateful to G. Bračko, S. Polak, S. Prevorčnik, B. Sket, M. Zagmajster, and V. Zakšek for their help with fieldwork; S. Dattagupta and J.-F. Flot for furnishing specimens from Frasassi Cave; A. Moškrič and V. Zakšek for their help in the lab; M. Zagmajster for helping with GIS work. We thank D. Schluter for thoughtful discussions on earlier versions of the manuscript, B. Sket for information about microhabitats, and two anonymous reviewers for their helpful comments. A part of the work by PT was carried out during his sabbatical at Dolph Schluter's lab at the Biodiversity Research Centre, University of British Columbia. This research was supported by the Slovenian Research Agency Program P1–0184 and the Slovenian Research Agency Project J1–0676.