Molecular Phylogenetic Analyses

Mitochondrial and Nuclear Gene Trees: Phylogenetic analysis of the 162-speci-men matrix of Cytb sequences recovered a large, well-supported clade (bootstrap = 98%) that circumscribes tanala-like specimens representing the nominal taxa E. tanala, E. tanala-A and E. tanala-B, to the exclusion of other species of Eliurus (fig. 3). The concatenated analysis of mitochondrial genes Cytb and Nd2 (fig. 4) recovered this same clade with equally strong support (bootstrap = 100%), as did each of the individual nuclear-gene trees (fig. 5), and the species tree based on all four nuclear genes (fig. 6). Three haplotype groups within this large clade are identifiable from the mitochondrial trees (figs. 3, 4). The largest includes specimens from the humid montane forests of the Central and Southern highlands (E. tanala; bootstrap = 98%); individuals collected near the type locality of E. tanala, Forêt de Vinanitelo (appendix 1: locality 65), are interspersed among the branches of this haplogroup (fig. 3). A second subclade (E. tanala-B; bootstrap = 97%) encompasses individuals from humid forests of northern Madagascar; the partial DNA sequence from the type specimen of E. ellermani (MNHN 1981.871) is associated with this haplogroup. The third haplogroup (E. tanala A; bootstrap = 100%) comprises specimens collected from tsingy regions around Namoroka and Bemaraha (E. tanala-A); no species name is presently available for this subclade.

Unexpectedly, six individuals collected in the Forêt d'Analamavo, a tsingy-associated, dry forest in on the Namaroka Formation of western Madagascar (fig 2; see also fig. 16: locality 35), have a mitochondrial haplotype that aligns them with typical E. tanala, a montane species from eastern humid forest in the Central and Southern highlands (figs. 3, 4). Based on geographic proximity alone, we had expected that the Forêt d'Analamavo sample would be allied with the western dry-forest form E. tanala-A, which it otherwise resembles morphologically (see next section). Analyses of the nuclear loci provide additional insight into this apparent conflict. When analyzed independently, the three diploid loci (Rbp3, Vwf, Fgb) do not show a clear pattern of coalescence along meaningful geographic or taxonomic lines; however, the X-linked locus, Opn1mw, clearly allies the populations from Forêt d'Analamavo with the western, dry-forest form E. tanala-A (fig. 5). Moreover, a species-tree analysis of the four nuclear loci, which treated the Forêt d'Analamavo sample as a distinct entity, places this population as the sister-group to E. tanala-A with high posterior probability, not to E. tanala sensu stricto (fig. 6). This result is not driven solely by the X-linked locus: a species-tree analysis based on only the three diploid loci recovers the same relationship, albeit with lower posterior probability (PP = 0.77 vs. 1.0). The increased support for this relationship from the X-linked locus reflects expectations of the coalescent process, whereby genes on the X chromosome, which have a relatively lower effective population size than do those on autosomes should coalesce relatively more recently.

Multilocus Species Delimitation: When the Forêt d'Analamavo sample is treated as an equivalent species-level taxon in the nuclear-gene BPP analysis, it either forms a unit independent from all other Eliurus species or forms a unit with E. tanala-A; no evidence supports the combination of the Forêt d'Analamavo sample with any other Eliurus, including E. tanala proper, to form a species-level unit (table 1). This result bolsters the allocation of the Forêt d'Analamavo sample to E. tanala-A, notwithstanding the population's retention of a mitochondrial haplotype that relates it more closely to E. tanala. When samples from the Forêt d'Analamavo are excluded from analysis, the model with the highest posterior probability includes nine independent species-level units; this model receives high posterior probabilities in analyses that assume a small effective population size, regardless of assumptions about divergence time (PP = 0.989 or 0.964; table 1). For analyses that assume a larger effective population size, there is no clear evidence for this nine-species model over one that combines antsingy and carletoni into a single species-level unit (table 1). All nine species-level groupings are recovered with high posterior probability (PP >0.95) across most combinations of prior values for population size and divergence time (table 2). Even under the difficult scenario of large effective population size and shallow divergence times, BPP analyses uniformly provided strong evidence for recognition of E. tanala, the tanala-like population from the Northern Highlands (E. tanala-B), and an undescribed species from western landscapes (E. tanala-A) as separate entities.

FIG. 3.

The maximum-likelihood (lnL = -6820.4) tree of Eliurus resulting from a codon-partitioned analysis of 162 Cytb sequences (rooted with Gymnuromys roberti, not shown). Numbers at nodes are bootstrap support, support values <50% not shown; the dashed line indicates the phylogeny has been broken across two columns, the actual branch length is shown with a solid line (branch length units in substitutions/site as indicated by scale bar below). The arrow indicates the type specimen of E. ellermani (MNHN 1981.871). Terminal branches are labeled with locality name, locality number in parenthesis (appendix 1, fig. 16) and museum voucher number.

Continued.

FIG. 4.

The maximum-likelihood (lnL = -9767.1) tree of Eliurus resulting from a partitioned analysis of concatenated Cytb and Nd2 sequences (rooted with Gymnuromys roberti, not shown). The best partitioning scheme included four partitions: one each for first codon positions of Cytb and Nd2, one for all second codon positions, and one for all third codon positions. Numbers at nodes indicate bootstrap support; support values <50% not shown, 100% shown with solid circles. Terminal branches are labeled with locality name, locality number in parenthesis (appendix 1, fig. 16), and museum voucher number.

Patterns of Morphometric Differentiation

In the following sections, we first explore sexual dimorphism and age-related size differences, the two sources of nongeographic variation typically accessible with museum material. Subsequently, we use principal component analysis (PCA) to illuminate patterns of variation in two regions where populations of the E. antsingy and E. tanala complexes occur in sympatry or close allopatry. Finally, we apply discriminant function analysis (DFA) to assay differentiation among geographically defined OTUs of the E. tanala complex that covers its entire distribution.

Nongeographic Variation: According to one-way ANOVAs that designated sex as a categorical effect, males and females do not differ in cranial size within the large sample of Eliurus carletoni from the PHP de Loky-Manambato (table 3). The nearly identical skull size of males and females is remarkably fine: most continuous variables are equal, the remainder differing by only 0.1–0.3 mm.

In contrast, a majority of cranial dimensions of E. carletoni disclosed moderate to highly significant differences with age class as categorical effect in one-way ANOVAs (table 3). These variables are characterized by regular, incremental increases in mean size across the three adult age cohorts we defined, producing age-correlated differences that measurably affect variation within our sample. In general, F values and attained significance levels are greatest for the three largest variables (ONL, PPL, ZB) and those measured on the facial region (BIF, LR, WR, WZP); dimensions of the neurocranium (BBC, BOC, IOB) generally yielded lower F values. Measurements of the maxillary toothrow (LM1–3, WM1) constitute a notable exception to the pattern of age-correlated size increase; once erupted, the rooted molars of Eliurus may decrease in crown height during trituration, but do not grow in length and width (table 3). Some postweaning ontogenetic variation is usually captured by the first principal component, often characterized as a size factor, in craniometric studies of closely related (congeneric) muroid species. This relationship explains the elongated constellations of specimen scores commonly observed in plots of the first two principal components extracted, the major axis of each taxon ellipse typically oriented obliquely to the PC eigenvectors (e.g., Voss et al., 1990; Voss and Marcus, 1992; Carleton et al., 1999; Carleton and Stanley, 2012).

Given these results, we combined measurements of all adult males and females in computing descriptive statistics and conducting morphometric comparisons. We also united measurements of the three adult age classes (Y, A, and O), notwithstanding the potential variation introduced by postweaning cranial growth, rather than restrict ordination to a single age class and further diminish sample sizes. Although some age-related size increase may contribute to within-sample variation of Eliurus, it consistently emerged as negligible relative to those extracted factors that proved to be taxonomically informative.

Principal Component Analysis of Eliurus antsingy and E. tanala-A: Carleton et al. (2001) characterized the first individual (MNHN 1966.2221) provisionally identified as a western tanala-like form (herein as E. tanala-A) as large in cranial size, resembling the newly described E. antsingy. This cranial similarity is borne out by certain dimensions of the skull (ONL) and molars (LM1–3), a bivariate scatterplot that randomly intermixed specimens of E. tanala-A, including the series from the Forêt d'Analamavo, with those of E. antsingy (fig. 7A). Other cranial dimensions, however, offer sharp distinction between examples of E. tanala-A and E. antsingy, especially those involving the size of the incisive foramina (LIF) relative to the bony palate (LBP), as plotted in fig. 7B. The latter variables figured prominently in multivariate discrimination of the two western forms.

A PCA of all 18 craniodental variables produced nonoverlapping, unambiguous segregation of examples of E. antsingy from those of E. tanala-A (fig. 7C). Phenetic cohesion of both gene-sequenced specimens and nonsequenced specimens is apparent within each elliptical spread; the series from the Forêt d'Analamavo broadly intersects the constellation representing E. tanala-A, a placement consistent with analyses of nuclear-gene sequences (figs. 5, 6). Each constellation also encloses scores from localities where E. antsingy and E. tanala-A occur in sympatry; such specimens are appropriately grouped consistent with their genetic affinity (specimens not flagged in fig. 7C, but see appendix 1: localities 33, 40, 42). The nonsequenced holotype of E. antsingy (MNHN 1966.2220) is obviously associated with sequenced and nonsequenced specimens that have a similar cranial morphology, confirming application of that name to the clade recovered in our molecular trees (fig. 3) and in past investigations (Goodman et al., 2009; Rakotoarisoa et al., 2010). The nonsequenced specimen (MNHN 1966.2221) first reported as a western tanala-like form clustered with others later collected on the Bemaraha and Namoroka Massifs for which molecular data are available, but for which no name is currently applicable.

FIG. 5.

Individual nuclear-gene trees (above and opposite page) inferred during a species-tree analysis using *BEAST. In each case, trees are maximum-clade-credibility trees showing relationships among members of the Eliurus tanala-like clade (other species were included in the analysis but are omitted from the figure for clarity). Terminal branches are labeled with locality number in parenthesis (appendix 1, fig. 16) followed by museum voucher number; for heterozygous individuals, alleles are labeled “a” or “b” after the specimen number. Specimens were assigned species identity (shown by symbols) according to haplogroup membership in the Cytb phylogeny (fig. 3). Numbers at nodes are posterior probability values ≥ 0.99.

Continued.

Subsequent Oblimin rotation of the variable loadings comprehensibly decomposed age effects (RPC I) and taxon effects (RPC II). A majority of measurements loaded positively and highly significantly on RPC I (r = 0.6–0.9), except those taken on the molars (table 4A). Variation along this axis captured substantial within-sample age variation, as reflected by a post hoc one-way ANOVA with age class as effect (df = 2, 36; F = 26.6; P ≤ 0.001); on the other hand, taxon (sample) as categorical variable yielded insignificant results in a one-way ANOVA of RPC I scores (df = 2, 37; F = 2.4; P = 0.101). Discrimination between samples is clearly appreciated by those variables that strongly correlate with the rotated second component, which is oriented perpendicular to the major axes of the sample ellipses (fig. 7C); logically, a one-way ANOVA of RPC II scores with taxon as effect is highly significant (df = 2, 37; F = 118.0; P ≤ 0.0001). Compared with specimens of E. antsingy, the sign and magnitude of loading coefficients on RPC II (table 4A) describe the E. tanala-A skull as having a smaller neurocranium (BBC, BOC, IOB), shorter and narrower incisive foramina (BIF, LIF), but longer hard palate (LBP), and less inflated auditory bulla (DAB).

FIG. 6.

The maximum-clade-credibility species tree resulting from species-tree analysis of four nuclear genes in *BEAST. Numbers at nodes represent posterior probabilities, error bars are 95% credibility intervals on node height.

The sample from the Forêt d'Analamavo warrants additional comment in view of the contradictory relationships apparent in analyses of the mitochondrial genes (figs. 3, 4) and nuclear genes (fig. 5). Visually, the major axes of the three sample ellipses appear more or less parallel to one another, and the scatters of E. tanala-A and the Forêt d'Analamavo animals overlap appreciably based on their factor scores (fig. 7C). Statistically, the slopes of the major axes of all three samples were inseparable (df = 2, 37; F = 0.2; P = 0.849). The Y-intercept of E. antsingy contrasted substantially with both E. tanala-A (df = 1, 30; F = 157.1; P ≤ 0.001) and Forêt d'Analamavo (df = 1, 17; F = 149.2; P ≤ 0.001); whereas, the Y-intercept of the latter two samples failed to attain a significant difference (df = 1, 27; F = 2.0; P = 0.169). Based on the 18 cranial variables we measured and on the PCA performed, the Forêt d'Analamavo animals are quantitatively inseparable from other tanala-like specimens obtained from western dry forests. In the DFA reported below, we treat the series from Forêt d'Analamavo as part of the E. tanala-A OTU.

Principal Component Analysis of Eliurus carletoni and E. tanala-B: In the first taxonomically broad molecular study of Nesomyinae, Jansa et al. (1999) identified three specimens from the Northern Highlands (appendix 1: localities 1, 22) as Eliurus sp. B because of their substantial genetic divergence, revealed by Cytb sequences, from vouchers of E. tanala collected in the Central (appendix 1: localities 58, 69–71) and Southern highlands (appendix 1: locality 82). The much denser geographic sampling used herein reveals that this genetically distinctive, tanala-like form (E. tanala-B) is widely distributed in the Northern Highlands (fig. 3). As in western Madagascar, the northern tanala-like form may occur in sympatry or contiguous allopatry with a morphologically similar member of the E. antsingy group, E. carletoni, described by Goodman et al. (2009) from the Ankarana formation and distributed broadly within the Loky-Manambato region (Rakotoarisoa et al., 2010, 2012, 2013). Below we expand the morphological discrimination of E. tanala-B from E. carletoni based on a PCA of log-transformed, craniodental measurements, including most specimens used in our molecular analyses (figs. 3–5) along with the gene-sequenced holotypes of E. carletoni and E. ellermani.

TABLE 1

Results from eight different species-delimitation analyses in BPP

Analyses either include or exclude the Forêt d'Analamavo sample, and assume one of four combinations of priors¹ for divergence time (τ, shallow or deep) and effective population size (θ, large or small). Numbers indicate posterior probabilities for the species-delimitation model shown in the first column; results are shown for only the top-scoring models.

Two discrete clouds of specimen scores are evident in a scatterplot of the first two principal components extracted from ordination of all intact adult skulls representing E. carletoni and E. tanala-B (fig. 8). Whether based on sequenced or nonsequenced specimens, the separate groups are readily evident in morphometric space, and their membership conforms to the distinctive species-level clades recovered in the molecular trees (figs. 3, 4). Slopes of the major axis within each constellation are statistically equivalent (df = 1, 101; F = 0.2; P = 0.628), but their Y-intercepts differ markedly (df = 1, 101; F = 482.7; P ≤ 0.0001). The holotype of E. carletoni (FMNH 173105) is nestled predictably near the center of the E. carletoni ellipse; the holotype of E. ellermani (MNHN 1981.871) is associated with specimens of E. tanala-B, concordant with its affinity based on the partial Cytb sequence recovered from the MNHN prepared skin (see fig. 3).

Oblique rotation of PC correlations usefully enhanced interpretation of the extracted eigenvectors. As in the morphometric comparison of E. antsingy and E. tanala-A, nearly all variable correlations (loadings) with RPC I are large and positive (table 4B). Age class as a categorical effect explains much of the spread in a one-way ANOVA of RPC I scores (df = 2, 102; F = 27.3; P ≤ 0.001), but not taxon as a categorical variable (df = 1, 103; F = 0.3; P = 0.587). Conversely, age class contributed trivially to separation of specimens along RPC II (df = 2, 102; F = 0.3; P = 0.775), whereas taxon as group effect was singularly explanatory (df = 1, 103; F = 482.7; P ≤ 0.0001). According to the sign and magnitude of RPC II loadings (table 4B), the skull of E. tanalaB possesses relatively short and narrow incisive foramina (BIF, LIF) but a longer bony palate (LBP) compared with E. carletoni, a proportional relationship that mirrors results of the E. tanalaA vis-à-vis E. antsingy morphometric comparison. Nevertheless, the skull of E. tanala-B is overall stoutly constructed compared with that of E. carletoni, a size difference conveyed by the large, positive correlations of the largest cranial dimensions (ONL, ZB), the longer rostrum (BR, LD, LR), and broader hard palate (BM1s) and zygomatic plate (BZP).

TABLE 2

Results from species-delimitation analysis in BPP

The Forêt d'Analamavo sample was not included in this analysis, and four combinations of priors¹ were assumed for divergence time (τ, shallow or deep) and effective population size (θ, large or small). Numbers indicate posterior probabilities for recognizing the mitochondrial haplogroup in the first column as a species-level unit. Posterior probabilities > 0.95 are in bold.

Discriminant Function Analysis of the Eliurus tanala Group: DFAs conducted in earlier studies, using smaller sample sizes, were inconclusive in resolving the status of E. ellermani compared with E. tanala (Carleton, 1994; Carleton and Goodman, 1998). The locality samples now available substantially expand our geographic perspective for addressing differentiation among tanala-like populations broadly distributed in eastern humid forests, from the Southern to the Northern highlands, and for assessing that differentiation in view of the well-defined mitochondrial haplogroups evident across this same region (figs. 3, 4). Too few specimens of tanalalike animals from western tsingy-associated forests, E. tanala-A, were heretofore available to apply multivariate analyses; we included the series from the Forêt d'Analamavo within the E. tanala-A OTU because the BPP and PCA results, as reported above, demonstrated their genetic and morphometric inclusion.

TABLE 3

Sex and age variation in Eliurus carletoni

Arithmetic means (in mm) of 18 craniodental variables and results of one-way ANOVAs for sex and age cohorts in adult Eliurus carletoni collected within the PHP de Loky-Manambato region (N = 58, localities 7–15, 17; F and M = female and male, respectively; Y, A, and O = young, full, and old adult age classes, respectively; F (sex) and F (age) = univariate F values; see Materials and Methods for variable abbreviations)

The first two canonical variates extracted from the nine-group DFA summarized 72.7% of among-group craniodental variation (table 5), and a scatterplot of the two conveyed informative taxonomic and geographic structure (fig. 9). Eigenvalues of CVs 1 and 2 are ≥ 1, and multivariate means among the nine predefined groups differ significantly (Wilks' Lambda = 0.043; df = 8, 149; F = 3.6; P ≤ 0.0001). One-way ANOVAs on CV scores uncovered no significant age-class effect that influenced the dispersion of specimens along CV 1 (df = 2, 157; F = 2.9; P = 0.060) or CV 2 (df = 2, 157; F = 0.1; P = 0.944); the same one-way ANOVAs did disclose highly significant taxon (OTU) effects for both CV 1 (df = 8, 151; F = 40.6, P ≤ 0.001) and CV 2 (df = 8, 151; F = 24.2; P ≤ 0.001).

In general, CV 1 partitions the western moiety of the E. tanala group (E. tanala-A) from the eight OTUs distributed along the island's eastern side (E. tanala-B and E. tanala); whereas, CV 2 captures size trends among the eastern population samples (fig. 9). Two dimensions, interorbital width (IOB) and bullar size (DAB), load prominently on CV 1 (r ≥ 0.6), the former negatively and the latter positively (table 5), and primarily account for the isolation of E. tanala-A on that factor. The predominant influence of these two variables was unexpected in view of their nearly imperceptible difference when comparing skulls side by side; these same two variables logically generated the largest F values in univariate ANOVAs conducted across all nine OTUs. In a previous DFA iteration (not illustrated), we had treated the Forêt d'Analamavo sample as a separate OTU; this 10-group DFA affirmed the congruence of the Forêt d'Analamavo OTU with that of E. tanala-A, both set apart from the eight eastern OTUs along CV1.

FIG. 7.

Scatterplots of select raw cranial dimensions (A and B) and of extracted principal components (C) for examples of Eliurus antsingy (N = 11), E. tanala-A (N = 21), and the genetically enigmatic sample from Forêt d'Analamavo (N = 8). A, Bivariate plot of molar size (LM1–3) by occipitonasal length (ONL). B, Bivariate plot of palatal dimensions, LIF × LBP. C, Projection of specimen scores, based on all 18 log-transformed cranial variables, onto the first two principal components (RPC) extracted from ordination of all specimens representing the three samples, followed by oblique factor rotation (Oblimin). Confidence ellipses (1 SD) around sample centroids and the major axis of sample constellations were computed post hoc; highlighted specimen scores include the holotype of E. antsingy (MNHN 1966.2220) and the individual compared with E. tanala (MNHN 1966.2221) by Carleton et al. (2001). Closed symbols denote specimens used in gene-sequence analyses; open symbols represent nonsequenced individuals. See table 4A for variable correlations (loadings) and variance explained by the latent principal components.

In contrast to CV 1 loadings, a majority of variables correlate moderately and positively (r = 0.4–0.6) with CV 2, in effect grading OTUs by size, from smaller in the Southern Highlands (t6–PN de Befotaka-Midongy du Sud, t7–PN d'Andohahela) to larger toward the north (t1–PN d'Analamazaotra, E. tanala-B). The same geographic groupings are suggested by the pair-group associations portrayed in the phenogram generated from Mahalanobis' distances between OTU centroids (fig. 9C). The OTU E. tanala-B, represented by specimens from the RS d'Anjanaharibe-Sud and vicinity, is superimposed in morphometric space with OTU t1 (PN d'Analamazaotra) from the northern area of the Central Highlands (fig. 9A, B); their strong phenetic linkage is repeated in the cluster analysis (fig. 9C). Although our DFA involved more OTUs and larger sample sizes, these results reprise those of Carleton and Goodman (1998) regarding the lack of clear phenetic separation of population samples from the Northern Highlands. The indistinguishable morphometric footprint of our two northernmost OTUs, E. tanala-B and t1 (PN d'Analamazaotra), contradicts their unambiguous genetic separation, using Cytb and Nd2 sequences (figs. 3, 4), between examples broadly drawn from the Northern Highlands and those of E. tanala proper, the latter including specimens from the PN d'Analamazaotra and vicinity (OTU t1).

The disposition of certain specimens employed in the DFA deserves mention based on a posteriori calculations of their group membership. The CV score for the holotype of E. ellermani (MNHN 1981.871) is positioned approximately equidistant to the centroids of E. tanala-B and OTU t1–PN d'Analamazaotra (fig. 9B); consequently, its post hoc classification is statistically inconclusive, although the predicted membership is actually greater for t1–PN d'Analamazaotra (P = 0.76) than E. tanala-B (P = 0.23). The post hoc assignment of the holotype of E. tanala (BMNH 97.9.1.154) is most strongly affiliated with the OTU embracing its geographic origin, t3–Forêt de Vinanitelo; nonetheless, the probability of its membership, although high (P = 0.83), falls below a comfortable level of statistical certainty (i.e., P ≥0.95). The series from the Forêt d'Ianasana (Itremo), an isolated site at the western margin of the Central Highlands (see appendix 1: locality 55), was not defined as an OTU because of the small sample (N = 3), but all three individuals were entered as unknowns in the computation. Concordant with their relationship as divulged by Cytb (fig. 3), the specimens from Forêt d'Ianasana were classified among various OTUs of E. tanala, albeit none with a P ≥ 0.5 (FMNH 166053 with t7–PN d'Andohahela; 166054 with t6–PN de Befotaka-Midongy du Sud; 166055 with t5–RS du Pic d'Ivohibe); moreover, all three displayed negligible affinity with the OTUs E. tanala-B or t1–PN d'Analamazaotra.

TABLE 4

Results of two principal components analyses that compare regional samples of the Eliurus antsingy and E. tanala groups

A, E. antsingy and E. tanala-A in western Madagascar; B, E. carletoni and E. tanala-B in northern Madagascar. See figures 7 and 8 for plots of rotated principal component (RPC) scores and Materials and Methods for variable abbreviations.

Eliurus tanala Species Group

The genetic and morphometric results collectively support recognition of three species within the Eliurus tanala group, i.e., that node that defines the common ancestor of E. tanala-A, E. tanala-B, and E. tanala in the several molecular trees (figs. 3, 4, 6). Applicable names are available for two of those, E. tanala Major, 1896, for populations in the Central and Southern highlands, and E. ellermani Carleton, 1994, for those in the Northern Highlands (Eliurus tanala-B). The third well-defined clade, Eliurus tanala-A, represents several localities in dry forest associated with tsingy formations of western Madagascar and lacks a scientific name; we describe it below as a new species.

FIG. 9.

Three graphic depictions of discriminant function analysis performed on nine OTUs of the Eliurus tanala group, based on 18 log-transformed cranial variables as measured on 161 intact adult specimens. A. Projection of OTU centroids, each enclosed by maximally inclusive polygons of specimen scores, onto the first two canonical variates (CV) extracted. B. Projection of OTU centroids, each surrounded by 95% confidence envelopes of the n-dimensional mean, onto the first two canonical variates extracted. Highlighted scores in A and B represent the holotypes of E. ellermani (MNHN 1981.871) and E. tanala (BMNH 97.9.1.154), both represented by stars; figure B indicates the centroid placement of the seven OTUs (t1–t7) of E. tanala; see table 5 for variable correlations (loadings) and percent variance explained by the latent canonical variates. C. Cluster diagram (UPGMA) based on Mahalanobis distances between the nine OTU centroids (coefficient of cophenetic correlation = 0.959). Bootstrap values are indicated for nodes that were consistently recovered (≥ 70%, 10,000 iterations).

TABLE 5

Results of discriminant function analysis and univariate one-way ANOVAs based on nine OTUs representing populations of the Eliurus tanala group (N = 161)

See figure 9 for plot of first two canonical variates and Materials and Methods for variable abbreviations; F (OTU) = univariate F values.

As so defined, the E. tanala species group is narrower than the arrangement proposed by Carleton and Goodman (2007), who also included E. webbi. Although some authors have viewed E. tanala and E. webbi as closely related, perhaps sister species (Carleton, 1994; Jansa et al., 1999), other molecular studies have suggested that E. webbi is distantly related to the E. tanala assemblage (Rakotoarisoa et al., 2010, 2013; this study). Three other species groups, sensu Carleton and Goodman (2007), are represented among the species we considered—E. antsingy and E. carletoni, E. majori, E. myoxinus, and E. minor; all are distantly related to the E. tanala assemblage based on our gene trees (figs. 3, 4, 6). Although our results reinforce certain sister-group patterns disclosed in past studies (e.g., Jansa et al., 1999; Goodman et al., 2009), no study to date has included a sufficiently broad sampling of taxa or traits to resolve evolutionary relationships at the deeper nodes within the phylogenesis of Eliurus species. The improved understanding of species limits and diversity afforded by this study sets the stage for a more rigorous phylogenetic study of Eliurus, an investigation that must simultaneously address the relationships and generic validity of Voalavo.

The three species of the E. tanala group share the cardinal generic traits that are now well understood to differentiate Eliurus from other genera of nesomyine rodents (Ellerman, 1949; Carleton, 1994, 2003). Members of this species group can be distinuished from most other Eliurus species through a combination of qualitative differences in pelage and quantitative and qualitative differences in the cranium, mandible, and dentition (see Carleton and Goodman 2007, and comparisons below). However, members of the E. tanala species group are remarkably similar morphologically to members of the E. antsingy group (E. antsingy, E. carletoni), despite their relatively distant phylogenetic relationship as revealed by molecular data (figs. 3, 4, 6). Therefore, to clarify differences between these two species groups and reduce repetition in the species descriptions below, we first contrast traits shared by members of the E. tanala group with those of the E. antsingy group.

Eliurus antsingy Group vis-à-vis E. tanala Group

The five species belonging to these two species groups are among the largest forms within Eliurus, e.g., as indexed by the size of body (HBL ≈ 145–155 mm), skull (ONL ≈ 38–42 mm), or molars (LM1–3 > 5.5 mm); their univariate ranges overlap considerably, such that most individual variables offer little utility when used alone as key characters for species identification. Crania of members of the E. tanala group generally display relatively longer rostra (LR ≈ 36% of ONL) than do species of the E. antsingy group (LR ≈ 33% of ONL); however, certain other proportional features of the diastema and palate are more readily discernable (table 6).

Carleton and Goodman (2007: 15) emphasized the relatively long and wide incisive foramina possessed by E. antsingy in isolating the species in its own group. Accordingly, these variables (BIF, LIF), and their proportions relative to the upper diastema (LD) and bony palate (LBP), figured prominently in multivariate discrimination of crania of the two species groups, whether in the western or northern sectors of their distributions (figs. 7, 8; table 4). Thus, when identifying specimens from western tsingy-associated habitats, E. antsingy and the new species described below can be readily segregated by inspecting the size of the incisive foramina (LIF) relative to the bony palate (LBP), shorter in the latter species and longer in the former (fig. 10A); these proportional contrasts are also obtained between examples of E. carletoni and E. ellermani in northern Madagascar. Similarly, members of the E. tanala species group possess noticeably shorter incisive foramina (LIF) compared with diastemal length (LD) than do specimens of E. antsingy and E. carletoni (fig. 10B). Other cranial variables were less informative for consistently contrasting members of the two species groups, whether used in univariate comparisons (compare tables 7 and 8) or multivariate analyses (see table 4).

Carleton and Goodman (2007) also highlighted the length of the lower incisor alveolus and its terminal expression on the ascending ramus as another trait for separating species of the tanala and antsingy groups. In examples of the former, the alveolar tract of the lower incisor is longer, usually forming a knobby capsular projection that passes just beyond the base of the coronoid process and terminates immediately below the ventral rim of the sigmoid notch (fig. 11A; also see Carleton, 1994: fig. 20). In members of the antsingy group, the lower incisor is shorter, ending at the level of the coronoid process, appreciably below the sigmoid notch and forming an indistinct mound (fig. 11B). Other cranial characteristics of the species groups are here expressed as population tendencies, although we expect that careful scoring of character states would reveal quantitatively significant trait-frequency differences. Noteworthy are the degree of fenestration of the bony palate and the patency of the subsquamosal fenestra and correlative definition of a hamular process (table 6).

TABLE 6

Qualitative characters useful for identifying members of the Eliurus antsingy and E. tanala species groups

FIG. 10.

Proportional differences between the palatal regions in members of the Eliurus antsingy and E. tanala groups. A. (above) Ventral view of the bony palate and diastema: left, Eliurus tsingimbato n. s. (FMNH 172723, holotype; LM1–3 = 5.94 mm), an adult male from 3.5 km E Bekopaka, PN de Bemaraha; right, E. antsingy (FMNH 172721; LM1–3 = 6.09 mm), an adult male from 3.5 km E Bekopaka, PN de Bemaraha. The left photograph displays the landmarks defined for our LBP and LIF measurements, whose proportions offer distinctive ratios for separating the two species. Abbreviations: LBP, length of bony palate; LIF, length of incisive foramen; mpf, mesopterygoid fossa; mx, maxillary; pa, palatine; pmx, premaxillary. B. (opposite page) Dice-Leraas diagrams comparing expanse of the incisive foramen (LIF) relative to the diastema (LD) in species representing the E. antsingy (E. antsingy, E. carletoni) and E. tanala (E. ellermani, E. tanala, E. tsingimbato) groups. Symbols: horizontal line = sample mean; vertical line = observed range; open rectangle = ± one standard deviation; closed rectangle = ± two standard errors of the mean.

Continued.

Members of the E. antsingy group possess bushier, more visually prominent tail tufts compared with those of the E. tanala group, a contrast resulting both from elaboration of a tuft over a greater length of the tail and from the longer caudal hairs that compose the terminal tuft (fig. 12). The longer tuft in species of the E. antsingy group comprises about 45%–55% of tail length versus about 35%–40% in species of the E. tanala group (table 6). The tuft is monocolored, dusky brown to blackish, over its full length in nearly all specimens of E. antsingy (fig. 12E) and E. carletoni, but bright white hairs accentuate the terminal tuft in most examples of E. ellermani and E. tanala (table 6; fig. 12D). The new species is exceptional among species of the E. tanala group in typically having a tail tuft with a brown tip (fig. 12B, C; see species accounts for description of variation).

Species of the E. tanala Group

FIG. 11.

Lateral view (ca. 4 ×) of left mandible in adult representatives of two Eliurus species groups. A, E. tanala species group as represented by E. tsingimbato (FMNH 172723, holotype), a male from 3.5 km E Bekopaka, PN de Bemaraha; B, E. antsingy species group as represented by E. carletoni (FMNH 173105, holotype), a female from 7.5 km NW Mahamasina, RS d'Ankarana. Arrows indicate the posterior terminus of the lower incisor's alveolus within the ascending ramus of the mandible. Members of the E. tanala group (A) possess longer incisors, their posterior extension producing a protuberant capsule below the ventral rim of the sigmoid notch; in species of the E. antsingy group (B), the incisor's alveolus forms an indistinct mound set appreciably below the sigmoid notch.

FIG. 12.

Variation in tail-tuft development and color pattern in gene-sequenced adult specimens of three Eliurus species covered herein: A, E. tsingimbato (FMNH 172723, holotype, TL = 192 mm), a male from 3.5 km E Bekopaka, PN de Bemaraha; B, E. tsingimbato (FMNH 209224), a female from 4.9 km S Ambinda, Forêt de Beanka; C, E. tsingimbato (FMNH 175911, TL = 171 mm), a male from Forêt d'Ambovonomby, RNI de Namoroka; D, E. tanala (FMNH 170836, TL = 172 mm), a male from 15.5 km SE Vohitrafeno, Forêt de Vinanitelo; E, E. antsingy (FMNH 175913, TL = 169 mm), a female from Forêt de Mahabo, RNI de Namoroka. Specimens figured in B and C illustrate typical development of the tail tuft observed in E. tsingimbato; however, a few specimens, including the holotype shown in A, possess a subterminal band of white hairs, although the tip itself is dark brown. Drawings by Velizar Simeonovski.

FIG. 13.

Color image of the Eliurus tsingimbato holotype (FMNH 172723; TOTL = 355 mm; WT = 118 g), taken soon after its capture on limestone tsingy in the PN de Bemaraha. The slight reflective coloration of the dorsum, tail and hindfoot is associated with the camera flash. Photograph by Harald Schütz.

FIG. 14.

Variation in ventral pelage color observed among adult examples of the Eliurus tanala group (about 0.7×). A, E. tanala, a female from Vatoharanana, PN de Ranomafana (FMNH 170827; HBL = 157 mm); B, E. tanala, a male from 8 km NE Ivohibe (FMNH 162086; HBL = 145 mm); C, E. tanala, a female from 6.5 km ESE Ivohibe, RS d'Ivohibe (FMNH 162082; HBL = 145 mm); D, E. tsingimbato, a female from 4.9 km S Ambinda, Forêt de Beanka (FMNH 209224; HBL = 160 mm).

FIG. 15.

Dorsal, ventral, and lateral views of the skull and lateral view of the left mandible of the holotype of E. tsingimbato (FMNH 172723, ONL = 41.7 mm), an adult male from 3.5 km E Bekopaka, PN de Bemaraha, 100 m.

FIG. 16.

Map illustrating the distributions of the three species of the Eliurus tanala group recognized as valid (E. ellermani, E. tanala, E. tsingimbato), together with distributions of the E. antsingy group (E. antsingy, E. carletoni). Numbered collecting localities (see appendix 1) represent all specimens used in our morphological examinations, morphometric comparisons and/or molecular analyses; inset maps A–C illustrate regions where species occur in sympatry or close allopatry. Outlined stippled areas indicate major Mesozoic limestone regions (after Bésairie, 1964); elevations above 800 m are shown in shades of grey, grading to darker gray at higher elevations.

FIG. 17.

Plot of elevational ranges (in m above sea level) of taxa representing the Eliurus antsingy (E. antsingy, E. carletoni) and E. tanala (E. ellermani, E. tanala, E. tsingimbato) species groups. Vertical line embraces the elevational range of each species and filled circles indicate elevations of individual collecting localities, including any duplicate sites recorded for the same elevation (see Specimens Examined within species accounts and appendix 3).

FIG. 18.

Two views of western dry deciduous forest habitat in the Forêt de Beanka (near sites 37 and 38, appendix 1), where individuals of Eliurus tsingimbato were collected. A, General view of the lower portion of the forest, in close vicinity to a large exposed outcrop of sculpted karst limestone, a formation known as tsingy. The red ribbon in the lower left is a trail marker. B, A view of the understory of tsingy-associated forest with the characteristic vegetation, often composed of spiny plants belonging to the genus Pandanus, and shallow soils as witnessed by the exposed bedrock limestone. Photographs by Laurent Gautier.

TABLE 7

Descriptive statistics for selected OTUs of the Eliurus tanala species group

Statistics include the sample mean, ± one standard deviation, the observed range and sample size (in parentheses); measurements from the holotype for each species are presented under the specimen number. See Materials and Methods for variable abbreviations and geographic coverage of OTUs.

continude.

TABLE 8

Descriptive statistics for selected OTUs of the Eliurus antsingy species group

Statistics include the sample mean, ± one standard deviation, the observed range and sample size (in parentheses); measurements from the holotype for each species are presented under the specimen number. See Materials and Methods for variable abbreviations and geographic coverage of OTUs.

continude.

FIG. 19.

Two views of montane eastern humid forest where individuals of Eliurus ellermani were collected, in close vicinity to Ambohitsitondroina, PN de Masoala, 1078 m, a site situated in close proximity to Hiaraka (fig. 16: locality 29), the type locality of this species. A, Panoramic view from a neighboring ridge of the dense canopy of montane forest and the topographic features of the Masoala Peninsula. B, The understory of montane forest showing characteristic vegetation, often composed of plants belonging to the genera Cyathea and Pandanus, and epiphytic vegetation on tree trunks and limbs. Photographs by Voahangy Soarimalala.

Taxonomic Issues and Species Delimitation

Genetic, morphological, morphometric and distributional evidence provided concordant evidence that three species exist within the Eliurus tanala group, one described as new to science (E. tsingimbato) and two whose distribution and definition were substantially revised (E. ellermani, E. tanala). Analyses of both mitochondrial (figs. 3, 4) and nuclear (fig. 6, table 2) DNA sequences united samples of western E. tsingimbato in a clade isolated from those of the two eastern species E. ellermani and E. tanala. Morphological traits and morphometric analyses consolidated samples of E. tsingimbato apart from those of E. ellermani and E. tanala (figs. 9, 12). Moreover, morphometric (Mahalanobis D²) and genetic (uncorrected pairwise) distance measures are strongly correlated among six selected Eliurus species we evaluated, including those of the E. tanala group (fig. 20; table 9). Distributional records of E. tsingimbato uniformly document the species in low elevation, dry forests of western Madagascar, an environmental setting that sharply contrasts with the middle to high elevation occurrence of E. ellermani and E. tanala in eastern humid forests (figs. 16, 17, 19). These independent lines of evidence offer pluralistic and robust justification for the description of E. tsingimbato as a new species, consistent with an integrated taxonomic approach (e.g., Padial et al., 2010; Galimberti et al., 2012; Dávalos and Russell, 2014; Pante et al., 2015).

Notwithstanding the usual reciprocal illumination supplied by molecular and morphometric data, two areas of apparent conflict require discussion. One issue involved contradictory evidence for recognition of E. ellermani as a species distinct from E. tanala. A second dilemma concerned the species allocation of the population sample from Forêt d'Analamavo, in western Madagascar (fig. 16: locality 35), as inferred from mitochondrial genes (= E. tanala) or from nuclear genes and morphological similarity (= E. tsingimbato).

Evidence for recognizing E. ellermani as distinct from E. tanala is strongly supported by molecular data (including multilocus, coalescent-aware analyses of species limits), which unambiguously delineate populations in the Northern Highlands (E. ellermani) from those distributed across the Central and Southern highlands (E. tanala). Based on mitochondrial Cytb sequences, the holotype of E. ellermani definitively clustered within the large clade that represents localities in the Northern Highlands (fig. 3). On the other hand, our morphometric comparisons, like those of Carleton and Goodman (1998), offered inconclusive results that could be interpreted as a latitudinal size cline, with larger animals in the north and smaller ones in the south. In fact, the two northernmost samples—one genetically identifiable as E. ellermani (fig. 16: localities 22–29) and the other as E. tanala (fig. 16: localities 43, 49, 53)—are indistinguishable in morphometric space based on discriminant function analysis (fig. 9). Having eliminated a dark tail tuft as strictly diagnostic of E. ellermani, the most perceptible external contrast between the taxa involves their ventral pelage color, bicolored ventral hairs and overall grayish appearance in examples of E. tanala versus monocolored ventral hairs and creamy white underparts in E. ellermani. We underscore that the taxonomic status of E. ellermani vis-à-vis E. tanala is amenable to clarification based on additional collecting in remaining forests intermediate to the southernmost locality of E. ellermani, near Marotandrano (fig. 16: locality 30), and the northernmost site recorded for E. tanala, around the Sihanaka Forest (fig. 16: locality 45), a hiatus of some 200 km. Pending new analyses of such potentially transitional samples, we accept E. ellermani as a distinct species based principally on its singular genetic integrity.

FIG. 20.

Scatterplot of distance values derived from Cytb sequences (uncorrected pairwise distances) and discriminant function analysis (Mahalanobis D²) of six species of Eliurus plus the Forêt d'Analamavo sample (see table 9). “Intraspecies group” denotes values derived between species within the same species group (E. antsingy and E. tanala groups), including sister species; “Interspecies group” denotes values derived between species assigned to different species groups (E. antsingy, E. myoxinus and E. tanala groups).

Mitochondrial gene sequences (figs. 3, 4) surprisingly grouped the six individuals from the Forêt d'Analamavo, a dry-forest site in western Madagascar (fig. 16: locality 35), with samples of E. tanala, a species broadly distributed across eastern montane forests in the Central and Southern highlands (fig. 16). In contrast to the genetic affinity conveyed by mitochondrial loci, however, the X-linked gene affiliates the Forêt d'Analamavo specimens with examples of E. tsingimbato (E. tanala-A in fig. 5; PP = 1.00). Independent analyses of the three autosomal loci did not yield a comprehensible picture of species membership (fig. 5), but coalescent-aware, species-tree analysis of all four nuclear genes (fig. 6; PP = 0.99), as well as species-delimitation analysis based on these data (table 1), affirms the close relationship of the Forêt d'Analamavo series with E. tsingimbato (E. tanala-A). With their creamy white ventral pelage and dusky brown tail tuft, the morphological appearance of the Forêt d'Analamavo animals does not evoke an image of E. tanala. Nor are qualitative cranial traits and cranial morphometry equivocal: the Forêt d'Analamavo sample decidedly resembles other specimens of E. tsingimbato (figs. 7C, 9). In summary, these several independent lines of evidence persuasively explain the conflicting signal from mitochondrial DNA as an instance of incomplete lineage sorting (e.g., Avise et al., 1983; Pamilo and Nei, 1988; Hudson and Turelli, 2003). As in the case of E. ellermani and E. tanala, directed collecting may illuminate the stability and/or geographic scope of this tanala haplotype within E. tsingimbato. The Forêt d'Analamavo is situated only 14 km from the Forêt d'Analanomby (fig. 16: locality 36), where genetically and morphologically typical specimens of E. tsingimbato were obtained.

TABLE 9

Matrix of morphometric and genetic distances between species of Eliurus (see fig. 20), including the sample from the Forêt d'Analamavo

Above diagonal: uncorrected pairwise distances (x 100) based on Cytb sequences. Below diagonal: Mahalanobis D² values based on seven-group* discriminant functional analysis.

Although the evidence in combination is compelling for recognizing three species in the E. tanala group, hierarchical relationship among the three is not decisively supported in any phylogenetic analysis of the genetic data. The mitochondrial tree based on Cytb (fig. 3) and nuclear-gene species tree (fig. 6) both recovered E. tsingimbato and E. tanala as sister species, with negligible nodal support. The concatenated mitochondrial gene tree (fig. 4) instead portrays E. ellermani as sister to E. tanala, again with uninspiring support. Among the discrete morphological characters marshaled by Carleton (1994), E. tanala and E. ellermani share several derived traits—presence of white tail tuft, greater fenestration of the bony palate, reduction or closure of the subsquamosal fenestra—that suggest their more recent common ancestry compared with E. tsingimbato; although not strictly phylogenetic, craniodental morphometry demonstrates the closer phenotypic similarity of E. ellermani and E. tanala to one another and their equal degree of divergence from E. tsingimbato (fig. 9; tables 6, 9).

This weak phylogenetic resolution can be attributed to multiple factors, including a lack of informative sites, conflict among gene partitions, or closely spaced speciation events (Hoelzer and Meinick, 1994; Walsh et al., 1999; Winkler et al., 2015). In this case, the lack of coalescence for individual nuclear genes, along with the relatively short branches describing relationships among the three species, suggests that speciation was both recent and rapid (Hudson and Turelli, 2003; Rosenberg, 2003; Degnan and Rosenberg, 2009). A scenario of rapid speciation is consistent with the nature of our morphological diagnoses, which are polythetic formulations (or “cluster classes,” as used by Zachos, 2016). That is, traits cited as diagnostic are exhibited by the majority of specimens within locality samples, but not one is necessarily fixed within the species (class); their diagnostic value is operationally realized when used in combination with one another to effect a unique taxonomic definition.

A case in point, as exemplified by our experience with the E. tanala group, involves the color of the distal tail tuft, the presence of a terminal brush of elongated hairs being a cardinal trait of the genus (fig. 12). Based on two specimens, Major (1896a) had overlooked the significance of a white terminal tuft in his description of E. tanala; based on two specimens, Carleton (1994) emphasized possession of a fully dark tuft as central to his diagnosis of E. ellermani. The accumulation of vastly improved species samples (see fig. 1), in number and in geographic coverage, has revealed that a white caudal pencil is common within both species, each strongly supported by inclusive molecular clades that represent many localities over broad geographies. In retrospect, the underappreciation of population variation or stability in terminal tuft color is partially attributable to the encumbrance of taxonomic history. In the early descriptive era of Malagasy biodiversity, Oldfield Thomas and C.I. Forsyth Major had meager museum series at hand to forge their new species discoveries. Even by 1994, Carleton documented about 200 specimens in his generic revision of all species of Eliurus. Our current study utilized some 370 vouchered specimens to clarify the systematics of three species, representing one of five species groups within the genus. As integrated with genetic data, qualitative cranial traits and ventral pelage color, possession of a white caudal tuft has emerged as an informative feature for identification of two of those species, E. ellermani and E. tanala.

Biogeography

Several hypotheses have been advanced to explain the patterns and processes of diversification of species endemic to Madagascar, as synthetically reviewed by Vences et al. (2009). A refined taxonomic foundation, derived from a well-documented understanding of species limits and their distributions, is integral to testing these hypotheses. Fortunately, the taxonomic framework is much improved for Madagascar's native rodent fauna (Nesomyinae), in particular Eliurus, thanks to the last two decades of directed biological exploration (e.g., Goodman and Rasolonandrasana, 1999; Goodman et al., 2007, 2013b), revisionary work (e.g., Carleton, 1994; Carleton et al., 2001; Carleton and Goodman, 2007), and phylogenetic study (Jansa et al., 1999; Jansa and Carleton, 2003; Goodman et al., 2009). These synergistic efforts have substantively enhanced our knowledge of how many nesomyine species exist, where they occur among the island's habitats, and how this diversity may have arisen.

Species of Eliurus are scansorial or semiarbo-real rodents and with few exceptions inhabit native forests, a behavioral habit and ecological reliance that make the genus an excellent model to examine phylogeographic structure and speciation patterns associated with Madagascar's complex geology, topography, and climate (Wilmé et al., 2006; Muldoon and Goodman, 2010; Rakotomalala and Goodman, 2010; Shi et al., 2013). Although discovery of the genus Eliurus and its type species, E. myoxinus, emanated from a western environment (Milne Edwards, 1885), descriptive activity over the next 120 years would concentrate on eastern humid forests, where 10 new species were eventually recognized. Not until 2001 was a second species of Eliurus, E. antsingy (Carleton et al., 2001), described from western dry forest, followed shortly thereafter by documentation of E. carletoni (Goodman et al., 2009). Both taxa are largely confined to dry forests that overlie areas of Mesozoic limestone, a geological stratum that is intermittently exposed along the island's western lowlands and extends to the far north and northeast (figs. 2, 16). At certain areas across this zone, the limestone formations are eroded to form unusual karst pinnacles, also known as tsingy (Bésairie, 1964; Du Puy and Moat, 1996; Veress et al., 2008).

Dry forests associated with tsingy formations contain a highly endemic biota whose diversity is only beginning to be appreciated (Rajeriarison et al., 2000; Cardiff and Befourouack, 2003; Bora et al., 2009; Glaw et al., 2009; Goodman et al., 2011, 2013b). Eliurus tsingimbato, the new species documented herein, represents another member of this unique biota, the fourth nesomyine rodent known to be endemic to tsingy environments, joining E. antsingy, E. carletoni, and Nesomys lambertoni.

Description of the new species E. tsingimbato from the island's dry forests brings the number of Eliurus known from western landscapes to three, including E. antsingy and E. myoxinus. Two of these, E. antsingy and E. tsingimbato, have been documented only in dry forests associated with tsingy (fig. 16); one, E. myoxinus, is more broadly distributed in western Madagascar, inhabiting not only tsingy-associated dry forest but also spiny-bush forest in the south, dry deciduous forest in the west and humid forest with an extended dry season in the north (Goodman et al., 1999, 2009; Carleton et al., 2001; Soarimalala and Goodman, 2003; Shi et al., 2013). Each of the three western Eliurus has been provisionally allocated to different species groups (Carleton and Goodman, 2007); none of the three was recovered as sister species in phylogenetic analyses of molecular data (figs. 3, 4, 6). These phylogenetic results indicate at least two independent origins of tsingy-forest denizens among Eliurus species, once in the E. tsingimbato lineage and once in the ancestor of the E. antsingy species group, E. antsingy and E. carletoni. The latter two species are allopatric, latitudinally isolated in tsingy-associated forests of the west (Bemaraha and Namoroka formations) and north (Ankarana formation).

Distributional records of sympatry between members of different species groups within Eliurus are commonplace: for instance, the cooccurrence of E. tsingimbato with E. antsingy and E. myoxinus in western Madagascar; of E. ellermani with E. carletoni or E. majori in northern Madagascar; of E. tanala with E. majori and E. minor in the Central Highlands. To date, there are few records of strict syntopy between species within the same species group, such as that recorded for E. minor and E. myoxinus along the northwestern slopes of the PN de Marojejy (Soarimalala and Goodman, 2003); in most species groups, member species geographically replace one another (Goodman et al., 2013a). Demonstration of sympatry usefully presupposes isolated gene pools and removes ambiguity about species status. Within Eliurus, genetic distances calculated between species of different groups exceed by twofold those derived between species within a particular group (table 9; fig. 20).

Finally, patterns of phylogeographic structure within species groups of Eliurus, as so far understood, conform to three of the species-diversification patterns distilled by Vences et al. (2009). All three invoke a process of geographic isolation instigated by bioclimatic oscillations, ultimately culminating in vicariant speciation (their ecogeographic constraint model also considers parapatric speciation across a steep east-west environmental gradient).

(1) According to the ecogeographic constraint model of Vences et al. (2009), a phylogenetic division separates sister species or clades in western and eastern Madagascar, respectively corresponding to dry- versus humid-forest associations. Isolation of E. tsingimbato in western dry forest, wholly allopatric to the eastern humid-forest clade E. tanala and E. ellermani (fig. 16), is consistent with their east-west phylogeographic pattern, which suggests origination from a formerly widespread, habitat-generalist species, subsequent adaptation to locally arid (western) or mesic (eastern) conditions, and finally divergence into two species clades. Preliminary gene-sequence evidence indicates that E. myoxinus, broadly distributed in western dry forest, and E. minor, equally broadly distributed in eastern humid forest, may also comprise an east-west kinship group (figs. 3, 4; also see Jansa et al., 1999: fig. 1; Rakotoarisoa et al., 2010: fig. 2); although this relationship requires additional investigation with a broad geographic and taxonomic sampling of nuclear genes.

(2) Vences et al. (2009) applied their montane refugia model to north-south divisions within the eastern humid forest, whereby widespread populations restricted to high elevations become isolated in neighboring mountain systems during geological periods of drier climate. Within the E. tanala group, differentiation of the eastern clade, E. ellermani and E. tanala, plausibly transpired within such montane refugia, centered in the Northern versus Central highlands and separated by the lowlands forming the Mandritsara Basin (fig. 1). According to this view, the record of E. ellermani in the northernmost Central Highlands, to the south of the Mandritsara Basin (fig. 16: locality 30), would represent an instance of secondary population expansion. Because divergence among E. tanala, E. ellermani, and E. tsingimbato appears to be coeval (fig. 6), their initial isolation may be attributable to a single, recent episode of climatic drying; however, a rigorous test of this hypothesis will require a more comprehensive, time-scaled phylogeny.

(3) Vences et al. (2009) identified a second east-west diversification mechanism, involving the formation of western rainforest refugia. In this model, a widespread, humid-forest species becomes isolated in relictual pockets of western humid forest during cool, dry climatic periods, eventually differentiating from sister taxa in the continuous block of eastern humid forest. Speciation among members of the E. majori species group (E. danieli, E. majori, E. penicillatus) merits consideration in this context. Available molecular data have allied E. danieli, isolated on the Isalo Massif in south-central Madagascar, with E. majori, a montane species broadly distributed in eastern humid forest (Jansa et al., 1999: fig. 1, sp. A = E. danieli; Goodman et al., 2009: fig. 6). Here again, taxon and geographic sampling must be augmented, in particular to include examples of E. penicillatus, a species localized along the eastern flank of the Central Highlands (also see Carleton and Goodman, 2007: 13).

With respect to conservation issues, Eliurus tsingimbato is fortuitously restricted to forest habitats that occur on limestone base rock, several of which are now incorporated within Madagascar's system of protected areas. This official protection, and the general lack of agricultural productivity on tsingy karst, should minimize human despoliation of these western dry forests, factors that bode well for the conservation of this distinctive species into the near future. However, none of Madagascar's pristine forests, including those within protected areas, is proving exempt from the recent wave of hardwood exploitation; such overharvesting would impinge on the limited range of this species, as well as many other tsingy-dependent taxa, and impact survival of local populations.