Institutions and Specimens

Specimens cited by institutional abbreviations and catalog numbers are stored in the collections of the Amathole Museum (formerly the Kaffrarian Museum), King William's Town (AM); American Museum of Natural History, New York (AMNH); Natural History Museum, London (BMNH); Carnegie Museum of Natural History, Pittsburgh (CMNH); Field Museum of Natural History, Chicago (FMNH); Museum of Comparative Zoology, Harvard University, Cambridge (MCZ); Muséum National d'Histoire Naturelle, Paris (MNHN); Michigan State University, East Lansing (MSU); the Natural History Museum, Bulawayo, Zimbabwe (NHMZ); Royal Belgian Institute of Natural Sciences, Brussels (RBINS); Royal Museum for Central Africa, Tervuren (MRAC); National Natuurhistorisch Museum, Leiden (RMNH); Université d'Anvers, Rijksuniversitair Centrum Antwerpen, Antwerpen (RUCA); Transvaal Museum, Pretoria (TM); National Museum of Natural History, Smithsonian Institution, Washington, D.C. (USNM); Staatliches Museum für Naturkunde Stuttgart, Stuttgart (SMNS); Zoologisches Forschungsinstitut und Museum Alexander Koenig, Bonn (ZFMK); and Zoologisches Museum und Institut für Spezielle Zoologie, Museum für Naturkunde der Humboldt-Universität zu Berlin, Berlin (ZMB). The abbreviations preface catalog numbers referring to specimens listed in the gazetteer, tables, text, and figure legends. Most of the material consists of standard museum preparations (a stuffed skin accompanied by a cranium and mandible); a few specimens are preserved in fluid.

Measurements

Values (in millimeters) for total length and length of tail are those recorded by collectors on labels attached to skins; subtracting length of tail (LT) from total length produced a value for length of head and body (LHB). Proportion of tail length relative to length of head and body is expressed in percent derived from LT/LHB. Values for length of hind foot (LHF), which includes claws, were obtained either from skin labels or from Holden's measurements of dry study skins and specimens preserved in fluid; those for length of ear (LE) come from collector's measurements notated on skin labels (presumably they represent greatest length, from the notch to the distal margin of the pinnae). Weights in grams (WT) were sometimes recorded by collectors and these are included in some descriptive statistics.

With the aid of a dissecting microscope, Holden used handheld digital calipers to measure (values were recorded to the nearest 0.1 mm) the following cranial and dental dimensions listed in the sequence they appear in tables:

Limits of these measurements are illustrated in figure 1 and reasons behind their selection are elaborated by Holden (1996; one dimension shown in the figure, LIP, length of interparietal, was not measured). Only the cranial and dental measurement values were incorporated in multivariate analyses; values from external measurements portray general body size and lengths of appendages.

Figure 1

Limits of cranial and dental measurements imposed upon the outline of an adult Graphiurus christyi. Abbreviations are explained in the text.

Age and Sexual Dimorphism

Data used in the tables of descriptive univariate statistics and multivariate statistical analyses were derived from “adults” only, which includes the following three age classes.

Young adult: Occlusal surface of fourth premolar is level with surfaces of the molars. The premolar and molars show slight or no wear.

Adult: Occlusal surfaces of all teeth are moderately worn (lingual and buccal facets are worn, transverse ridges are worn but not obliterated).

Old adult: Chewing surfaces of all teeth are very worn. The labial margins are often sharply beveled or eroded, many or all of the transverse ridges are obliterated or distorted, and the teeth appear flattened, worn nearly to tops of their roots.

Although stage of tooth eruption and wear were the primary criteria for assigning specimens to the age classes, we also evaluated degree of fusion of cranial sutures along with pelage characteristics. All specimens identified as “adults”, for example, have shed the juvenile coat and are clothed in full adult pelage. Uniting such a broad range of adult stages was the only way to obtain adequate samples for analysis; “series of African dormice are, with a few exceptions, characteristically small and from scattered localities” (Holden, 1996: 7).

The multivariate statistical analyses and tables of descriptive statistics are derived from samples containing both sexes combined. Males and females of most species of Graphiurus are about equal in body size and closely similar in morphological features not associated with sexual traits. Our data demonstrate no consistent patterns of sexual size dimorphism within available samples of the species of Graphiurus discussed here (see Holden, 1996, for a broader discussion covering the nature of intergender sexual variation within samples of Graphiurus).

Statistical Analyses

We calculated standard univariate descriptive statistics (mean, standard deviation, and observed range). Values are listed in table 4 for the sample of G. walterverheyeni, n. sp., and samples of G. crassicaudatus, G. lorraineus, and G. surdus to which the new species is compared.

Multivariate statistical results take the form of principal components analyses employing values from 11 (fig. 11), 13 (fig. 12), or 15 (fig. 13) of the 24 cranial and dental measurements taken from adults. The reduction from 24 was necessary because the incidence of broken skulls is high and complete sets of measurements for the samples were available only for those subsets of variables; also a very few measurements were excluded from the analyses because they revealed no discriminatory resolution (usually highly correlated with other measurements). All measurement values were transformed to natural logarithms, and principal components were extracted from a covariance matrix. Plots of specimen scores in cluster analyses provided visual patterns that reflect similarity or contrast in the combined cranial and dental dimensions surveyed among samples of different species, each composed of specimens from more than one collection locality. SYSTAT® software for Windows® (version 11.00.01, 2004) was used for processing all morphometric analyses.

Ratio Diagram

Proportional craniometric relationships among three species of Graphiurus are illustrated using the ratio diagram in figure 10. For each measurement, the absolute value of the mean and plus and minus two standard errors of the mean were converted to logarithms. For each dimension, the logarithm of the mean of the standard (G. crassicaudatus from Cameroon, Nigeria, Togo, Ghana, Côte d'Ivoire, Liberia, and Equatorial Guinea; N  =  36) was subtracted from the logarithm of the mean of G. lorraineus (from Cameroon, Gabon, D.R. Congo, and Equatorial Guinea; N  =  96) and the new species (from the central Congo Basin; N  =  2; the holotype and paratype). The logarithms of plus and minus two standard errors of the mean of the standard were subtracted from the logarithms of plus and minus two standard errors of the mean of the two comparative species. A species with the same proportions as the standard will be represented by mean values on a line parallel to that of the standard regardless of absolute size. If values for the species being compared with the standard are similar in absolute size, they will be close together on the diagram. Proportional differences will be indicated by deviation from a comparative species' hypothetical parallel line along the horizontal scale. Data are derived from values for mean, standard deviation, and sample size of variables listed in table 4.

Anatomical Definitions

Most anatomical terms for cranial foramina and osseous elements are illustrated and described for dormice by Wahlert et al. (1993). Other features (sphenopterygoid vacuities, mesopterygoid and pterygoid fossae) are depicted in Musser and Holden (1991: 346).

Ecological Niche Modeling

Sources of data and algorithms used to generate the predicted distribution maps in figures 14–Figure 1516 and pertinent discussion are described below.

Provenances for the specimens of Graphiurus: Collection localities for G. crassicaudatus, G. walterverheyeni, n. sp., “true” G. lorraineus (samples from west-central, south-central, and east-central forest regions), and the G. lorraineus complex were compiled through Holden's comprehensive examination of preserved specimens (skins, skulls, and fluid-preserved material) in the 18 museums listed in the Institutions and Specimens section. Limits of species are derived from qualitative anatomical data combined with morphometric information. Decimal longitude and latitude coordinates, accurate to at least one minute, of all collection localities were determined utilizing one or more of the following sources for each specimen: localities and coordinates recorded on specimen tags by collectors; collectors' field notes; personal communication with collectors, and scientific publications providing coordinate data and maps of Graphiurus collecting localities (referenced in Holden, 2005, in press). Coordinate data were corroborated when possible by utilizing the Alexandria Digital Library Gazetteer (1999–) and other reference sources. Collection localities were considered only once with no weighting to account for multiple collections of the same species at a single place, as our goal in performing ecological niche modeling (ENM) was to identify only the species (or species complex) distribution and not its abundance within any zones of favorable habitat. The final set of data derived from collection localities contained a total of 29 (fig. 14), 2, 26 (fig. 15), and 106 (fig. 16) verified and unique points (each referring to a georeferenced collection site) for G. crassicaudatus, G. walterverheyeni, true G. lorraineus, and the G. lorraineus complex, respectively. These localities are listed in the gazetteer for G. crassicaudatus, G. walterverheyeni, true G. lorraineus, and the G. lorraineus complex.

Geographic and environmental data: Thirteen environmental data parameters were selected as ecological input layers in ASCII raster grid format for ENM using Genetic Algorithm for Rule-set Production (GARP) software (Version 1.1.3, 2003). All data layers were resolved (pixel size) to ∼10 × ∼10 km (0.1 × 0.1°) for analysis. Of these 13 parameters, seven were annual climatic data layers averaged over the period 1961–1990 and obtained from the Intergovernmental Panel on Climate Change (IPCC; native resolution ∼50 × ∼50 km;  http://ipcc-ddc/cru.uea.ac.uk/): mean temperature, mean maximum temperature, mean minimum temperature, diurnal temperature range, mean precipitation, wet days, and ground frost days (see table 1). Five land-surface data layers were obtained from the U.S. Geological Survey's Hydro-1K data set (native resolution 1 × 1 km;  http://edc.usgs.gov/products/elevation/gtopo30/hydro/): elevation, flow direction, flow accumulation (number of pixels in watershed), topographic aspect (direction of pixel slope), and compound topographic index (a measure of the tendency of water to pool). The remaining data layer was obtained from the University of Maryland Global Land Cover Facility ( http://glcf.umiacs.umd.edu/index.shtml): land use/land cover.

Table 1

Annual Values for Environmental Variables at Geo-referenced Collection Localities (Occurrence Points) for Graphiurus crassicaudatus^a (29 locations), G. walterverheyeni^a (2 locations), and G. lorraineus^b (26 locations) in Equatorial Africa

(Data are summarized by mean, followed by median in parentheses, and range.)

Modeling method: We applied a machine-learning algorithm to model the ecological niches occupied by two of the three dormouse species of interest, G. crassicaudatus and G. lorraineus (both true G. lorraineus and the G. lorraineus complex; see discussion below), and predict their geographic distributions based on the spatial locations of this niche, where the ecological niche is defined as set of environmental conditions within which the species is able to maintain populations without immigration (Grinnell, 1917; Holt and Gaines, 1992). Ecological niche modeling of the third species, G. walterverheyeni, n. sp., was not attempted due to small sample size (only two known occurrences). Ecological niche models were created using Genetic Algorithm for Rule-set Production (GARP) software (Version 1.1.3, 2003), an algorithm specifically designed for predicting and modeling the ecological niches of species (Stockwell and Peters, 1999), and publicly available for download as a desktop application ( http://www.nhm.ku.edu/desktopgarp/).

GARP models ecologic niches of species by relating the previously described point-occurrence data to the environmental data grids and produces a diverse set of logical, range, and logistic rules that describe the potential distribution of species in ecological dimensions (Stockwell and Peters, 1999). As a genetic algorithm, it is designed for problem solving based on “natural selection,” and therefore the rules describing the species niche are automatically evolved using simulated processes of mutation, recombination, and selection, thereby creating an optimized set of distribution rules which subsequently predict the potential distribution of the species in geographic space (Peterson et al., 2002). GARP models have been shown to have predictive value that is statistically significant by both rigorous intrinsic tests, performed on a subset of the data during model development, and extrinsic tests, performed using data not included in model development (Peterson, 2001: Levine et al., 2004a, 2004b, 2007; Benedict et al., 2007).

We created 10 GARP models for each of the two dormouse species and summed them using ArcView GIS 3.3 ( http://www.esri.com/) to create each of the final prediction maps. Therefore, individual pixels of all maps were assigned values of 0–10 representing the number of models that predicted ecological suitability in that pixel, which are represented by color gradations. Darker shades of color indicate the increasing proportion of times out of 10 that specific areas (pixels) were included in the predicted distribution. All models were created using a maximum of 100,000 iterations and a convergence criterion (degree of change between each model during the development process) of 0.0001.

Holotype and Type Locality

The holotype (BMNH 27.3.1.35) is a young adult female collected by Douglas C. Fraser (original number 46) on 21 October 1926. The specimen preparation consists of a stuffed skin, cranium, and mandible. The skull and mandible are in good condition (fig. 4) as is the skin, although the limbs are slightly contorted and the pinnae are flattened (fig. 3). The tail is unnaturally short or “bobbed” due to an injury sustained during the animal's life; that the terminal tail hairs had time to cover this region and undergo a change in pigmentation (the end of the bobbed tail is tipped with white hairs) indicates that the injury did not occur immediately prior to or during this animal's capture. Holden's personal experience derived from studying hundreds of specimens of Graphiurus confirms that tail injuries are often accompanied by a bushy, white-tipped growth of hair over the truncated tail tip. This condition has also been recorded for Eliomys by Miller (1912) and for Muscardinus by Juškaitis (2006). External, cranial, and dental measurements of the holotype are listed in table 2.

Figure 4

Cranial views of the holotype of Graphiurus walterverheyeni (BMNH 27.3.1.35) captured near the Lukenie River, Democratic Republic of the Congo. Values for cranial and dental measurements are listed in table 2. ×3.5.

Figure 5

Cranial views of the paratype of Graphiurus walterverheyeni (MRAC 88-005-M12), a subadult male from Wafanya, Democratic Republic of the Congo. Values for cranial and dental measurements are listed in table 2. ×3.5.

The type locality, noted on the specimen tag by Fraser, is “River Lukenie (Meridian 22.30E), 500 meters,” with the additional annotation that the dormouse was “caught close to the river.” No village name was given for a point of reference; the coordinates are approximated to be 03°30′S, 22°30′E (fig. 2Figure 2(Continued.)). The Lukenie River flows westward to its confluence with the waters draining Lac Mai-Ndombe; after this junction the river is known as the Fimi, which later joins the Kasai River; together these form the westward flowing Kwa River, one of the large tributaries of the Congo River.

Referred Specimen

Only one other specimen of G. walterverheyeni is known to exist (MRAC 88–005–M12), a young adult male, captured at “Wafanya” (Wafania), 01°21′S, 20°20′E, Democratic Republic of the Congo, on 4 December 1987 by E.P.P. Lootens (original number 2110). Wafania is located near the left bank of the Luilaka River, an eastern tributary of the Ruki River that flows westward into the Congo River. The specimen consists of a skull (fig. 5), mandible, and carcass in alcohol.

Figure 6

Dorsal views comparing adult crania from three species of Graphiurus. Top left: G. crassicaudatus (holotype of G. c. dorotheae, BMNH 12.1.12.1, southern Nigeria). Top right: G. walterverheyeni (paratype, MRAC 88-005-M12, Wafania D. R. Congo). Bottom left: G. lorraineus (holotype, BMNH 7.7.8.115, D.R. Congo). Bottom right: G. lorraineus (cf. haedulus, CMNH 17921, southern Cameroon). ×3.

Diagnosis

Graphiurus walterverheyeni shares several morphological similarities with Graphiurus crassicaudatus (for example, a stubby snout, short rostrum and nasals, broad skull, vaulted braincase, and wide interorbital and postorbital regions with their dorsolateral margins delimited by ridges), but is most easily distinguished from that species by its overall smaller size, which is reflected in the often markedly absolutely smaller external, cranial, and dental dimensions (tables 2–Table 34, figs. 6–Figure 78), and proportional differences (fig. 10).

Figure 7

Ventral views of the same crania from Graphiurus illustrated in figure 6. Top left: G. crassicaudatus dorotheae. Top right: G. walterverheyeni. Bottom left: G. lorraineus. Bottom right; G. lorraineus cf. haedulus. All ×3.

Figure 8

Lateral views of the same crania from Graphiurus portrayed in figures 6 and 7. Top left: G. crassicaudatus dorotheae. Top right: G. walterverheyeni. Bottom left: G. lorraineus. Bottom right; G. lorraineus cf. haedulus. ×2.7.

Figure 9

Scanning electron micrographs of left maxillary (left pair) and right mandibular (right pair) tooth rows from adult Graphiurus. Left row in each pair is G. lorraineus (CMNH 42203, southern Cameroon); the right row of each pair is G. crassicaudatus (ZMB 71325, Côte d'Ivoire). Approximately ×21. Occlusal patterns of the teeth in G. crassicaudatus resemble those of G. walterverheyeni.

Geographic Distribution

G. walterverheyeni is known only from the two localities described above, one along the Lukenie River, the other near the Luilaka River, both places within an area circumscribed by the western arm of the Congo River and by its large eastern tributary, the Lomami River. The species may be endemic to that vast equatorial tropical rainforest stretching south from the great bend of the Congo River between its western arm and the Congo-Lualaba River to the east within the Central Congo Basin, an area known as the south-central forest region (Grubb 1978, 1982, Grubb, 2001, and references therein; Colyn, 1991; Colyn and Van Rompaey, 1994; Happold 1996; White, 2001) and the Salonga-Lukenie-Sankuru landscape (Congo Basin Forest Partnership, 2006).

Etymology

The patronym honors the late Dr. Walter Verheyen, whose extensive field, laboratory, and museum research on African mammals, particularly rodents, resulted in many significant and influential publications in the fields of ecology and systematics. Verheyen generously provided Holden access to specimens and data during her research on Graphiurus systematics.

Description and Comparisons

The description is presented in the context of comparisons among G. walterverheyeni, G. crassicaudatus (its closest morphological and possible phylogenetic relative), and the sympatric true G. lorraineus, which it resembles in body size. In its external features, G. walterverheyeni superficially resembles both G. crassicaudatus and G. lorraineus. It has a much shorter head and body than G. crassicaudatus, and averages slightly smaller than G. lorraineus, and the hind feet are shorter relative to length of head and body than is typical of either G. crassicaudatus or G. lorraineus (fig. 10, table 4). The dorsal pelage is uniformly dull rufous brown, soft, and short; it is composed of soft overhairs (4 mm long over the body) and short, fine guard hairs (6–7 mm long near the rump) inconspicuously projecting beyond the overhair layer. Overhairs are steel gray for most of their length, and tipped with a rusty-yellow band, which darkens to a darker brown tip on some hairs; guard hairs are dark brown. The skin, while in good condition, was perhaps molting at the time of capture, so that many gray bases of the overhairs are visible; it is likely that color of the dorsal pelage is similar to the richer rufous brown of G. crassicaudatus. The short, soft ventral coat is dark gray washed with buff. Dorsal and ventral pelage colors are not clearly delineated, a character shared with most specimens of true G. lorraineus; most specimens of G. crassicaudatus exhibit clear delineation of the dorsal and pelage colors. Head coloration is similar to that of the dorsal pelage, and each eye is encircled by an inconspicuous, narrow eye ring. Coloration of the tail generally matches that of the dorsal pelage, although white hairs are scattered conspicuously throughout the tail, and on the holotype (which has a bobbed tail –see discussion under Holotype and Type Locality) the tail is conspicuously tipped in white due to prior injury. The external pinnae are short and ovate, clothed in fine, dark brown hairs, and the pelage color behind the ears matches that of the dorsal pelage (postauricular patches are not present; they are exhibited in some samples of G. lorraineus and other species of Graphiurus). Dorsal surfaces of the hind feet are white with a dark metatarsal streak. No mammae count is available from the single female specimen, but it is presumed to be concordant with that of all other species of Graphiurus as well as other glirids: females usually have four pairs of nipples (one pectoral, one brachial, and two inguinal), although some variation has been documented in the non-African Glis (Kryštufek, 2004).

Comparison with G. crassicaudatus

The skull of G. walterverheyeni superficially connotes a miniature of G. crassicaudatus, but exhibits contrasts related to proportions of certain cranial and dental dimensions (figs. 3, 4). Although G. walterverheyeni is smaller than G. crassicaudatus in nearly all cranial and dental dimensions (table 4), the two species share: a short and narrow rostrum; vaulted braincase; dorsal maxillary roots of the zygoma that flare out at almost a 90° angle from the rostrum with ventral zygomatic roots originating anterior to the dorsal roots; pronouncedly wide interorbital and postorbital regions, with their dorsolateral salients defined by ridges transforming to beading along dorsolateral margins of the braincase and fading out above squamosal roots of the zygomatic arch; a conspicuous lacrimal; an anterior bullar chamber that is notably less inflated than the posterior chambers; and incisor tips that usually form a straight cutting surface (rather than V-shaped). Occlusal patterns of ridges and valleys on premolar and molars were not closely studied, but seem similar to those characteristic of G. crassicaudatus (fig. 9).

Figure 10

Ratio diagram illustrating proportional relationships in cranial and dental dimensions among samples of adult Graphiurus. The standard is G. crassicaudatus from Cameroon, Nigeria, Togo, Ghana, Côte d'Ivoire, Liberia, and Equatorial Guinea (N  =  36) contrasted with samples of G. walterverheyeni from the Central Congo Basin (N  =  2, the holotype and paratype) and G. lorraineus from Cameroon, Gabon, D.R. Congo, and Equatorial Guinea (N  =  96).

Proportional differences (in addition to overall smaller size) contrast the skulls of G. walterverheyeni and G. crassicaudatus (fig. 10). Compared with that West African endemic, G. walterverheyeni has a markedly significantly shorter and wider rostrum (indexed by length of nasals) relative to length of skull (using condyloincisive length as the measure for skull length), which is also reflected in its absolutely shorter incisive foramina; the zygomatic breadth is broader relative to length of skull; the interorbital region is wider relative to length of skull (but the dorsolateral ridges are not as pronounced); the braincase is deeper relative to its breadth (G. walterverheyeni also exhibits greater cranial flexion, perhaps correlated with the deeper braincase); the entire skull behind the rostrum is smaller relative to overall skull length (relatively shorter diastema, bony palate, and molar row; and shorter palatal length); wider bony palate relative to its length; broader first molar relative to length of the molar row; markedly narrower incisors and smaller fourth premolar relative to nearly all other cranial and dental dimensions; and much larger auditory bullae relative to size of skull, with a correlated decrease in breadth of the basioccipital between the bullar capsules. The two examples of G. walterverheyeni are young adults, but the distinctions between that species and G. crassicaudatus do not reflect those associated with age as the same contrasts can be demonstrated comparing young adults of each species.

Figure 11

Principal components analysis (performed on cranial and dental variables) showing dispersion of specimen scores representing Graphiurus crassicaudatus and G. walterverheyeni along the first two principal components. Arrow identifies score for the holotype of G. c. dorotheae. See discussion in text and table 5.

The size discrepancy that distinguishes G. walterverheyeni and G. crassicaudatus can be appreciated by viewing results of the principle components analysis (complete sets of measurements were available only for 15 of the 24 variables measured) in figure 11 where the distribution of specimen scores form two discrete clusters along the first axis; size predominantly affects the spread of scores. The isolation of the two points representing G. walterverheyeni from those scores for G. crassicaudatus results from size differences in nearly all included variables, but particularly breadth across the incisor tips, breadth of incisive foramina, and lengths of nasals, diastema, bony palate, and tooth row (table 5). Length of bulla (and breadth of basioccipital, which is not shown) is negligible in its effect on dispersion of scores. An ordination employing second and third principle components (not illustrated) also yields two separate clusters, each representing one of the two species. Here the divergence of scores is along the third axis, where the dispersion of scores is influenced primarily by interorbital breadth (proportionally wider in the new species than in G. crassicaudatus), palatal length (proportionally shorter), length and breadth of the incisive foramina (breadth is wider relative to the proportionally shorter length), and breadth of the bony palate (relatively wider relative to palate length) (table 5).

Figure 12

Principal components analysis (performed on cranial and dental variables) showing dispersion of specimen scores representing Graphiurus lorraineus and G. walterverheyeni along the first two principal components. See discussion in text and table 6.

Table 5

Results of Principal Components Analysis of Adult Graphiurus walterverheyeni (N  =  2)and G. crassicaudatus (N  =  24)

(Principal components are extracted from a covariance matrix of 15 log-transformed cranial and dental variables; see table 4, fig. 11.)

Side-by side comparisons of skulls (figs. 6–8), combined with results from morphometric analyses that incorporate univariate statistics (table 4), a ratio diagram (fig. 10), and a principle components analysis (fig. 11; table 5) document the primary distinctions between the two species of dormice. Not only is G. walterverheyeni markedly physically smaller than G. crassicaudatus, but regions of the skull are proportionally different, as is highlighted by the relatively stubbier rostrum of G. walterverheyeni, its relatively shorter skull anterior to the braincase, wider interorbital region, higher braincase, and relatively much larger auditory bullae.

The mandible of G. walterverheyeni was not studied in detail, but generally resembles that of G. crassicaudatus.

Comparison with G. lorraineus

The sympatric true G. lorraineus is physically smaller than G. crassicaudatus and closer in body size to G. walterverheyeni, but the latter has a shorter head and body, hind feet, and ears compared with true G. lorraineus (table 4). Many cranial and dental dimensions are either similar or only slightly different in the two species; however, G. walterverheyeni has a greater zygomatic breadth, wider interorbit and braincase, higher braincase (reflecting markedly greater cranial flexion), and longer bony palate (table 4). Coordinated with the longer palate, the incisive foramina are strikingly shorter (and narrower) in G. walterverheyeni and the postpalatal dimension less. The new species also exhibits appreciably narrower incisors and shorter nasals. These similarities and distinctions between the two species portray G. walterverheyeni as closer in body size to G. lorraineus than to G. crassicaudatus, but differing rather spectacularly from G. lorraineus by its attenuated, stubby rostrum; wider interorbital region with the dorsolateral margins marked by ridging, which is generally absent in G. lorraineus; vaulted braincase as part of greater cranial flexion; narrower incisor tips; and longer bony palate with attendant lesser values associated with length of incisive foramina and the postpalatal region. The contrasts in conformation of rostrum, interorbital region, and braincase are easily observed in illustrations of skulls in figures 6–8.

These dimensional dissimilarities between the two species that are reflected in absolute values are also evident in proportional context and can be visualized in the ratio diagram (fig. 10). For example, relative to size of skull, G. lorraineus has a significantly much longer rostrum than does G. walterverheyeni; a narrower interorbital region and breadth across the zygomatic arches, lower braincase, a markedly shorter and wider bony palate, larger incisive foramina, longer postpalatal region, and wider incisor tips. Proportions of the teeth (breadths of fourth premolar and first molar, length of molar row) seen in G. lorraineus match those for G. walterverheyeni, but size of the braincase, breadth of rostrum, and length of diastema relative to length of skull in G. lorraineus fits with the proportions in G. crassicaudatus and not G. walterverheyeni.

The ordination of specimen scores in multivariate space rendered by principal components analysis shows two discrete clusters, one representing the two examples of G. walterverheyeni, the other depicting samples of true G. lorraineus from Cameroon, Gabon, Congo, Central African Republic, and Democratic Republic of Congo (fig. 12). Size, as usual, is primarily responsible for separating the two groups of scores along the first axis. Nearly all variables contribute to the dispersion of scores, but condyloincisive and postpalatal lengths, length of nasals, breadth of incisive foramina, and breadth of bony palate are most influential (table 6). The low value for interorbital breadth (0.001) indicates the wider interorbital region in G. walterverheyeni, and its relatively higher braincase evidenced by the negative value for that variable. Separation of the two species-clusters along the second axis is of lesser magnitude. Here zygomatic breadth and breadth of incisive foramina weigh heavily in the scatter of scores, but most of the other variables also contribute to some extent. The relatively shorter rostrum of G. walterverheyeni, and its relatively narrower incisive foramina and bony palate is reflected in the negative values. The ordination of first and third, and second and third components also demonstrated two constellations of scores (not illustrated here), each representing the two species. Breadth of incisive foramina and length of nasals are basically responsible for the separation of clusters (table 6).

Figure 13

Principal components analysis (performed on cranial and dental variables) showing dispersion of specimen scores representing Graphiurus crassicaudatus, G. lorraineus, and G. walterverheyeni along the first two principal components. See discussion in text and table 7.

Table 6

Results of Principal Components Analysis of Adult Graphiurus walterverheyeni (N  =  2) and G. lorraineus (N  =  22)

(Principal components are extracted from a covariance matrix of 13 log-transformed cranial and dental variables; fig. 12.)

Graphiurus walterverheyeni and G. lorraineus share small body size and general pelage coloration along with certain cranial and dental dimensions, but configurations of the rostrum, interorbital region, and braincase easily distinguish the two. Additionally, the incisor tips of specimens of G. lorraineus (both true G. lorraineus and the complex) form a V-shaped cutting edge, a trait exhibited by most species of Graphiurus; the incisor tips of almost all individuals of G. crassicaudatus and both specimens of G. walterverheyeni form a straight cutting surface.

Comparison Summary

In summary, the three discrete clusters of specimen scores, derived from all samples, in the ordination of first and second principle components summarized in figure 13 acknowledges three entities defined by morphometric traits. These groups translate to the reality of G. crassicaudatus, true G. lorraineus, and G. walterverheyeni, and supports the hypothesis that the two small-bodied specimens from south of the great bend of the Congo River, although similar to G. crassicaudatus in particular cranial features, do not mark merely a population of G. crassicaudatus occurring in the Congo basin, but instead represent a species as distinct from G. crassicaudatus as that species is from G. lorraineus. Condyloincisive and palatal lengths, lengths of bony palate and tooth row, and interorbital breadth partly influence the scatter of specimen scores along the first principle component (table 7); all these dimensions are markedly greater in G. crassicaudatus than in either of the other two species, differences readily apparent in the table of descriptive statistics (table 4). The large negative values associated with breadth across the incisor tips and breadth of incisive foramina reflect their smaller size in G. walterverheyeni compared to either G. crassicaudatus or G. lorraineus (table 4). Specimen scores for the two G. walterverheyeni that are isolated from the other clusters along the second axis is mostly influenced by condyloincisive and postpalatal lengths, lengths of nasals and diastema, and breadths of incisor tips and incisive foramina, all dimensions that are relatively less than in G. crassicaudatus and G. lorraineus. An ordination pitting the second principle component against the third left the two specimen scores for G. walterverheyeni isolated, but scores for the other two species melded together into a single cluster; no informative structure emerged by plotting first versus third components (these diagrams are not illustrated here).

Figure 14

The predicted geographic distribution of Graphiurus crassicaudatus derived from ecological niche modeling. Darker shades signify the most hospitable regions for the species as predicted by the model. White circles identify actual collection localities (which are listed in the gazetteer, along with the specimens collected from each place).

Table 7

Results of Principal Components Analysis of Adult Graphiurus walterverheyeni (N  =  2), G. crassicaudatus (N  =  22), and G. lorraineus (N  =  32)

(Principal components are extracted from a covariance matrix of 11 log-transformed cranial and dental variables; fig. 13.)

Karyotype and Molecular Data

No data are available. The holotype and paratype do not have karyotype, tissue, or molecular preparations associated with them.

Ecology

The two collecting localities are within the South Central faunal region (Grubb, 1978, 1982, 2001; Colyn, 1991; Colyn and Van Rompaey, 1994; Happold, 1996; White, 2001), and the Salonga-Lukenie-Sankuru landscape (Congo Basin Forest Partnership, 2006) of equatorial Africa. Aside from the scant habitat information written on the specimen tag of the type specimen, and from general habitat descriptions derived from Fraser (1927), no direct, detailed habitat information is available; however, utilizing vegetation maps as well as habitat information reported for sympatric species (Colyn, 1991; Colyn and Van Rompaey, 1994; Happold, 1996; Gauthier-Hion et al., 1999), conservation efforts (Congo Basin Forest Partnership, 2006), and other sources (e.g., Inogwabini, 2005; Seyler et al., 2005), the new species habitat can be generally characterized. Most of the region (94%) is covered by dense, intact, diverse tropical forest formations (on poorly drained, hydromorphic soil) that sustain major seasonal flooding: Swamp or floodplain forests comprise 23.6%, and terra firma forests account for 70.8% of the forest formations (Bailey, 1986; White, 2001; Congo Basin Forest Partnership, 2006). As much as one-half of the forest may be inundated during the rainy seasons, particularly during the September–December interval, with the flood plain extending up to 2.5 km from the river bank in certain areas (Bailey, 1986; Gauthier-Hion et al., 1999; Inogwabini, 2005). Vegetation is described by Evrard (1968), Gauthier-Hion et al. (1999), White (2001), Inogwabini (2005), and the Congo Basin Forest Partnership (2006).

Sympatric Associations with Other Graphiurus

Of the five equatorial species of Graphiurus, the ranges of only two overlap the geographic distribution of G. walterverheyeni, but examples of those two have not been collected at either of the sites yielding G. walterverheyeni. Twenty of the known 23 specimens in museum collections identified as G. surdus are from the western portion of the Central Forest Block in southern Cameroon, Equatorial Guinea, and Gabon (see map in Holden, 1996). Three others are from farther east in the Democratic Republic of the Congo. One was collected along the Sankuru River at Inkongo, in the watershed south of the great bend of the Congo River, and approximately 175 air km east-southeast of the site along the Lukenie River where the holotype of G. walterverheyeni was obtained, and about 500 air km southeast of Wafania, the provenance of the paratype. The other D.R. Congo specimen of G. surdus examined by Holden was collected at Masako Forest Reserve northeast of Kisangani in a loop of the Tshopo River (Hutterer and Dudu, 1990) (only one of two specimens collected at Masako reported by Mukinzi et al., 2005, was available for study). Masako is approximately 550 air km northeast of the type locality (Lukenie River), and about 580 air km east-northeast of paratype collection site (Wafania). Holden (1996: 39) compiled what meager ecological information is tied to specimens of G. surdus.

With a documented range extending from Guinea and Sierra Leone to northern Zambia (Holden, 2005, in press; fig. 16), the G. lorraineus complex has an appreciably more expansive geographic distribution than does G. surdus. Besides G. walterverheyeni and G. surdus, it is the only other species of dormouse documented from the Congo basin south of the great bend of the Congo River. The new species and G. lorraineus have not been collected at the same locality, and it is possible that their habitat requirements are distinct; G. lorraineus seems to be associated with secondary forest and anthropogenic habitats (see discussion under Kinship and Zoogeography below).

Endemicity

While two localities do not provide the strongest data bolstering the hypothesis that a distinctive faunal region exists, the known distribution of Graphiurus walterverheyeni is concordant with ranges of a small suite of equatorial nonvolant rainforest mammals that are unknown outside the South Central faunal region: one species of elephant shrew, seven or more primates, one mongoose, and several species of murid rodents.

The taxon, tordayi, marks a very morphologically distinctive population of the four-toed elephant shrew, Petrodromus tetradactylus, that is endemic to this region (see maps in Kingdon, 1974a: 66, 1974b: 55). Among populations of P. tetradactylus, which ranges beyond the Congo and Lualaba rivers to East Africa, tordayi is the only shrew that occurs in tropical evergreen rainforest, and although now treated as a subspecies of P. tetradactylus, is likely a separate species (Kingdon, 1974a, 1974b, 1997; Schlitter, 2005).

Eighteen taxa representing species and distinctive subspecies of primates are endemic to the south-central region: the galago Galago demidoff phasma (after taxonomic revision of the “demidoff species group,” phasma may prove to be a separate species endemic to the Congo basin south of the great bend of the Congo River [Groves, 2001; phasma's range is rendered in blue on Kingdon's map, 1997: 109]); the cercopithecines Allenopithecus nigroviridis, Cercocebus chrysogaster, Cercopithecus ascanius (the subspecies ascanius, atrinasus, katangae, and whitesidei), Cercopithecus dryas, Cercopithecus mitis heymansi, Cercopithecus wolfi (the subspecies elegans, pyrogaster, and wolfi), Lophocebus aterrimus, and Lophocebus opdenboschi; the colobines, Colobus angolensis angolensis, Piliocolobus foai parmentierorum, and Piliocolobus thollon; and the hominid, Pan paniscus (Colyn et al., 1991; Groves, 2001, 2005; Kingdon, 1997, provides informative figures depicting ranges of some of these primates).

The only carnivore apparently endemic to the South Central faunal region of the Congo basin is the herpestid, Crossarchus ansorgei nigricolor (Colyn and Van Rompaey, 1994; Kingdon, 1997: 249).

Two murid rodents have been recorded in rainforests south of the great bend of the Congo River and nowhere else. One is the deomyine, Lophuromys huttereri, a sister species to L. nudicaudus, which occurs on the right side of the Congo River and ranges from there to the western coast in lowland tropical evergreen rainforest (Musser and Carleton, 2005; Verheyen et al., 1996). The other is Praomys minor, a murine still known only by three specimens collected at Lukolela, on the left bank of the Congo River in the South Central faunal region (Nicolas et al., 2005; Musser and Carleton, 2005: 1448).

Future studies may alter the endemic status of the mammals listed above based on right-bank versus left-bank localities, which, for example, has already been realized for the murines Praomys lukolelae and P. muttoni. The former was known only by the type series collected at Lukolela on the left bank of the Congo River, within the South Central faunal region, but a large series has been obtained in the Kisangani region on the right bank of the Congo River, within the East Central faunal region (Musser and Carleton, 2005: 1447). Praomys muttoni was first collected in the Masako Forest Reserve on the right side of the Congo River (East Central faunal region), and has now been found in the Kisangani region on the left bank of the river (South Central faunal region) (Mukinzi et al., 2005; Nicolas et al., 2005).

How Graphiurus walterverheyeni's membership in the mammalian guild endemic to the South Central faunal region may be altered, or not, will depend upon results of future surveys concentrating on diversity of small mammals south of the Congo River. One long-term inventory already reinforces the distribution of G. walterverheyeni as restricted to the South Central faunal region. Results from a survey of small mammals (mostly rodents and shrews) around Kisangani in various kinds of forests as well as anthropogenic habitats undertaken from 1979 to 2003 has yielded only small samples of G. surdus and G. lorraineus (Mukinzi et al., 2005). The absence of G. walterverheyeni in historic samples consisting of hundreds of specimens of Graphiurus collected in West Africa and the West Central faunal region in particular, especially in southern Nigeria and southern Cameroon, suggests that the distribution of the new species does not extend into these regions. It must be noted, however, that certain species of African dormice (e.g., G. crassicaudatus and G. surdus) are uncommon at best, and the hypothesized distribution of G. walterverheyeni may well be an artifact of inadequate sampling.

Kinship and Zoogeography

These are subjects for future reports. Through our simple inspection of skulls, particular cranial conformations suggest a closer morphological link between G. walterverheyeni and G. crassicaudatus than between G. lorraineus or any other equatorial forest species of dormouse. Both G. walterverheyeni and G. crassicaudatus, to the exclusion of G. lorraineus, share a short rostrum relative to length of skull, relatively wide interorbital and postorbital regions bordered by some expression of ridging, and a vaulted braincase. On the other hand, proportional relationships involving dimensions of individual molars, complete molar row, and auditory bulla recall G. lorraineus. Recovering the phylogenetic affinities of G. walterverheyeni will entail a cladistic inquiry using a variety of anatomical traits and molecular data obtained from samples of many species of Graphiurus, both those occurring in equatorial forests and species found primarily in bordering savanna woodlands. That tropical lowland forest species such as G. crassicaudatus, G. walterverheyeni, and G. lorraineus may be closer relatives of one another to the exclusion of savanna species seems intuitively sensible but may not reflect phylogenetic reality. Possibly a forest dweller is closer kin to a savanna inhabitant than to another equatorial forest species. While the core of its geographic range is in the belt of equatorial lowland tropical forest, G. lorraineus has apparently never been trapped in primary forest but instead inhabits gallery forests, forest margins, woodland savanna, and disturbed areas, especially those formed by anthropogenic activities resulting in plantations of banana, cocoa, palmyra, and pawpaw, as well as in occupied and abandoned buildings (Holden, in press). Such habitats are structurally similar to woodland savanna.

If G. walterverheyeni is the closest relative of G. crassicaudatus among equatorial species of Graphiurus, the phylogeographic pattern (G. crassicaudatus in western Africa, G. walterverheyeni in the Congo basin) formed would be similar to those of a host of other equatorial rainforest mammals, among them species of murid rodents (Carleton and Robbins, 1985: Musser and Carleton, 2005), primates and artiodactyls (see review by Happold, 1996; Groves, 2001), and carnivores (for example, Colyn and Van Rompaey, 1994). These distributions overlaid with patterns of phylogeographic relationships among species of equatorial Graphiurus should add substantial information to the evolutionary history of Africa's tropical forest endemics.

Ecological Niche Modeling

Incorporating ecological niche modeling in the present report allowed us to compare the known geographical range versus predicted geographical range for two of the three species discussed in this report. We have examined most, and in one case all, existing museum specimens of the three Graphiurus species discussed in this paper. The collection localities associated with those specimens provide a general idea or hypothesis of the geographic range for each species. For species that are fairly commonly captured in faunal studies, the known geographic range based on specimen collecting localities may closely approximate the actual geographic range. For species that are uncommonly (G. crassicaudatus) or rarely (G. walterverheyeni) encountered, it is likely that the actual geographic distribution exceeds the known geographic range based on specimen data. It is of benefit to model the possible distribution of such species, as ecological niche models provide “first-pass hypotheses [of areas of occurrence] that may be the only usable information for many rare and poorly known taxa” (Peterson, 2001). This first pass allows researchers to identify potentially critical sampling sites for certain species, and when many species' predicted distributions are overlaid, it provides a cogent prediction of community composition for any analyzed site. Such cross-species prediction of community composition aids in pinpointing areas as conservation priorities (Peterson et al., 2000; Peterson, 2001). Distributional modeling also provides clues regarding the ecology of poorly known species.

Predicting species distributions has obvious applications in epidemiology. Certain African dormouse species, including G. lorraineus, G. crassicaudatus, and G. kelleni, are considered as possible hosts of monkeypox virus (Orthopoxviridae). Accurate or at least reasonable predictions of host distributions and ecology allow for enhanced understanding and analyses of virus ecology and epidemiology. If one or more species of Graphiurus is the primary host (or one of several hosts) of monkeypox virus, accurately predicting the distributions of these dormouse species would help predict where outbreaks of monkeypox may likely occur.

The predicted distribution of G. crassicaudatus (fig. 14) is somewhat surprising. All historical collection localities are in West Africa (western and west-central forest regions), but the model indicates strong ecological suitability (indicated by darker-shaded pixels) in the south-central region. This is interesting because (1) it may be that suitable habitat exists, but for a variety of other reasons, such as chance, geographic barriers, population extinction, climate change leading to isolation of populations (Colyn et al., 1991), the species does not currently occur there; (2) the species may actually occur there, but has not been captured due to low population density, trapping technique, fewer comprehensive faunal studies, and/or chance (if so, it would be sympatric with G. walterverheyeni); and (3) the new species, G. walterverheyeni, which we hypothesize to be closely relate to G. crassicaudatus, may occupy a similar ecological niche—its distribution falls within the more likely predicted range of G. crassicaudatus in the south-central region. Niche modeling unfortunately could not be performed to predict the distribution of G. walterverheyeni due to too few occurrence points (only two collecting localities).

Figure 15

The geographic distribution of true Graphiurus lorraineus (includes haedulus from southern Cameroon but excludes spurrelli from West Africa) as predicted from ecological niche modeling. Darker shades signify the most hospitable regions for the species as predicted by the model. White circles identify actual collection localities yielding specimens (places and specimens are listed in the gazetteer).

We produced two models for the G. lorraineus complex: one incorporating only true G. lorraineus (see gazetteer; fig. 15), and the other including all samples of the G. lorraineus complex from Guinea and Sierra Leone to northern Zambia (see gazetteer; fig. 16). Figure 15 bolsters the idea that what we refer to here as true G. lorraineus is in fact a distinct species from the other samples included in the G. lorraineus complex. Based on actual collection localities and the ecological parameters shared by those localities, the known geographic distribution for true G. lorraineus is within the West Central, South Central, and East Central faunal regions (historical collection locality points are plotted in fig. 15). Interestingly, ecological niche modeling for true G. lorraineus predicts few areas (indicated by dark shading of pixels) of strong ecological suitability outside its known range, indicating that the suite of ecological requirements for that species is uniquely found in the three forest regions in which it has been documented to occur. There are areas within the known range where occurrence is most likely based on strong ecological suitability, especially in eastern Gabon, northern Congo, and northwestern D.R. Congo (fig. 15), and these areas should be considered when identifying possible study sites for this species.

Figure 16

The geographic distribution of the Graphiurus lorraineus complex (includes haedulus from southern Cameroon and spurrelli from West Africa) derived from ecological niche modeling. Darker shades signify the most hospitable regions for the species as predicted by the model. White circles identify actual collection localities, which are listed in the gazetteer, along with the specimens collected from each place.

When all specimens referred to the G. lorraineus complex (a complex that we believe contains several separate species) are included in the analysis, the distributional model predicts strong ecological suitability over large areas far outside the known geographic range of the complex (even much of Madagascar). If we are correct that several distinct species have been modeled together under the umbrella of the G. lorraineus complex, then it makes sense that the usefulness of the model breaks down. In this case, the dominant ecological requirement of the entire complex would be tropical rainforest or deciduous tropical forest, and would not aid in identifying important survey or conservation efforts, or in epidemiological studies. This illustrates some of the limitations of distributional modeling. For genera that have not been adequately systematically revised, where several species are in fact grouped together in the same taxon, ecological niche modeling will probably not yield informative results. For those taxa that have been revised or at least sorted into more refined units, the application of ecological niche modeling provides potentially useful predictions of geographical distributions (providing enough historical collection localities exist), and overlain with other species' models, aids in prioritization of conservation sites.