Introduction

During Paleozoic times, a succession of waterways covered what is now northern Vietnam. Lithification, followed by tectonic deformation and uplift, and spurred by subsequent erosional processes, transformed those ancient tidal flows and marine landscapes into a spectacular terrestrial topography of tower karst riddled by fissures and honeycombed with caverns. Beautiful cycad rosettes secured high on sheer limestone cliffs, and a forest cover in which many of the components are adapted to karstic habitats, are the living botanical link to those ancient seas. A dark gray rat, its fur coloration indistinguishable from the limestone, its lair concealed deep within fissured cliffs and limestone talus, is the zoological connection.

We discovered this limestone rat during a survey (April, 2004) of small mammals in a forested tower karst region that forms much of Lang Son Province in northeastern Vietnam (Głazek, 1966; Tuyet, 1998) (fig. 1). Our intent was to sample, using a variety of traps and other techniques, as many native species as possible. Our past field experience in Indochina had taught us that setting traps on the ground in regular lines would quickly sample the generalists of the more common terrestrial habitats but not those specialized for life on substrates above the forest floor or in unusual petricolic environments. With this in mind we made special efforts to set a portion of our traps on tree limbs, woody vines looping through the understory, and on high limestone ledges. At one place, chunks of limestone had long ago fallen from two high forested towers, filling the space between them with huge blocks, their surfaces now partially covered with tropical vegetation, and riddled beneath with tunnels and fissures detectable at the surface only by small scattered openings (fig. 2). We set traps on top of this block pile, and also tied traps to lengths of rope and carefully lowered them through the openings to the bottom, to depths of at least 8 m. Unsure if we would catch anything different from the common rats already encountered in nearby different habitats—Niviventer fulvescens, N. langbianis, and Rattus andamanensis—we were very pleased, and somewhat surprised, to capture a dark gray rat on top of the rubble and two from deep within the labyrinth. Additional individuals were captured during the days that followed.

We knew nothing like this animal had ever been caught in Vietnam or elsewhere in Southeast Asia. Except for their fur coloration and short tails, the rats resembled species of Leopoldamys, but no member of that genus has dark gray fur and a tail shorter than head and body. Those two attributes also pointed to Berylmys, but none of them have dark gray underparts. We knew our sample of the limestone rat represented an undescribed species, but only after study in the museum did we confirm our suspicion that it could not be placed in any named genus.

Here we describe the limestone rat as a new genus and species, and summarize our limited observations about its natural history. We focus on describing the animal and comparing it to murines with which it is morphologically most similar, members of a cluster containing genera that Musser and Carleton (2005) refer to as the Dacnomys Division: extant species of the Indomalayan Chiromyscus, Dacnomys, Leopoldamys, and Niviventer; the Sri Lankan Srilankamys; and the recently described Laotian Saxatilomys (Musser et al., 2005). We also contrast Tonkinomys with members of the Dacnomys Division represented by fossils recovered from Indochinese Plio-Pleistocene sediments: extinct species of Leopoldamys and Niviventer, and the extinct genera Wushanomys and Qianomys. Of these, the new rat requires close and careful comparisons only with Leopoldamys, Niviventer, Chiromyscus, Saxatilomys and the Plio-Pleistocene Wushanomys and Qianomys.

Descriptions of new taxa, and in particular those occurring in gradually diminishing tropical forest refuges, should be published as soon as possible. Phylogenetic inquiries take time—incorporating the new rat into phylogenetic analyses employing samples of all native Indochinese murines will be the subject of a future report.

Our discovery joins those of several other small-bodied mammals living in Indochina that have been discovered and described, with little fanfare, during the past five years: two shrews, Chodsigoa caovansunga and Crocidura kegoensis (Lunde et al., 2003, 2004); three bats, Hipposideros scutinares (Robinson et al., 2003), Rhinolophus stheno microglobosus (Csorba and Jenkins, 1998) and Myotis annamiticus (Kruskop and Tsytsulina, 2001); and a rabbit, Nesolagus timminsi (Averianov et al., 2000; Can et al., 2001). These can be added to an assemblage of large mammals occurring in Vietnam, and some in nearby Laos, named and described as new within the last decade or so, but with attendant worldwide publicity: two primates, Trachypithecus ebenus and Pygathrix cinerea (summarized in Groves, 2001); three muntjacs, Muntiacus truongsonensis, M. vuquangensis, and M. puhoatensis (see reviews in Amato et al., 2000; Groves and Schaller, 2000; Grubb, 2005); and the saola, Pseudoryx nghetinhensis (see review and references in Groves and Schaller, 2000).4 All are a testimony to the significant results derived from past biodiversity surveys, discoveries that provide compelling stimulus for zoologists to continue detecting the real diversity of small mammals in Vietnam.

Materials and Methods

All 14 examples of the new rat were fixed in 10% formalin in the field and subsequently transferred to 70% ethanol for storage. Liver samples were collected prior to fixation and placed in lysis buffer in the field, and later stored in liquid nitrogen at the American Museum of Natural History (AMNH). Skulls were subsequently removed from 10 of 13 specimens (one of the 14 preserved is represented by only an ear with a bit of fur-covered skin attached, the only remnant left in the trap by a hungry predator) and cleaned by a dermestid colony, American Museum. The elements of each specimen are identified by a catalog number in the Department of Mammalogy, AMNH. Each liver sample was transferred to the American Museum Cryo Collection (AMCC), where it is registered under a second number in addition to its departmental catalog number. Some fluid-preserved specimens and accompanying skulls will be retained at AMNH; others will be returned to the Department of Zoology at the Institute of Ecology and Biological Resources in Hanoi.

Comparative material we consulted, primarily samples of Leopoldamys, Niviventer, Chiromyscus, and Saxatilomys, are in collections at AMNH; the Field Museum, Chicago; and the National Museum of Natural History, Washington, DC.

Values (in millimeters, mm) for total length, length of tail (LT), length of hind foot, including claw (LHF), and length of ear from intertragal notch to crown (LE) are those we obtained in the field and recorded in our field journals. Values for length of head and body (LHB) were determined by subtracting length of tail from total length. Weights were obtained in the field with a Pesola spring scale (300-gram capacity). The distal portion of the tail is white (unpigmented) in all specimens of the limestone rat. We measured the length of this white tip on the dorsal surface of each tail and calculated it as a percentage of total tail length. Values for cranial and dental measurements were obtained by Lunde using digital calipers accurate to the nearest 0.01 mm; however, values were rounded to the nearest 0.1 mm. The following cranial and dental dimensions (listed in the sequence they appear in tables) were measured; their limits are illustrated in figure 3 and defined in Musser and Newcomb (1983).

ONL occipitonasal length (= greatest length of skull)

ZB zygomatic breadth

IB interorbital breadth

LR length of rostrum

BR breadth of rostrum

BBC breadth of braincase

HBC height of braincase

BZP breadth of zygomatic plate

LD length of diastema

LIF length of incisive foramina

BIF breadth of incisive foramina

LBP length of bony palate (palatal bridge)

BBP breadth across bony palate at first molars

PPL postpalatal length

BMF breadth of mesopterygoid fossa

LB length of bulla

CLM1-3 crown length of maxillary molar row

BM1 breadth of first upper molar

Old adults, adults, and young adults (as defined by Musser and Heaney, 1992: 5) were lumped as “adults” and measurements derived from them were used to calculate standard descriptive statistics (mean, standard deviation, and observed range). Small sample size dictated that we combine sexes in the analyses. Anatomical terminology follows Brown (1971) and Brown and Yalden (1973) for external features of the head and limbs; Bugge (1970) for the cephalic arteries; Wahlert (1985) for the cranial foramina; and Carleton (1980), Carleton and Musser (1984), Musser and Durden (2002), Musser and Heaney (1992), Musser et al. (1998), and Voss (1988) for cranial morphology. Names of cusps and cusplets of upper and lower molars are noted in figure 13; sources of this terminology are explained by Musser and Newcomb (1983: 332).

Systematic Description of the New Genus and Species

Tonkinomys, new genus

Comparisons

The morphological attributes of Tonkinomys daovantieni place it in the Dacnomys Division of Musser and Carleton (2005), which contains the Indomalayan Chiromyscus (one species), Dacnomys (one), Leopoldamys (six), and Niviventer (17); Sri Lankan Srilankamys (one); and the recently described Lao PDR Saxatilomys (one species). Musser (1981), Musser and Newcomb (1983), Musser and Heaney (1992), Musser and Carleton (2005), and Musser et al. (2005) have chronicled the geographic distributions of these genera, along with their taxonomies, descriptions of morphologies, and postulated phylogenetic relationships. In conformation of body and limb proportions, and in characteristics of skull and teeth, T. daovantieni most closely resembles species of Leopoldamys, Niviventer, Chiromyscus, and Saxatilomys. Our comparisons contrast Tonkinomys first with extant examples of Leopoldamys, then with Niviventer and Chiromyscus, and finally with Saxatilomys.

A large body of literature has accumulated over the past two decades reporting fossil murines, mostly isolated molars and incisors, which have been recovered from Pliocene, Pleistocene, and Holocene sediments in the Indomalayan region. Some material has been described as new taxa whose diagnostic traits identify them as members of the Dacnomys Division, and other samples have been regarded as examples of extant members of that group. We examined pertinent publications to determine whether any of the new taxa that are based on fossils represent examples of Tonkinomys. The significance for nomenclature is obvious. Equally important is the zoogeographic context. If some samples of fossils that have been identified as either extinct or extant species are actually examples of Tonkinomys, or close phylogenetic relatives, they would provide a window through which could be glimpsed the past distribution of this unique limestone rat, and the geologic depth of its evolutionary history.

Leopoldamys: The smallest extant member of the six currently recognized species of Leopoldamys (Musser and Carleton, 2005) is the Thai L. neilli (only the extinct L. minutus is smaller, described from samples recovered from late Pliocene–early Pleistocene fissure and cave sediments; Chaimanee, 1998), and the head and body proportions, body size, weight, and posture of the live animal closely resemble Tonkinomys daovantieni (Marshall, 1988: 485, provided a photograph of a live Leopoldamys). Although the two species are similar in body size (table 2), they are strikingly dissimilar in other particulars of external morphology. Leopoldamys neilli has: (1) short, sleek brown fur over upperparts of head and body that is sharply demarcated from white underparts (long, semispinous, grayish black fur over the dorsum and dark gray venter in T. daovantieni); (2) slightly smaller external pinnae, both absolutely and relative to head and body length; (3) a long, slender, tapered tail that is much longer than length of head and body, LT/LHB  =  124% on average, and similar to an inverted U in cross-section—flat ventral surface and arched across sides and top (less tapered in T. daovantieni, chunky and much shorter than head and body, LT/LHB  =  83%, and round in cross-section); (4) a very slender, pointed tail tip that is covered with short scale hairs no longer than 1−2 mm (tip of tail is stubby in T. daovantieni, covered by much longer hairs, 5−7 mm, and brush-like in appearance); and (5) swollen plantar tubercles situated in distal two-thirds of plantar surface, as seen in Tonkinomys (figs. 9, 10), but slightly smaller relative to plantar area—the thenar pad, for example, forms about 17% of planter surface (T. daovantieni has relatively more expansive plantar pads—the thenar constitutes about 20% of plantar surface). Short, sleek fur with the coat that covers upperparts of the head and body strikingly contrasting with the ventral covering is characteristic of all extant species of Leopoldamys. Also common to all extant species in that genus is a very long and tapered tail, with its cross section resembling an inverted U, and its tip covered by very short bristles. Elongate hind feet are common to all Leopoldamys, as are the number of plantar tubercles, their size relative to plantar surface, and distribution in the distal two-thirds of the plantar surface.

The skull of L. neilli is similar to that of T. daovantieni in overall size (table 2), and the two are remarkably alike when viewed from a dorsal perspective (compare the cranial photographs of Leopoldamys in Musser, 1981: 259, and the drawing on p. 264, with the renditions in fig. 10). In ventral view, however, the two species are strikingly dissimilar. Compared with the Vietnam species, L. neilli (and all other extant species of Leopoldamys) has: (1) significantly shorter incisive foramina, their posterior margins well anterior to faces of first upper molars (longer in T. daovantieni, the posterior edges located just before anterior margins of first molars, even with them, or slightly projecting between the first molars); (2) a much shorter bony palate, its posterior margin aligned somewhere between the middle or end of each third upper molar, depending upon the species and variation within each sample (longer in T. daovantieni, extending beyond the molar rows to form a wide platform); (3) longer postpalatal length, which reflects the shorter bony palate (much shorter in T. daovantieni); (4) wider, chunky molars relative to size of skull, which is correlated with an absolutely narrower bony palate (smaller, gracile molars relative to size of skull, and an attendant absolutely wider bony palate in T. daovantieni); (5) molar rows that barely diverge posteriorly (molar rows diverge appreciably along anterior-posterior axis in T. daovantieni); (6) a small, moderately inflated ectotympanic bulla relative to size of the skull (more inflated capsule in T. daovantieni, and absolutely larger in most examples); and (7) a very long, bony eustachian tube, especially contrasted with the small bulla (much shorter eustachian tube in T. daovantieni). These differences also separate T. daovantieni from specimens we examined representing the other five species of Leopoldamys. Our study of those comparative samples also reveals that L. neilli and the other species of Leopoldamys have wider and sturdier alisphenoid struts relative to skull size, and these struts are present in every individual of each species (relatively weaker struts in T. daovantieni, and absent from some specimens).

Patterns of roots beneath certain molars that are characteristic of T. daovantieni are dissimilar to those exhibited by species of Leopoldamys. Both extant samples (material in AMNH and USNM; also see Musser, 1981: 260) of Leopoldamys and those represented by fossils (Chaimanee, 1998; Zheng, 1993) have four-rooted upper molars: large posterior, single large lingual (we have never encountered a specimen with two lingual roots), large anterior, and a small labial, which is sometimes divided (also four roots in T. daovantieni, but there are two large lingual holdfasts instead of one, and no labial root). All Leopoldamys have a second upper molar that is anchored by either three or four roots (only three roots in all our T. daovantieni). Both Leopoldamys and Tonkinomys have three-rooted third upper molars. All species of Leopoldamys have four-rooted first lower molars: two large anterior and posterior anchors, and smaller labial and lingual holdfasts (all examples of T. daovantieni have only two roots, an anterior and posterior coequal in size). Second and third lower molars in Leopoldamys have two roots (large anterior and posterior), which is also the pattern in T. daovantieni.

In addition to the marked contrasts in robustness and width of molars between Leopoldamys and Tonkinomys, occlusal patterns of molars are different (compare illustrations of molar rows for Leopoldamys in Musser, 1981: 262, with those of Tonkinomys in fig. 14). Coronal surfaces of maxillary molars are characterized by simple patterns in both genera (no cusp t7 or posterior cingulum on any molar, no accessory cusplets, no longitudinal crests connecting the primary cusps), but the conformations of laminae forming the occlusal surfaces are dissimilar. The anterior lamina of the first molar in Leopoldamys consists of a nearly horizontal segment, formed by the complete fusion of cusps t2 and t3, and a discrete cusp t1 on its posterolingual margin. The second lamina is not as straight but gently arcuate. These are diagnostic molar traits for Leopoldamys. In contrast, the two comparable laminae in Tonkinomys are tight chevrons—cusps t1 and t3 are opposite each other in the first lamina, as are cusps t4 and t6 in the second. Furthermore, cusps t4 and t6 in T. daovantieni, although coalesced with the central cusp t5, are large relative to overall dimensions of the lamina, and retain part of their identity as cusps. Comparable cusps in the second lamina in Leopoldamys are relatively smaller and so broadly merged with cusp t5 that their boundaries are not determinable. The large chunky posterior lamina in Leopoldamys is wider than long, but roughly diamond-shaped in Tonkinomys. Contrasts in second upper molars are similar to those seen in the first. The third upper molars in Leopoldamys have a stretched, ridge-like cusp t4 compared with the shorter version in Tonkinomys; otherwise, the occlusal patterns in the two genera are similar.

Cusps form either nearly straight or gently arcuate lamina in the lower (mandibular) molars of Leopoldamys. The posterior cingula are wide and elliptical in cross-section in all members of that genus, prominent posterior cusplets adorn the labial margin of the first and second molars, and two nearly straight laminae form the occlusal surface of each third molar (see the illustrations in Musser, 1981: 260). In parallel with the upper molars of Tonkinomys, the two cusps forming the laminae of its lower molars are diagonally oriented so that after wear they form chevrons in which the arms are very close, the posterior cingula are roughly triangular, labial cusplets are indistinct or not present on most specimens (prominent only in the holotype), and the occlusal surface of the third molar is formed by a distinct anterior chevron-shaped lamina and a posterior oblong chunk.

Niviventer and Chiromyscus: We examined samples of all 17 of the currently recognized species of Niviventer and the single species of Chiromyscus (taxonomy and geographic distributions for both genera are summarized in Musser and Carleton, 2005; specimens forming the comparative samples we utilized are listed in the Appendix by Musser et al., 2005). We preface our comparisons by explaining that the present contents of Chiromyscus and Niviventer will likely be altered after careful systematic revision of these genera. Musser (1981) characterized them, provided provisional diagnoses and contents, but his exposition was more of a taxonomic summary and meant to be a working hypothesis, not a systematic revision. That kind of study has yet to be realized. Most species will remain in Niviventer, but two sets may eventually be differently allocated. Except for its nail-like claw on the hallux, shelf-like ridges bordering dorsolateral margins of the postorbital region and braincase, pattern of roots beneath the first upper molar (table 3), color pattern of the dorsal pelage, and large and very bulbous plantar pads set close together, other external and most cranial and dental characteristics of the arboreal C. chiropus are similar to those features in the species of Niviventer, particularly N. langbianis (Musser, 1981; Musser and Lunde, ms.). The latter is also arboreal (observations by Lunde and Musser derived from trapping surveys in Vietnam), and shares with C. chiropus comparable prominent postorbital and temporal ridges, the number of roots anchoring each first upper molar, a nail-like claw on the hallux, a pelage color pattern that does not match in details but is similar, and equal expression of plantar pads. A revisionary inquiry may either move N. langbianis to Chiromyscus, or C. chiropus will be subsumed within Niviventer (that name would then be a synonym of the older Chiromyscus).

Niviventer andersoni and N. excelsior form another set now allocated to Niviventer that will probably be removed. Musser (1981) described a suite of cranial and dental traits held by these two species that contrast with the characters common to the other species in Niviventer. Should information from other anatomical systems, and that derived from chromosomal and biochemical surveys, support the cranial and dental distinctions, he (p 328) noted “… that the phylogenetic relationship between Niviventer and the N. andersoni division [would be] better expressed by excluding the latter from Niviventer and placing N. andersoni and N. excelsior in a genus of their own.”

We accept the current, although provisional, contents of Niviventer for our comparative context. We will also refer only to “Niviventer”, understanding that the contrasts between species in that genus and Tonkinomys are much the same as between the latter and C. chiropus.

While the body conformation of Tonkinomys daovantieni recalls small-bodied Leopoldamys, no such resemblance exists between the Vietnamese animal and any species of Niviventer. Among the largest species of Niviventer in body size is N. andersoni, which is still appreciably smaller than T. daovantieni (table 2). In all species of Niviventer, dorsal head and body fur ranges from buffy or brownish gray through brown to reddish brown (Musser, 1981). Niviventer brahma and N. eha have grayish white underparts (Musser, 1970), N. hinpoon has a buffy pale gray venter (Marshall, 1988), but all other species have white venters. No example of Niviventer has Tonkinomys's unique dorsal and ventral pelage coloration. Some species of Niviventer, however, do have semispinous pelage similar in texture to that possessed by T. daovantieni (N. fulvescens and N. tenaster are good examples). The conformation of tail and proportional relationship between lengths of tail and body in Tonkinomys (stocky, with a stubby tip, and significantly shorter than combined head and body length) does not occur in any species of Niviventer. The tail is tapered to a pointed tip and gracile in appearance in all species of Niviventer, and much longer than length of head and body in all species except N. hinpoon, in which the tail is equal to or slightly longer than head and body length (table 2; Marshall, 1988; Musser, 1970, 1973, 1981; Musser and Chiu, 1979; Lunde and Son, 2001). The plantar pads of Tonkinomys are larger, closer together, and more swollen relative to plantar surface than in any terrestrial or scansorial species of Niviventer, including N. hinpoon, which is thought to be closely associated with limestone cliff habitats (Marshall, 1988), or even the arboreal N. langbianis and Chiromyscus chiropus (fig. 9); N. fulvescens (fig. 9) exhibits the configuration of plantar pads that is common to most species of Niviventer.

The skull of Tonkinomys is appreciably larger (as indexed by occipitonasal length) and more robust than all species of Niviventer, even N. andersoni, which has the largest skull among the species in the genus (table 2). The skull of Tonkinomys, however, is not a giant version of Niviventer; compared with Niviventer, Tonkinomys has: (1) insignificant basicranial flexion (most species of Niviventer display marked basicranial flexion); (2) maxillary molar rows that are situated farther forward relative to posterior margin of the zygomatic plate (front of molar row either even with back of zygomatic plate or well behind it in most specimens of Niviventer); (3) the squamosal root of the zygoma originating higher on side of the braincase (set lower in Niviventer, particularly N. andersoni and N. excelsior); (4) more slender alisphenoid struts relative to size of skull, with some specimens lacking struts (relative to skull size, alisphenoid struts are more substantial in all species of Niviventer, and present in all specimens we examined); (5) strongly posteriorly diverging molar rows (parallel or only slightly diverging in Niviventer); (6) a bony palate projecting appreciably beyond posterior margins of third molars to form a wide platform (posterior margin of bony palate ends about 0.5 mm anterior to posterior surfaces of upper third molars, even with them, or no more than about 0.5 mm beyond in all species of Niviventer); (7) a posterior margin of the bony palate that is smooth or only slightly thickened (the palatal border is defined by a prominent and thick bony ridge in Niviventer); and (8) a thick, bony, and flat posterolateral projection of the bony palate covering the anteromedial portion of each pterygoid plate, its sharp diagonal border (ppm in fig. 12) defining the anteriomedial edge of the pterygoid fossa (no such conformation in any species of Niviventer). Other than the larger size of each dentary in Tonkinomys, we did not detect any significant differences with those of Niviventer in overall shape; positions of foramina; expanse of coronoid, condyloid, and angular processes relative to body of the dentary; expression of masseteric and lingual ridging; or position of the posterior terminus of the incisor capsule.

Shapes of the upper and lower incisors in Tonkinomys, their enamel color and thickness, and the opisthodont conformation of the uppers resemble the incisors in all species of Niviventer. The larger size of these structures in Tonkinomys provides the obvious contrast. Reflecting the much greater skull dimensions of Tonkinomys when compared with species of Niviventer are the elongate molar rows of the former, which are much longer than even the largest-bodied species of Niviventer (table 2). Occlusal patterns of the upper molars fit within the range of variation we observed among most species of Niviventer, except that cusps t4 and t6 on the second lamina of the first molar in Tonkinomys tend to retain more of their identity, and cusp t9 is smaller and indistinct relative to the huge central cusp t8. The anterior lamina on the first upper molar in N. andersoni and N. excelsior resembles the shape in Leopoldamys, and contrasts with the comparable chevron-shaped lamina in Tonkinomys, but those two species also differ from most other species of Niviventer in this trait. Coronal patterns of lower molars in Tonkinomys are also closely similar to those in the species of Niviventer.

While occlusal patterns on upper and lower molars in our sample of Tonkinomys fall within the range of variation of the patterns we observed among the species of Niviventer, the number of roots anchoring most of those molars do not—this set of structures by themselves identifies Tonkinomys as the representative of a monophyletic group separate from Niviventer, a position only strengthened by the external and cranial contrasts between the two genera already described. Except for the first upper molar, which has four roots, Tonkinomys exhibits the primitive number of roots anchoring upper (three per tooth) and lower (two per tooth) molars. All species of Niviventer, in contrast, have multi-rooted molars (table 3; also see Chaimanee, 1998, and Zheng, 1993). First upper and lower molars can be used as examples. Most species of Niviventer have five-rooted first upper molars: large anterior, two smaller lingual, large posterior, and small labial (some specimens have one or more labial rootlets in addition to the primary labial anchor). In N. langbianis, and also in Chiromyscus chiropus, four roots anchor each molar: large single lingual, large anterior and posterior roots, and small labial. Tonkinomys also possesses four roots beneath each first upper molar, but the pattern is different from that in N. langbianis and Chiromyscus. No specimen in our sample of Tonkinomys has a labial anchor, and the four-rooted pattern is formed by large anterior and posterior roots along with two prominent lingual roots. A small labial root, not a divided lingual, forms the fourth root in N. langbianis and Chiromyscus. Although our survey results for second and third upper molars are not summarized in table 3, we also determined that second upper molars in Niviventer have four roots (three in Tonkinomys), and third upper molars have three (as in Tonkinomys, which is the usual number for the third molar in most murines, even when the first and second molars have more than three roots).

Four roots (large anterior and posterior, smaller labial and lingual) anchor each first lower molar in all species of Niviventer (and Chiromyscus), but only two are present in the Vietnamese limestone rat. Two to four is the range in number of roots anchoring each second lower molar among species of Niviventer, and each third lower molar has either two or three roots (two roots beneath second and third molars of Tonkinomys).

Saxatilomys: Described from two whole animals along with 14 individuals represented by cranial, mandibular, and dental fragments recovered from owl pellets, S. paulinae has been recorded only from the Khammouan Limestone National Biodiversity Conservation Area in Lao PDR (Musser et al., 2005).

Body size, build, and proportions of limbs, pinnae, and tail that are characteristic of S. paulinae resemble most species of Niviventer; “S. paulinae simply looks like a Niviventer and not like any other species in any other genus” (Musser et al., 2005: 20). That Lao species strikingly contrasts with the large-bodied and short-tailed Vietnamese animal. Tonkinomys daovantieni is more robust in build and much larger in body size than S. paulinae (range in length of head and body is 184–217 mm for 13 examples of T. daovantieni versus 144–150 mm for two S. paulinae). Its thick, stubby tail is shorter than head and body length (LT/LHB  =  83%, average of 13 specimens) compared with the slim, tapered, and gracile tail of S. paulinae that is appreciably longer than body length (LT/LHB  =  112%–116% for two individuals). The tail of both species is bicolored, but the dark brown and white pattern is different in each. In T. daovantieni, the tail is dark brown except for the ventral surface, which is white, and the distal one-fourth to one-half, which is white on all surfaces; the tail of S. paulinae is dark brown on dorsal and lateral surfaces from base to tip, and white (with slight speckling) over only its ventral surface. Both species have dense and dark gray fur covering upperparts and underparts of the head and body. Saxatilomys paulinae, however, lacks a white forehead blaze and white pectoral patch, a pattern exhibited by most examples of T. daovantieni, and its venter is lightly frosted in contrast to the total gray of the Vietnamese animal. Tonkinomys and Saxatilomys do share bulbous plantar pads that are very large relative to plantar surface and arranged in a tight cluster (compare the plantar surface of T. daovantieni in fig. 8 with that of S. paulinae in fig. 9).

Tonkinomys and Saxatilomys possess an elongate skull, wide zygomatic plate relative to cranial size, narrow alisphenoid struts present in most specimens, bony palate projecting past the third molars to form a broad shelf with either a smooth or slightly thickened (not ridged) posterior margin, strongly divergent maxillary molar rows, and a similar mandibular conformation (compare figs. 10 and 11 with the views of Saxatilomys in Musser et al., 2005: fig. 6).

The two species differ in size and some qualitative traits. Tonkinomys has a more robust and larger skull (occipitonasal length, for example, ranges from 47.1 to 52.4 mm in seven T. daovantieni, and 40.3 to 40.5 mm for the two intact skulls of S. paulinae). Size of the gracile skull in the Lao genus resembles the larger-bodied species of Niviventer (the Indochinese N. tenaster, for example; compare the values listed in tables 1 and 4 of Musser et al., 2005, with those presented in table 1). Tonkinomys further contrasts with Saxatilomys by having a less tapered rostrum (when viewed from dorsal or ventral aspect), significantly less basicranial flexion (marked flexion in Saxatilomys; see fig. 6 in Musser et al., 2005), and spacious incisive foramina (not the narrow slits of Saxatilomys.

Three other qualitative features provide marked contrasts between the two genera. In Tonkinomys, the squamosal root of the zygomatic arch is situated high on the side of the cranium where its ridge-like portion runs horizontally just below the posteroventral margin of the parietal, which projects ventrad to form the dorsolateral side of the braincase; the low ridge of the squamosal root disappears 10–15 mm anterior to the vertical ridge formed by the squamosal-exocciptal suture (fig. 11). The configuration is similar to that characteristic of Leopoldamys and Niviventer. The squamosal root of each arch in Saxatilomys also originates high on the side of the cranium, but is oriented diagonally so that its posterior ridge-like portion extends diagonally to merge with the squamosal-parietal suture (which forms the temporal ridge) well anterior to the vertical ridge created by the squamosal-exoccipital suture (see figs. 6 and 8 in Musser et al., 2005). The parietal does not drop below the temporal ridge, and the upper portion of the braincase consists only of the squamosal.

Tonkinomys daovantieni and S. paulinae have a wide and long bony palate that projects beyond the molar rows to form a broad platform. In Tonkinomys, the posterolateral edges of this shelf extends as a flat, bony surface onto the anterior portion of each pterygoid plate where its sharp diagonal border (ppm in fig. 12) marks the anteromedial margin of each pterygoid fossa. No such conformation exists in S. paulinae (or in any species of Niviventer).

The globular and only slightly inflated ectotympanic (auditory) bulla of Tonkinomys does not conceal all of the periotic. A wide, wedge-shaped segment of that bone is exposed between the dorsal capsular margin and basioccipital, and this wedge of periotic extends anteriorly to form the dorsal and slanting posterodorsal wall of the carotid canal (fig. 12). The conformation is primitive among murines and closely resembles the morphology depicted by Carleton and Musser (1989: 33) for the sigmodontine, Oligoryzomys, and by Musser and Heaney (1992: 99) for some of the Philippine Old Endemic murines. By contrast, the globular ectotympanic capsule exhibited by Saxatilomys covers all but a narrow wedge of the periotic, and the anteromedial margin of the capsule meets the basioccipital, so that it is the ectotympanic (or auditory capsule) that forms the posterior wall of the opening into the carotid canal. This structural relationship between ectotympanic and periotic is specialized, and is particularly well illustrated in Carleton and Musser (1989: 33) for the sigmodontine Oryzomys palustris. Species of Rattus and its allies have a comparable pattern (Musser and Heaney, 1992: 99).

Primary dental differences between Tonkinomys and Saxatilomys involve size, one proportional feature, and pattern of molar roots. Tonkinomys has larger upper and lower incisors than does Saxatilomys, but otherwise pigmentation and thickness of enamel, their shapes, and opisthodont configuration of the uppers relative to the rostrum are similar. Tonkinomys also has longer molar rows (as indexed by lengths of maxillary rows: 8.0–8.8 mm in 7 specimens of Tonkinomys, 7.2–7.3 mm for three Saxatilomys); length of maxillary row relative to occipitonasal length for the two genera, however, is, on average, not significantly different (the ratio, CLM1-3/ONL, using mean values from each generic sample, is 17% for Tonkinomys and for Saxatilomys). In contrast to molar length, the individual molars in Tonkinomys are not significantly wider than are the molars of Saxatilomys (using breadth of the first upper molar as a guide, the mean is 2.1 mm for 7 examples of Tonkinomys and for 3 specimens of Saxatilomys; contrast the values listed in table 1 with those figs. in table 4 in Musser et al., 2005: 21), indicating significantly narrower molars relative to length of the maxillary row (the ratio, BM1/CLM1-3, using mean values from each generic sample, is 25% for Tonkinomys and 29% for Saxatilomys). Two roots (large anterior and posterior) anchor each lower molar in all specimens examined of Tonkinomys and Saxatilomys, but the root patterns for the first and second maxillary molars are dissimilar in the two genera. In Tonkinomys, the first maxillary molar has four roots (anterior, posterior, and divided lingual), the second has three. Saxatilomys has five-rooted first molars (table 3) and four-rooted second molars; in samples of both Tonkinomys and Saxatilomys, three roots anchor each third molar.

Tonkinomys and Saxatilomys (along with most species of Niviventer) share similar maxillary and mandibular molar occlusal patterns. Distinctions between the Vietnamese and Lao limestone rats are present but not spectacular (contrast fig. 14 with the occlusal views of Saxatilomys in Musser et al., 2005: fig. 10): compared with Saxatilomys, cusps t4 and t6 on the second lamina of the first maxillary molar in Tonkinomys tend to retain more of their identity (less distinct in Saxatilomys and in most species of Niviventer), second and third mandibular molars lack anterolabial cusps in most specimens (most species of Niviventer also lack these cusps, but they are present in Saxatilomys), and the cusps on upper and lower molars form more distinct and tighter chevrons (cusp rows more arcuate, with greater space between arms of each chevron in Saxatilomys).

Other genera: We also compared examples of Dacnomys and Srilankamys, other Indomalayan members of the Dacnomys Division, with our sample of Tonkinomys. Dacnomys is a giant version of Niviventer, especially N. andersoni, in its external and cranial traits, as well as root patterns of the molars (see description and illustrations in Musser, 1981). Those molars, however, are large, chunky, and hypsodont, with complicated cuspidate occlusal patterns, and strikingly contrast with the brachydont and gracile molars with their simple occlusal patterns possessed by all species of Niviventer. Tonkinomys bears no close resemblance to Dacnomys in external, cranial, or dental traits.

A combination of traits, some of which recall features in Niviventer, others that resemble Maxomys, and some that are unique, characterizes the morphological definition of Srilankamys (see the description and illustrations in Musser, 1981). Its membership in the Dacnomys Division is provisional, pending reassessment of its phylogenetic relationships by additional morphological characters in combination with molecular data (Musser and Carleton, 2005). Srilankamys shares the same pattern of molar roots with Tonkinomys, but otherwise its diagnostic external and cranial traits, along with molar occlusal patterns, are unlike those of Tonkinomys.

Musser and Carleton (2005) included the Philippine Mindoran endmic Anonymomys mindorensis in the Dacnomys Division. Its placement there, however, is very provisional. Some morphological traits of Anonymomys recall those of genera in the Dacnomys Division, particularly Niviventer, but others suggest a closer affinity to Bornean and Sulawesian Haeromys, a member of the Micromys Division of Musser and Carleton (2005). Tonkinomys is morphologically distant from Anonymomys and requires no comparisons with it.

Our comparative study also included the Indomalayan Maxomys, although that genus is not contained within the Dacnomys Group (Musser and Carleton, 2005). Species of Maxomys are prominent members of the Indochinese murine fauna (Corbet and Hill, 1992), and some share with Tonkinomys the same number of roots anchoring each molar (Musser et al., 1979). Any close similarity ends with that character complex. No species of Maxomys has the pelage coloration possessed by Tonkinomys, or the thick, short tail, or the crowded, fleshy palmar and plantar pads. Additionally, all Maxomys have short, wide incisive foramina that have the outline of an inverted heart, a bony palate ending well anterior to posterior margins of the molar rows, a very long postpalatal region, and other cranial attributes, as well as molar occlusal patterns (see descriptions and illustrations in Musser, 1991) that are not shared with Tonkinomys. We found no derived characters that identified Tonkinomys as either a member or close relative of Maxomys.

Fossils: Reports covering murines recovered from late Pliocene to Pleistocene sediments in caves and limestone fissures in Thailand and China contain descriptions of samples identified as species of Leopoldamys, Niviventer, and related extinct genera. The bulk of the material is comprised of isolated molars; a few samples contain cranial and mandibular fragments. Only some of the publications contain information pertinent to our inquiry.

In her detailed and comprehensive treatise on the Plio-Pleistocene rodents of Thailand, Chaimanee (1998) described samples representing extant species of Leopoldamys (L. sabanus) and Niviventer (N. fulvescens) and new species in those genera that are extinct and represented only by fossils (L. minutus and N. gracilis). Judged from her descriptions, measurements, and illustrations, all these samples are correctly allocated to genus; none represents Tonkinomys.

Pleistocene samples from Guangxi Province in southern China were determined by Chen et al. (2002) to contain examples of Niviventer and Leopoldamys, but no identifications to species were provided. Zheng et al. (1997) reported the extant Niviventer confucianus and extinct N. preconfucianus from Plio-Pleistocene fissure fillings in Shandong Province. Measurements and drawings of molar occlusal patterns presented in these reports convince us their generic attributions are correct.

Zheng (1993) has provided perhaps the most comprehensive and thorough survey of fossil murines recovered from late Pliocene and Pleistocene cave and fissure fillings. All come from the northern margin of the Yunnan-Guizhou Plateau within an area bounded by 104° to 110°E and 26° to 31°N in the provinces of Sichuan and Guizhou. By clear descriptions and a wealth of measurements, Zheng meticulously documented those parts of the samples representing extant Leopoldamys edwardsi and extinct L. edwardsioides; extant Niviventer fulvescens, N. confucianus, N. andersoni, and possibly N. excelsior; and the extinct N. preconfucianus. His descriptive and quantitative data, along with clear illustrations, provide unquestionable support for his identifications.

Zheng (1993) also described three new species in two new genera that seem morphologically similar to Niviventer and other members of the Dacnomys Division. Wushanomys brachyodus is represented by a large cranial fragment containing intact molar rows, dentary fragments with complete or partial molar rows, and numerous isolated upper and lower molars. The fossils come from sediments thought to be no older than 2.5 million years ago, and no younger than 1.0 million years ago. Judged by length of upper (9.30 mm, 1 specimen) and lower (8.80–9.20 mm, 5 specimens) molar rows, this extinct species is much larger than Tonkinomys daovantieni (uppers, 8.03–8.78 mm, 7 specimens; lowers, 7.64–8.52 mm, 10 specimens). Range in measurements of first upper (4.00–4.90 mm by 2.14–2.66 mm, 85 examples) and lower (3.40–3.95 mm by 1.90–2.20 mm, 97 specimens) molars of W. brachyodus overlaps with those of T. daovantieni (table 4), but average longer and wider (see Zheng, 1993: 190). Most of the first upper molars in W. brachyodus have three roots, but Zheng noted that some molars have a divided lingual (see Zheng, 2004: 175), which is the pattern in T. daovantieni. First lower molars have two roots, the primitive pattern that is also typical of the Vietnamese limestone rat.

Molar occlusal patterns of Wushanomys are unlike those of Tonkinomys. The anterior lamina on the first upper molar in Wushanomys consists of a straight segment formed by the complete fusion of cusps t2 and t3, and a posteriorly offset cusp t1; the second lamina is arcuate in outline; and cusp t9 is distinct in the posterior part of the tooth (see the illustration in Zheng, 1993: 192). These occlusal outlines are strongly dissimilar to the strong, tight chevrons formed by the two front laminae on the first upper molar in Tonkinomys (fig. 14) and its cusp t9, so coalesced with the large central cusp t8 that it resembles a low ridge. Laminar outlines on lower molars of Wushanomys are also nearly straight or form wider-armed chevrons, a contrast with the tight chevrons in Tonkinomys.

The cranial and mandibular fragments of W. brachyodus provide additional comparative information. Zheng did not illustrate the specimen, but described it as having very long incisive foramina that terminate 1.7 mm posterior to anterior faces of first molars, a bony palate extending about 0.7 mm beyond the molar rows to form a short platform, a very narrow interorbit (6 mm), especially for what seems to be a large skull judging from lengths of molar rows, and a flat, wide zygomatic plate (no measurement listed). Zheng did include a drawing of a partial dentary (the body of the ramus behind the molar row is gone, p. 204). The portion anterior to the molar row, along with the incisor, is long and slender, and the dorsal outline from anterior margin of the molar to incisor tip is gently concave. This conformation usually accompanies a long rostrum, and the mandibular configuration recalls that typical of the long-nosed Sulawesian murines, Bunomys prolatus and B. chrysocomus (Musser, 1991: 12–13). In Tonkinomys, posterior margins of the incisive foramina end just in front of anterior faces of the first molars, even with them, or extend only slightly beyond (no more than 0.5 mm); the bony palate extends 1.5–2.0 mm beyond the molar rows forming a much more expansive platform, and the interorbit is wide (6.6–7.3 mm; table 1), which is the usual proportion in skulls the size of Tonkinomys (and other murines, Leopoldamys neilli, for example). The bony projection containing the lower incisor in the dentary of Tonkinomys is shorter and thicker, and the dorsal outline from molar row to incisor tip is more deeply concave, a shape common to species of Niviventer, Leopoldamys, and other murines with less specialized dentaries (Musser, 1981: 272; Zheng, 1993: 204).

Zheng's description of the cranial traits in Wushanomys brachyodus tantalizingly resembles the conformation in other extinct murines, such as the Canary Islands Malpaisomys insularis, which was likely adapted to living and climbing about in the creviscular system of lava fields (Boye et al., 1992; Hutterer et al., 1988). Perhaps W. brachyodus lived and climbed about in karstic fissures, had a long rostrum, and included invertebrates in its diet. Whatever its habitat and habits, the summation of morphological traits characterizing the complement of fossil remains identifies, in our view, a monophyletic group (at the level of genus) separate from that represented by Tonkinomys daovantieni; other than Wushanomys likely being a member of the Dacnomys Group, there seems no closer phylogenetic link between Wushanomys and Tonkinomys.

Zheng (1993) described a second species of Wushanomys, W. hypsodontus, based on a few isolated molars that are high-crowned, have slightly more complex occlusal surfaces, and greater dimensions than found in W. brachyodus. The time span in which the fossils lie is 2.0 to 1.8 million years ago.

Zheng's other new genus and species, Qianomys wui, is based on a small series of dentary fragments containing full or partial molar rows, an isolated first lower molar, and two upper molars. These were recovered from sediments thought to be late Pleistocene, 400,000 to about 200,000 years ago. The two upper molars (4.40 mm and 4.60 mm long) fall within the high end and slightly beyond the range of variation in molar length we recorded for Tonkinomys daovantieni, but they are much wider (2.35 mm and 2.40 mm) than the molars of T. daovantieni (table 4). The occlusal pattern does not match that of Tonkinomys, and five roots are present. The uppers are not examples of the Vietnamese limestone rat. Even Zheng noted their close similarity to Niviventer in crown outline and number of roots, and suggested they may not be correctly associated with the lower molars, to which the generic and specific name is nomenclaturally attached. We agree. For us, Zheng's diagram (p. 199) of a first upper molar closely resembles N. andersoni in occlusal pattern; however, the size is slightly outside the range of variation we obtained from our sample of that species (table 4) and that which Zheng derived from his material of N. andersoni (p. 181).

The three intact lower molar rows of Qianomys for which measurements are provided (8.60, 9.40, and 9.43 mm) are appreciably longer than those in Tonkinomys (7.64–8.52 mm, 10 specimens) as are length and width dimensions of the first lower molars (3.66–4.20 mm by 2.10–2.34 mm, 8 specimens of Qianomys; 3.40–3.50 mm by 1.90–2.00 mm in Tonkinomys, table 4). If molar dimensions are any indication of skull and body size, Q. wui was a much larger rat than T. daovantieni. Qianomys and Tonkinomys share the same primitive root pattern, two beneath each lower molar. Occlusal patterns formed by the rows of laminae fall within the range of patterns exhibited by Tonkinomys as well as species of Niviventer. Zheng, however, claimed that none of the molars he assigned to Qianomys have anterolabial cusps on the second and third molars or posterolabial cusplets on any molar, features, in addition to number of roots, which he used to distinguish Qianomys from Niviventer. Comparable cusps do occur on the lower molars of T. daovantieni. An illustration of a partial dentary (Zheng, 1993: 204) indicates the portion anterior to the molar row to be similar in shape to that of Tonkinomys, Niviventer, Leopoldamys, and Saxatilomys.

Qianomys wui is certainly a different species than Tonkinomys daovantieni. Whether the gross similarities between the two in molar occlusal patterns signal close phylogenetic relationship or convergence cannot be determined until we can compare the fossils directly with our extant samples, and until cranial material representing Qianomys is recovered and described. Resolving the ambiguous association between the present sample of upper molars and lowers allocated to Qianomys is also critical. That genus would resemble only Saxatilomys among members of the Dacnomys Division if the five-rooted first upper molars with their Niviventer-like occlusal patterns actually belong to Qianomys with its two-rooted lower molars.

A recent report by Zheng (2004) documents identities of rodents, represented mostly by isolated molars, collected from sedimentary layers in Longgudong Cave (30°39′14.9″N/110°04′29.1″E), the “Jianshi Homind Site” in western Hubei Province of central China. Geological age was given as Early Pleistocene, but correlation between the 11 sedimentary layers sampled and the geomagnetic polarity timescale indicated a range extending from the Olduvai Event of the Matuyama Chron (the Pleistocene/Pliocene boundary) back to earlier than 2.15 million years before the present, reflecting initial deposition during the Late Pliocene (see Flynn et al., 1997, and Qiu, 1989, for illuminating expositions of Chinese geochronology at the end of the Neogene). Among the murines recovered were three species in the Dacnomys Division. One is the extinct Niviventer preconfucianus, which Zheng himself had named and described in 1993, and another is represented by isolated molars identified only as an undetermined species of Leopoldamys. The third is a new species of the extinct Wushanomys, W. ultimus, described from molars recovered from all but the top two layers of the cave sediments, most from layers dated between 1.95 to earlier than 2.15 million years before the present (Late Pliocene). Molars from W. ultimus are intermediate in size between the earlier described W. brachyodus and W. hypsodontus (Zheng, 1993), express greater hypsodonty, and differ slightly in coronal patterns. Ranges in length and breadth values for first upper (3.90–4.92 mm by 2.20–2.65 mm, 22 examples) and lower (2.90–4.30 mm by 1.98–2.30 mm, 18 specimens) molars of W. ultimus (see Zheng, 2004: 173) overlap with those of Tonkinomys daovantieni (table 4), but average slightly longer and wider. Each first lower molar of W. ultimus has two roots, the pattern common to the other two species of Wushanomys, but first upper molars have either four (anterior, divided lingual, posterior) or five roots (a small labial holdfast or nubbin in addition to the four primary roots). The pattern of the four primary roots beneath the first upper molars and two anchoring first lower molars also characterizes Tonkinomys daovantieni, but molars of that Vietnamese rat are not hypsodont, and although the extinct and extant species share molars similar in size, their occlusal patterns are dissimilar. The occlusal cusp patterns characteristic of W. ultimus resemble, except for minor details, those in W. brachyodus and differ from T. daovantieni in the configurations we previously described contrasting W. brachyodus and the Vietnamese species.

These new samples of the Dacnomys Division recovered from the Jianshi Homind Site provide additional information on Late Pliocene geographic distributions of Niviventer and Leopoldamys. The geographic range of Wushanomys is also expanded as is the diversity of species in that genus. Judged from Zheng's (2004) descriptions, measurements, and illustrations, none of the identified fossil molars represent the Vietnamese Tonkinomys.

Our final survey of fossils assesses the identification of murines recovered from Pleistocene sediments in Ma U'Oi cave in northern Vietnam, south of Hanoi (Bacon et al., 2004). One mandible containing a complete molar row was identified as Niviventer fulvescens; an upper first molar and a lower first molar were attributed to N. andersoni; and a dentary containing all molars, and isolated first and second molars, were determined as Leopoldamys sabanus. The molars were identified by comparing them with Chaimanee's (1998) treatise on fossil Plio-Pleistocene Thai rodents. Unfortunately, we are not treated to drawings of molar occlusal cusp patterns, or to measurement values for the sample of N. fulvescens.

We cannot verify the identification of N. fulvescens. That species is common in northern Vietnam and in many places occurs together with N. tenaster (results of our trapping surveys in several different regions of northern Vietnam), which is somewhat larger in dimensions of body and skull, as well as length of molar row and breadths of molars (Musser and Lunde, ms.). Niviventer tenaster also occurs in northern Thailand (Musser and Carleton, 2005), but Chaimanee (1998) did not include it in her report. The material reported by Bacon et al. (2004) should be reexamined to determine whether it is really N. fulvescens or possibly N. tenaster.

Measurements listed by Bacon et al. (2004) for two first lower molars of Leopoldamys (4.0 by 2.6 mm and 4.7 by 2.8 mm) fall within the range of variation for the larger species of that genus (Chaimanee, 1998; Zheng, 1993), and well outside the range for our sample of Tonkinomys (table 4). The lower second and third molars that were identified as Leopoldamys are also large and likely represent that genus. But these fossil molars may not belong to L. sabanus. Leopoldamys requires taxonomic revision because more species exist than are usually recognized (four were listed by Musser and Carleton, 1993, and six by Musser and Carleton, 2005, which still underestimates the number of species), and the situation in northern Vietnam is especially complex. At least two species occur there. One may be L. edwardsi, the other is usually identified as L. sabanus, but is actually a different species (Gorog et al., 2004; Musser and Carleton, 2005).

An upper first molar and lower first molar were identified as Niviventer andersoni in Bacon et al. (2004). Ma U'Oi cave is far south of the present geographic distribution of N. andersoni, which extends from eastern Xizang (Tibet) through northern Yunnan, western Sichuan, and northern Ghizhou to southern Shaanxi, and over an altitudinal range from 6000 to 10,000 ft (Musser and Chiu, 1979; Musser and Carleton, 1993, 2005). We sought to determine whether these two fossil molars belong to Tonkinomys. Dimensions of the first upper molar (3.7 by 2.0 mm) do fall within the range of variation of our values (table 4) and those tabulated by Zheng (1993: 181) for large sample sizes of N. andersoni. Breadth of the fossil molar fits within our observed range for first upper molars of Tonkinomys, but the length is much too short—based on these data, the molar did not come from the Vietnamese limestone rat. However, it may not represent N. andersoni. The width, for example, matches our range of values for N. excelsior, and the length is only 0.1 mm shorter (table 4). Niviventer excelsior is a close relative of N. andersoni that presently occurs only in high mountains north of the Yangtze River (Musser and Chiu, 1979). For N. excelsior to have once occurred in northern Vietnam, however, seems as implausible to us as the presence of N. andersoni there, especially because all the fossils of other mammals found in situ in the Ma U'Oi cave define “… a relatively modern fauna …” (Bacon et al., 2004: 312). Niviventer tenaster currently lives in northern Vietnam in forest habitats. Lengths and breadths of first upper molars from this species in our sample from that country nearly embrace dimensions of the fossil molar (table 4), and that tooth just might be from a large individual of N. tenaster. Both N. andersoni and N. excelsior can easily be distinguished from N. tenaster by occlusal patterns. In the former pair, the anterior lamina on the first upper molar consists of a nearly straight segment, formed by the coalesence of cusps t2 and t3, and a posteriorly displaced lingual cusp t1 (see figs. in Musser, 1981, and Zheng, 1993). The comparable lamina has an arcuate outline on the upper molar in N. tenaster, similar to that found in N. confucianus (Musser, 1981). The fossil molar needs to be reexamined; it is not an example of Tonkinomys, and we question its identity as N. andersoni.

The first lower molar determined to be N. andersoni in Bacon et al. (2004) is large (4.1 by 2.5 mm), and its dimensions fall far outside the range of variation in our samples of Tonkinomys, Niviventer andersoni, N. excelsior, and N. tenaster (table 4). The tooth is actually about the same size as the two first lower molars (4.0 by 2.6 mm, 4.7 by 2.8 mm) they identified as Leopoldamys. The values are either misprints, or the molar does belong to Leopoldamys; it, and the first upper molar assigned to N. andersoni, warrants restudy.

Summary: There would have been an elegant symmetry in discovering ancient, fossilized remnants hidden in limestone fissures that represented the same murine species living today in creviscular habitats within the tower karst landscape forming northern Vietnam. Unfortunately, we cannot unambiguously tie any published identifications of fossilized fragments representing Indochinese murines to the extant Tonkinomys daovantieni. Its past distribution and phylogenetic alliance to extinct taxa, as might have been illuminated by fossils, remain a mystery.

Unresolved also is the phylogenetic position of Tonkinomys with respect to Leopoldamys, Niviventer, and Saxatilomys, the three extant genera with morphologies most like that of Tonkinomys. The differences enumerated above between Tonkinomys daovantieni on one hand and species of Leopoldamys and Niviventer on the other reflect a combination of primitive and derived features. Relative to the pelage coloration, shape of tail, and its length relative to length of head and body in Leopoldamys and Niviventer, the contrasting condition in Tonkinomys is derived. Also representing specializations in Tonkinomys are the position of the squamosal zygomatic roots high on the sides of the braincase, slender or no alisphenoid struts, its long incisive foramina, a very long bony palate that projects beyond the molar rows to form a broad shelf, posteriorly diverging maxillary molar rows, simple molar occlusal patterns, and a divided lingual root beneath the first upper molar. Presence of three roots beneath each second and third molar, and two roots anchoring each lower molar represent the primitive condition in Tonkinomys compared with the derived state in Leopoldamys and Niviventer. Tonkinomys daovantieni resembles the Indochinese Leopoldamys neilli in body size, and it along with the other Indomalayan species of Leopoldamys in general cranial conformation and a few cranial particulars (position of the squamosal zygomatic roots, for example); with Niviventer, Tonkinomys shares long incisive foramina, comparable inflation of the auditory bulla relative to the underlying periotic bone, and similar molar occlusal patterns. Other traits exclude T. daovantieni from each of those genera as they are currently diagnosed (Musser, 1981). No species of Leopoldamys or Niviventer (or Chiromyscus), whether occurring in Indochina or on the islands on the Sunda Shelf, has Tonkinomys's combination of dark grayish black upperparts and dark gray underparts; thick tail with a stubby tip that is much shorter than length of head and body; large, bulbous, and contiguous plantar pads; bony palate projecting well beyond the third molars to form a shelf with a smooth posterior margin; strongly diverging (along the anteroposterior axis) maxillary molar rows; derived root pattern for the first upper molar coupled with the primitive root patterns underlying the other uppers and all the lower molars.

Both Tonkinomys daovantieni and Saxatilomys paulinae are associated with forested karstic habitats. Both genera display dark gray pelage, large and swollen plantar tubercles, a bony palate that projects beyond the molar rows to form a broad platform, maxillary molar rows that appreciably diverge posteriorly (along the anterior-posterior axis) and are long relative to length of skull, and two-rooted mandibular molars. All but the root pattern are specialized traits. This combination of attributes is not combined in any species of either Leopoldamys or Niviventer (or Chiromyscus). We do not know whether the morphological specializations shared by T. daovantieni and S. paulinae phylogenetically link these species closer to each other than to species of Leopoldamys and Niviventer, or rather reflect independent evolutionary adaptations to forested karstic environments in the Indochinese tropics. In opposition to the shared specializations, external morphology of Tonkinomys (except for relative length of tail) recalls a small-bodied Leopoldamys, while that of Saxatilomys resembles most species of Niviventer. Overall cranial form in Tonkinomys also resembles Leopoldamys, and except for those cranial traits that distinguish S. paulinae from species of Niviventer (comparisons are enumerated by Musser et al., 2005), the general cranial conformation of Saxatilomys is not unlike most species of Niviventer. Inferring the phylogenetic relationship of Tonkinomys among species now allocated to the Dacnomys Division in general, and especially to species of Leopldamys, Niviventer (including Chiromyscus), and Saxatilomys will rest on results derived from future phylogenetic analyses of information derived from both morphological and molecular sources.

Habitat and Habits

The Dao village of Lan Dat lies within a spectacular tower karst landscape, its houses and adjoining agricultural clearings scattered over flat depressions nestled among high limestone hills with steep sides and sheer, expansive cliffs. The area epitomizes the classic description of karst landforms: “Karstic terrain is commonly characterized by closed depressions, subterranean drainage and caves” (Gillieson, 1996: 59). The climate is monsoonal, with most of the year's rain concentrated from April to October. Average annual rainfall (measured at Lang Son over 50 years) is 1395 mm, and even though cold temperatures dominate from November through March, humidity remains high, the annual relative humidity averaging 81% (Điê ´u, 1995).

Most of the flat areas, once supporting tropical moist deciduous forest, have been cleared by the Dao for rice cultivation and shifting agriculture. Forest remains at the base of the cliffs, in ravines between the high towers, and on the steep slopes, and it is not unlike Whitmore's (1984: 174) description: “… limestone slopes have a dense, irregular forest with trees clinging precariously, their roots penetrating to great depths in crevices …” The forest consists of a low canopy (up to 15 m high) with taller (up to 25 m or more) emergents scattered over the steep slopes. Most trees are small (1–3 ft in diameter near the base), except in places allowing accumulation of sufficient soil where huge emergents are able to grow. Here and there a grove of emergent species forms a small area of high canopy. Palms and bamboo clumps are scattered through the understory, strangler figs are uncommon, and woody vines drape over high limestone ledges and form tangles beneath the forest canopy. The limestone towers are densely forested except for the sheer cliffs. Clumps of Pandanus are common at the base of these walls, and sometimes grow higher where they can obtain purchase in joints and fissures. We did not encounter cycads in the forest, and saw them only high on the vertical cliffs. Isolated clumps of bamboo remain in some of the clearings, and dense bamboo forest covers the steep slopes and tops of some limestone towers.

Most of the hillsides support primary forest, old and tall secondary growth covers hillsides close to houses and fields, and scrub habitats dominate clearings not utilized for crops (fig. 1). Forest once growing on the few flat or gently sloping places above the valley and between limestone towers has been either removed or thinned. The primary forest cover is also not undisturbed by human activity. Trails cut through the understory usually lead to remnants of a huge emergent tree felled and processed for lumber. Natural disturbance in the form of treefalls and landslips is eventually covered by shrubs and then taller second-growth.

The day we broke camp and climbed out of the valley our usual trail chatter was silenced by a loud rumble, which had been generated by a massive rockslide. This is a periodic, unpredictable event, according to the villagers, and potentially dangerous. Weathering of the high cliff faces through the combined forces of water dissolving the limestone, differences in temperature, and pressure from large tree roots eventually dislodge chunks of cliff wall, some small, others huge. Such an event formed the habitat in which we trapped most of the limestone rats. Above the valley floor, a wide ravine between two contiguous limestone towers had been filled by a chunk of cliff wall the size of a small house along with smaller pieces—basically a talus of large blocks between the towers. The openings among the pieces were eventually enlarged, mostly by water flowing into the rubble pile from the top, and underneath it along the bottom of the old ravine. The result is a honeycomb of passageways, some extending vertically to small openings on top of the rubble, others opening to an uncovered section of the ravine. Walls of the labyrinth are wet and the air is cool. Shrubs, vines, and moss cover the surface of the rubble, hill forest persists at the cliff bases of each tower, and small trees are scattered over the talus (fig. 2). The surface of the rubble is difficult to negotiate. Rain has sculpted the limestone into small depressions surrounded by knife-sharp edges; small to large openings are everywhere, some as deep as 8 meters that expand into large cavities; and some openings are concealed by loose limestone fragments that shift underfoot.

Thirteen of the T. daovantieni were caught in this talus. Two were trapped on the surface, one beneath a shrub, the other one at the margin of the rubble beneath a rotting trunk. Five were taken in traps set at the bottom of the ravine beneath the main overhanging limestone chunk (the configuration resembled a small cave, but was really only a low passage beneath the straddling rock). Another was captured on a moss-covered ledge just outside of this cave-like passage. Five others were caught in traps lowered by ropes through small and large holes in the talus surface to the bottom of the labyrinth, which ranged from 4 to 8 m, depending on how close the vertical passages were to sides of the ravine. In these situations, most live and snap-traps were placed directly on the passageway floors. Traps placed on limestone ledges above this rubble near bases of the towers, or in the dense forest covering steep slopes away from the talus, yielded only Niviventer fulvescens and N. langbianis.

Tonkinomys daovantieni was not the only inhabitant of this talus. We captured a Rattus andamanensis and a Niviventer fulvescens at the mouth of the cave-like passage, an N. fulvescens on the surface, and one at the bottom of the deepest chamber.

The only limestone rat not trapped in or on the talus was taken in mature second growth hillside forest behind camp, which is at the other end of the valley from the rubble where all the others were caught. Here between the massive cliffs and valley floor is a steep middle region of limestone talus, formed of small to massive chunks that broke away from the cliff above. This is the only talus-like structure we encountered here. While soil has filled in most of the spaces among the blocks, apparently enough interstices and underground extensions exist to provide the proper habitat for T. daovantieni. Either that or the animal wandered from places on the tower we did not survey. The forest covering the talus is dense and festooned with woody vines. Niviventer fulvescens and the arboreal N. langbianis were common.

Populations of T. daovantieni apparently have a patchy distributional pattern. Despite our sampling efforts on other hillsides, where traps were placed beneath rocky outcrops, on ledges, at the bottom of deep fissures in cliffs, and on the forest floor, we did not encounter any T. daovantieni. A large cave was accessible, but it was being used as a corral for cattle; the interior was dry, and mostly scrub and young secondary forest persisted near the entrance. Traps placed on ledges inside the cave and outside near the entrance yielded nothing. We also set traps in a small, wet cave used as a roosting site by bats. No limestone rats were caught. Caves occur throughout the area, but the openings to most are high in the cliffs and inaccessible to us.

Certainly more field study is required to assess the habitat requirements of T. daovantieni, but our trapping experience suggests this forest species to be petricolous. It seems to require large areas of rock debris or talus that provides sufficient passageways and chambers for protected den sites, and a cool, damp environment. Long vibrissae; large, bulbous palmar and plantar pads; and a tail that is considerably shorter than head and body, that is thick and stubby-tipped, and that has a well-haired distal half and tip are adaptations important in a creviscular environment. The vibrissae monitor the cracks and walls for prey, surface information, and passage width; the fleshy pads provide purchase on rock surfaces; and the tail is used for balance as the rat scurries up and over the rocks and through the crevices and passageways, or as a fifth appendage when pressed against the limestone (the hairs would aid in purchase by pressing against the rough surface). We see similar adaptations in other petricolous muroid rodents. The bushy-tailed woodrat, Neotoma cinerea, for example, is a North American cricetid analog to the Vietnamese murid, T. daovantieni. This species of Neotoma dens in caves, fissures in high cliffs, and deep in talus slopes (Finley, 1958; Grayson, 1999; Musser's trapping experience). Size and fleshiness of its footpads relative to foot size, length of tail relative to head and body length, and even overall body size is closely similar to these qualities in the limestone rat (see measurements provided by Finley, 1958: 322–325).

A cool and damp environment of crevices and larger passageways may be important to the diet of T. daovantieni because the diversity and numbers of arthropods is probably higher there than in dry situations. Of two limestone rats kept captive, one devoured two kinds of cockroaches: a large-bodied species caught at the talus, and a smaller species collected in village houses and at camp. This rat refused offers of corn, manioc, snails, earthworms, grasshoppers, beetles, and small frogs. Our other captive rejected cockroaches offered during daytime, but headless bodies we piled in its cage were gone by morning. We are not sure whether they were dragged away by insects or eaten by the rat. We extracted stomach contents from two rats. In one, the stomach was distended with bait (mixture of peanut butter, bacon pieces, oatmeal, and raisins) in which were scattered pieces of very small beetles, which were likely on the bait and inadvertently ingested by the rat. The stomach of the other was partly full, containing fragments of sclerites intermingled with soft unidentifiable animal tissue. Tonkinomys daovantieni includes insects in its diet, but other components remain to be discovered.

The limestone rat is probably nocturnal, but may also be active at times during the day. We were drawn to the talus because during one late morning Lunde had glimpsed a large, dark gray rodent scurrying across the limestone and disappearing into the rubble. Results from the first night's traps revealed the rodent to be the limestone rat. We kept our traps in place during day and night, and often checked them in the evening, but never caught any rats during the day.

Closing Observations

Other species of Indochinese murids live in the tropical forests of karst regions but, except for Saxatilomys paulinae, none have any of the diagnostic morphological attributes of Tonkinomys daovantieni and none are so ecologically tied to the kind of creviscular habitat in which we discovered the Vietnamese species. Two Thai murids are generally considered closely associated with limestone. Leopoldamys neilli is documented by a few specimens captured on the upper sections of forested limestone cliffs in Saraburi Province and adjacent Kanchanaburi Province in southwestern Thailand (Marshall, 1988; specimens in AMNH and USNM). Wiles (1981: 49), however, captured three individuals in lowland bamboo forest in a different part of Kanchanaburi Province. The trap sites averaged 115 m from the base of the nearest limestone mountain, and L. neilli was not encountered in the other habitat types Wiles surveyed, indicating the species to prefer lowland bamboo forest, which “… challenges the belief that L. neilli is strictly a cliff-dwelling rat.”

Niviventer hinpoon is apparently endemic to the Korat Plateau in Saraburi Province of southern Thailand (Marshall, 1988; specimens in AMNH and USNM). One sample, which includes the holotype, was collected halfway up the face of a forested limestone cliff outside the entrance to a cave. Another series came from “… high above the valley floor, in scrubby vegetation at the base of vertical limestone cliffs … ,” (Marshall, 1988: 459). Other species of murids were captured in the forested stream valley below the cliffs, but not N. hinpoon. Marshall (1988) also noted that one individual had been taken inside a limestone cave at a different locality. Whether N. hinpoon dens in clefts in cliffs or in talus is not recorded. Except for its tail, which is equal to or slightly longer than length of head and body, its buffy gray underparts, and its smaller body size (Marshall, 1988), the morphological traits of N. hinpoon are similar to those of N. fulvescens; our multivariate analyses of morphometric cranial and dental data place it closer to N. fulvescens than to any other species (Musser and Lunde, ms.). Relative to palmar and plantar areas, the pads are no larger or set no closer together than in N. fulvescens (see fig. 9) or any of the other species of Niviventer. Nothing in the morphology of N. hinpoon points to an obligate limestone cliff-dweller. We need to learn details of its actual microhabitat and other aspects of its life history.

Nothing comparable to the Vietnamese Tonkinomys, or to the Lao Saxatilomys, in morphology or habits has so far been found in the karst landscapes of Thailand.

Whitmore (1984: 148) noted that karst landscapes covered only a small part of tropical Asia. In Indochina they occur in southern China, north and central Vietnam, central Lao PDR, southern Cambodia, Thailand, and northeastern Burma (Middleton and Waltham, 1986; Zhang, 1989; Tuyet, 1998; also see the references cited in Whitmore, 1984). Except for the large limestone plateau in southern China, karst landscapes in Indochina are small in area, scattered over the region, and constitute isolated geological islands (see the informative distribution maps in Middleton and Waltham, 1986). They are difficult places in which to work, and most Indochinese karst habitats have been inadequately sampled for murid rodents.

Saxatilomys paulinae and Tonkinomys daovantieni are the only petricolous Indochinese murines that to date have been found to be endemic to tropical karstic environments. The Khammouan Limestone in central Lao PDR defines the known distribution of S. paulinae. That tower karst landscape is part of the Quy Dat massif (or Kebang karst block; Tuyet, 1998: 189), which extends eastward at the same latitude into Bienh Tri Thien Province of north-central Vietnam. In the Lao segment, S. paulinae, the gymnure, Hylomys megalotis, and the hystricomorph rodent, Laonastes aenigmamus form a small community that may be restricted to forested, rocky habitats within the extensive and massive karst landscape; this same mammalian assemblage may also occur in the Vietnamese segment of the Quy Dat massif (Musser et al., 2005). The Quy Dat outcrop forms a karstic island approximately 475 km south of the expansive Vietbac tower karst region in northern Vietnam where Tonkinomys daovantieni occurs. That limestone rat, isolated on a geological island separate from the one on which S. paulinae was found, has been recorded only from the type locality in northeastern Vietnam but may have a wider geographic distribution in forested habitats throughout the karst landscapes of northern Vietnam and adjacent southern China. Unlike the Khammouan Limestone region, no other petricolous mammalian limestone endemics have been discovered in the same landscape occupied by T. daovantieni.

We have asked ourselves if the Vietnamese and Lao limestone rats are the only murine components of a pattern reflecting evolutionary adaptation of ancestral murines to forested karstic environments, or if other karstic islands in Indochina support different and unique murine endemics. Surveys for small mammals, such as ours that yielded Tonkinomys, would uncover the mammalian diversity in these extraordinary isolated landscapes and provide an answer to our question.