Sheath-tailed bats (Chiroptera: Emballonuridae) from the early Pleistocene Rackham’s Roost Site, Riversleigh World Heritage Area, and the distribution of northern Australian emballonurid species

Materials

The fossil emballonurid specimens described here were recovered by acetic acid processing of fossiliferous limestone collected from Rackham’s Roost Site in the Riversleigh World Heritage Area, north-western Queensland. Fossil specimens are from the Queensland Museum fossil collection, Brisbane, Australia (prefix QM F), and the University of NSW vertebrate fossil collection, Sydney, Australia (prefix AR).

Comparative material used in this study included the skulls of 90 modern emballonurids as follows: Saccolaimus flaviventris (seven specimens), S. saccolaimus (2), Taphozous australis (9), T. georgianus (55), T. hilli (7) and T. troughtoni (10) (Supplemental Datas S1 and S2). These specimens are from the American Museum of Natural History, New York, NY, USA (prefix AMNH), Australian Museum mammal collection, Sydney, Australia (prefix AM M), and the Museum and Art Gallery of the Northern Territory, Darwin, Australia (prefix U).

Cranial and dental measurements

Measurements of 71 cranial and dental variables, for 90 extant adult emballonurid specimens, were collected for quantitative analysis (Fig. 1; Supplemental Data S3 (Fig. S3.1; Tables S3.2 and S3.3)). Skull characters were those used by Chimimba & Kitchener (1991), and tooth and dentary characters followed Hand (1985). Mean measurements of left and right dentitions were used. Because the raw data from Chimimba & Kitchener’s (1991) morphometric analysis of Australian emballonurids was not published, and has subsequently been lost, measurements of many of the same specimens used in that study were repeated but with additional dental characters included.

Specimens were measured at the University of New South Wales using a Wild 5MA stereomicroscope with Wild MMS235 Digital length measuring set (accurate to 0.01 mm) and Mitutoyo M.N. 85 dial calipers (accurate to 0.05 mm) for larger measurements (greatest skull length, basicranial length and cranial height). Specimens with prefix AM M were measured at the Australian Museum using a Leica MZ95 stereomicroscope with graticule (accurate to 0.01 mm) and digital TESA CAL IP65 calipers (accurate to 0.01 mm). Measurements produced by these devices were cross-checked for repeatability by measuring a subset of emballonurid specimens at both institutions. Fossil specimens were imaged using a QUANTA 200 Scanning Electron Microscope in the Mark Wainwright Analytical Centre, UNSW Sydney, and modern specimens by Leica M205 C stereo microscope with Leica DFC290 digital microscope camera or Nikon DSLR D5100 digital camera.

Analytical methods

Univariate analyses of modern and fossil species were conducted to examine levels of variability in craniodental measurements within modern taxa and the fossil sample.

The Coefficient of Variation (CV) was used to compare variability in the fossil sample with that found in modern emballonurid species: $CV={\displaystyle \frac{100s}{\mathrm{X}}}$ (x is the mean and s is the standard deviation).

The statistical program PAlaeontological STatistics (PAST; Davis, 1986; Hammer, Harper & Ryan, 2001) was used for multivariate analysis. Several variables from Chimimba & Kitchener (1991) and Hand (1985) were excluded from final analyses due to the high degree of missing data in fossil or modern samples, or because they were identified as uninformative in preliminary analyses (Supplemental Data S3 (Fig. S3.1; Glossary)).

To test for sexual dimorphism in craniodental measurements of Australian emballonurid taxa, a PERMANOVA (non-parametric MANOVA or NPMANOVA) of the best-represented species in the study, T. georgianus, was performed on 43 characters (Table 1). Due to the increased likelihood of misidentification in specimens both from the east of Mount Isa and from its surrounding area, this analysis was restricted to 32 specimens from west of Mount Isa (2 of 34 available specimens were not used due to missing sex data; Supplemental Data S1 (Table S1.1)). PERMANOVA was considered the best analysis to test for sexual dimorphism, instead of parametric ANOVA or MANOVA, due to: (1) 35% of characters being non-normally distributed in raw format and 40% when log-transformed; (2) the data failing Box’s M test for homoscedasticity (raw data: F = −7.18, p = 0; log-transformed: F = −1.56, p = 0); and (3) the data failing multivariate normality tests determined on PCA scores (raw data: Mardia kurtosis = −7.57, p <<< 0.001, Doornik & Hansen omnibus test, Ep = 105.7, p = 0.0005; log transformed data: Mardia kurtosis = −7.57, p <<< 0.0001, Doornik & Hansen omnibus test, Ep = 91.11, p = 0.009). Homogeneity of variances, an assumption of PERMANOVA, was confirmed using the ‘permutest.betadisper’ function (a multivariate analogue of Levene’s test) in R (F = 1.60, p = 0.24; Anderson (2006); R Core Team (2017)).

A Principal Components Analysis (PCA) was undertaken to investigate whether currently recognised modern Australian emballonurid species could be differentiated in morphospace, using craniodental characters, and to investigate multivariate differences within fossil taxa suggested by univariate analyses as well as qualitative analyses (see “Systematics”). All characters for all modern taxa were included in these analyses. Further analyses, combining modern and fossil taxa, included amalgamated cranial and dental characters, as well as a dataset of dental characters alone (canines, premolars and molars).

Distributions of T. georgianus and T. troughtoni overlap in north-western Queensland, with both species occurring in the Mount Isa Inlier bioregion (Environment Australia, 2000). The identities of 18 Taphozous specimens (identified as T. georgianus) collected previously from this geographic area, and currently held in museum collections, were checked in this study using Canonical Variate Analysis (CVA) classification with a training set of T. troughtoni, T. georgianus (only specimens from west of Mount Isa), T. australis and T. hilli (Table 2). The training data is used to determine discriminant functions that exhibit maximum variation between a priori groups, while also minimising variation within these groups, which then allow for classification of unknown specimens (Harper & Owen, 1999). Two additional specimens (AM M27892 and AM M8550), previously identified in the PCA as outliers, were also included.

Canonical Variate Analysis (CVA) was used on the complete character dataset for modern Australian emballonurids to investigate generic and species boundaries, particularly in the T. georgianus/T. troughtoni complex, and on dental variables alone for fossil and modern species combined. CVA allowed for examination of morphometric differentiation of taxa, determination of potentially useful discriminating variables, and classification of fossil specimens.

Quantitative analysis of modern Australian emballonurids

A PERMANOVA (using the Euclidean similarity index) performed on specimens of Taphozous georgianus, from the west of Mount Isa, revealed no sexual dimorphism in 11 analyses incorporating from 2 to 43 cranial and dental characters (Table 1), justifying combining males and females in subsequent analyses. The pattern of clustering in a PCA of craniodental characters supported these results, with substantial overlapping of 95% elliptical confidence regions and centroids (Fig. 7).

Simpson, Roe & Lewontin (1960) proposed that, in individual linear measurements of anatomical elements of mammals, CV values between 4 and 10 indicate reasonably unified samples (i.e. single species), while values lower than 4 suggest sample size was inadequate to assess variability and values greater than 10 indicate that the sample probably represents more than one species (see “Discussion” in Cope & Lacy, 1992, 1995; Plavcan & Cope, 2001). Minimal variability was found within each modern emballonurid species across 71 measurements of the Coefficient of Variation (CV) (Supplemental Data S3 (Table S3.2)). Between 78% (S. flaviventris) and 96% (T. troughtoni) of characters within each species examined, with a sufficient sample size (CV > 4), were within the range suggesting a unified sample (CV < 10). M³ characters (BUC L, LIN L and W), however, displayed considerable variability, with values of up to 41.41 for example in S. flaviventris. Similarly, variables relating to paracristid and metacristid length exhibited greater variation. Examination of T. georgianus specimens from west of Mount Isa only, to reduce possibility of mis-identified T. troughtoni, produced similar results to the sample including all T. georgianus specimens (Supplemental Data S3 (Table S3.2)). The values recorded here suggest characters relating to M³ size and cristid length are not reliable for distinguishing Australian emballonurid species.

The fossil assemblage exhibited CV values greater than 10 for several variables (e.g. M² width and heel width gave results of 11.56 and 10.57, respectively) (Supplemental Data S3 (Table S3.3)). The greatest CV values, however, were for M₃, with all five variables giving values between 11.41 and 17.48. These results suggest that more than one species may be represented in the fossil sample.

Principal Components Analysis of modern emballonurid specimens separated most currently recognised species on the first two principal components (representing 72.1% and 8.0% of the total variation, respectively), with only partial overlap of 95% confidence ellipses among species (Fig. 8). Taphozous hilli, T. georgianus, T. troughtoni and S. flaviventris are separated primarily on PC1, with the latter only partially overlapping with T. troughtoni. Saccolaimus saccolaimus grouped closer to species of Taphozous than it did to S. flaviventris but, along with T. australis, was differentiated from other species on PC2. Characters with the highest loadings for PC1 include, in order of decreasing contribution: dentary length (DL); basicranial length (BL); greatest skull length (GL); zygomatic width (ZW); lower tooth row length (C₁–M₃); and rostrum length (ROL). PC2 characters with significant negative loadings are distance outside promontorium (OP), braincase width (BW) and mastoid width (MW), while characters with high positive loadings include DL, distance of M₃ to dentary condyle (RC) and least interorbital width (LOW). No further separation was obtained using other Principal Component combinations (e.g. PC 1v3, 2v3), with even more overlap occurring between species.

A Canonical Variate Analysis (CVA) of cranial and dental variables grouped all Taphozous species together, while S. saccolaimus and S. flaviventris were isolated from each other, and the Taphozous cluster (Fig. 9). Characters with high loadings on the first axis (explaining 61.9% of the variation) are primarily cranial: DL, posterior nose-shield width (PNS), MW, GL, BL and ZW. C₁–M₃ is the only dental variable with a loading comparable to these. BL, DL, GL and ROL also have high loadings for axis two (24.5% of variation), while LOW and OP are inversely significant contributors to axis two loadings.

Reardon & Thomson (2002) significantly revised the expected distributions of T. georgianus and T. troughtoni in northern Australia. More recently, Armstrong et al. (2017) and Armstrong & Reardon (2017) re-examined the distributions for these species (Fig. 10). A CVA to confirm the identity of T. georgianus specimens from east of Mount Isa, using T. troughtoni, T. hilli, T. australis and T. georgianus from west of Mount Isa as training data, successfully classified 100% of specimens within the training set (Fig. 11). Of the 18 specimens from east of Mount Isa, and identified as T. georgianus, the CVA classified 14 as T. georgianus, one as T. hilli (AMNH107770) and three as T. australis (AM M6024, AMNH183557, AMNH66145). The outliers AM M8550 and AM M27892, both also originally identified as T. georgianus, were classified in the CVA as T. troughtoni and T. hilli respectively. AM M8550 is positioned close to the T. troughtoni ellipse, while AM M27892 was isolated from all other specimens, approximately equidistance from T. hilli and T. georgianus ellipses. One of the three specimens re-classified as T. australis, AMNH66145, plotted near the centroid of the 95% ellipse for T. australis, while the other two were situated between the ellipses for T. australis and T. troughtoni (Fig. 11).

The main discriminators for species of Taphozous, on axis one (69.9% of variation), were length-related (Table 2; Fig. 11). Taphozous troughtoni exhibits the longest dentary, basicranium, skull, rostrum and palate, while T. hilli has the shortest. Specimens of T. georgianus and T. australis were of intermediate length for these variables but were separated on axis two (16.6% of variation) by variables relating to width; T. georgianus has a wider basicranium and mastoid, as well as greater distance outside promontorium and interorbital width, relative to T. australis.

Quantitative analysis of fossil taxa

At least three emballonurid taxa, T. georgianus, T. troughtoni and Saccolaimus sp., appear to be represented in the Rackham’s Roost Site fossil sample. In a PCA of extant and fossil specimens, excluding six specimens (two outliers and four T. georgianus reclassified by CVA as other Taphozous species), all fossil specimens clustered tightly within the T. georgianus 95% ellipse (Fig. 12). However, this PCA used dental variables as well as cranial, for which data is mostly absent for the fossil specimens (e.g. the ten variables with the highest loadings were not represented in the fossils).

A PCA was also run using only dental variables for which more than one data point was represented in the fossil sample (Fig. 13). Most fossil specimens clustered close to the T. georgianus centroid, although some specimens also overlapped with T. troughtoni and T. australis ellipses and near Saccolaimus saccolaimus. First upper molar width and length had the highest loadings on PC1, separating Taphozous species from each other, and Saccolaimus flaviventris from all other species.

A CVA performed on the same dataset also indicated the congruity of T. georgianus and most fossil specimens, with only three specimens (QM F23868, QM F23846 and QM F23845) being classified as T. troughtoni (Fig. 14). Upper first molar length and width were again the greatest contributors to separating species on axis one (77.4% of variation), while upper canine and upper fourth premolar length were important for separating T. hilli and T. troughtoni from other Taphozous species on axis two (12.2% of variation).

Saccolaimus species were isolated from each other, as well as all Taphozous and the fossil specimens. Removal of Saccolaimus spp. from the CVA increased separation of Taphozous species (Fig. 15; for example T. australis partially overlaps only with T. hilli). Lower canine length and upper first molar length were the largest contributors to species differentiation on axis one (72.3% of variation), while upper canine length and upper first molar width were important for separating T. australis from other Taphozous species on axis two (20.8% of variation). One specimen, previously classified as T. troughtoni (QM F23846), was isolated from Taphozous species (Fig. 15). These results therefore support the morphological analysis conclusion that QM F23846 may represent a new species of Saccolaimus.

Quantitative assessment of species boundaries in extant Australian emballonurids

Chimimba & Kitchener (1991) used morphological systematics, protein electrophoresis and immunology to separate the eight extant Australian emballonurid species. The examination here of six emballonurid species (T. kapalgensis and S. mixtus having been omitted due to insufficient data) supports the taxonomic validity of species proposed by Chimimba & Kitchener (1991).

Although Principal Components Analysis (PCA) does not test hypotheses, it can help visualise natural groupings that correspond to interspecific differences and/or similarities in multivariate data (Hammer & Harper, 2006). In the PCA of six extant Australian taxa, Saccolaimus flaviventris had the greatest magnitude of separation from other emballonurid species, while S. saccolaimus grouped closer to all species of Taphozous (being separated only by the less significant PC2). Although the 95 percent ellipses at least partially overlapped for all species of Taphozous, the centroids for T. hilli, T. georgianus and T. troughtoni were separated on PC1, while T. georgianus and T. australis separated on PC2. This reflects a close morphological similarity among most Australian Taphozous species, with absolute size (PC1) contributing most to the separation. In all analyses of modern taxa here, S. flaviventris also separates from other emballonurid species mostly on the basis of its much larger size. Chimimba & Kitchener (1991) recognised that S. flaviventris and S. saccolaimus are morphologically similar, while also being morphologically distinct from all other Australian emballonurids.

Canonical Variate Analysis (CVA) of combined cranial and dental characters provided the best means of discriminating all Australian emballonurid species examined. However, the most significant discriminators of emballonurid species were primarily cranial. Characters relating to skull and dentary length (BL, GL and DL) and width of the posterior nose-shield, mastoid and zygomatic arch (PNS, MW and ZW) were the most important for distinguishing: (1) Saccolaimus species from each other; (2) Saccolaimus from Taphozous; and (3) T. troughtoni from other Taphozous species. For differentiation of Taphozous species, rostrum length (ROL) and, inversely, distance outside promontorium (OP) and least interorbital width (LOW) were also important variables. At one end of the morphocline, T. troughtoni is characterised, for example, by a relatively long and narrow snout, while T. hilli possesses a shorter and wider rostrum.

The CVA results largely reflect size variation between species, with the highest loadings supporting the conclusions of other studies that have investigated size variation (Freeman, 1981; Chimimba & Kitchener, 1991). This is not surprising because, for instance, the mean maxillary tooth row length in T. hilli (8.89 mm), T. georgianus (10.29 mm) and S. flaviventris (12.38 mm) appears to reflect a size cline. It has been suggested that such clines (possibly the result of character displacement to reduce competition between sympatric congeners) can result from resource partitioning among otherwise morphologically similar and closely related species of insect-eating bats (Freeman, 1981; Chimimba & Kitchener, 1991). Evidence for sexual dimorphism could not be found, at least in T. georgianus, supporting the findings of both Chimimba & Kitchener (1991) and Carpenter, McKean & Richards (1978) that cave-dwelling bat species display less pronounced sexual dimorphism than their forest-dwelling counterparts.

Canonical Variate Analysis confirmed that at least 14 of the 20 extant Taphozous georgianus specimens included here, but collected east of Mount Isa, were correctly identified, with the range extending north as far Chillagoe and east to Rockhampton, Queensland. This is consistent with some previously proposed distributions for this species (Chimimba & Kitchener, 1991). Two specimens, AM M27892 and AM M8550, were consistently outliers and re-classified as T. hilli and T. troughtoni respectively. AM M27892 is from the Alligator River in the Northern Territory, which is consistent with the central and western Australian distribution for T. hilli, albeit a northern outlier. AM M8550 is from Barry Cave, in the Northern Territory, approximately 150 km west of Camooweal. This supports the range extension for T. troughtoni suggested by Reardon & Thomson (2002). CVA also indicated that at least one other specimen identified as T. georgianus from east of Mount Isa, AMNH107770, may also be T. hilli. This specimen was re-classified as T. hilli primarily based on a very large occipital region (OP) and relatively long M¹. However, AMNH10770 is from Pentland, near Charters Towers, Queensland, which would extend the known distribution of T. hilli east by several hundred kilometres. CVA results also placed this specimen close to the T. australis cluster. This specimen should be re-examined to confirm the identification as T. hilli.

A further three specimens, originally identified as T. georgianus from east of Mount Isa, were re-classified in the CVA as T. australis: AMNH183557 and AMNH66145 from Chillagoe, Qld and AM M6024 from Morinish, near Rockhampton, Qld, sites within 150 km of the east coast. Taphozous australis is usually described as having a coastal distribution (Duncan, Baker & Montgomery, 1999). However, records are known from inland localities, including central Cape York (Atlas of Living Australia, 2020), and it is feasible that the three specimens are T. australis.

For modern emballonurid taxa, intraspecific variation in the metric characters, as measured by the Coefficient of Variation (CV), was generally low (falling within that expected for single species samples (Cope & Lacy, 1992)). However, the dimensions of M³ and molar cristid characters were particularly variable within species compared to all other craniodental measurements (Supplemental Data S3 (Table S3.2)). High variability in posterior upper molar metrics has been identified in many mammal groups including, for example, canids (Gingerich & Winkler, 1979), viverravids (Gingerich & Winkler, 1985) and primates (Gingerich & Schoeninger, 1979; Morita, Morimoto & Ohshima, 2016). Polly (1998), however, cautioned that a correlation between CV and size may artificially inflate apparent variability, although this could not be confirmed for carnivores in subsequent studies (Dayan, Wool & Simberloff, 2002; Meiri, Dayan & Simberloff, 2005). Similarly, Wolsan et al. (2015) found that hypotheses concerning developmental fields, the timing of tooth formation, occlusal complexity and sexually dimorphic hormonal activity, could not explain size variation along the pinniped toothrow. Until such relationships are examined in emballonurids, the M³ should be considered unreliable for discriminating between Australian species.

The similarity in size and morphology between T. georgianus and T. troughtoni precludes separation of these species based on qualitative and quantitative analyses of dental characters alone. However, the inclusion of cranial characters increased taxonomic resolution. It is clear that cranial material is essential for the correct identification of these two often-confused species.

Qualitative and quantitative assessment of fossil emballonurids from Rackham’s Roost Site

The results indicate that CV values for M₃ show the most variability of all measurements (Supplemental Data S3 (Table S3.3)). This reflects variation in both width and length of M₃ and, combined with CV values for a suite of other craniodental characters, is consistent with recognition of more than one emballonurid species within the Rackham’s Roost Site fossil sample. Qualitative and quantitative analyses indicate the presence of three emballonurid taxa: Taphozous georgianus, T. troughtoni and Saccolaimus sp. These are the first emballonurids described from the Australian fossil record.

Although generally larger, fossil specimens retained morphology typically found in species of Taphozous rather than Saccolaimus (see “Systematics”). Multivariate analyses could not differentiate most fossil specimens from T. georgianus. Two specimens were identified as T. troughtoni based on morphometrics of P⁴ and, to give greater confidence, this assignment should be re-examined once further material becomes available.

The large number of T. georgianus specimens (n = 59), relative to only a few T. troughtoni (n = 2) and Saccolaimus sp. (n = 1), may reflect actual relative abundance of the three taxa at that time in the immediate area of the fossil site. Alternatively, it could reflect differences in the proximity and/or susceptibility to predation by the cave-dwelling Ghost Bat Macroderma gigas (see “Palaeoecology” below).

The occurrence of Saccolaimus sp. (QM F23846) in the Rackham’s Roost assemblage indicates that both genera of Australian emballonurids were present by at least the early Pleistocene. Overall molar dimensions in the fossil taxon differ significantly from the smaller S. saccolaimus and S. mixtus, with other morphological differences from S. flaviventris including the presence of a paracingulum, a paraloph, a well-developed parastyle and a depression between the protocone and heel.

Additional comparative specimens of Saccolaimus species are required for quantitative analysis, together with more fossil material, before the identity of the fossil species can be confirmed. However, morphometric analyses presented here indicate that it is unlikely to be S. saccolaimus or S. flaviventris. The restriction today of both S. saccolaimus and S. mixtus to north and north-eastern coastal Australia is associated with their habitat preference for tropical forests along the coastal fringe (Churchill, 2009). The absence of wet forest at Riversleigh during the early Pleistocene (Archer, Hand & Godthelp, 2000), also weighs against the Rackham’s Roost Saccolaimus representing either of these species.