Specimens, taxonomic sampling, and abbreviations

Fresh tissues of Stochomys longicaudatus, Dephomys defua, Typomys planifrons, and T. trivirgatus were acquired through expeditions to Ghana (in 1999; Decher et al. 2021), Côte d'Ivoire (in 2002; Decher et al. 2005), Guinea (in 2003 and 2008; Norris 2006), Sierra Leone (in 2006; Decher et al. 2010), and Liberia (in 2010). Tissues of Hybomys basilii were obtained from the Field Museum of Natural History (FMNH), Chicago, Illinois, and those of H. univittatus from the Muséum National d'Histoire Naturelle (MNHN), Paris, France. Voucher specimens are deposited in the U. S. National Museum of Natural History (USNM), Smithsonian Institution, Washington, D. C., in the FMNH or in the MNHN (Table S1); tissues of the USNM vouchers are maintained at the University of Vermont Natural History Museum. Gene sequences for all genera and species of the Hybomys division sensu Musser & Carleton (2005) other than D. eburneae were obtained either from these tissues or from accessions downloaded from GenBank (Table S1). Mitogenomes assembled as by-products of sequence data obtained by anchored hybrid enrichment by Mikula et al. (2021) were used to obtain sequences for two mitochondrial genes (Cytb and 12S rRNA) from six taxa of the Hybomys division (Genbank Accession numbers MN807597-MN807602). The quality of the Cytb sequences obtained from these mitogenomes were assessed by Kimura-2-parameter distance (Kimura 1980) and amino acid p-distance intraspecific comparisons using MEGA6.0 (Tamura et al. 2013). Sampling localities of sequences of the Hybomys division used in this study are shown in Fig. 1B.

Musser & Carleton (2005) did not recognize formal tribes of the Murinae (see Discussion) but allied the 126 murine genera they considered valid into 29 divisions, including the Hybomys division. Rowe et al. (2019) subsequently refined this number to 33 divisions, excluding 12 extant and extinct genera, which Rowe et al. (2019) considered incertae sedis. To inform selection of sister groups and second-order outgroups, we relied upon hierarchical relationships divulged by recent, taxonomically broad molecular phylogenies of Murinae and accompanying tribal reclassifications (Lecompte et al. 2008, Schenk et al. 2013, Missoup et al. 2016, Steppan & Schenk 2017, Rowe et al. 2019). Thus, we included an exemplar of Deomyinae (Acomys) as sister-group to Murinae and selected generic representatives of several murine tribes (Apodemini, Millardini, Murini, Otomyini, Phloeomyini, and Rattini) as potential outgroups to Arvicanthini, all based on sequences obtained from GenBank (Table S2). We underscore that the tribe Arvicanthini, which includes the genera of the Hybomys division (Lecompte et al. 2008), the hypothesized ingroup, is well covered among our genetic samples and numbers 49 species representing 19 arvicanthine genera (Table S1, Table S2).

Morphological terms used to describe cranial and dental features of murine rodents, as mentioned in the Discussion, are defined and/or illustrated in Rosevear (1969) and Carleton & Robbins (1985). We abbreviated the International Code of Zoological Nomenclature, 4^th edition (International Commission on Zoological Nomenclature 1999), to the “Code” and abbreviated its authorship as ICZN, together with the year and relevant article.

DNA extraction, sequencing, and alignment

DNA extraction was carried out using the Gentra Puregene Mouse Tail Kit (QIAGEN). Approximately 5-10 mg of liver or other tissue stored in 95% ethanol was soaked briefly in sterile distilled water (Kilpatrick 2002) before being ground to a fine powder in liquid nitrogen and incubated overnight at 55 °C in 300 µl cell lysis solution and 1.5 µl Proteinase K. Extracted DNA was air dried overnight and rehydrated with 50 µl of sterile water. The quantity and quality of the extracted DNA was assessed on a NanoDrop 1000 (Thermo Scientific) spectrophotometer.

Cytochrome b (Cytb) was sequenced in three parts with the primer pairs Cytb A and Cytb E, Bath3 and 752R, and Ru13 and End2 or Cytb G and Cytb J (Table S3). Primers 12S-1S and 12S-3'GW were used to sequence the entire 12S rRNA or 12S-1S and 12S-2'NS, and 12S-2NS and 12S-3'GW were used to sequence in two parts (Table S3). Interphotoreceptor retinoid-binding protein (Rbp3) was sequenced with the primer pair IRBP119A2 and IRBP8F and growth hormone receptor exon ten (Ghr) was sequenced in two parts with primer pairs GHREXON10 and GHR8, and GHR7 and GHR2 (Table S3). Cytb and 12S rRNA were amplified with 35 cycles with denaturation at 94 °C for 1 min, annealing at 50 °C for 1 min and extension at 72 °C for 1 min. Rbp3 and Ghr were amplified with 40 cycles with denaturation at 94 °C for 1 min, annealing at 58 °C for 1 min and extension at 72 °C for 1 min.

Amplification reactions were conducted in 25 µl volumes with PuRe Taq Ready-To-Go PCR Beads (GE Healthcare Life Sciences). PCR products were visualized with ethidium bromide after gel electrophoresis on 1.2% agarose gels. ExoSAP (Exonuclease and Shrimp Alkaline Phosphatase) (ThermoFisher Scientific) was used to purify the PCR products. A few amplifications were conducted in 25 µl reactions using Titanium® Taq (Takara Bio USA) following the manufacturer's protocol. Sequencing reactions (15 µl) were conducted with 1-2 µl of ExoSAP product, 1 mM primer (forward or reverse), nuclease free water and fluorescent tagged terminator (BigDye v3.1; Applied Biosystems). Sephadex columns were used to purify the products of the sequencing reaction before they were fractionated on an Applied Biosystems 373 automated DNA sequencer. Some of the later sequences were obtained in 16.4 µl reactions containing 1.5 µl of a 1:10 to 1:20 dilution of the ExoSAP product, 1.6 µl of 2 mM primer and purified with the BigDye X Terminator (Applied Biosystems) reactions. Chromatograms were visualized and edited in Chromas 2.6.2 (Technelysium:  http://technelysium.com.au/wp/chromas/).

Two mitochondrial (Cytb and 12S rRNA) and two nuclear (Rbp3 and Ghr) genes were included in the analyses. Cytb (1,140 bp), Rbp3 (1,236 bp) and Ghr (921 bp) were aligned using ClustalW (Thompson et al. 1994) in Mesquite 2.75 (Maddison & Maddison 2011); whereas, 12S rRNA (1,094 bp) was aligned in two steps, first using MAFFT (Katoh & Standley 2013) and then adjusted by eye.

Genetic distance, phylogenetic analyses, and divergence times

Kimura 2-parameter genetic distances (K2P; Kimura 1980) between genera and within species were estimated using MEGA6.0 (Tamura et al. 2013), based on Cytb alignments of individuals in Table S1. To calculate genetic distances, species were grouped into their respective species as currently recognized (i.e. Happold 2013, Missoup et al. 2016, 2018) and by sampling localities (Fig. 1B).

The concatenated data set of sequences in Table S1 and Table S2 was divided into four individual partitions a priori with each gene (Cytb, 12S rRNA, Rbp3, Ghr) treated as a single partition. Partition finder 2.1.1 (Lanfear et al. 2016) was used to determine the best partitioning scheme and best model for each partition based on the AIC criterion under a likelihood framework using PhyML (Guindon et al. 2010) and the greedy algorithm (Lanfear et al. 2012).

Maximum likelihood (ML) analysis was conducted with RA×ML (Stamatakis 2014) for the concatenated sequences with 1,000 bootstrap replicates using the GTR+I+G model on all partitions. The majority rule consensus tree (MRC) of bootstrap results was constructed in Mesquite. A partitioned Bayesian analysis was conducted on the CIPRES portal (Miller et al. 2010) using the GTR+I+G for the concatenated dataset in MrBayes 3.2.3 (Ronquist & Huelsenbeck 2003). Two simultaneous runs of 10,000,000 generations with sampling every 1,000 generations were carried out. The MrBayes log files for both runs were examined in Tracer 1.6 (Rambaut et al. 2014) and a burn-in of 2,500,000 generations was set for each run. The runs were combined after discarding the burn-in and the MRC tree with posterior probability (PP) values was constructed in Mesquite.

Timing of divergence among clades was estimated using the Bayesian relaxed-clock model implemented in BEAST 1.8.4 (Drummond & Rambaut 2007) using the concatenated dataset with four partitions. Three calibration points from Kimura et al. (2017) and Aghová et al. (2018) were used: crown Murinae (lognormal prior distribution; mean: 3.2; Log(Stdev): 1.0; offset: 13.24; median: 15.2 million years before present (mya); 95% interval: 13.5-27.0 mya), Arvicanthis/Mus (lognormal prior distribution; mean: 4.0; Log(Stdev): 1.0; offset: 10.47; median: 12.9 mya; 95% interval: 10.8-27.7 mya), and Arvicanthini/Otomyini/Millardini (lognormal prior distribution; mean: 4.6; Log(Stdev): 1.0; offset: 8.52; median: 11.3 mya; 95%: 8.9-28.3 mya).

Aghová et al. (2018) also calibrated their analysis by applying a date at a stem position for the clade containing all extant Arvicanthis (lognormal prior distribution; mean: 5.5; Log(Stdev): 1.0; offset: 5.34; median: 8.7 mya; 95% interval: 5.8-29.0 mya) based on the Lemudong'o Formation, Kenya (Ambrose et al. 2007, Deino & Ambrose 2007, Manthi 2007). Because Lemniscomys is recovered as a well-supported sister taxon to Arvicanthus (Aghová et al. 2018), their approach effectively uses these fossils to establish a calibration at Arvicanthis/Lemniscomys. In addition to Arvicanthis, Manthi (2007) also identified material that he assigned to Lemniscomys with the same age at the Lemudong'o Formation, but Aghová et al. (2018) excluded these from consideration due to the absence of the first upper molar (M1). Aethomys is also present at Lemudong'o (Manthi 2007, Aghová et al. 2018). Mikula et al. (2021) opted against applying the Lemudong'o Arvicanthis to set a minimum age for the Arvicanthis/Lemniscomys divergence, but instead used it to establish a minimum age for Arvicanthis/Aethomys, which is a much more basal node. Under this scenario, they recovered an Arvicanthis/Lemniscomys date (∼3 mya) that is considerably younger than the Lemudong'o Formation (∼6 mya). A literal interpretation of Manthi (2007) indicates that the date associated with Lemudong'o should be applied to the Arvicanthis/Lemniscomys divergence, but the results of Mikula et al. (2021) suggest otherwise. Pending a more detailed evaluation of the Lemudong'o fossils in the context of arvicanthine morphological evolution, we chose to analyse our data both with and without this calibration at a stem Arvicanthis position.

The BEAST analysis was run for 100,000,000 generations, sampling trees and parameters every 5,000 generations. The BEAST log file was examined in Tracer 1.6 (Rambaut et al. 2014) and a burn-in of 10,000,000 generations was set. TreeAnnotator (Drummond & Rambaut 2007) was used to generate a maximum clade credibility tree by discarding the first 10,000,000 generations. Geological time elapsed since a most recent common ancestor is expressed either in million years ago (= mya – Miocene, Pliocene, and Pleistocene) or in years before present (= yr BP – Holocene). Posterior age estimates from BEAST analyses are expressed with associated 95% highest posterior densities (= 95% HPD). African rainforest classification follows Hardy et al. (2013).

The Hybomys division, tribe Arvicanthini

Musser & Carleton (2005) classified the genus-group taxa Dephomys, Hybomys (Hybomys), Hybomys (Typomys), and Stochomys within the Hybomys division, a grouping that was convincingly supported as monophyletic based on the four genes used to generate our phylogeny (Fig. 2). Musser & Carleton (2005) refrained from specifying formal tribes within Murinae when they formulated their family-group classification of Muroidea. At the time (editorial deadline: June 2003), no taxonomically dense, synthetic molecular studies addressed the Murinae and illuminated its major lineages, in contrast to the substantial literature then available for other murid subfamilies and ample evidentiary bases for delineating formal tribes (e.g. see their introductory commentary to the subfamilies Arvicolinae, Neotominae, Sigmodontinae, and Gerbillinae). Rather than burden systematists with a word-scrabble of newly named tribes and complicate future nomenclatural issues, they elected to borrow a generic aggregative term from the classical literature on murid rodents, this being the “division” of Misonne's (1969) African and Indo-Australian Muridae (Musser & Carleton 2005: 1,248). The division per se is not specifically governed by the Code (International Commission of Zoological Nomenclature 1999), which expressly denotes family-group ranks (Article 35.1) and their prescribed endings (Article 29.2) – i.e. the superfamily (-oidea), family (-idae), subfamily (-inae), tribe (-ini), and subtribe (-ina).

Taxonomically and geographically inclusive molecular studies have proliferated since 2005 and immensely improved our understanding of inter-generic relationships within Murinae. The innovative contribution by Lecompte et al. (2008) is especially relevant in this regard given its emphasis on African genera and murine tribal assemblages. According to their phylogenetic perspective, a view repeatedly corroborated by others (Fabre et al. 2013, Missoup et al. 2016, 2018, Steppan & Schenk 2017, Rowe et al. 2019, Mikula et al. 2021, this study), the Hybomys division of Musser & Carleton (2005) hierarchically corresponds to a sub tribal rank within the tribe Arvicanthini (see Table S4, footnote^a). Should systematic mammalogists ultimately determine that these lesser clades within Arvicanthini deserve formal names, then proper definitions, designation of type genera, and circumscription of generic contents must ensue, a purpose beyond the scope of our investigation. In the meanwhile, the following discussion continues to employ “division” in the non-binding nomenclatural sense intended by Musser & Carleton (2005).

The monophyletic (ML = 97, PP = 1.00) Hybomys division is a part of a large clade that includes the core Arvicanthini (minus Golunda and Oenomys). The Hybomys division represents the earliest diverging clade within this group, though this position receives only poor support (ML < 50, PP = 0.80). Within the Hybomys division, our results corroborate certain generic relationships reported in past molecular studies, which depicted H. univittatus as sister group either to Stochomys (Lecompte et al. 2008, Schenk et al. 2013, Bryja et al. 2017) or to Dephomys and Stochomys (Missoup et al. 2016, Steppan & Schenk 2017). With addition of samples of Typomys, the molecular analyses of Missoup et al. (2018), Mikula et al. (2021), and this study recovered four well defined subclades within the Hybomys division; however, the phylogenetic structure inferred from these data failed to portray Typomys as a subgenus and junior synonym of Hybomys, the generic construct long accepted as valid (Ellerman 1941, Rosevear 1969, Musser & Carleton 2005, Carleton 2013, Aplin 2017). On this basis, Missoup et al. (2018) proposed that Typomys be recognized as a genus distinct from Hybomys s.s., a taxonomic recommendation also adopted by Burgin et al. (2018), supported by Mikula et al. (2021) and one with which we are entirely in agreement.

Although they provided ample discussion of taxonomic and genetic components, Missoup et al. (2018) provided only a brief discussion of the morphological context of their result and we expand on that here. Morphological contrasts between species of Hybomys and Typomys are numerous and substantial as so far documented (here consolidated from Thomas 1911, Ingoldby 1929, Rosevear 1969, Carleton & Robbins 1985 and Carleton 2013). Species of Typomys lack the pectoral pair of mammae (0 + 2 = 4); those of Hybomys possess a pectoral pair (1 + 2 = 6). The t9 (metacone) in the upper molars of Typomys is diminutive in size, usually absent in the M2; a well-defined t9 occurs on both the M1 and M2 in nearly all specimens of Hybomys. Accessory cusps (anteromedian, anteorlingual, posterolingual) irregularly occur along the m1 cingula in examples of Typomys; such supplementary enamel structures commonly embellish the m1 cingula in members of Hybomys. The upper and lower first molars in Typomys are anchored by three and two roots, respectively, a formula interpreted as the ancestral character state in Muroidea; accessory rootlets uniformly occur in samples of Hybomys, the M1 4- or 5-rooted and the m1 4-rooted. Most examples (< 7%) of Typomys lack an alisphenoid strut; this strut, a bridge of the alisphenoid bone dividing the masticatory-buccinator foramen and foramen ovale accessorius, is present in most Hybomys (> 90%). The optic foramen is large, approximating the size of the sphenoidal fissure, in examples of Typomys; the optic foramen is smaller, noticeably less than the area of the sphenoidal fissure, in Hybomys.

Specimens of Typomys and Hybomys differ in their cranial architecture, displaying visually obvious shape contrasts that were captured in Rosevear's (1969) characterization of the “Typomys-type” vs. “Hybomys-type” skull. Key shape features involve the development of the incisive foramina (relatively shorter but wider in Typomys; long and narrow in Hybomys), zygomatic plate (narrower, anterior edge slanted, forming a shallow zygomatic notch in Typomys; wider, anterior edge vertical, incising a deep zygomatic notch in Hybomys), interorbital region (broader with amphoral supraorbital borders in Typomys; narrower with cuneate borders in Hybomys), and mandible (gracile and slender, with deep angular notch in Typomys; robust and deeper, with shallow angular notch in Hybomys). Predictably, such conspicuous shape differences would find quantitative rigor in variable loading coefficients derived in traditional morphometric investigations, in which the first factor extracted, whether in principal component or discriminant function analyses, sharply segregated specimens of Typomys and Hybomys without overlap in multivariate space (Van der Straeten & Verheyen 1982, Van der Straeten 1984, Carleton & Robbins 1985).

Hybomys and Typomys may also be distinct in their chromosomal complement based on the species karyotyped to date (Carleton & Robbins 1985). Karyograms of Typomys are characterized by lower diploid (2n) of 35-43 and fundamental numbers (FN) of 40-43 compared with the higher figures reported for populations of Hybomys (2n = 44-48, FN = 46-48).

In summary, the phylogenetic, genetic, morphological, morphometric, and chromosomal data currently marshalled, unambiguously reinforce one another and advise recognition of Typomys Thomas, 1911, as a genus distinct from Hybomys Thomas, 1910, bringing the number of genera in the Hybomys division to four (Table S4).

Systematics of the taxa of the genus Hybomys

In this study, representatives of all six of the described species of Hybomys sensu stricto were included. Hybomys badius Osgood, 1936; H. basilii Eisentraut, 1965; H. lunaris (Thomas, 1906); and H. univittatus (Peters, 1876) are currently considered valid species (Aplin 2017), whereas H. eisentrauti Van der Straeten & Hutterer, 1986 is considered a synonym of H. badius, and H. rufocanus (Tullberg, 1893) a synonym of H. univittatus (Musser & Carleton 2005, Carleton 2013). Aplin (2017) referred to recently collected molecular data showing a close relationship between material referable to H. badius and H. rufocanus. Missoup et al. (2018) reported close affinities among H. badius, H. eisentrauti, and H. rufocanus and proposed that H. badius and H. eisentrauti were junior synonyms of H. rufocanus. Our phylogenetic analyses recovered all species of Hybomys in a single well supported clade (ML = 100, PP = 1.00) containing four subclades (Fig. 2).

Thomas (1906) described lunaris as a subspecies of H. univittatus from specimens collected from Mubuku, Ruwenzori East, in Uganda at an elevation of 6,000 ft. Van der Straeten et al. (1986) concluded that lunaris was a distinct species and expanded its range to include specimens from eastern Democratic Republic of the Congo (DRC) and Rwanda. However, their analysis in 1986 did not include specimens from the type locality or any other parts of Uganda. While the morphometric analyses of Van der Straeten et al. (1986) failed to examine the type and holotypes of H. lunaris, the body measurements of the type were within the ranges reported for the specimens examined from Rwanda and the DRC. In most recent revisions (Musser & Carleton 2005, Dieterlen 2013, Taylor 2017), the distribution of H. lunaris has been restricted to the Rwenzori Mountains of Uganda. However, Musser & Carleton (2005) noted that two specimens from Kanyawara, Uganda, matched Thomas's (1906) description of the type specimen has a small, delicate form and differ from the larger and more robust forms from Rwanda. Huhndorf et al. (2007) expanded the range to the montane forest along the Albertine Rift north of the Virunga Volcanoes and demonstrated substantial genetic differentiation with the populations to the south in Burundi (H. cf. lunaris). Sequences from specimens of H. lunaris from Uganda and the eastern DRC, previously reported by Huhndorf et al. (2007) and Missoup et al. (2018) respectively, along with sequences from H. cf. lunaris reported by Huhndorf et al. (2007) and H. sp. n. reported by Mikula et al. (2021) were recovered in a well-supported (ML = 100, PP = 1.00) clade (Fig. 2, clade I). The mean genetic distance (K2P) observed between the H. lunaris subclade samples and the H. cf. lunaris subclade samples for Cytb was 0.0580, whereas the intraspecific differentiations were 0.0180 and 0.000, respectively. Our analyses support a broader geographic range for H. lunaris than described by Huhndorf et al. (2007), as the specimen from Bangole in the DRC (recorded as Masako, DRC in the MNHN database) and the specimen identified as H. sp. n. (recorded as H. lunaris from Masako, DRC in the MNHN database) are recovered in the clade with samples from the type locality of H. lunaris. The Cytb K2P distance of 0.022 between these specimens and those sampled in the Rwenzori Mountains, suggest that the range of H. lunaris extends at least 550 km to the NW into north-eastern portion of the DRC. This H. lunaris species complex of central Africa is estimated to have diverged about 3.1 mya from other forms of Hybomys (Fig. S4). More complete molecular and morphological analyses are needed to better define the geographic range of H. lunaris and to further test the hypothesis of Huhndorf et al. (2007) of an unnamed species occurring in Burundi.

Hybomys univittatus is distributed in the rainforest of the Congo Basin and from the River Niger to the Albertine Rift and the shores of Lake Victoria (Carleton 2013, Aplin 2017). Considerable morphological variation has been reported among populations of this taxon (Rosevear 1969, Carleton & Robbins 1985) leading Musser & Carleton (2005) to suggest that H. univittatus consists of several morphologically similar species. In addition, three different karyotypes with diploid numbers of 44, 46, and 48 have been reported (Carleton & Robbins 1985, Verheyen & Van der Straeten 1985) among populations in Cameroon and Gabon. Our phylogenetic analyses of sequence data recovered H. univittatus sampled from locality 9 (Fig. 1B) north of the River Sanaga in Cameroon, locality 6 between the Sanaga and Ogooue rivers in Gabon, locality 11 south of the River Ogooue in Gabon, and locality 12 from the Congo (Table 3, Fig. 1B) as polyphyletic (Fig. 2). The sample from north of the River Sanaga from the Southwestern Province of Cameroon (Fig. 1B, locality 9) was recovered within a well-supported (ML = 80, PP = 0.90) H. rufocanus clade. The mean level of genetic differentiation (Cytb K2P) between this H. univittatus sample and samples of H. rufocanus was 0.029 (Table 3), lower than the level for interspecific differentiation reported by Baker & Bradley (2006). These results suggest that populations currently recognized as H. univittatus from Southwestern Cameroon are part of the rufocanus-badius-eisentrauti lineage. Low levels of genetic differentiation among rufocanus, badius and eisentrauti based on a somewhat larger sampling of these taxa was reported by Missoup et al. (2018) to range between 0.010 and 0.028. Our analyses not only support the conclusion of Missoup et al. (2018) that H. badius and H. eisentrauti should be considered as synonyms of H. rufocanus (Tullberg, 1893) but provide data to support their suggestion that populations from lowland forest north of the River Sanaga previously reported as H. u. univittatus (Eisentraut 1968, 1973, Denys et al. 2009) should be included within this taxon. Eisentraut (1963) and Sanderson (1940) had noted the similarity of the lowland forms of H. univittatus north of the Cameroon Mountains in size with H. badius and further noted that these two taxa were only differentiated by colour.

The sequence data from Missoup et al. (2018) from a specimen of H. univittatus from Mvoum (Fig. 1B, locality 6), between the Sanaga and Ogooue rivers and only 15 km NE of Dongila (type locality of univittatus), was recovered in our analyses as sister to the H. rufocanus clade with strong support (ML = 91, PP = 1.00). However, this specimen from locality 6, which we assume to represent true univittatus as it was collected near the type locality for this taxon, demonstrates considerable genetic differentiation from the taxa of its sister clade to the north (Table 3) with a mean Cytb K2P = 0.087 and a range of 0.079-0.094. We hypothesize that additional sequences from samples between the Sanaga and Ogooue rivers would result in a distinct clade, representing true univittatus, sister to the H. rufocanus clade.

Fig. 3.

The Guineo-Congolian Region, sensu White (1983), and specific distributions of the genera Hybomys and Typomys, Hybomys division. A) Current extent of Guineo-Congolian rainforests (green – range shapefiles imported from White 1983) compared with the hypothesized maximal extent of rainforests during wet periods of the Early Pliocene (chequered, following Hardy et al. 2013). Also depicted are hypothesized Late Quaternary rainforest refugia (yellow, after Maley 1996) and centres of endemism (red, after Happold 1996), which substantially overlap in their geographic location (orange). B) Distributions of currently recognized species of Hybomys and Typomys (range shapefiles acquired from Terrestrial Mammal dataset, IUCN Red List –  http://www.iucnredlist.org/technical-documents/spatial-data) and modified with the results of this study. Type locality shown with a dot within a circle.

Our phylogenetic analyses recover the remaining samples of H. univittatus, from Gabon south of the River Ogooue (Fig. 1B, locality 11) and from the Congo (Fig. 1B, locality 12), in a well-supported clade (ML = 76, PP = 1.00) that is sister to the H. rufocanus – true H. univittatus clade. This latter H. univittatus clade has a K2P distance ≥ 0.088 from H. univittatus from localities 9 and 6 and a distance between 0.086-0.111 to other taxa of the genus (Table 3).

One solution to the paraphyly observed among population of H. univittatus would be to consider univittatus, rufocanus, badius, basilii, and eisentrauti as synonyms and recognize this taxon as H. univittatus (Peters, 1876). However, this would lump several populations that demonstrate considerable morphological variation and genetic differentiation (K2P > 8-9%). Musser & Carleton (2005) noted that the degree of morphological variation observed among populations of H. univittatus suggests that it consists of several morphologically similar species.

Pending further geographic sampling of H. univittatus we would suggest the recognition of samples previously recognized as H. univittatus from north of the River Sanaga, including badius and eisentrauti, as H. rufocanus as previously suggested by Missoup et al. (2018). Furthermore, we would suggest restricting the distribution of H. univittatus to the region between the Sanaga and Ogooue rivers and suggest that the specimens from south of the River Ogooue currently recognized as H. univittatus represent an undescribed species (Table 3). The eastern boundaries of either of these species are not known, nor are the areas of contact with populations of H. lunaris (Fig. 3B).

Biogeography and emergence of the Hybomys division

The middle to late Miocene was an eventful period that saw major faunal interchange between Africa and Eurasia, a movement of ancestral stocks facilitated by lower sea levels, a global trend toward cooler and drier climates, and the intermediacy of the Arabian Plate between southern Asia and north-eastern Africa (see summary in Morley & Kingdon 2013). Certainly, African landscapes were integral to the phylogenesis of murine rodents as the continent's Sub-Saharan region contains representatives of five of the ten tribes formally recognized to date (sensu Lecompte et al. 2008). Two of these tribes, Malacomyini and Otomyini, are endemic to Sub-Saharan Africa; the taxonomic diversity of two, Arvicanthini and Praomyini, is principally confined to this region; the tribal distribution of only one, Murini, extends across Africa and Eurasia. Arvicanthini numbers some 88 species representing 19 genera (including one genus and species in Asia – Hoffman et al. 2009, Happold 2013, Bryja et al. 2017) and is the largest of the five African tribes, compared with the Malacomyini (one genus, three species – Happold 2013), Otomyini (two genera, 31 species – Taylor et al. 2011, Happold 2013), Praomyini (nine genera, 55 species – Hoffman et al. 2009, Happold 2013, Carleton et al. 2015), and Murini (two genera, 20 species native to Africa – Happold 2013). The initial entry of murine rodents into Africa transpired around 9 to 12 mya, an interval bracketed by both fossil records (Jacobs et al. 1990, Jacobs & Flynn 2005) and molecular chronograms calibrated to key “palaeontological events” (Lecompte et al. 2008, Steppan & Schenk 2017). Like the rapid diversification of core Murinae into its tribal-level lineages (Fabre et al. 2013, Kimura et al. 2017, Steppan & Schenk 2017), the murine stem groups that entered Africa also underwent an explosive radiation commensurate with the continent's vastness and its geological, climatological, and ecological heterogeneity. Fossils assignable to living arvicanthine genera date from the late Miocene through early Pliocene, around 5 to 7 mya (e.g. Aethomys – Denys 1990, Manthi 2007; Arvicanthis – Denys 1999, Winkler 2002; Lemniscomys – Manthi 2007).

The basal radiation of Arvicanthini coincides with a prolonged period of cooler, drier climate, about 6 to 8 mya (Fig. S4), and the accompanying proliferation of grasslands and open woodlands (Maley 1996, Plana 2004, Morley & Kingdon 2013). The expansion of grasslands and concomitant retreat of high forests over this period are firmly documented, drawing upon palynological records (Morley & Richards 1993, Morley et al. 2003) and isotopic evidence that tracks the pronounced increase in C₄ grasses (Poaceae) that now dominate tropical ecosystems (Cerling et al. 1997, Sage et al. 2012). Sub-Saharan biomes support by far the greatest biodiversity within Arvicanthini (see species accounts in Happold 2013), with only three species recorded outside this region – two in northern Africa and the single Asian extralimital, Golunda ellioti, in the Indian subcontinent. Many arvicanthine genera – such as Aethomys, Arvicanthis, Grammomys, Lemniscomys, Rhabdomys, and Thallomys – typically constitute the ecologically abundant, small mammal guild that populates Africa's grasslands and moorlands, bushlands and shrublands, savannahs and woodlands, and all ecological gradations between these habitats.

In contrast to such open environments, species of the Hybomys division are denizens of closed-canopied and deep forest, the Guineo-Congolian Region of White (1983). According to our fossil-calibrated phylogeny (Fig. S4), the origin of the Hybomys division dates from 7.4 (95% HPD = 6.9-9.4) mya, consistent with a late Miocene (Tortonian) appearance. The Hybomys division is among the earliest clades to separate within Arvicanthini, following the divergence of the Golunda and Oenomys lineages and preceding the many branches that evolved into the extant genera that inhabit Africa's expansive grasslands and savannahs. The same chronological order of basal divergences within Arvicanthini was captured in other molecular studies (Lecompte et al. 2008, Fabre et al. 2013). Nonetheless, the critical basal branch lengths are short and nodal support weak, statistically rendering early branches as a polytomy (e.g. as depicted by Missoup et al. 2016). Appeal to other genes, perhaps coupled with more precise determinations of common ancestry illuminated by palaeontological investigation (Kimura et al. 2015, 2017, 2021), may amplify statistical confidence in the early cladogenesis of Arvicanthini and its primal association with Guineo-Congolian rainforest.

Closed-canopied, tropical rainforest developed by the late Cretaceous, and as of the middle Miocene climatic optimum, it extended coast to coast across the equatorial zone of central Africa (Maley 1996, Morley & Kingdon 2013). The maximum extent of the African rainforests has been shrinking since the Eocene and Miocene (22 to 15 million km²), Pliocene (10 million km²) to current day (3.4 million km²) (Hardy et al. 2013). The long-term contraction of rainforests was punctuated with shorter cycles of range expansion and contraction in the Late Miocene, Early Pliocene, and glacial-interglacial cycles of the Quaternary (Hardy et al. 2013). The antiquity of these lowland forests and their relatively stable climatic regime are thought to have provided optimal conditions for sustaining ancient lineages (palaeoendemics) and minimizing extinctions (Moritz et al. 2000); these expectations are generally supported by the concentration of genealogically older species within Guineo-Congolian rainforests, sometimes characterized as “museums” of biodiversity (Fjeldså & Bowie 2008, Murienne et al. 2012). Those sectors within the Guineo-Congolian Region that harbour exceptional species richness and high endemism (Fig. 3A), whether of plants or animals, decidedly support the notion of former rainforest refugia (Maley 1996, Plana 2004, Fjeldså & Bowie 2008, Hardy et al. 2013). While the hypothesized refugia (pictured in Fig. 3A) are from the Late Quaternary (Maley 1996), they may roughly represent the refugia in the Pleistocene, Pliocene and Miocene due to the congruent placement of the areas of endemism (Happold 1996). The Miocene emergence of the Hybomys division (Fig. S4) and the geographic complementarity of its generic and specific distributions apropos subdivisions of the Guineo-Congolian Region (Figs. 1B, 3B) intimate a similar biogeographic history.

Diversification within the Hybomys division in African rainforests

The following discussion entails three assumptions: 1) the common ancestor of the Hybomys division inhabited lowland rainforest, whether dwelling in forest proper or in environments dependent upon evergreen moist forest. This supposition seems reasonable in view of the association of all living species with rainforest environments, primary or secondary, and the tight distributional coalignment of the division with the Guineo-Congolian Region (Figs. 1B, 3B); 2) speciation has transpired as an allopatric process, with vicariant barriers such as dry savannahs or impassable rivers fragmenting ancestral populations and promoting genetic differentiation. Authoritative synopses of many tropical organisms encourage acceptance of this assumption (e.g. Moritz et al. 2000, Plana 2004, Fjeldså & Bowie 2008). We lack requisite sample sizes and fine-scaled geographic representation to adequately test whether our cladogram fits a parapatric speciation model consistent with differentiation along some environmental gradient; 3) descendant species within the Hybomys division have retained the essential ecological characteristics and physiological tolerances of their distant Miocene ancestors. The question of niche conservatism must remain an inference in view of our fragmentary ecological knowledge of the forest species under study and the absence of any fossils of the four genera that might support paleoenvironmental interpretation.

According to our fossil-calibrated consensus tree (Fig. S4), the four descendant genera within the Hybomys division arose during the late Miocene (Messinian) and early Pliocene, sequentially delineated by three major lineage splits at approximately 7.4 (Typomys), 6.7 (Hybomys), and 4.8 mya (Dephomys-Stochomys). Only the youngest of these phyletic divisions gave rise to genera that are now wholly allopatric, the distributions of Dephomys and Stochomys being divided by the River Volta (Fig. 1B). Living descendants of the two older branching episodes are partially or fully overlapping in their areal distributions (Fig. 1B), indicating considerable range expansion and secondary contact following their divergence (again, assuming an allopatric speciation process). Thus, Typomys is broadly sympatric with Dephomys in Upper Guinea forest and with Stochomys in Lower Guinea west of the River Niger; whereas, the distribution of Hybomys overlaps that of Stochomys in Lower Guinea and Congolia (Fig. 1B). Post-isolation range expansion and secondary geographic overlap are similarly necessary to account for the specific distributions of Typomys, following their separation in the middle Pliocene (Fig. S4). The ranges of T. planifrons and T. trivirgatus intersect in the western reaches of the Upper Guinea forest block, in the area of the hypothesized Liberian refuge (Fig. 3).

Genera within the Hybomys division likely derive from common ancestors that spanned multiple forest blocks, if not the full range of Upper Guinea to Congolia. These early Pliocene ancestral taxa lived through wet cycles involving extensive forestation across Africa (chequered region in Fig. 3A). By the end of the early Pliocene, however, modern patterns of generic distribution may have emerged, with Typomys excluded from Congolia, Dephomys restricted to Upper Guinea, and Hybomys and Stochomys excluded from Upper Guinea.

The collection of S. longicaudatus from eastern Ghana (Decher et al. 2021) highlights the limited role that the Dahomey Gap appears to play in the evolutionary history of the Hybomys division. The Dahomey Gap is a wide swath of Sudan-like savannah that reaches the Atlantic coast and separates West African rainforest into the Upper Guinea and Lower Guinea sectors (Fig. 1A). Although earlier syntheses of mammalian distributions had minimized the significance of the Dahomey Gap as a vicariant barrier in West Africa (Booth 1954, Robbins 1978), recent floristic investigations have revitalized the biogeographic role of the Gap and underscored its recurrent formation, demonstrably in younger epochs (Salzmann & Hoelzmann 2005, Duminil et al. 2013) and inferentially in deeper time (Plana 2004, Couvreur et al. 2008). Pollen analyses document several Holocene fluctuations between rainforest and savannah vegetation in this region, the last encroachment of savannah occurring around 1,100 yrBP(Salzmann&Hoelzmann2005).Asurprisingly recent divergence time separates S. longicaudatus from eastern Ghana and a clade uniting samples from Gabon and western Cameroon (0.50 (95% HPD = 0.23 to 0.78) mya). Among remaining members of the Hybomys division, the isolation of populations of T. trivirgatus in western Nigeria, separated from the main species distribution to the west of the Dahomey Gap (Fig. 3B), is plausibly interpreted as a another very recent occurrence.

Riverine barriers have clearly played an important role in isolating and restricting the geographic spread of rodents in the Hybomys division, but most examples of isolation are accompanied by counter examples. Typomys planifrons does not extend east of the River Sassandra in Côte d'Ivoire, but both T. trivirgatus and D. defua are found on both sides. The River Volta separates Dephomys from Stochomys, but does not separate T. trivirgatus (though this needs testing). Typomys trivirgatus on either side of the Volta exhibit different striping pattern with forms east of the Volta displaying longer dorsal stripes compared to those from the west (Happold 2013). The Niger represents the easternmost edge of the distribution of Typomys and the westernmost edge for Hybomys, but the range of S. longicaudatus spans it. Finally, the Sanaga and Ogooue rivers appear to delineate potential species within Hybomys. Overall, these riverine barriers broadly align with intrageneric studies on other small mammals of African forests such as Malacomys (Bohoussou et al. 2015), Hylomyscus (Nicolas et al. 2020), Grammomys (Bryja et al. 2017), and Crocidura (Jacquet et al. 2014, 2015). The foundation for a comparative phylogeography model for the evolution of sylvan small mammals has been laid. Such a model may involve allopatry via forest refugia (Fig. 3A) followed by subsequent maintenance of isolation due to large rivers (Haffer 1982, Maley 1996, Bohoussou et al. 2015, Nicolas et al. 2019). More extensive intraspecific sampling of species in the Hybomys division, particularly, S. longicaudatus, D. defua, and T. trivirgatus will better illuminate these patterns.

Our results fail to support the distinctiveness of montane forms of Hybomys (badius and eisentrauti) found along the Cameroon Volcanic Line (Mount Oku, Bamenda Highlands) as compared with neighbouring lowland Hybomys. In contrast, the Albertine Rift taxon, H. lunaris, is the earliest diverging species of Hybomys. This divergence dates to the late Pliocene, during a dry phase (3.0 to 3.5 mya) when forests would have retreated and conceivably isolated the H. lunaris progenitor in a montane refugium. The Albertine Rift region has long been recognized as a biodiversity hotspot (Kerbis Peterhans et al. 1998, Mazel et al. 2014). Biogeographic studies of Afrotropical resident birds have consistently identified the Albertine Rift as a region of rich endemism where phylogenetically distinct species are concentrated (Fjeldså et al. 2007, Fjeldså & Bowie 2008). Likewise, the montane forests of the Albertine Rift exhibit numerous mammalian endemics (Kityo et al. 2003, Plumptre et al. 2007) including many that have been described recently such as bats in the genus Rhinolophus (Kerbis Peterhans et al. 2013a), shrews in the genera Crocidura and Myosorex (Kerbis Peterhans et al. 2010, 2013b); rodents in the genus Hylomyscus (Demos et al. 2014a, Kerbis Peterhans et al. 2020) and Lophuromys (Verheyen et al. 2007); and a primate (Davenport et al. 2006). Additional potential Albertine Rift species have been identified based on unique evolutionary lineages in Sylvisorex (Demos et al. 2014b, 2015), Miniopterus (Demos et al. 2020), and H. lunaris (Huhndorf et al. 2007).

Located in the Gulf of Guinea near the Cameroon coast, Bioko Island is an amalgam of three stratovolcanoes. Bioko is part of the Cameroon Volcanic Line, a string of volcanoes that include the islands of Annobón, São Tomé, Príncipe, and Bioko along with several highlands on the African continent such as Mount Cameroon, Mount Oku, and the Ngaoundéré and Biu Plateaus (Jones 1994, Aka et al. 2004). Although our results do not support the distinctiveness of Hybomys from the Cameroon Line highlands on the African continent (badius on Mount Cameroon, rufocanus s.s. from the Bakossi Mountains, and eisentrauti from Mount Oku and Mount Lefo), the Bioko Island species, H. basilii, is quite distinctive, having diverged from mainland Hybomys 2.0 (95% HPD = 1.5 to 2.5) mya. Although our data allow for the arrival of Hybomys on Bioko any time between its divergence from mainland Hybomys until the most recent common ancestor of the two haplotypes of H. basilii sampled (∼100 kya), our 1.2 (95% HPD = 0.83-1.66) mya estimate is surprisingly congruent with the formation of the stratovolcanoes that comprise the island, which took place 1.3 mya (K-Ar dating – Aka et al. 2004). Bioko was a peninsula during glacial cycles, was repeatedly connected to mainland Africa near Mount Cameroon, and the most recent connection was, ∼10 kya, after the Last Glacial Maximum (Jones 1994). Hybomys basilii is not related to H. rufocanus, the mainland species found in the part of Africa to which Bioko has shared a past physical connection but is instead related to Hybomys from Congo and Gabon. This rare animal is probably an old endemic to Bioko and is one of only three endemic mammal species found on the island (Amori et al. 2008). The others are both shrews, Sylvisorex isabella and Myosorex eisentrauti (though see Groves & Grubb 2011, who consider the Bioko blue duiker, Philantomba melanorheus, to be its own species). Several Bioko endemic subspecies of primate, rodent, and hyrax have been recognized along with distinct intraspecific clades of small mammal (Missoup et al. 2012, Nicolas et al. 2019), but these taxa likely arrived much later in the Pleistocene compared to H. basilii. Our findings highlight the importance of conservation efforts to protect this endangered and evolutionary distinctive rodent (Kennerley 2016).

Future research needs

The molecular evidence presented above supports the monophyly of the Hybomys division (Fig. 2), evaluates relationships and divergence times among its members (Fig. S4), and recommends their taxonomic rank (Table S4). We also have formulated a preliminary interpretation of the historical biogeography of the Hybomys division, highlighting its intimate association with forests of the Guineo-Congolian Region since the late Miocene. Several research avenues would improve our biogeographical understanding of these forest-dwelling rodents.

Foremost is the need to markedly augment geographic sampling in order to compare patterns of intraspecific genetic variation with interspecific relationships. In spite of having a similar range, Stochomys and Hybomys yield markedly different results. Based on only three localities, we found surprisingly little genetic variation in S. longicaudatus. Broader geographic sampling, especially from eastern Congolian forests and south of the River Congo are needed to confirm this result and to determine what patterns, even recent ones, are present in this species. Taxonomic problems within H. univittatus have been identified (Carleton & Robbins 1985, Musser & Carleton 2005, this study). Table S4 should be viewed as a temporary taxonomic solution for H. univittatus, as the species as delineated there remains paraphyletic. Additional sampling across more of Congolia is also needed in H. univittatus. Expansion of geographic representation within Hybomys must include additional samples of H. lunaris from the Albertine Rift. Morphological and genetic evidence casts doubt that only one Hybomys species inhabits montane forests of the Albertine Rift (Musser & Carleton 2005, Huhndorf et al. 2007). We also lack sampling from a single species, D. eburneae.

The genetic signature of T. trivirgatus in western Nigeria, Lower Guinea, invites attention in view of its isolation from the main distribution of the species in Upper Guinea forest, west of the Dahomey Gap (Fig. 3B). The Nigerian isolate had been named as a subspecies, pearsei Ingoldby, 1929, distinguished by its weaker definition of trilinear dorsal striping compared with animals from Upper Guinea forest, where T. trivirgatus overlaps T. planifrons. Does the Nigerian segment represent a recent expansion (Holocene) eastward across the Dahomey Gap or an older relict (Pliocene) that dates from the divergence of T. trivirgatus and T. planifrons in separate refugia?

Comparative phylogeographic analysis of multiple murine taxa indigenous to the Guineo-Congolian Region, with largely congruent distributions, may assist in delimiting contemporaneous episodes of speciation and in pinpointing older zones of vicariance. Along with members of the Hybomys division, potentially informative candidates include the genus Malacomys (Bohoussou et al. 2015), Colomys (Giarla et al. 2020), the Thamnomys poensis group (Bryja et al. 2017), the Hylomyscus alleni group (Nicolas et al. 2006, 2020), the Hylomyscus anselli group (Kerbis Peterhans et al. 2020), and the Praomys tullbergi complex (Nicolas et al. 2008, Missoup et al. 2012).