Rare crested rat subfossils unveil Afro–Eurasian ecological corridors synchronous with early human dispersals

Significance

The extent and timing of paleoenvironmental connections between Africa and Eurasia during the last glacial and interglacial periods are key issues in relation to early dispersals of Homo sapiens out of Africa. However, direct evidence of synchronous faunal dispersals is sparse. We report the discovery near the Dead Sea of subfossils belonging to an ancient relative of the eastern African crested rat dated to between ∼42,000 and at least 103,000 y ago. Morphological comparisons, ancient DNA, and ecological modeling suggest that the Judean Desert was greener in the past and that continuous habitat corridors connected eastern Africa with the Levant. This finding strengthens the hypothesis that early human dispersals were prompted by climatic changes and Late Pleistocene intercontinental connectivity.

Abstract

Biotic interactions between Africa and Eurasia across the Levant have invoked particular attention among scientists aiming to unravel early human dispersals. However, it remains unclear whether behavioral capacities enabled early modern humans to surpass the Saharo–Arabian deserts or if climatic changes triggered punctuated dispersals out of Africa. Here, we report an unusual subfossil assemblage discovered in a Judean Desert’s cliff cave near the Dead Sea and dated to between ∼42,000 and at least 103,000 y ago. Paleogenomic and morphological comparisons indicate that the specimens belong to an extinct subspecies of the eastern African crested rat, Lophiomys imhausi maremortum subspecies nova, which diverged from the modern eastern African populations in the late Middle Pleistocene ∼226,000 to 165,000 y ago. The reported paleomitogenome is the oldest so far in the Levant, opening the door for future paleoDNA analyses in the region. Species distribution modeling points to the presence of continuous habitat corridors connecting eastern Africa with the Levant during the Last Interglacial ∼129,000 to 116,000 y ago, providing further evidence of the northern ingression of African biomes into Eurasia and reinforcing previous suggestions of the critical role of climate change in Late Pleistocene intercontinental biogeography. Furthermore, our study complements other paleoenvironmental proxies with local—instead of interregional—paleoenvironmental data, opening an unprecedented window into the Dead Sea rift paleolandscape.

Situated at the gateway of Africa, the Levant witnessed major Afro–Eurasian biotic exchanges during the Neogene-Quaternary (1–3), including multiple hominin dispersal events (4–11). The extent and timing of potential paleoenvironmental connections between these regions during the Late Pleistocene have received considerable attention due to their role in the global expansion of Homo sapiens out of Africa (9, 12–17). Due to multiple dating uncertainties, the paucity of the fossil record, and the disparate scale and resolution of paleoenvironmental proxies (17), it remains unclear whether technological and behavioral capacities enabled early modern humans to surpass the Saharo–Arabian biogeographic barrier (6, 18–20) or if climatic changes triggered punctuated dispersals out of Africa through the creation of ecological corridors (14–16, 21, 22). The southern part of the Levant is a geographical bottleneck, longitudinally divided by mountain ranges into a wooded Mediterranean zone and the more arid, inland regions of the Dead Sea rift. Under monsoon-dominated climatic conditions, these arid regions could have supported savanna-like environments that funnelled the dispersal of African faunas by linking the Arabian and Sinai regions with the eastern Mediterranean (16, 23). While the paleoenvironmental interpretation of the well-known Mediterranean Levantine fossil record continues to be scrutinized and debated (16, 17, 24–26), there is no record of faunal movement in the Dead Sea arid areas during the Late Pleistocene (1, 27).

Here, we report evidence that the southeastern Levant, today blocked by a rain shadow desert, had more densely vegetated habitats in the Late Pleistocene that supported African faunal immigrants synchronously with early human dispersals. Recent excavations (28) and surveys (29) in the Cave of the Skulls (CoS), in the southern Judean Desert, have yielded exceptionally well-preserved fossils of a subspecies of the eastern African crested rat, Lophiomys imhausi maremortum subspecies nova (subsp. nov.) (Figs. 1–3), dated to between ∼42,000 y ago (ka) and at least 103 ka. The subspecific status is sustained by paleogenomic data, which indicate low molecular divergence (<2%) between the Levantine population and the extant populations from eastern Africa. A few specimens dated to >44 and ∼112 ka from Sodmein Cave (SOD), in northeastern Egypt [Fig. 1A (30)], are also ascribed to this subspecies.

Fig. 1.

Geographical setting of this study. (A) Map of Africa and the Middle East showing fossil sites with Lophiomys: 1, Irhoud Ocre (Jebel Irhoud); 2, Khemis; 3, Lissasfa; 4, Salobreña; 5, Amama; 6, Oued Ahtmenia; 7, Sheikh Abdallah; 8, SOD; 9, Alayla, Middle Awash; and 10, CoS. (B) Paleoartistic reconstruction of L. i. maremortum subsp. nov. Artwork by Aya Marck, used with permission. (C) Elevation map showing the location of the CoS. (D) Planar map of the CoS, showing the different sectors excavated (sectors A to Q) and the areas with higher density of L. i. maremortum subsp. nov. (E) The CoS main entrance, opening to a vertical cliff in Nahal Tze’elim. During excavations, the cave had to be approached with the use of ropes. Photograph by Guy Fitoussi (Israel Antiquities Authority), used with permission. (F) A complete skull of L. i. maremortum subsp. nov. (SMNH-TAU M17121) found in situ in sector Q during a survey of the cave. Note that the hemimandibles are in anatomical connection. Photograph by I.A.L.

Fig. 3.

Dental metrics and first upper molar geometric morphometrics of Lophiomys imhausi maremortum subsp. nov in comparison to extant populations of L. imhausi and extinct relatives. (A) Dental metrics of upper (M1 to M3) and lower (m1 to m3) molars. Note that the specimens of L. i. maremortum subsp.nov. (CoS, in blue) are consistently larger than the extant populations (northeastern, NE, in orange; southwestern, SW, in green). (B) Geometric morphometric of the first upper molar (M1) visualized by means of a between-group principal component analysis. Centroid size is calculated as the sqrt of the sum of squared distances of all landmarks from the center of the shape before Procrustes analysis. Independently of size, the three populations are well distinguished by M1 shape.

The crested rat is a large (>800 g) eastern African rodent that shows some of the most extraordinary adaptive features among living mammals, including a granulated helmet-like skull, a poisonous pelt, and a three-chambered stomach with hindgut fermentation (31, 32). The crested rat has the ability to become toxic to predators by chewing up the roots and bark of the poison-arrow tree (Acokanthera ssp.) and using its mouth to spread the toxin—a cardiac glycoside called ouabain—on strips of specialized hairs on its body flanks (33, 34). In general, the crested rat inhabits steep, rocky valleys within woodlands and montane forests under variable but typically seasonal rainfall, and its diet includes leaves, fruits, and shoots (32). Populations are apparently stable in relatively wet and densely vegetated habitats in the montane forests of Kenya and the highlands of Ethiopia, referred to here as southwest populations (SW; Fig. 4A), where the mean annual precipitation can reach 2,500 mm (35). It is also known from drier and more sparsely vegetated mountain ranges or lowland woodlands in Somalia, Djibouti, Eritrea, and southeastern Sudan, referred here as northeast populations (NE; Fig. 4A), where the mean annual precipitation can be as little as 350 mm (36). Despite its apparently broad climatic tolerance, the crested rat is rare in the wild, and it usually prefers riparian vegetation, where it can hide and den in holes under trees (34). Therefore, it is likely that the potential distribution and dispersal capability of Lophiomys is limited by the availability of trees and by vegetation density.

Fig. 4.

Estimated habitat corridors derived from SDMs of Lophiomys in relation to paleoclimatic, paleobotanical, and archaeological evidence in the Late Pleistocene. (A) Map of the study region colored by NDVI, with areas in green indicating more vegetation cover and areas in orange showing sparsely vegetated areas. The circles are confirmed occurrences of extant L. imhausi. The southwest (SW) population is colored in green, and the northeast (NE) population is colored in orange. (B) Climatic niche envelope of extant L. imhausi SW and NE populations and at the location of the CoS with the present climate. The variables are the same used in the SDMs. Units have been scaled to fit the chart. For an explanation of variables, reference SI Appendix, Text S13 and SI Appendix, Table S17. (C–E) Maps of the study region showing potential habitat suitability areas for Lophiomys based on Maxent-based species distribution modeling during (C) current time, (D) LGM, and (E) the LIG. The red colors indicate higher habitat suitability estimates. (F) Diagram showing δ¹⁸O curves in the Southern Levant and pollen records of the Dead Sea. From Left to Right, chronology and δ¹⁸O record of Soreq, Peqi’in, and Tzavoa Caves (data from refs. 47 and 64), arboreal and sclerophyllous pollen concentration, Pistacia pollen concentration, and Artemisia and Amaranthaceae pollen concentration (data from refs. 23 and 46). (G) A photograph of Nahal Tze’elim taken from the CoS main entrance, with the Dead Sea on the top of the image. Photograph by I.A.L. (H) Photo of Iran’s Zagros Mountains. Photograph by Marijn van den Brink, distributed under a CC-BY 3.0 license. (I) A photo of Djibouti’s National Day Park, showing how the Judean Desert could have looked like during part of the Late Pleistocene. Photograph by Steven Dessein, distributed under a CC-BY 3.0 license. (J) Map of the southern Levant showing current mean annual precipitation (in blue) and Late Pleistocene archaeological sites (yellow squares). The spatial arrangement of sites suggests that the nowadays-arid regions—including the Dead Sea basin—could have funneled early human dispersals into the Levant. Precipitation curves are redrawn from ref. 65.

The evolutionary and biogeographic history of crested rats is poorly documented. Nowadays, the Lophiomyinae is represented by only one eastern African species. However, the group was more diverse in the past, and its distribution spread over areas of eastern and northern Africa and Spain (Fig. 1; see SI Appendix, Table S1, for a list of all sites and references). The subfossils described here represent the most extensive paleontological collection (n > 250) of Lophiomys ever found. Based on the known habitats occupied by extant Lophiomys and paleoclimatic data, this finding suggests that continuous habitat corridors connected eastern Africa with the Levant during the Last Interglacial (LIG, ∼129 to 116 ka) along the western Red Sea mountains. In contrast to the present-day hyper-arid conditions, the Late Pleistocene Judean Desert was likely wetter and included a variable but relatively persistent arboreal component. Altogether, our study provides further evidence of the expansion of African biomes into Eurasia and the role of habitat corridors in early hominin dispersals out of Africa.

Results

Depositional and Chronological Setting.

Large-scale excavations in the CoS were held in 2016 (28). Recent animal and human activities in the cave, combined with a low rate of sedimentation typical of the Judean desert caves (37), have resulted in the mixing of materials from different time periods (SI Appendix, Text S1). Therefore, the bioturbated sediments retain almost no stratigraphic context (29). Remains of L. i. maremortum subsp. nov. were identified from nearly all areas of the cave but are concentrated mainly in the narrower passages, predominantly in sectors K, L, and M (Fig. 1D). The bones belong to at least 16 individuals based on the number of left mandibular m1 alveoli, of which 8 are subadult or not fully adult, and 4 are old individuals (based on the degree of dental wear). Almost all craniodental and postcranial elements are represented, there are no signs of predation on the bones, many of the skeletal elements are intact, and some of the remains were found in anatomical connection (Fig. 1F). The depositional characteristics, the presence of nesting material and fecal pellets, and the absence of carnivore or digestion marks suggest that the rodent used the cave for refuge and habitation. Given the lack of stratigraphic control due to the mixed nature of the CoS faunal assemblage, the remains had to be dated directly. Radiocarbon dating performed on the inorganic (bioapatite) fraction of the bone (38) yielded one date of 42,000 ± 690 calibrated years before present (calBP). Six other specimens were beyond the range of radiocarbon dating and are dated to >50 ka (SI Appendix, Tables S2 and S3). Complementary to radiocarbon dating, uranium–thorium (U-Th) analyses were used to date a small gypsum crystal associated with the in situ L. i. maremortum subsp. nov. skull SMNH-TAU M17121 (Fig. 1F). Assuming a closed system condition and minimal initial Th contribution (with a mean crustal value), the ²³⁰Th/²³⁸U age for this gypsum crystal is 103.3 ± 0.44 ka (SI Appendix, Table S4). The gypsum crystal likely formed after the deposition of SMNH-TAU M17121, so the dates indicate the minimum temporal span of L. i. maremortum subsp. nov. in the region (SI Appendix, Texts S2 and S3).

Systematic Paleontology.

Order Rodentia Bowdich, 1821

Family Muridae Illiger, 1811

Subfamily Lophiomyinae Milne-Edwards, 1867

Genus Lophiomys Milne-Edwards, 1867

Species Lophiomys imhausi Milne-Edwards, 1867

Type.

NMNH-ZM-MO 1910-51, a male purchased alive at the Port of Aden but of unclear provenience (likely northern Somalia).

Genus and species diagnosis.

SI Appendix, Texts S4 and S5.

Subspecies Lophiomys imhausi maremortum subsp. nov. (Fig. 2).

Fig. 2.

Selected craniodental and postcranial remains of Lophiomys imhausi maremortum subsp. nov. from the CoS. Holotype SMNH-TAU M16897, a cranium in (A1) dorsal, (A2) ventral, with augmented detail of the upper dentition, (A3) anterior, (A4) right lateral, and (A5) posterior views. SMNH-TAU M17121, a skull with cranium in (B1) dorsal and (B2) ventral views and (B3) left hemimandible in lateral view. SMNH-TAU M16877, left hemimandible in (C1) lateral, and (C2) occlusal views, with augmented detail of the lower dentition on occlusal view. SMNH-TAU M17120, a right hemimandible in (D1) lateral and (D2) medial views, with augmented detail of the lower dentition in occlusal view. SMNH-TAU M16888, left hemimandible in (E1) lateral and (E2) medial views. (F) SMNH-TAU M16911, right hemimandible in lateral view. (G) SMNH-TAU M16876, left maxilla with M1 and M2 in ventral view. SMNH-TAU M16878, cranium in (H1) posterior, (H2) dorsal, (H3) right lateral, and (H4) ventral view. SMNH-TAU M16871, partial cranium in (I1) dorsal and (I2) posterior views. SMNH-TAU M16886, left scapula showing the (J1) glenoid cavity and (J2) the scapula spine. SMNH-TAU M16900, lumbar vertebrae in (K1) anterior and (K2) dorsal views. (L) SMNH-TAU M17009, lumbar vertebrae in ventral view. The vertebrae are in anatomical connection and still joined but are not fused or pathological. (M) SMNH-TAU M17041, left pelvis in medial view. SMNH-TAU M17014, sacrum in (N1) posterior and (N2) anterior views. (O) SMNH-TAU M16902, left radius in posterior view. (P) SMNH-TAU M17055, right ulna in lateral view. SMNH-TAU M16884, right humerus in (Q1) anterior and (Q2) posterior views. (R) SMNH-TAU M17081, right tibia in lateral view. Note the pathological bone exostosis on the anterior proximal end. SMNH-TAU M16997, right femur in (S1) anterior and (S2) posterior views. Note the pathological bone exostosis on the shaft medially. SMNH-TAU M16889, right femur in (T1) anterior and (T2) posterior views. (U) SMNH-TAU M16873, left femur in anterior view. Left femur of a juvenile individual in anterior view with unfused (V1) shaft and (V2) distal epiphysis.

zoobank.org/urn:lsid:zoobank.org:act:C584F155-8CAF-45DB-A3B1-B625B91892C4

Common name.

Dead Sea Crested Rat.

Etymology.

maremortum, for the location of the type locality, near the Dead Sea.

Temporal Distribution.

Late Pleistocene, from ∼42 to ∼112 ka.

Geographical Distribution.

The type locality and SOD in northeastern Egypt (Fig. 1A).

Type Locality.

CoS (31.359542°N, 35.305132°E), Judean Desert, Israel.

Type Specimen.

SMNH-TAU M16897, complete cranium with left and right M1 to M3 (Fig. 2A). Housed at the Steinhardt Museum of Natural History, Tel Aviv University.

Other Subspecies.

The number of subspecies is unclear, but our analyses suggest the existence of two morphotypes, Lophiomys imhausi imhausi, Milne-Edwards, 1867 and Lophiomys imhausi bozasi, Oustalet, 1902 (SI Appendix, Text S10).

Diagnosis.

A subspecies of L. imhausi of large body size with craniodental affinities to modern populations of L. imhausi but differing in having a relatively flattened and elongated skull; parietals nearby or in contact with the posterior ridge of the orbit; lateral occipital crests directed laterally instead of posteriorly; supraoccipital processes directed posteriorly and roughly in level with the transverse plane; zygomatic arches less rotated and flattened ventrally; cranial granulations generally less marked and not extending prominently into the anterior part of the nasals, the premaxillae, and the anterolateral portion of the jugal bone; larger molars on average; M1/m1 and M2/m2 characterized by flared labial and lingual cingulae that produce a straight, rather than a sinuous, lateral outline; and M3/m3 relatively larger, with two distal cusps of similar size and labiolingually aligned. In addition to the above, L. i. maremortum subsp. nov. differs from L. i. imhausi in being significantly larger and from L. i. bozasi in having elongated limbs with proportionally narrow epiphyses.

Material, Description, Measurements, and Taxonomic Discussion.

See Figs. 2 and 3, SI Appendix, Texts S6–S10, Figs. S1–S8, and Tables S5–S14, and Datasets S1 and S2.

Paleogenetic Analyses.

DNA was extracted from four CoS Lophiomys samples following the protocols described in refs. 39 and 40 (SI Appendix, Text S11). Only one sample (SMNH-TAU M16883) had minimum coverage ∼2×, yielding an almost-complete mitogenome (SI Appendix, Table S15). With >42 ka, this sample is, to our knowledge, the oldest mitogenome sequenced in the Levant. We used 12 mitochondrial genes and at least three nuclear genes (up to five for other species) of selected rodent species (n = 33) representing the major subfamilies to build a phylogenetic tree (SI Appendix, Figs. S9 and S10 and Dataset S3). Our results revealed an uncorrected pairwise distance of 0.8% between the ancient Lophiomys sample and a modern specimen from Kenya (Dataset S4). This difference is considered not sufficient to warrant species status to the fossil population (usually >2%). The average estimates for the split between the ancient Levantine and the modern Kenyan Lophiomys samples using two different Bayesian molecular dating methods is 165 and 226 ka, respectively, with a 95% CI ranging between 72.2 and 337.2 ka (SI Appendix, Text S12, Figs. S11 and S12, and Table S16).

Species Distribution Models.

We used species distribution models (SDMs) to evaluate the distribution of potentially suitable habitat corridors for Lophiomys between eastern Africa, the Arabian Peninsula, and the Levant during the Late Pleistocene. The climatic envelope of extant L. imhausi was estimated based on 19 bioclimatic variables, three topographic variables (slope, roughness, and topographic position index [TPI]), and three proxies for vegetation cover and habitat productivity (normalized difference vegetation index [NDVI], net primary productivity [NPP], and percentage of tree cover [PTC]) (SI Appendix, Tables S17 and S18). Available records of L. imhausi in the wild were compiled from the literature and were subdivided into the two broad biogeographical groups, the SW and NE populations. After corrections for variable autocorrelation and pseudoabsence background selection, the input of the SDM comprised 14 variables, including 7 bioclimatic variables that reflect quarterly means of temperature and precipitation (bio8, bio9, bio11, bio16, bio17, and bio18) and 3 topographic variables (slope, roughness, and TPI; SI Appendix, Text S13). A multistep procedure involving variable transformation, variable selection, model selection, and Maxent analyses was performed with this data. Two models were selected. The best model with the lowest number of variables consisted of bio9, bio11, bio16, bio17, bio18, and roughness. Finally, several alternative models were run by testing all possible combinations of a formula that included 2 temperature variables, 2 precipitation variables, and 2 topographic variables out of the initial 10 variables. The model with the best area under the receiver operating curve (AUC) was selected and had a formula with roughness, slope, bio9, bio11, bio16, and bio18. This model has the advantage of being ecologically balanced and meaningful without losing much predictive power. Both the simple model and the alternative model worked optimally after 100 replicates, with values of the AUC over 0.93. Predicted habitat suitability estimations based on these two models were projected in paleoclimatic simulations of different time frames, including the Last Glacial Maximum (LGM, ∼21 ka) and the LIG (∼129 to 116 ka; Fig. 4 C–E and SI Appendix, Figs. S13–S16). The paleoclimatic simulations are based on global, coupled ocean–atmosphere–land–sea–ice general circulation models [NCAR Community Climate System Model (CCSM)] (41, 42).

The SDM results indicate, as expected, that the habitat is not suitable for Lophiomys at the location of the CoS and SOD with the present climate (Fig. 4C). The habitat would not have been suitable during the LGM either (Fig. 4D). However, during the LIG, habitat suitability estimates range between 0.21 and 0.63 in CoS and between 0.34 and 0.95 in SOD in a logistic probability scale from 0 to 1 (Fig. 4E). The alternative model gives slightly higher estimates than the best model, but in both models, the habitat suitability during the LIG is much higher than during the LGM or currently (values ∼0; SI Appendix, Figs. S14–S16 and Table S20). The mountain ranges on both the eastern and western sides of the Red Sea are predicted to be potential habitat corridors for Lophiomys, mainly because they include areas with high topographic relief that are the preferred habitat of this rodent (Fig. 4E). The SOD Lophiomys remains dated to ∼112 ka are roughly contemporaneous or slightly younger than the LIG. The fact that Lophiomys remains have been recovered from these localities within the modeled time frame supports the reliability of the SDMs.

Discussion

The Lophiomys skeletal assemblage from CoS is remarkable because crested rat fossil findings are extraordinarily rare, and the size of the collection is much larger than the taxon’s entire fossil record. The general preservation of the bones is outstanding despite being older than 42 ka, with several specimens preserving mummified tissue. The collection includes hundreds of the rodents’ coprolites, which open the door for further paleodietary and paleoenvironmental reconstructions. The comparative morphological analyses of abundant craniodental, mandibular, and postcranial remains and paleogenomic analyses (which include the oldest sequenced paleogenome in the Levant) demonstrate that the ancient populations from the Judean Desert and Egypt belong to a crested rat subspecies, L. i. maremortum subsp. nov., with several diagnostic characters in the skull and the dentition, larger body size, and elongated limbs. Furthermore, extant eastern African crested rats can be grouped in two distinct morphotypes that we hypothesize represent two different subspecies: 1) the individuals that inhabit the more arid and less-vegetated areas of Somalia, Sudan, Djibouti, Eritrea, and Ethiopia, tend to be small and are ascribed to L. i. imhausi (including the type specimen) and 2) the populations inhabiting the relatively forested and wet areas of Kenya, Uganda, and Ethiopia, tend to be larger and are ascribed to L. i. bozasi (Fig. 3 and SI Appendix, Fig. S4 and Tables S6–S9). More DNA samples, however, are needed to examine the relationships among the three hypothesized morphotypes and to further test its subspecies rank.

The CoS subfossils add critical information to the relatively recent biogeographic history of crested rats because this is the only account of their presence out of Africa (Fig. 1). In view of the small molecular divergence among relatively recent populations and the ancient distribution of Lophiomys over areas of northern Africa and the Middle East, the fragmentation and reduction of the geographical distribution of crested rats to eastern Africa must have occurred relatively recently. The molecular divergence estimates between the extant and fossil populations to the late Middle Pleistocene ∼226 to 165 ka (SI Appendix, Figs. S11 and S12) is compatible with a subsequent isolation of these populations during the Late Pleistocene <129 ka. The SDMs suggest that during the LIG (∼129 to 116 ka), the climatic and environmental conditions present in the Saharo–Arabian belt were favorable for the presence of Lophiomys in the Levant and the connection of this population with those of eastern Africa. It is currently not possible to determine how much time this Afro–Eurasian bridge existed during the Late Pleistocene, but paleoenvironmental records suggest it was brief. Regional speleothem δ¹⁸O curves from Soreq, Peqi’in, and Tzavoa caves and Dead Sea deep sediment cores show that rapid and drastic climatic fluctuations occurred during Marine Isotope Stage 5e (16, 43, 44), and pollen analyses from Dead Sea drills confirm that there was a notable increase in arboreal cover and sclerophyll vegetation during the LIG (23).

The African monsoon influence retreated quickly (44), and between ∼116 and 110 ka, a significant lake-level drop marks the return to extreme regional arid conditions in the Dead Sea basin, which likely resulted in a rapid decrease in woody vegetation (23). L. i. maremortum subsp. nov. likely survived in refugial areas, taking advantage of water springs, developed Acacia woodland, and patches of Mediterranean and tropical vegetation relics in the Dead Sea area (SI Appendix, Text S14). A few humid and vegetated refugia persist nowadays in association with water springs and Sudanian floral elements, like the Ein Gedi oasis (29, 45). Suitable habitats for Lophiomys must have been present at least until ∼42 ka as the climate deteriorated with the advent of the LGM and the spread of Irano–Turanian steppe biome (46). Since extant populations of Lophiomys generally prefer relatively wet and densely vegetated areas, the presence of a population in the Judean Desert suggests that the Dead Sea basin was greener than today and likely sustained patches of woodland/shrubland habitats during the early and middle parts of the Late Pleistocene, including during cycles of extreme regional aridity (44). Even the drier and less-productive areas occupied by extant crested rats are more productive than the Judean Desert today (Fig. 4G). One of these areas, the National Day Park in Djibouti, is shown in Fig. 4I as a conservative example of how the Judean Desert may have looked when L. i. maremortum subsp. nov. occurred there.

Our study provides independent paleoenvironmental data operating at a different resolution and spatial scales than previous research approaches (e.g., pollen, speleothem δ¹⁸O curves, and Dead Sea sedimentation), by providing local—instead of interregional—paleoenvironmental data. For example, the oxygen isotope ratios in the speleothems reflect primarily the east Mediterranean source-water composition (47, 48), and changes in weather systems from the Mediterranean had likely a weaker effect on precipitation in the Judean Desert due to local conditions of rain shadow (49). The Dead Sea level and that of its precursor, the Lisan Lake, were mainly regulated by the water input of the Jordan River and by Mediterranean precipitation regimes so that the Dead Sea basin sedimentation reflects the hydrological activity in a large and highly heterogeneous drainage area (50, 51). Similarly, a large part of the pollen from Dead Sea drills was likely transported by the Jordan River or wind-blown from the Judean Highlands, thus reflecting more the conditions in the watersheds of Cis- and Transjordan than those of the Judean Desert or a mixture of both (48, 52). In sum, these paleoclimatic and paleoenvironmental proxies do not necessarily reflect the local environmental conditions of the Judean Desert. Our study provides independent paleoenvironmental data, thereby filling the gap between cause (climate change) and effect (local environmental change). The window of opportunity for the migration of crested rats and early humans may have been short-lived (44), but we show that at least during the first half of the Late Pleistocene, the local conditions of vegetation cover (and arboreal composition) were sufficient to maintain the population of Dead Sea crested rats.

The sensitivity of the Dead Sea area to climate changes, the biogeographic admixture of vegetation and fauna, and the presence of refugia generated a mosaic of unique ecosystems in this northern part of the African Rift Valley during the Late Pleistocene. Our finding of long-distance habitat continuity between eastern Africa and the Levant proves that the dramatic climatic changes and the creation of ecological corridors during the Late Pleistocene were relevant in explaining faunal and early human dispersals out of Africa (7, 9, 15, 16, 44, 53). The most immediate implication is that the technological and sociocultural advancements of early modern humans were likely not a prerequisite for a dispersal during MIS5e; it is more likely that the dispersal was the ecological result of a regular expansion of their preferred habitats. There has been a tendency in anthropological literature to interpret the Levantine evidence for an early migration of H. sapiens out of Africa during MIS5e as a “failed excursion”—a punctuated dispersal that was both short and local (20, 54). However, accumulating archaeological evidence from Saudi Arabia, southeast Asia, and Australia indicate that human populations were present in these areas throughout most of the Late Pleistocene (13, 55–57). Whether these populations were survivors of earlier dispersal waves or newcomers, they ultimately disappeared without leaving a significant genetic trace in modern humans.

Though paleoclimatic and human population models are useful, fossils provide hard evidence of dispersal viability. The archaeological and fossil discoveries in Manot Cave indicate that the modern human populations that afterward colonized Europe were present in the northern Levant ∼55 ka (58, 59). The fact that Lophiomys’ populations survived in the Dead Sea area at least until 42 ka, suggests that The Dead Sea area could have supported savannah-like habitats that would have funneled hominin dispersal from the Arabian deserts into the Judean Desert before moving west to the Levantine Mediterranean coast. This could explain why most early– to middle–Late Pleistocene sites are located either in the Negev Desert or in the midnorthern Mediterranean region of the Levant (Fig. 4J). Current and future work in the Judean Desert caves will hopefully shed light on human movement through the arid portion of the Levantine corridor.

Materials and Methods

Comparative Material and Metrics.

Available adult specimens of L. imhausi were examined from the Natural History Museum, the Muséum National d’Histoire Naturelle, the Kenyan National Museums, and the Museum of Zoology, University of Cambridge. Only specimens with third molars fully erupted were considered. Reference SI Appendix, Fig. S2 for a visual guide to dental anatomical terms. Measurements are explained in SI Appendix, Fig. S3 and Table S5 and were taken to the nearest 0.01 mm with a Mitutoyo digital caliper by I.A.L. Multiple Mann–Whitney U tests with Bonferroni correction were used to test differences among populations for each tooth. All statistical analyses were carried out in R version 4.04. Results are summarized in SI Appendix, Tables S6–S11.

Radiocarbon Dating.

Given the lack of stratigraphic control and the mixed nature of the CoS faunal assemblage, the remains had to be dated directly. The preservation of organics in the cave, the presence of seemingly mummified tissue, and the low bone mineralization suggested that the specimens could preserve collagen. Collagen extraction was first attempted on a batch of 11 samples of L. i. maremortum subsp. nov. at the Oxford Radiocarbon Accelerator Unit. The pretreatment and radiocarbon dating procedures are described in detail in ref. 60. The % N was higher in all samples (average of 3.0 ± 0.8%; SI Appendix, Table S2) than the standard value (>0.7%) used to determine potential collagen preservation (61). However, when collagen extraction was attempted, all the 11 bones yielded no collagen (effectively 0% collagen). As a consequence, radiocarbon dating was performed on the inorganic (bioapatite) fraction of the bone (38). A total of seven specimens were analyzed in the Centre for Applied Isotope Studies (CAIS) at the University of Georgia. The inorganic fraction of the bone is susceptible to external contamination and molecular exchange with soil carbonates. However, it was shown that in desert caves like CoS, radiocarbon bioapatite dating does not significantly differ from traditional collagen radiocarbon dating (62). See extended methods in SI Appendix, Text S2.

U-Th Series.

Complementary to radiocarbon dating, U-Th analyses were used to date small gypsum crystals that were found within the CoS deposits. On a few occasions, these crystals had formed on bone surfaces, being more frequent on bones of larger taxa (e.g., caprines, cervids, and medium-sized carnivores). The Lophiomys remains did not show significant gypsum growth. When the in-situ skull SMNH-TAU M17121 was excavated from a small cavity in cave area L1, a small sample of the associated sediment (no more than 5 cm around the skull) was taken. This sediment was mostly composed of dry organic matter and fossilized feces. Two samples of these feces were radiocarbon dated at CAIS, and both were beyond radiocarbon calibration (>50 ka). Upon examination of the SMNH-TAU M17121–associated sediment in the laboratory, an isolated gypsum crystal was recovered. U-Th dating was performed on this sample at the Geological Survey of Israel. For optimal detrital correction, 10 other gypsum samples from the same cave were analyzed. See extended methods and results in SI Appendix, Text S3 and Table S4.

Geometric Morphometrics.

Geometric morphometric analysis of upper M1 molar shape (n = 47) was carried out on two-dimensional photographs of M1 occlusal surfaces. Three landmarks were digitised in the middle of each loph, where a rounded enamel feature could be seen in the dentine revealed in worn teeth. Additional 100 semisliding landmarks were digitized along the teeth parameter using the curve and resampled by length functions in TPSDig version 2.3 (SI Appendix, Fig. S5). Differences in shape between the CoS/SOD, SW, and NE populations were visualized by means of a covariance PCA (SI Appendix, Fig. S6) and a group PCA on Procrustes-transformed coordinates using the morpho library in R (Fig. 3 and SI Appendix, Fig. S7). Centroid size is calculated as the sqrt of the sum of squared distances of all landmarks from the center of the shape before Procrustes analysis. It is obtained as part of the output of the gpagen function in the library geomorph in R. A Procrustes statistical analysis of variance (ANOVA) with permutation procedures using procD.lm function implemented in geomorph was used to examine M1 shape differences among populations, and a statistical analysis of covariance (ANCOVA) was used to evaluate the interaction of size. Results are summarized in SI Appendix, Tables S12 and S13.

Molecular Analyses.

Four samples from the CoS were processed in the genetic analysis (SI Appendix, Table S15). DNA was extracted according to a method modified from ref. 39, as described in ref. 40. Genomic libraries were built following the BEST (Blunt‐End‐Single‐Tube) protocol with a few modifications, and sequencing was performed on an Illumina NextSeq 500 system for 75 cycles in single-read mode. Reads were mapped using Burrows–Wheeler Aligner (BWA) against a mitogenome of Lophiomys imhausi from Kenya (extended methods in SI Appendix, Text S11). Only one sample (SMNH-TAU M16883) had minimum coverage ∼2×, yielding an almost-complete mitogenome, and therefore was the only one used for further phylogenetic analyses. Phylogenetic inferences were carried out by building a supermatrix of 33 species, with representatives from the families Cricetidae, Muridae, and Spalacidae. Dipus was used as an outgroup. The supermatrix included sequences of 12 protein-coding genes (CDS) (Atp6, Atp8, COI, COII, COIII, Cytb, Nad1, Nad2, Nad3, Nad4L, Nad4, and NAd5) and 5 nuclear genes (Acp5, BRCA1, GHR, RBP3, and RAG1), which were downloaded from GenBank (Dataset S3). The best-fit partitioning scheme was identified by PartitionFinder2 run on the CIPRES Science Gateway version 3.3. Phylogenetic relationships were estimated using both maximum likelihood (ML) and Bayesian inference (BI). The ML analysis was carried out using IQ-TREE version 2.1.2, while BI analyses were carried out using Beast 2.6.3. Divergence time estimates were obtained using MCMCtree, contained in the PAML software and with Beast 2.6.3. Reference SI Appendix, Texts S11 and S12 and Table S16 for extended methods and results.

Climatic and Topographic Variables.

The ecological niche of Lophiomys and its potential distribution in the past was examined through climatic and topographic variables based on known occurrences (Fig. 4A and Dataset S5). After dealing with autocorrelation problems and other issues (SI Appendix, Text S13 and Tables S17 and S18), the resultant dataset included seven bioclim variables that reflect quarterly means in temperature and precipitation, including bio8 (mean temperature of wettest quarter), bio9 (mean temperature of driest quarter), bio11 (mean temperature of coldest quarter), bio16 (precipitation of wettest quarter), bio17 (precipitation of driest quarter), bio18 (precipitation of warmest quarter), and bio19 (precipitation of coldest quarter). Data on these bioclimatic variables was obtained for the current period (average of the years 1979 to 2013), the LGM (∼21 ka), LIG (∼129 to 116 ka), and other periods during the Late Pleistocene and Holocene (SI Appendix, Text S13 and Fig. S12). The topographic variables initially considered were slope, roughness, and TPI and were calculated from a 30 arc-s digital elevation model (DEM) grid using the function terrain in the library raster in R (references in SI Appendix, Text S13).

SDMs.

The construction of SDMs was based on maximum entropy fitting, which is widely used in species distribution modeling, using procedures implemented in the MIAmaxent R library. The model is built by subset selection in a multistage procedure. A set of variables are derived from the original variables using five transformation types: linear, quadratic, threshold, forward hinge, and reverse hinge (references in SI Appendix, Text S13). The most parsimonious set of derived variables is then selected by means of nested model comparison using χ² tests, with a threshold level (α) of 0.01 and no interactions allowed. This process is repeated to select a set of explanatory variables that constitute the best model (with the lowest Akaike information criterion [AIC] predictor) out of the selected derived variables, considering not only predictive power but also simplicity. A Maxent model is then trained based on the formula obtained in the previous step and projected to the desired time slice. The values of the generated rasters in raw Maxent format were transformed to logistic scale (0 to 1), with higher values indicating higher habitat suitability. The AUC is used as a measure of the model’s predictive power, ranging from 0.5 (random) to 1.0 (perfect discrimination). The resultant best and simplest formula included bio11, bio17, bio16, bio18, and roughness. Finally, an alternative model was selected by testing all possible combinations of a formula that included 2 temperature variables, 2 precipitation variables, and 2 topographic variables out of the initial 10 variables. The model with the best AUC was selected and had a formula with roughness, slope, bio9, bio11, bio16, and bio18 (SI Appendix, Text S13 and Dataset S6). The alternative model is more complex but has the advantage of being ecologically balanced and meaningful without losing predictive power. The average AUC of the best model is 0.932 ± 0.002, and the average AUC of the alternative model is 0.939 ± 0.001. Both the best and the alternative models were projected in different time slices. Habitat suitability at CoS and SOD was then calculated as a logistic value from 0 to 1 for each time slice (SI Appendix, Text S13, Figs. S13–S16, and Tables S19 and S20).

Data Availability

The current bioclimatic rasters were downloaded from https://chelsa-climate.org, which also hosts a DEM of 30 arc-s resolution (42). The paleoclimatic rasters were downloaded from Paleoclim.org, https://doi.org/10.1038/sdata.2018.254 (63). The PTC dataset was downloaded from https://globalmaps.github.io/ptc.html. The NPP dataset was downloaded from https://sedac.ciesin.columbia.edu/data/set/hanpp-net-primary-productivity/data-download. The NDVI dataset was downloaded from https://land.copernicus.eu/global/products/ndvi. The data are supported by the Socioeconomic Data and Applications Center in NASA’s Earth Observing System Data and Information System, hosted by Columbia University. The DNA sequence provided in this work is deposited into GenBank (accession nos. MZ159975–MZ159977 and MZ156017. The R code is accessible at Open Science Framework at https://doi.org/10.17605/OSF.IO/6Y7TB. The rest of the data used or generated in this work is provided in the SI Appendix and Datasets S1–S6.