The expansion of later Acheulean hominins into the Arabian Peninsula

The Acheulean is the longest lasting cultural–technological tradition in human evolutionary history. However, considerable gaps remain in understanding the chronology and geographical distribution of Acheulean hominins. We present the first chronometrically dated Acheulean site from the Arabian Peninsula, a vast and poorly known region that forms more than half of Southwest Asia. Results show that Acheulean hominin occupation expanded along hydrological networks into the heart of Arabia from Marine Isotope Stage (MIS) 7 until at least ~190 ka – the youngest documented Acheulean in Southwest Asia. The site of Saffaqah features Acheulean technology, characterized by large flakes, handaxes and cleavers, similar to Acheulean assemblages in Africa. These findings reveal a climatically-mediated later Acheulean expansion into a poorly known region, amplifying the documented diversity of Middle Pleistocene hominin behaviour across the Old World and elaborating the terminal archaic landscape encountered by our species as they dispersed out of Africa.

at depths of 1.33-1.49 m, and degrading granite bedrock at 1.62 m. Uranium-thorium ages on calcrete adhering to the artefacts suggested that the deposits were at least 200,000 years old, but the ages could not be considered secure (see Methods) and were not further reined.
Given the importance of Safaqah, a re-assessment of the site's stratigraphy, archaeology and chronology was conducted by the Palaeodeserts Project in 2014 through re-excavation and extension of Trench 1, and a study of the surrounding landscape 10 . As reported here, new ield investigations now also provide the irst secure dates for what is still the only known stratiied Acheulean site in the Arabian Peninsula.

Results
he Dawadmi region of central Saudi Arabia is characterized by a lat plain of Proterozoic igneous bedrock intruded by younger felsic and maic dykes, and overlain in places by Quaternary aeolian and luvial sediments. he climate is currently arid, although an intensiication of the African Summer Monsoon and its incursion into Arabia brought increased summer rainfall to the Peninsula in the past 12,13 . Palaeoenvironmental records and climate model simulations indicate that increased humidity occurred during insolation maxima, in particular during interglacials and interstadials, such as those during Marine Isotope Stage 7 (MIS 7, ~240-190 ka) and the interstadials of MIS 5 (~130-75 ka) 12,14 . During these periods, the landscape of Arabia featured extensive river networks, lakes and wetlands with widespread vegetative increases. he Safaqah site itself is situated near the Central Saudi Arabian town of Dawadmi. It is located beside the most prominent andesite dyke in the region, which rises ~60 m above the surrounding plain and is located on the northern lank of the dyke, below its highest point (Fig. 1). Our systematic survey revealed that Safaqah is the largest Acheulean site yet documented in Arabia and is surrounded by a dense Acheulean landscape 10,15 .
Palaeohydrological reconstruction demonstrated that the site sits at the interface between two major extinct river systems: the Wadi al Batin and the Wadi Sahba, major riparian corridors that lowed into the northern and southern ends of the Gulf, respectively ( Fig. 1) 15,16 .
We excavated several test trenches running north downslope from Whalen and colleagues' Trench 1 that conirmed increasingly attenuated stratigraphic sequences lacking buried artefacts. Re-excavation of Trench 1 itself revealed a shallow depression holding in situ material, which has inilled to give a thicker stratigraphic sequence and limited post-depositional artefact movement. his depression and its inilling is also clearly visible in piece-plots of the artefacts from Whalen's excavation (Fig. 2). We therefore extended this central part of Trench 1 by 2.5 × 1.5 m (Fig. 3).
Our excavation revealed 1.53 m of sediments within which we identiied seven stratigraphic units (A to G, Fig. 3). Both Layer G and the overlying Layer F are archaeologically sterile. A high density accumulation of artefacts in fresh condition, including relatively small pieces, are found concentrated within Layer E, the earliest  . Layer E predominately comprises andesite artefacts but also included the rare exploitation of rhyolite (Fig. 2). One of the large lakes from Layer E, which was resting on the core from which it had been struck, still had its eraillure lake adhering to it (Fig. 4a) indicating that the artefacts appear to be in efectively the same positions in which they were discarded by hominins. his initial occupation was buried by iner sediment with lower artefact densities (Layer D). he overlying deposit (Layer C) is the irst of three colluvial horizons that includes a distinct concentration of both large and small artefacts, as well a second concentration of rhyolite artefacts. Lower artefact densities are observed in Layer B, which contains discrete clusters of larger artefacts and dispersed evidence for quartz exploitation.
Sediment samples were collected for luminescence dating within Layers F and D of the freshly excavated trench, two layers bracketing the basal high-density in situ lithics in Layer E (Figs 2, 3 and 5, Tables 1-3). Samples were analysed using the post-IR elevated temperature (290 °C) infrared protocol (pIRIR 290 ) [17][18][19] and the infrared-radioluorescence protocol at controlled temperature (RF 70 ) 20,21 (See Methods). pIRIR 290 and RF 70 De dispersions are shown in Fig. 5 and corresponding age estimates are consistent at 2 sigma, providing an average age of ~276 ka for the sterile Layer F and ~197 ka for Layer D.
hese dates, together with the fact that central Arabia was predominantly arid during global glacial periods, indicate that artefact deposition in the intermediary Layer E occurred at the beginning of the MIS 7 interglacial. During this time, riparian and lacustrine networks of the Arabian interior were activated following an eastward extension of the African Summer Monsoon system 15,22,23 . he sharp contact between Layers F and E, and a very difuse contact between Layers E and D also supports the notion of a quasi-continuous phase of deposition between E and D during MIS 7 and into the beginning of MIS 6.
Our new excavations at Safaqah resulted in the recovery of over 500 lithic artefacts in stratiied context. hese are described in detail in Shipton and colleagues 10 . his assemblage is dominated by large lakes (the largest measuring 310 by 285 mm) that acted as biface blanks (Fig. 4b). hese lakes were detached from large blocks of andesite with considerable force (Fig. 4d). Marginally and ventrally retouched lakes, cleavers and discoidal cores were also present. Biface blanks were also produced through bifacial laking of appropriate edges on large blocks of andesite. Despite the late Acheulean date, the handaxes are minimally laked, long (x 16 mm, σ 38) and thick (x 5 mm, σ 15) (Fig. 4c).    he artefact positioning and stratigraphic observations indicate that the lower, chronologically constrained Layer E represents an occupational surface with in situ knapping, while the upper layers represent successive occupations, with some minor colluvial deposition in Layers C-B likely dating to MIS 6. Excavated material exhibited no size sorting, preserving lakes as small as 13 mm, which are not found on the surface. Our spatial plotting of artefacts from Whalen's excavation also showed that materials were unevenly distributed throughout: a small amount of quartz artefacts occurred in the upper layers, and rhyolite artefacts occurred as distinct horizontal clusters in the middle and lower parts of the excavation (Fig. 2).

Discussion
Our results demonstrate an Acheulean presence in the Arabian Peninsula during MIS 7, and illustrate how hominins moved into marginal areas with the onset of environmental amelioration. Safaqah also documents the youngest yet known Acheulean in Southwest Asia 24 .
he positioning of Safaqah at the drainage divide between both the Wadi al Batin and the Wadi Sabha palaeodrainage systems, suggests that Acheulean hominins were using luvial networks as dispersal corridors into the Peninsula's interior (Fig. 1) 13,16,23,25,26 . Technological similarities between Safaqah and other undated Acheulean sites strongly suggest this dispersal may have been widespread in Arabia. For example, minimally laked handaxes with large lake blanks and cleavers have been documented in both Wadi Fatima, and the Jubbah area ( Fig. 1) 27,28 , in addition to recently reported Acheulean sites in southern Arabia 29,30 .
he Acheulean technology from Safaqah can be contrasted to Acheulean surface assemblages found in the southwestern Nefud Desert 9 . Although undated, these assemblages document very diferent technological characteristics within an Acheulean tradition. he southwestern Nefud assemblages lack cleavers and any large lake component, despite the apparent availability of large blocks of raw material, and feature handaxes that are highly reined, morphologically pointed and shaped from raw material blocks (façonnage). his constellation of technological features is similar to the Late Acheulean in the Levant 31 .
In contrast, the large lakes and cleavers of Safaqah difer from the pene-contemporaneous Acheuleo-Yabrudian technocomplex of the Levant 10 and are instead features more typical of the African Acheulean 32,33 . Late Acheulean sites such as Mieso (Ethiopia) are of a similar age to Safaqah as well as manifesting large lakes and cleavers 34,35 , although it lacks the ine marginal trimming on its bifaces seen at Mieso. It therefore seems possible that Safaqah relects a hominin dispersal from the Horn of Africa following the northeastward migration of the African Summer Monsoon during MIS 7 34,35 . he similarities in the lithic typologies at Safaqah across all layers spanning MIS 7 and very probably MIS 6 in the case of layers C-B, are also indicative of cultural continuity at least within the general area of the site. Light colluvial activity within Layers C-B plausibly relects insolation peaks within MIS 6 (i.e. at ~170 and/ or 155 ka), providing climate mechanisms to support a sustained hominin presence in the region during generally arid conditions. he results presented here therefore both extend the known spatial and temporal distribution of Acheulean hominins and document variation in behaviour across an expanded range and timeframe. While the age of the Acheulean sites in other parts of Arabia is not yet conirmed, the techno-typological and chronological similarities between the later Acheulean in Africa and Safaqah indicate that several waves of dispersal may have structured the Acheulean record of Eurasia 36 .
Finally, because Safaqah also represents the youngest yet documented Acheulean in southwest Asia, further insights are provided into the persistence of the last Acheulean hominins, the youngest of which have been documented in India in MIS 6 37 . At Safaqah, this Acheulean presence was late enough to overlap with an emerging Middle Palaeolithic, both in the Peninsula as well as in surrounding regions, adding to the spatial diversity of Middle Pleistocene hominin behaviour 22 , and relecting the complexity and breadth of biogeographical exchange across the Eurasian gateway. he date of ~188 ± 11 ka in Layer D provides a terminus post quem for the artefacts in the overlying Layers B-C, suggesting that cultural overlap continued into MIS 6. he broader regional evidence, including the presence of Homo sapiens in association with early Middle Palaeolithic tool assemblages at Misliya Cave in the Levant during this time 38 , provides strong indications that Safaqah formed part of the terminal archaic landscape irst encountered by our species as they dispersed out of Africa. Future research should seek to clarify the spatio-temporal character of the inal Acheulean and early Middle Palaeolithic/Stone Age, as well as exploring the reasons for this complex transition.

Methods
Sediment samples for luminescence dating were collected from the freshly opened section by inserting opaque metal tubes. Preparation and analyses were carried out at the Luminescence Dating Laboratory of the Research Laboratory for Archaeology and the History of Art, School of Archaeology, at the University of Oxford under iltered laboratory lighting (low intensity LED lighting with peak emission at 594 nm). Ater the sample tubes were opened, the light-exposed ends were removed for dose rate determination and the interior, light shielded sediment was retained for luminescence dating. Each sediment sample was prepared in a conventional manner in order to extract sand sized (180-255 µm) potassium (K−) feldspar mineral grains. his included wet sieving, hydrochloric acid (10%) and then hydrogen peroxide (30%) digestions, followed by heavy liquid density separation using a solution of sodium polytungstate (2.58 g cm −3 ), and a inal second sieving.
For the external dose rate contribution, the beta dose rates were calculated from the uranium, thorium and potassium concentrations determined from a homogenized portion (10 g) of sediment by inductively coupled mass spectrometry (ICP-MS) and inductively coupled atomic emission spectroscopy (ICP-AES) outsourced to an accredited specialist laboratory (Actlabs, Canada). Gamma dose rates were measured on site for both samples with an Inspector 1000 gamma-ray spectrometer itted with a 2.5′ NaI probe and calibrated against the Oxford blocks 39 using the threshold technique 40,41 . he cosmic-ray dose rates were estimated from the equations provided by Prescott  the location of the site (altitude, latitude). Water content corrections were calculated assuming a burial average of 10 ± 4%, with correction factors from Zimmerman 43 . Assuming that each grain comprises 12.5 ± 0.5% potassium and 400 ± 100 ppm rubidium-87 44,45 , with absorption factors calculated from Brennan 46 and the rubidium dose rate from Readhead 47 , an internal feldspar dose rate of 0.91 ± 0.15 Gy. ka −1 was included in the dose rate calculations (Table 1). Equivalent doses (D e ) were measured for ten multigrain K-feldspar aliquots (1 mm diameter) from each sample using the pIRIR 290 protocol of hiel et al. 19 . Measurements were performed on a Lexsyg Smart luminescence reader manufactured by Freiberg Instruments 48 and itted with a calibrated beta source delivering 0.134 ± 0.003 Gy.s −1 . Aliquots were stimulated with infrared LED's emitting at 850 ± 20 nm (200 mW cm − ²) and the 410 nm IRSL emission signal was detected with a blue detection window (Schott BG39 3 mm and AHF Brightline HC 414/46 Interference 3.5 mm).
Net pIRIR 290 signals were calculated by subtracting a mean background (last 50 s of stimulation) from the total signal emitted over the irst 2 s 49 . Equivalent doses were calculated from each aliquot using the Analyst (v. 4.11) sotware developed by Duller [50][51][52] and meeting the following criteria: i) detectable net natural test signal greater than three sigma above the background signal; ii) test dose error greater than 20% of the calculated test dose; iii) recycling ratios >10%; iv) calculated zero-ratio >5% of L N /T N . Based on previous results from samples collected in the Nefud Desert and discussions in hiel et al. 19 and Buylaert et al. 53 , fading rates were considered to be negligible and ages are uncorrected. All twenty measured aliquots had measurable signals and good SAR characteristics; no aliquots were excluded ater application of the rejection criteria and none were found to be in saturation. Figure S1 (see supplementary information) shows a typical feldspar pIRIR 290 signal and corresponding growth curve. Overdispersion values 54 were low for both samples, with a value of 23 ± 6% calculated for DWD-OSL2, and 9 ± 6% for DWD-OSL3 ( Table 1). Investigation of a modern, aeolian-deposited sample has also indicated that well-bleached feldspars have negligible residual pIRIR 290 signal (~1 Gy). D e 's were also measured using infrared-radioluorescence (IR-RF). he IR-RF signal of K-feldspar is believed to provide a promising alternative to more established luminescence dating approaches. he technique was introduced by Trautmann et al. 55 21 and the D e values were obtained using the function analyse IRSAR.RF from the R 'Luminescence' package 21,59 . Figure S2 (see supplementary information) shows the results of the RF 70 measurements for the two samples. For both samples, the overdispersion values are less than 5%, indicating that the RF 70 signal has been suiciently bleached. We note that the pIRIR 290 and RF 70 age estimates are consistent at 2 sigma and in stratigraphic order ( Table 2).
he only independent age data available for this site consists of uranium-series dating performed upon three samples of 'caliche rind' collected from the underside of lithics excavated by Whalen and colleagues 11 . hese dates are summarized in Table 3. Little information is available concerning the methodology of this study, as it was published as a work in progress. herefore, the large age diference obtained for subsamples from the same tool is not explicable at the moment, but may be due to varying pre-treatments upon contaminated caliche rinds 11 . Because of this, and due to the nature of the samples (post-depositional carbonate concretion), these ages must certainly be considered as minimum estimates. While not fully reliable, these calculated ages do not conlict with the current pIRIR 290 and RF 70 age estimates. It is also important to note that, although these values indicate that there is post-depositional enrichment of the uranium component in the sediment, the proportion of the dose rate contributed by the uranium series in this environment is quite low. herefore, any decay chain disequilibrium is expected to have a negligible efect on the calculated ages. Interestingly Whalen and colleagues 11 observed that this 'caliche rind' exclusively occurred on the underside of artefacts, suggesting that however old it is, the artefacts have not moved since deposition.

Data Availability
Luminescence data generated during the current study are available from M.F. and J.-L.S. upon reasonable request.