Changing environments during the Middle-Upper Palaeolithic transition in the eastern Cantabrian Region (Spain): direct evidence from stable isotope studies on ungulate bones

Environmental change has been proposed as a factor that contributed to the extinction of the Neanderthals in Europe during MIS3. Currently, the different local environmental conditions experienced at the time when Anatomically Modern Humans (AMH) met Neanderthals are not well known. In the Western Pyrenees, particularly, in the eastern end of the Cantabrian coast of the Iberian Peninsula, extensive evidence of Neanderthal and subsequent AMH activity exists, making it an ideal area in which to explore the palaeoenvironments experienced and resources exploited by both human species during the Middle to Upper Palaeolithic transition. Red deer and horse were analysed using bone collagen stable isotope analysis to reconstruct environmental conditions across the transition. A shift in the ecological niche of horses after the Mousterian demonstrates a change in environment, towards more open vegetation, linked to wider climatic change. In the Mousterian, Aurignacian and Gravettian, high inter-individual nitrogen ranges were observed in both herbivores. This could indicate that these individuals were procured from areas isotopically different in nitrogen. Differences in sulphur values between sites suggest some variability in the hunting locations exploited, reflecting the human use of different parts of the landscape. An alternative and complementary explanation proposed is that there were climatic fluctuations within the time of formation of these archaeological levels, as observed in pollen, marine and ice cores.

Sites sampled. The eastern sector of the Cantabrian region contains several key sites pertaining to the late Mousterian, Châtelperronian, Aurignacian and Gravettian technocomplexes including Bolinkoba and Axlor in Bizkaia and Labeko Koba, Amalda, Aitzbitarte III, Lezetxiki and Ekain in Gipuzkoa (Table 1). Although there are other contemporaneous sites in those provinces, the sites analysed were selected because they are considered key regional sites with well-established stratigraphies that have been reviewed and recently dated by ultrafiltration [33][34][35][36] (Table 2). Additionally, these sites have been the subject of technological studies and some taphonomic analyses that proved human presence [38][39][40][41][42][43] and, also have environmental datasets [44][45][46][47][48][49] . Sites and levels sampled are presented in Table 1, with AMS radiocarbon dates shown in Table 2. Briefly, we describe the sites and levels included in this study and their chronology, from west (Bizkaia) to east (Gipuzkoa).
Axlor in Dima (Bizkaia) is a rock-shelter that was excavated by J. M. Barandiarán between 1967 and1974, revealing a long sequence of Mousterian Levels I-VIII 50,51 . Subsequent interventions between 1999 and 2004 linked a detailed stratigraphy (Levels B-N) to the levels recorded by Barandiarán 52,53 . In this study, Levels VIII, VI and IV from the Barandiarán's excavations were analysed, and correlated to new Levels N, M and D, respectively 38 . New ultrafiltration dates were available for the uppermost of these levels, Level D, but were beyond the radiocarbon limit (>49,300 OxA-32428 and >49,900 OxA-32429). These dates are considerably older than previous ones achieved for this level using AMS but without ultrafiltration: 42,010 ± 1,280 BP (Beta-144,262) and >43,000 (Beta-22586) 37,54,55 . These new results suggest that the lower levels at Axlor are significantly older than previously thought. Based on artefact typology and technology, they are now estimated to date to around 60-50ka BP 38 . Macromammals represented are red deer, followed by Spanish ibex and large mammals (horse and bovines), with evidence of butchering marks and fresh breakage patterns 50,51,56,57 . Although a complete taphonomic analyses of the complete assemblage has not been undertaken, a recent the review of carnivore and bird bones have revealed evidence of manipulation by Neanderthals 38 (Table 3) Bolinkoba, in Abadiño (Bizkaia), was originally explored by J.M. Barandairán and T. Aranzadi in the early 1930s 58 , followed by excavations by J.M. de Urquijo in the 1940s and again recently, between 2008 and 2014 by M.J. Iriarte 59 . The site stratigraphy reveals a series of levels from the Gravettian to the Azilian. Pertinent to this project, Gravettian Level VI (F) from the Barandiarán excavations was selected and a new date indicates a Gravettian occupation (25,280 ± 210 OxA-32519) 33 , although AMS dates from new excavation have provided dates of 29,950 ± 120 (Beta-426854) and 21,020 ± 90 (Beta-302981) for the same level 59 ( 2015). Recent review of macromammal remains from the Barandiaran and Aranzadi collection and, from the recent excavations, indicate similar proportions and presence of ungulates between both assemblages, with Spanish ibex being the most common hunted species followed by red deer 60 (Table 3). Despite the small representation of micromammals, these show cold palaeoenvironmental conditions during the Gravettian that included a modest decline in forest 61 . Palynological results did not provide palaeoenvironmental information due to the poor spore and pollen preservation 62 (Table 4).  33 29,950 ± 120 Beta-426854 AMS - 59 21,020 ± 90 Beta-302981 AMS - 59 Amalda VI 27400 ± 1000 I-11-665 C14 - 74 27400 ± 1100 I-11-664 C14 - 74 28540 ± 310 OxA-32426 AMS AF AMA26 33 28710 ± 300 OxA-34934 AMS AF AMA25 33 Aitzbitarte III IV 28,320 ± 605 Ua-18465 AMS - 136 22,420 ± 290 Ua-24965 AMS - 136 27,580 ± 550 Ua-18464 AMS - 136 26,260 ± 480 Ua-37961 AMS - 136 24,240 ± 365 Ua-11146 AMS - 136 25,815 ± 475 Ua-11148 AMS - 136 29  Labeko Koba (Arrasate, Gipuzkoa) was discovered during the construction of a road and was the object of a programmed salvage excavation during 1987-88. The now-destroyed cave held one of the few Châtelperronian levels (Level IX lower) together with Morin Level 10 and Aranbaltza, in the entire Cantabrian region 35,63,64 . There is a lack of human remains in sites attributed to the Cantabrian Châtelperronian, although recent research has suggested that this techno-complex was the work of Neanderthals 33 . The Labeko Châtelperronian is followed by Early Upper Palaeolithic levels, with a sequence of Protoaurignacian (Level VII) and Early Aurignacian technocomplexes (Levels IV-VI) [65][66][67] . These have been recently dated, producing an accurate chronology of the modern human occupation at the site 35 . The archaeozoological and taphonomic analyses revealed that the site was alternatively used as an occasional hunting camp and a carnivore den 68,69 (Table 3). Levels sampled for stable isotope analysis were IX lower (Châtelperronian) and Early Aurignacian Levels VI, V and IV. No well-preserved faunal remains were available from Level VII and Level IX upper was not sampled as it was archaeologically sterile 35,66 . Palynological remains from Level IX lower reveal the presence of mesothermophilic species, including Castanea, while the other levels display characteristically stadial botanical associations, with a low representation of arboreal taxa 45 (Table 4).

Site and Level
Cuon alpinus Meles meles   74 . Mousterian Level VII was once interpreted to have been an anthropogenic occupation level. Subsequent, taphonomic analysis proposed that this level was the deposit of a carnivore den visited sporadically by Neanderthals 40,41 . However, recent reanalysis reaffirmed that the level's contents were mainly human-derived, but with occasional carnivore activity 42,74,75 (Table 3). Recent AMS ultrafiltered dates of 44,500 ± 2100 (OxA-32500) and 42,600 ± 1600 (OxA-34933) confirm its Mousterian chronology 33 . The Gravettian technocomplex is present in Level VI, which is dated to between 28,540 ± 310 (OxA-32426) and 28,710 ± 300 (OxA-34934). All animal samples analysed from this site were selected from areas where stratigraphic integrity remained intact. Despite poor pollen preservation, the spectrum from Level VII is represented predominantly by pine trees (c.60%), with Cichorioideae accounting for c. 20% of total pollen and low frequencies of Chenopodiaceae.
There are more temperate pollen species in Level VI including Corylus and Betula 76 (Dupré 1990). The scarce micromammal remains recovered in Level VII are from Microtus arvalis-agrestis and Pliomys lenki, Arvicola terrestris, Apodemus sp. and Sorex and include cool climate indicator species. In Level VI the representation of cold adapted species is lower and there is an increased frequency of species that thrive in moist/temperate conditions including Sorex araneus (7%), A. terrestris (5%), with Talpa europea and N. fodiens also present 77 (Table 4).
Aitzbitarte III is part of a karstic system with five different caves in Errenteria, Gipuzkoa, near San Sebastián. The most recent excavations and analyses of material from the site were undertaken in the entrance area of the cave between 1994 and 2002, revealing a cultural sequence of Evolved Aurignacian (Level Vb central) and Gravettian (Levels Vb upper, Va and IV) 78 . Ultrafiltration dates reveal a rapid transition from the Aurignacian to the Gravettian. The dates at the cave entrance show that these levels represent a short span of time and were formed of discrete periods of occupation. The dates for Level Vb Central are 31,130 ± 390 (OxA-34932) and 31,600 ± 400 (OxA-32418) 33 which are coherent with previous AMS dates 78 and the cultural attribution to an Evolved Aurignacian phase. Dates obtained from Level Vb upper of 31,950 ± 450 (OxA-32419) and 30,990 ± 390 (OxA-32416) are similar to those obtained in Level Vb Central, indicating either an Evolved Aurignacian occupation that already shows some features transitional to the Gravettian or the first manifestation of an Early Gravettian at the site (Rios- Garaizar et al. 2013). New dates in Level Va, previously identified as Early Gravettian with Noailles burins 33,71 , appear to support a quick transition towards the Gravettian, with both dates being very similar: 31,300 ± 400 (OxA-32421) and 31,090 ± 400 (OxA-32420), and are consistent with other dates obtained in Levels Vb Central and Vb Upper. Finally, for Level IV, also classified as Early Gravettian with Noailles burins, the new dates of 29,130 ± 310 (OxA-32422) and 29,020 ± 320 (OxA-32499) are consistent with other Early Gravettian dates in the region. Archaeozoological and taphonomic analyses indicate an intense exploitation of red deer and bovines, followed by chamois and horse during the Aurignacian, while during the Gravettian bovines are the most common taxa, followed by red deer, chamois and horse 43 (Table 3). Micromammals have a similar faunal composition during both periods; Microtus agrestis-arvalis is the dominant group and Pliomys lenki is sparsely found. These species reflect harsh climatic conditions, cold and humid, with open landscapes and scarce woodlands 79 . Pollen also reflect cold conditions with variations in the moisture, that indicate a landscape dominated by herbaceous-shrub taxa, with very little representation of trees 44 . Sedimentological studies correlate with the environmental proxies suggesting severe conditions 80 (Table 4).
Lezetxiki in Arrasate, Gipuzkoa, was excavated by J. M Barandiarán between 1956 and 1968 and since 1996 has been excavated by A. Arrizabalaga  was poor with these levels and samples did not yield sufficient collagen for radiocarbon dating and isotopic analysis, and will not be discussed further, but this serves to provide caution to other researchers contemplating analysis of remains from this site. Besides, Aurignacian Level IIIa, from the original excavation, was not sampled due to its possible mixing with Mousterian Level IIIb 34 , in addition to it having a possible Solutrean component.

Materials.
Bones of red deer (Cervus elaphus) and horse (Equus sp.), two of the most common mammal species represented during the regional Middle and Early Upper Palaeolithic 43,84-86 , were sampled. These specimens were selected strategically to measure the impact of broader climate on the isotopic values of two ecologically different species; grazers (horse) and intermediate feeders (red deer) to be observed. No contemporary browsers were available for comparison as they are scarcely represented during the period of study in the region. Specimens with evidence of anthropogenic modification (i.e., cut marks and/or anthropogenic breakage), associated with stone tools were targeted. Fused long bones of mature individuals were selected to prevent analysis of juveniles that might have prevailing weaning signatures 87,88 . In total, 138 animal individuals belonging to Mousterian (n = 61), Châtelperronian (n = 11), Aurignacian (n = 31) and Gravettian (n = 35) levels at six archaeological sites were analysed for δ 13 C and δ 15 N and 30 of these specimens were also analysed for δ 34 S ( Table 1). The 12 horse and 18 red deer specimens selected for δ 34 S all contained >5 mg of collagen required for analysis and had δ 15 N values from the lower, middle and upper ranges of the dataset to explore the relationship between δ 15 N values and feeding locations. At the beginning of this project, radiocarbon dating using the AMS ultrafiltration method was undertaken to confirm the chronological attribution of the levels sampled 33 . Although some of the archaeological sites already had radiocarbon dates (mostly completed in the 1990s and early 2000s), new dates were run with ultrafiltration method (Table 2), which has greatly improved the reliability of the ages obtained due to the more effective removal of low molecular weight contaminants 89,90 , especially during this period of study.

Results
Collagen preservation and sample integrity. Collagen  Results of the δ 13 C and δ 15 N analyses. Broad temporal trends. The δ 15 N values of both species range between 1.3‰ and 9.2‰, which is higher than typically observed for animals feeding within the same trophic level in the same geographical location 94 . The δ 13 C values range between −21.6‰ and −19.2‰. To explore wider broader environmental trends, specimens from each cultural period, defined by archaeological assemblage characteristics were grouped and populations were compared statistically, as outlined in the methods section at the end of this article.
There is an increase in mean red deer δ 15 N values from the Mousterian (3.4‰) and Châtelperronian (3.3‰) to the Aurignacian (4.3‰) and Gravettian (4.7‰), although these means are likely influenced by the presence of individuals with elevated δ 15 N values (Table 5; Figs 2-5). The Mousterian red deer populations were statistically significantly different from those in the Aurignacian (p = 0.02) and Gravettian (p = 0.01) ( Table 6). For the horse, the highest mean δ 15 N values are in the Mousterian (4.5‰), decrease in the Châtelperronian (2.9‰) and stay at a lower level throughout the Aurignacian (3.6‰) and Gravettian (3.8‰), but no statistically significant differences between δ 15 N values of horse populations is observed between any of these cultural levels ( Table 5).
There is little temporal change in δ 13 C values. In the Châtelperronian, red deer have lower δ 13 C values than in the Mousterian (Table 5) and the populations were statistically different (p = 0.04) ( Table 6). Horses had slightly lower δ 13 C values in the Châtelperronian compared to the Mousterian, but no statistically significant differences between populations within consecutive temporal periods were observed (Table 7).
When comparing the red deer and horse values, both species have similar mean δ 13 C values in the Mousterian and horses have a mean δ 15 N value 1‰ higher than the red deer (Fig. 2). In the Châtelperronian, Aurignacian and Gravettian, there is a shift in the diet of horses relative to the red deer and the mean horse values are lower in δ 13 C and δ 15 N values than red deer ones (Table 5) (Fig. 2).
Period-specific trends. Mousterian. Within the Mousterian, both the red deer and the horses display wide ranges in the δ 15 N and δ 13 C values observed. A group of horses and red deer have δ 15 N values between 1.8‰ and 5.1‰. There is another cluster of horse and red deer specimens with δ 15 N values within ± 1σ of the mean, ranging between 5.7‰ and 9.5‰ (Figs 2, 4), which have been identified as belonging to different groups using cluster analysis (horse cophenetic correlation coefficient = 0.91; red deer = 0.87) The latter is outside of the range typically expected for individuals feeding within the same geographical location 94 . This pattern is particularly observed in Mousterian Levels IV, VI, VIII of Axlor and in Amalda Level VII. There is little difference in the δ 13 C values of either species and a clear overlap is seen (Figs 3, 5).
Châtelperronian. Level IX lower at Labeko Koba is the unique Châtelperronian level in the eastern Cantabrian region dated by ultrafiltration 35 . All horses and red deer plot within the same δ 15 N trophic range (horse: 1.5-3.6‰, red deer: 2.7-3.8‰) (Figs 4, 5.). There is no overlap between the two species in the δ 13 C values (red deer min −20.1‰, max −19.8‰, horse min −21.0‰, max −20.2‰). Red deer δ 13 C values are consistently higher relative to the horse analysed in this level and difference in δ 13 C values between the two populations is statistically significant (p = 0.01) ( Aurignacian. Aurignacian Levels VI and V in Labeko Koba and Level Vb Central of Aitzbitarte III, exhibit a similar pattern to the Mousterian, with a group of individuals with lower δ 15 N values (range; 2.3‰ to 5.0‰) and another with elevated δ 15 N values (range; 6.7‰ to 9.2‰) (Fig. 4), and cluster analysis showed these individuals as belonging to different groups (horse cophenetic correlation coefficient = 0.93; red deer = 0.95). Within Level V at Labeko Koba all individuals analysed plot within the lower nitrogen range, except for one horse and Level VI contains two red deer with elevated δ 15 N values (Fig. 5). The δ 13 C ranges of the individuals with high and lower δ 15 N values fall within the same range (−21.2‰ to −20.5‰, Fig. 3) and there is a clear differentiation in the δ 13 C values of horse and red deer, with horse having consistently lower values (Fig. 2).
Gravettian. In the Gravettian horse and red deer are found to have lower δ 15 N values (range 1.3‰−5.1‰), apart from a group of red deer with higher values (6.9‰−9.5‰) observed within Aitzbitarte III Level IV, Amalda Level VI and at Bolinkoba Level VI(F) (Figs 4, 5). The red deer and horses with higher δ 15 N values from this period were identified using cluster analysis as belonging to a different group (horse cophenetic correlation coefficient = 0.94; red deer = 0.91). This is consistent in the preceding periods, although, unlike the earlier periods, no horses show higher values. As seen in the Châtelperronian and Aurignacian, horses have lower δ 13 C values than red deer (Fig. 2).
Results of the δ 34 S analysis. The δ 34 S values of the red deer and horse analysed ranged between 1.8‰ and 12.6‰, with most specimens falling between 5‰ and 11‰ and are consistent with animals inhabiting terrestrial ecosystems 20 . There is no linear correlation between δ 34 S and δ 15 N values (r = 0.049) or δ 34 S and δ 13 C values (r = 0.076), which would be expected if there was a link between the sulphur isozones and the carbon/nitrogen zones being exploited (Fig. 6)  In Amalda the values were generally higher than at Labeko Koba. In Amalda Level VII, δ 34 S values ranged between 9‰ and 12‰ and in Amalda Level VI, the δ 34 S values from the bulk of the individuals ranged between 8.7‰ and 11.9‰ (Fig. 7), with only one specimen showing an uncharacteristically low δ 34 S value of 1.8‰ with a δ 15 N value of 4.8‰ (Fig. 7). This was inconsistent with any of the other specimens analysed in this study.
At Aitzbitarte III, δ 34 S values ranged between 6‰ and 12.6‰ (Fig. 7). The Evolved Aurignacian samples in Level Vb Central ranged between δ 34 S values of 9.4‰ and 12.6‰. For the Gravettian at the site, Level Va one individual had a lower δ 34 S value of 6‰ with a δ 15 N value of 8‰. All other specimens from Levels Va, IV and Vb Upper fall within a δ 34 S range of 8.2‰−11.9‰.

Broad trends in bone collagen values and its impact for environmental reconstruction.
Within the red deer and horses sampled no δ 13 C value is higher than −22.5‰, meaning that they are consistent with the consumption of C3 plants in an open landscape, according to Drucker et al. 24 . They are also stable through time, indicating that there was little change in tree cover throughout the period of study (i.e., late MIS3). The higher δ 13 C values of horses, in comparison to red deer, observed in the Mousterian, and an opposite trend afterwards suggest a change in environment in the temporal sequence related to the horse niche. Typically horses persistently select poor quality, low-protein, high-fibre browse a fact which results in a relative depletion in the δ 15 N and δ 13 C values 96 , as seen in this study in the Châtelperronian onwards. A similar pattern has been observed in other Palaeolithic faunas in Central Europe, where horses are consistently depleted in 13 C and 15 N relative to red deer due to this dietary preference 21,22,[97][98][99] . This change in the horse relative δ 13 C values demonstrate a shift in the suite of vegetation, resulting in niche separation of red deer and horses, likely expressing environmental shifts, resulting from wider climatic changes between the Mousterian and the later periods studied. Other palaeoenvironmental records in the region during this period of study are sparse ( Table 4). The only Mousterian level with pollen information is Amalda VII, and it is mainly represented by Pinus and the Cichorioideae, although pollen preservation at the site was poor and pollen samples were selected from an area of the site with possible disturbance. During the Châtelperronian (represented by Level IX Lower at Labeko Koba) pollen evidence indicates that steppic vegetation dominated (Table 4). Similarly, trends based on the data from the sites with Aurignacian and Gravettian levels indicate that low tree cover, and steppic vegetation, such as the Poaceae/Gramineae grasses Cedrela tubiflora and Compositae liguliflor, dominated the landscape ( Table 4). The shift towards more open, steppic landscapes, after the end of the Mousterian, environments more favourable to the niche of horse, may explain this change in the diet of horses, although its representation in the macromammal assemblages remained stable through the period of study in the eastern Cantabrian region (Table 3)  both the red deer and horse requires further exploration. It is not seen during Châtelperronian, but this may be a product of the small sample, consisting of only one archaeological level from one site.
Typically differences between herbivorous individuals of up to 2‰ in δ 15 N are expected on an intra-site level 94 . In this study, however, the dataset shows a δ 15 N value range of 5‰ and above for both species. Furthermore, the site of Antoliñako Koba, also in Gipuzkoa, exhibited a similar phenomenon to the current study with six red deer dating to the Gravettian period having δ 15 N values greater than 7‰ (maximum value 8.1‰, range 4.8‰, σ = 1.8, Table 6 in 100 ), alongside a group of five individuals with lower values (3-5‰) 97 . Comparative red deer datasets of archaeological specimens, predominantly from the Late Upper Palaeolithic, all fall within lower δ 15 N ranges (typically between 3 and 5‰) with smaller standard deviations (Table 6) 100 .
The causes of elevated δ 15 N values in bone collagen can be related to either physiological or environmental factors. Regarding physiology, juvenile animals being nursed produce a δ 15 N value around one trophic level above that of their mothers 87,88,101 . However, all individuals analysed in this study were archaeozoologically determined to be adults, based on bone fusion, so this factor can be disregarded. Short-term stress episodes such as starvation or pregnancy can impact on δ 15 N values within an individual 102,103 , although to register in the long-term bone collagen signature, this had to have been experienced over long durations. Physiological factors are not sufficient to explain the pattern.
In environmental terms, altitude has been observed to have both positive 104,105 and negative effects 106 on δ 15 N values of plants and animal tissues. Positive correlations between plant δ 15 N values and altitude, related to soil activity and temperature differences, could exist, although a change in δ 13 C might also be expected. Altitudinal differences of up to 1,000-4,000 m are needed to have a substantial impact on isotopic values in plants 104,107 . Altitudes of up to 2,000 m in the Cantabrian Cordillera exist, but only in the west 31 , although none of the sites in our sample are located over 350 m.a.s.l. Additionally, feeding habits at high altitudes are not consistent with the ecology of horse and red deer, which typically do not inhabit higher elevations for prolonged periods of time. Modern ecological studies of red deer populations inhabiting mountainous regions demonstrate that red deer typically do not   reach altitudes above 500 m 108 , although caution must be considered when using modern ecological analogues. Nonetheless altitude differences are not sufficient to explain the large differences in δ 15 N values seen within this dataset. Another environmental factor is the difference in the parts of plants consumed by the animals. An increase in leaf consumption or a shift from eating shrubs or trees to grasses could both explain the increased δ 15 N values 109,110 . Vegetation change was cited as an explanation for increased ranges in δ 15 N values of red deer in the Jura uplands of eastern France during the Late Glacial period 23 . If regular consumption of different types of vegetation by individuals could cause differences in the δ 15 N values, we would expect this to typically affect the δ 13 C values as well 111 which is not observed in this study, meaning that other environmental factors must be at play. Differing mycorrhiza in ecosystems depending on localised conditions can impact on 15 N concentrations absorbed by plants 112,113 . Soil activity of nitrogen fixing mycorrhizal is reduced within colder environments and consequently, a temperature decrease is observed 114,115 , and decreased δ 15 N is often associated with lower mean annual temperatures 116 . Soil nitrogen content within different floodplains can also vary depending on water table    Consequently, animals habitually feeding in drier environments would be expected to have higher δ 15 N values than those consistently feeding in wetter environments. Given the above, to explain differing δ 15 N values in bone collagen within individuals in the same archaeological level, there are two plausible scenarios. Firstly, it could be that animals brought to the caves were being hunted in two isotopically distinct environments ('isozones'), producing a mixed assemblage of animals with   different isotopic signatures within the same archaeological levels, as has been recently proposed for contemporary regional sites in the central area of the Cantabrian region 125 . At early times in site occupation, humans might have consumed "trail food"-joints of transported meat acquired in their area of hunting-before they acquired local game. It is conceivable that human groups moved across the low Basque Mountains between the upper Ebro drainage and the Cantabrian coastal region as part of seasonal rounds (as suggested by the presence of Trans-Cordilleran flints (Urbasa, Treviño, etc.) in the sites of Bizkaia and Gipuzkoa. Secondly, it is conceivable that environmental fluctuations occurred during the formation of the archaeological levels, with changing environmental conditions of temperature and aridity, causing the accumulation of individual animals with distinctly different stable isotope signatures, as proposed by Bocherens et al. 21 for the ungulate assemblage studied in the SW France during the Middle to Upper Palaeolithic transition. The long-time formation of the archaeological levels analysed converted them into thick palimpsests which could represent multiple human occupations perhaps over centuries, and this fact might also explain the changes in the observed climatic conditions. Regarding the first hypothesis, the δ 15 N results could be a result of animals feeding in two isotopically distinct territories, with human activity being responsible for bringing these two groups of animals habitually feeding within different geographical regions: one, with lower δ 15 N values and another, with higher δ 15 Table 9). The explanation provided was that those animals might have been obtained from different hunting locations. Lower δ 15 N values are seen during the Châtelperronian of Labeko Koba (Level IX lower), where ungulates brought to the site by carnivores and alter scavenged by humans, are indicative of a local δ 15 N signature at that time. Another possibility is that the outlying individuals came from a drier region, such as the area to the south of the Cantabrian cordillera, potentially the Llanada Alavesa in Alava, Province of the Basque Autonomous Community (Fig. 1), in the rain shadow of the mountains and outside the reach of the Foehn effect. Proposed hunting ranges during the Middle and Upper Palaeolithic, based on the Optimal Foraging Theory, suggest that hunting territories were modified to target higher ranked prey, with movements between the coast and the mountains 86 , although the mobility in this particular region based on predictive models to calculate the potential distribution of ungulates, according to topography and related vegetation, indicate an area of exploitation within the two-hour walking territories of the sites 128 . At Axlor and Amalda studies suggest that, due to the steep topography surrounding these sites, occupants would have had extended territories to target specially selected prey 128 . Based on this, hunting in various locations within the wider sites was an option as reveal by the catchment territory of lithic raw materials in the sites analysed 38 . Thus, macromammal assemblages from the studied sites represent taxa related to the topographic location of each site, with high representation of montane (or at least, steep, rock environment) ungulates, fluvial valleys and plains or all of the above. In the sites studied, red deer is the most common taxon (24%), followed by bovines (18%), horses (7%), Spanish ibex (7%) ( Table 3) and exceptionally, chamois (at most 42%) is highly represented in Amalda Levels VI-VII attributed by one researcher to a carnivore accumulation with sporadically human occupations 40,42 . Taphonomic analyses in Aitzbitarte III show that prey transport was mostly dependent on body size. Age profiles indicate exploitation of prime-age individuals, although infantile and juveniles are also represented 43 . Bone marrow extraction has been documented in all the sites, in combination with butchering marks on the ungulates and even on some carnivores 41,42 . For instance, in Aitzbitarte III Level Vb, cut marks were identified in an ulna of cave bear and, in Axlor exploitation of carnivores (wolf and dhole) and birds (raven and golden eagle) by Neanderthals were also documented 39 .
To evaluate animal provenance further, δ 34 S analyses was undertaken. Sulphur studies of modern ecosystems note a 1.9‰ difference within ungulate populations, with higher variability in archaeological studies of 2.4‰ variation observed 20 . At Labeko Koba (Level IX lower) a 3‰ difference was observed between specimens, all of which resulted from a hyena-formed deposit, later scavenged by humans 68 and can be assumed to represent animals accumulated from the local area. For the other sites studied, excluding outliers, a range of 4‰ was observed. This suggests that the level of variation for animals feeding within the same region could be up to 3-4‰, particularly in landscapes within deep, narrow, sinuous valleys. The δ 34 S values from all levels at Amalda and Aitzbitarte III were typically higher than those seen within the levels analysed at Labeko Koba and may not have derived from the same locality. All three sites are in different, but adjacent valleys, which may explain the differences in baseline δ 34 S values observed between them (Fig. 7). One red deer from Amalda Level VI and one from Aitzbitarte III Level Va had very low δ 34 S values in comparisons to other analysed from the same levels, suggesting procurement of animals from at least two locations by the inhabitants of these sites (Fig. 7).
There was no linear correlation between δ 15 N and δ 34 S values, suggesting that if the animals with higher δ 15 N were being hunted from elsewhere, they were not necessarily coming from a different sulphur isoscape. Considering that 34 S systems are dictated by geographically determined factors, such as rock type and precipitation 20,95 and δ 15 N systems are controlled by more biological factors, such as nitrogen fixation by plants, nitrogen and sulphur isoscapes may not necessarily be the same.
The hypothesis of multiple catchment territories with different geographical and environmental conditions producing distinctive isotopic values (i.e. from the Llanada Alavesa) could be considered less feasible, when contemplating hunting efficiency for these hunter-gatherer groups in relation to the distance to the camp site and prey size hunted, among other factors 129 .
Regarding the second hypothesis, the high variability of δ 15 N values of the red deer and horses within the same archaeological levels could be a product of cycling environmental conditions within these archaeological levels (accumulated usually during multiple human occupations along hundreds of years and sometimes, thousands), potentially relating to fluctuating temperatures and cycling periods of aridity. The second part of MIS3 was a time of remarkable instability with abrupt and rapid climatic changes. The Heinrich events produced a sequence of cooling and warming cycles 2 and the Dansgaard-Oeschger cycles presented periods of climatic fluctuation 11 , occurring sporadically within millennial scales 1,4,130 . Offshore records from the NW of the Iberian Peninsula exhibit cycles of wetter-drier periods 13 . Marine core MD95-2042, from the SW Margin of the Iberian Peninsula, demonstrated millennial scale climatic changes, ranging from temperate and humid to cold and dry continental conditions 130 . Available environmental proxies for these sites do not exhibit evidence of intra-level fluctuations (Table 4), although these indicators may not be sensitive enough to reflect these smaller changes on this temporal scale. Isotopic values obtained in this study provide a direct indication of past environment, directly related to evidence of human occupation.
In terms of chronology of the timescales of the levels involved with the fluctuating cycling climatic conditions, the Mousterian levels within Axlor date beyond the radiocarbon limit, and it is possible that these represent thick deposits spanning several millennia. For the other archaeological levels analysed, the chronologies represented are much tighter.  Table 2). The tight chronological resolution of these levels, means that environmental oscillations would had to have occurred in a timeframe of hundreds of years, and the abrupt nature of these changes could have implications for human subsistence and behaviour. Other contemporaneous European sites with analyses of both red deer and horses 21,23,131 do not show evidence of rapid frequency of environmental shifts, a fact which may be related to site formation chronologies or local environmental conditions in the particular studied regions. Further exploration of this hypothesis is required, with more precise chronological timescale, in additional regions occupied by Neanderthals and AMH that represent similar timeframes, as presented here, would be beneficial in the future.

Conclusions
The results of this study of bone collagen δ 13 C, δ 15 N and δ 34 S values conducted on animal remains, with evidence of human manipulation, pertaining to the Middle-Upper Palaeolithic transition in the Cantabrian Region of Northern Spain provide a human-related reconstruction of the past environmental conditions at the time the replacement of late Neanderthals by the anatomically modern human populations took part in the Atlantic zone of Iberia. During the Palaeolithic, the Cantabrian region was special in providing a variety of habitats to exploit for both human species, and whilst MIS3 was a period of climatic instability in Europe, this did not prevent either human populations from successfully occupying this region. The δ 13 C values results show that after the conclusion of the Mousterian, a shift in climate, was expressed in the eastern part of the Cantabrian region by a change in the dominant vegetation, generating more open landscapes, with the horse niche changing to favour the lower-quality browse that they preferentially consume, an hypothesis which is supported by the other environmental proxies including sedimentology, pollen and micromammals. The δ 15 N results show high inter-individual variability within the same archaeological levels in the Mousterian, Aurignacian and Gravettian. This could be linked to animals hunted in different territories, although whether this represents micro-environments within the eastern Cantabrian region, and the same valley system, as proposed for the central region of Cantabria or animal carcasses (or joints of meat) being brought from further afield as trail food, is difficult to determine. However, we do not discard the possibility that these large differences in δ 15 N values are more a reflection of changing environments during the formation of these archaeological levels, representing several generations of human occupation activity. If so, changing environmental conditions across the transition would have had implications for human evolution and adaptive skills. Further research is required to explore, involving high resolution sampling and dating to observe the difference variation in isotopic values among the animals consumed by human groups during the period of study in this region.

Methods summary. Collagen extraction was undertaken following procedures outlined in Richards and
Hedges 132 with an ultra-filtration step 133 . Bone fragments between 0.5 and 0.8 g were cleaned using aluminium oxide air abrasion, before demineralisation in 0.5 M HCL at 6-8 °C for between 3-10 days,and were washed three SCIentIfIC REPORtS | (2018) 8:14842 | DOI:10.1038/s41598-018-32493-0 times using de-ionised water. Samples were gelatinised in a weak acidic solution (pH3 HCL) at 70 °C for 48 hours, then filtered with 5-8 μm Ezee ® filters, prior to ultrafiltration to separate out the larger >30ka collagen chains.
The >30ka fraction was frozen and lyophilized for 48 hours. Collagen was analysed in duplicate, using a Delta XP mass spectrometer coupled to a Flash EA 2112 elemental analyser at the Department of Human Evolution, Max Planck Institute for Evolutionary Anthropology (Leipzig, Germany). The δ 13 C values and δ 15 N values are reported relative to the V-PDB and AIR standards. International and internal standards were used to calculate analytical error which was ± 0.1‰ (1σ) or better. The mean difference observed between duplicate measurements was 0.03 for δ 13 C, and 0.01 for δ 15 N. Sufficient collagen was not available for duplicate analysis for specimens: AXL01, LAB06, and EK02, EK04, EK05 and EK06. Analysis for δ 34 S values was undertaken at the University of British Columbia Stable isotope laboratory in Vancouver using a MicroCube IsoPrime 100 DI mass spectrometer.
Isotopic values were analysed statistically using a Mann-Whitney U test, with a post-hoc Holm-Bonferroni correction 134 . A p value of <0.05 or less was deemed to be indicative of a statistically significant result. To compare nitrogen value groupings within the dataset classical cluster analysis was used. All tests were undertaken using the statistical package PAST 135 .
A summary of environmental proxies at the sites investigated in this study, including pollen, microfauna, macrofauna and sedimentology, when available, have been included in the discussion to enhance interpretation of the results and provide general environmental context (  Table 9. Red deer comparisons of mean, minimum, maximum and range δ 15 N values from existing isotopic studies in the Cantabrian Region. No horse specimens were available as a comparison.