Research Article

Last Interglacial Iberian Neandertals as fisher-hunter-gatherers

See allHide authors and affiliations

Science  27 Mar 2020:
Vol. 367, Issue 6485, eaaz7943
DOI: 10.1126/science.aaz7943

Fruits of the sea

The origins of marine resource consumption by humans have been much debated. Zilhão et al. present evidence that, in Atlantic Iberia's coastal settings, Middle Paleolithic Neanderthals exploited marine resources at a scale on par with the modern human–associated Middle Stone Age of southern Africa (see the Perspective by Will). Excavations at the Figueira Brava site on Portugal's Atlantic coast reveal shell middens rich in the remains of mollusks, crabs, and fish, as well as terrestrial food items. Familiarity with the sea and its resources may thus have been widespread for residents there in the Middle Paleolithic. The Figueira Brava Neanderthals also exploited stone pine nuts in a way akin to that previously identified in the Holocene of Iberia. These findings add broader dimensions to our understanding of the role of aquatic resources in the subsistence of Paleolithic humans.

Science, this issue p. eaaz7943; see also p. 1422

Structured Abstract


A record of the regular exploitation of aquatic foods has been lacking in Neandertal Europe. By contrast, marine resources feature prominently—alongside personal ornaments, body painting, and linear-geometric drawings—in the archeology of Last Interglacial Africa. A competitive advantage scenario of human origins is that the habitual consumption of aquatic foods and the fatty acids they contain, which favor brain development, underpins the acquisition of modernity in cognition and behavior. The resulting innovations in technology, demographic growth, and enhanced prosociality would therefore explain modern humans’ out-of-Africa expansion with regard to both dispersal process (along coastal routes and to southern Asia first) and outcome (the demise of coeval non-modern Eurasians). A corollary of this view is that the paucity of marine foods at Neandertal coastal sites is a genuine reflection of their subsistence behavior.


Europe’s Atlantic façade boasts resource-rich coastal waters comparable to those of South Africa. From Scandinavia to France, however, any evidence for the Last Interglacial exploitation of marine resources would have been lost to subsequent icecap advances and postglacial submersion of the wide continental platform. Conversely, the very steep shelf off Arrábida, a littoral mountain range 30 km south of Lisbon, Portugal, has enabled extant and submerged shorelines to be preserved short distances apart. Gruta da Figueira Brava, one of Arrábida’s erosion-protected, seaside cave sites, provides a singular opportunity to investigate whether any considerable Last Interglacial accumulations of marine food debris ever existed in Europe.


The Figueira Brava archeological sequence dates to ~86 to 106 thousand years ago (kya). Throughout, there is evidence of a settlement-subsistence system based on regular exploitation of all animal resources offered by the coastal environment: large crabs, marine mollusks, fish, marine birds and mammals, tortoise, waterfowl, and hoofed game. The composition of the food basket and the structure of the deposit vary as a function of the following: (i) sea-level oscillation, with implications for the ecosystems that were preferentially targeted; (ii) frequency of human occupation; (iii) site-formation process; and (iv) position of the archeological trenches relative to the changing configuration of the inhabited space. The initial occupations (phases FB1 and FB2), when the sea was closer to the cave (~750 m), include shell-supported accumulations. These occupations were followed by a period of infrequent use (phase FB3) and a final phase (FB4), when the shoreline was ~2000 m away but shellfish were again discarded at the site in substantial amounts. The density of marine food remains compares well to that seen in the regional Mesolithic and the Last Interglacial of South Africa and the Maghreb and exceeds the latter two in the case of crabs and fish. Figueira Brava also documents a stone pine economy featuring seasonal harvesting and on-site storage of the cones for deferred consumption of the nuts. The stability of this subsistence system suggests successful long-term adaptation.


Figueira Brava provides the first record of significant marine resource consumption among Europe’s Neandertals. Taphonomic and site-preservation biases explain why this kind of record has not been previously found in Europe on the scale seen among coeval African populations. Consistent with rapidly accumulating evidence that Neandertals possessed a fully symbolic material culture, the subsistence evidence reported here further questions the behavioral gap once thought to separate them from modern humans.

Gruta da Figueira Brava, Arrábida, Portugal.

Note the Mediterranean vegetation, like at the time of the Last Interglacial occupation, the MIS 5e marine abrasion terrace, and, under the overhang, the brecciated remnant dated to ~86 to 106 kya. Neandertal use of this cave space, which is currently unroofed due to Holocene erosion, has left an archeological record rich in fish, shellfish, and other coastal resources.



Marine food–reliant subsistence systems such as those in the African Middle Stone Age (MSA) were not thought to exist in Europe until the much later Mesolithic. Whether this apparent lag reflects taphonomic biases or behavioral distinctions between archaic and modern humans remains much debated. Figueira Brava cave, in the Arrábida range (Portugal), provides an exceptionally well preserved record of Neandertal coastal resource exploitation on a comparable scale to the MSA and dated to ~86 to 106 thousand years ago. The breadth of the subsistence base—pine nuts, marine invertebrates, fish, marine birds and mammals, tortoises, waterfowl, and hoofed game—exceeds that of regional early Holocene sites. Fisher-hunter-gatherer economies are not the preserve of anatomically modern people; by the Last Interglacial, they were in place across the Old World in the appropriate settings.

The major innovations of the Middle Stone Age (MSA) of southern Africa are widely seen as reflecting the emergence of cognitive and behavioral modernity. A feedback loop between the consumption of marine foods and the development of the brain might underpin this process (13). Reliant on dense and predictable resources, the “coastal adaptations” so engendered would represent a late Middle to early Upper Pleistocene “broad-spectrum revolution” (4, 5) that triggered demographic growth, social complexification, and, eventually, the out-of-Africa expansion of modern humans (6, 7). The strong prosocial behavior arising out of such adaptations would have provided the competitive edge that has been postulated to explain the disappearance of Neandertals and other anatomically archaic Eurasian humans (8, 9).

It has been proposed that shell middens, defined as shell-supported sediment where shells interfinger with other shells and the matrix fills the voids, are the archeological proxy for such South African coastal adaptations; by contrast—and reflecting a genuine and critical difference in subsistence and behavior—shellfish remains have been found at low density, if they occur at all, in Last Interglacial and later Middle Paleolithic sites of Eurasia (8, 9). An alternative view is that the difference between these two regional records is of degree rather than kind and affected by taphonomic bias (10, 11); therefore, humans might well have foraged for marine foods for much longer and across their entire range, as suggested by skeletal evidence (12).

In the Mediterranean basin, bivalve shell was used for functional purposes (1315), but low productivity may explain why shellfish consumption did not result in the formation of Middle Paleolithic shell middens; such accumulations are rare even in the regional Holocene, when the isotope evidence corroborates the minor role played by marine foods in hunter-gatherer subsistence (16, 17). Under this productivity explanation, one would expect things to be different in the marine resource–rich shores of the North Atlantic, as is indeed suggested by the documented consumption of shellfish at a string of Last Interglacial sites in Morocco (18). However, modern humans are assumed to be behind the formation of the Maghreb’s record and so it is the absence of Middle Paleolithic shell middens in the Atlantic coasts of Europe inhabited by Neandertals that is claimed to be of special significance (8).

Because of transport costs, shellfish consumption is tightly tethered to the point of acquisition (19, 20). Thus, when the sea level was lower, as it was during most of the Pleistocene, one can expect a marine mollusk–rich archeological record to be found only where the adjacent continental platform is very steep and extant and past shorelines are not separated by a large stretch of Holocene-submerged land. For peak interglacial periods, when the sea level was as high or higher than today, archeological site preservation necessitates that the original record formed sufficiently above the shore or in particularly shielded environments offering protection against marine erosion.

In South Africa, sustained tectonic uplift created the Cape Fold Belt’s abrupt coastline, which features caves and rock shelters located well above the present-day tidal range. This geomorphology has favored the preservation of archeological deposits that bear witness to the Last Interglacial exploitation of the seashore below. Such preservation conditions are generally not replicated along Atlantic Europe’s coastlines. In Scandinavia and Britain, Last Interglacial shell middens would have been wiped out by the ice caps of subsequent glacial maxima, whereas sea-level rise would have submerged any deposits formed off of the French coast, where the continental shelf is >200 km wide. Along the Cantabrian coast, the shelf is much narrower and, accordingly, marine food remains are found in Middle and Upper Paleolithic sites; none, however, feature accumulations in the shell-midden scale characteristic of the region’s later Mesolithic (21).

Along the west coast of Portugal, the continental platform is generally wide. However, during low-sea-level periods, a fjord-like landscape formed along the marine canyon off the ~20-km coastline of Arrábida, a mountain range 30 km south of Lisbon. Here, the submerged and extant shorelines are short distances apart, the karst setting provides for erosion-protected accumulation contexts, and coastal upwelling and large tidal amplitudes make for a highly productive littoral where resources were intensively harvested in the Mesolithic, as documented by the nearby shell middens of the Sado estuary (22). Exclusively in Europe, it is along this coastline that sites containing Last Interglacial shell middens stand a good chance of preservation. Gruta da Figueira Brava (38°28′14′′N, 8°59′10′′W; WGS84 datum), a cave first explored in the 1980s and the focus of our 2010 to 2013 archeological excavations, is one such site (23) (see the supplementary materials and methods and figs. S1 to S7).


Site, stratigraphy, and dating

In its current configuration, Figueira Brava features three entrances (Figs. 1 to 3). The Pleistocene fill has been almost entirely eroded away in Entrance 1 and interior Areas A and B but is preserved under flowstone behind the sediment-cum-speleothem blockages separating Entrances 2 and 3 from, respectively, Areas C and F. The exposed terrace in front is the marine abrasion platform of Marine Isotope Stage (MIS) 5e. Originally, this now-unroofed area was part of the cave and remained sediment filled until after the Last Glacial Maximum (LGM). An erosion-scarred, archeologically rich breccia—the external exposure of the sedimentary fill found interiorly in Areas C to F—is found at the back of the terrace, where preservation is explained by heavy cementation and the protection offered by the extant overhang. We estimate that only ~100 m3, or 5% of the original fill, remains.

Fig. 1 Setting.

(A) North–south transect of the Arrábida range through Figueira Brava, indicated by the red star; the plant cover is reconstructed from the site’s paleobotanical data. t, thalweg soil with Quercus deciduous and Vitis; c, calcareous soil with evergreens, e.g., Olea, Quercus, and Pistacia; s, sandy soil with P. pinea and Juniperus. (B and C) Oblique (B) and frontal (C) drone views of the marine abrasion platform of MIS 5e and the original, now-unroofed cave space. Note the cemented remnant between Entrances 2 and 3, which preserves a complete stratigraphic sequence. The triangles mark breccia-capping flowstone and other speleothems sampled for U-series dating.

Fig. 2 Excavation.

(A and B) Area F trench during the 2012 excavation of square T8 (A) and at the end of the last, 2013 field season (B). Stratigraphic depth reached in each square or quadrate is indicated; note the constrained space between capping flowstone and cave roof. (C and D) Entrance 3 during the initial, 2010 field season; (C) shows cutting of the rock-hard breccia for the extraction of a continuous column of blocks for soil micromorphology analysis spanning the sequence and (D) is a close-up view of Cut B, which samples the base of the UC complex, after extraction.

Fig. 3 Site.

(A) Plan with position of trenches (Area C, 1986 to 1989 paleontological work; Areas F and SEx, 2010 to 2013 archeological excavation) and indication of main features. (B) Stratigraphic outline along the T > S axis of the 2010 to 2013 grid. (C) Schematic of the evolution of cave and fill through the different occupation phases; note the changing position of the trenches relative to the center of human occupation (indicated by the asterisks). Elevations are shown in meters above sea level (m asl).

Figure 3 and Table 1 summarize the stratigraphic correlation between the different areas (see also supplementary text S1 and S2, figs. S8 to S21, and tables S1 to S17). The sequence, for which Bayesian modeling of the dating results offers a robust chronology and duration estimates (supplementary text S11 and Fig. 4), can be summarized as follows.

Table 1 Stratigraphy.

Correlation among the different areas of Figueira Brava, speleothem dating constraints, position of the sequence in the global Pleistocene record, and archeological phasing.

Embedded Image
View this table:
Fig. 4 Chronology.

Age boundaries for the different stratigraphic units and human occupation phases calculated by Bayesian modeling of the U-series and OSL results.

Middle Pleistocene: Fine white sands filling depressions in the substrate (unit IB2) accumulated in an endokarst setting before excavation of the cave by the sea, overlain by flowstone (unit IB1).

MIS 5e: Beachrock of the Last Interglacial highstand (complex CO).

MIS 5c: Trampled, anthropized beach sands exposed by sea-level retreat (unit LC3), overlain by colluvium with aeolian inputs (units LC1 to LC2 and complex MC, Entrance 3; complex IL, Area F; layers 3-4, Area C), in turn capped by flowstone (unit MC0, Entrance 3; unit IL1, Area F). Dating constrains this sedimentary package to between 92.0 and 94.0 ka ago (for the upper limit of the range) and 104.0 and 106.0 ka ago (for the lower limit of the range) and suggests a duration of minimally two millennia for the subsequent interval of sedimentation arrest and speleothem growth.

MIS 5b: Colluvium with aeolian inputs that eventually filled up the remaining cave space (complex UC, Entrance 3; complex IH, Area F; layer 2, Area C). Dating and stratigraphic constraints put this package in the 86.0 to 90.0 ka ago interval.

MIS 5a to MIS 2: Flowstone and associated speleothems (unit IH1, Area F; layer 1, Area C), which began to form no later than 76.9 ka ago, underwent continuous growth through MIS 4 to MIS 2 and, before the onset of post-LGM erosion, thoroughly sealed the deposit.

Tardiglacial–Holocene: Post-LGM flowstone and stalagmites (unit IT1, Area F; layer 0, Area C); in places, these speleothems are overlain by a thin, dark lens of fine, very recent organic sediments that also fill voids in their fabric (unit IT2, Area F); unit IT0 of Area F and layer 2a of Area C correspond to sediment reworked by subsurface burrowing and contain Holocene intrusions.

The MIS 5b and 5c sediments lack foraminifera tests, corroborating that the sea remained distant (fig. S22 and table S19), and their paleobotanical content (Fig. 1, fig. S23, and table S20) reveals a local vegetation cover of Mediterranean type (with Pinus pinea, Olea europaea, deciduous and evergreen Quercus spp., Ficus carica, and Vitis vinifera). Thus, local environmental conditions remained broadly like present. The most obvious factor of landscape change was sea-level fluctuation; because of the strict soil requirements of the stone pine and the continuous presence of pinewoods in the site’s catchment, we can nonetheless infer that, throughout, the cave’s limestone terrain remained separated from the seashore by a dune belt (supplementary text S3 and S4).

Formation process and phases of human occupation

In a stratified cave or rock shelter site with a sequence that spans many millennia, human usage of the place may change as a result of environmentally driven factors (with attendant implications for settlement-subsistence systems) as much as local factors; overhang collapse and sedimentary buildup, for instance, may bring about substantial change to habitability and spatial configuration, whereas variation in accumulation dynamics may cause syndepositional or postdepositional displacements and affect stratigraphic integrity. Thus, even when the different stratigraphic units exposed by excavation are in the same place relative to the site’s extant configuration, they may well correspond to loci of behavior and accumulation that represent quite distinct emplacements relative to the time of occupation. To offset the impact of these factors, we adopted a “phase” framework whereby stratigraphic units formed under conditions that remained similar at all scales (local, regional, and environmental) are grouped for meaningful comparison of change through time (Fig. 3 and fig. S18).

Phases FB1 to FB3 fall within MIS 5c. Dating constraints imply a duration of ~12 ka within the very long Greenland Interstadial (GI) 23 warm period; the three sedimentary packages are of comparable thickness and there is no reason to think that they do not represent broadly similar temporal intervals of about four millennia each. Such is also the span indicated by the much tighter chronological control available for the MIS 5b deposit containing the FB4 archeological remains. Although a shorter duration fully within the 2500 years of GI 22 cannot be excluded, the current arctic distribution of bird (the great auk, Pinguinus impennis) and mammal (the ringed seal, Pusa hispida) taxa represented in the IH complex of Area F and in Area C (23) (fig. S33) is consistent with FB4 extending into at least the initial stages of the colder phase that followed, Greenland Stadial (GS) 22. Distances to the shoreline, estimated from global sea-level curves and the local, seismically reconstructed bathymetry (24), could have oscillated between 250 and >2000 m, but on average were ~750, ~1500, and ~2000 m during phases FB1 to FB2, FB3, and FB4, respectively (see supplementary text S2.4 to S2.6, figs. S1 and S19, and table S18).

Phase FB1 is documented by the dense accumulations of mussel shell in the LC complex of Entrance 3 (Fig. 5 and fig. S25). Found along the site’s seaward erosional scarp, these exposures represent depositional contexts that, at the time of occupation, were interior and peripheral to the main activity area, which would have been located farther out, in the then-extant cave porch. The subhorizontal, well-layered disposition of the shell lenses is apparent in both field and soil micromorphology thin section (fig. S13); it reflects a largely in situ context and rules out the possibility that the shells are reworked from natural thanatocenoses formed by hydrodynamic processes. Human agency, otherwise demonstrated by the association with quartz artifacts and animal bone remains, is implied by the site-to-shore distance and the accumulations’ intrinsic features: most shells are broken, the fragments belong to edible-size specimens with valves 7 to 8 cm long or more, small-size specimens and articulated valves are scarce or altogether absent, and many valves lie on their convex side (in a seashore thanatocenosis, wave energy would have inverted most to a more stable position with the concave side facing down).

Fig. 5 Entrance 3.

Exposures and extent of the LC and lower MC complexes and their shell-midden lenses. (A and B) The charcoal-, burnt-bone-, and burnt-shell-rich deposit (units MC3 to MC5) exposed at the bottom of the SEx trench (A) and along the seaward erosional scarp of the brecciated fill (B), from which a representative soil micromorphology thin section (Cut E, sample 1002) was cut. (C) Horizontal exposure of the mussel shell bed in the upper LC complex from which a representative soil micromorphology thin section (Cut G, sample 1022) was cut. (D and E) Vertical exposures of the LC complex from which a representative soil micromorphology thin section (Cut G, sample 1021) was cut. (D) Before cutting. (E) After cutting. (F) Overview of the sedimentary fill preserved in Entrance 3; the exposures illustrated in (A) to (E) are positioned in the photo and against the excavation grid (inset).

Phase FB2 is represented in Entrance 3 by highly anthropized units MC3 to MC5, which overlie either the LC complex (along the seaward erosional scarp) or the MIS 5e beachrock [6 m inward, at the base of the Sondagem exterior (SEx) trench] (Fig. 5 and fig. S11). Because of the intervening recession of the porch, the location of these remnants then coincided with the main, daylight activity area, which is consistent with the abundant charcoal and burnt food debris that underpin the deposit’s dark color. The soil micromorphological thin sections document shell-supported structure in the MC5 unit and show that shell and shell fragments are the dominant component of the groundmass (Fig. 6, figs. S13 and S40, and tables S14 and S15).

Fig. 6 Micromorphology.

Units MC4 and MC5 under the microscope (thin section FB1002; PPL scan). The rectangles highlight areas in which the different microfacies (mF) present are readily apparent: mF1b, interconnected shells with calcitic matrix; mF3, horizontally oriented components; mF2a, heterogeneous coarse sands and shells.

Phase FB3 is represented by units MC1 and MC2 of Entrance 3 and the IL complex of Area F. From the marked decrease in the density of charcoal, mollusk shell, bones, and artifacts, we can infer that the site was infrequently visited during this period (Table 2 and tables S20 to S38). Beginning at this time and continuing into FB4, the abundant remains of which are contained in layer 2 (Area C) and the IH complex (Area F), sedimentary buildup pushed occupation outward of the drip line. However, through syndepositional, low-energy displacement mechanisms (mainly gravity and runoff), habitation debris were transported along the interior slope of the talus, eventually accumulating in the site’s interior areas, where they form the bulk of the deposit. A caprine scapula and wild cat maxillary cemented together in unit IH6 provide a good illustration of the site formation processes in operation: after the soft tissue decayed, the cat’s canine fell from its socket but remained close by, showing how, once its different components were set in place, postdepositional disturbance did not further affect the unstructured comingling of the remains (Fig. 7A).

Table 2 Density.

Shown is the intensity of marine resource exploitation versus the intensity of human occupation at Last Interglacial coastal sites of Iberia and Africa using shellfish and stone tools as proxies (19, 33, 34, 43, 4954).

View this table:
Fig. 7 Finds.

(A) Felis sylvestris left maxilla and its loose canine cemented to a caprine scapula from Area F (unit IH6). (B) Ruditapes decussatus shell fragments from the SEx trench (spit A51 of units MC3 to MC5). (C) Patella vulgata shells from Area F (left top and bottom, units IH4 and IH6, respectively; right, unit IH8). (D) Cracked-open and burnt fragments of Cancer pagurus pincers from Area F (left, units IT0 and IH3; right, unit IH8 both). (E to G) Vertebrae of eel (E), one thermo-altered, and shark [(F) and (G)] from Area F (units IH4 to IH6). (H) Stone tools from Area F: 1, Levallois core from unit IH3; 2 and 3, laminary Levallois flakes from, respectively, units IH4 and IH6. [Photos in (F), (G), and (H): J. P. Ruas.] (I to K) Scanning electron microscope images of Pinus pinea charred remains. Shown are a needle (I) and cone bract (J) from Area F (unit IH6) and a nutshell (K) from the SEx trench (spit A52 of units MC3 to MC5).


Once postglacial sea-level rise and attendant erosional processes unroofed the terrace in front and removed most of the infill deposit, burrowing animals were able to access the site’s interior; therefore, despite the extensive flowstone cap, the unconsolidated portions of the subsurface sediment were affected. This fact hinders the interpretation of Area C’s FB4 finds because its layers 2 (in situ) and 2a (reworked) were excavated as a single unit (23). In Area F, however, we carefully separated the disturbed areas (figs. S7 to S9); their Holocene-intruded components could thus be used as a standard of seaside, naturally accumulated faunal and plant remains against which to assess the Pleistocene material.

Stone tools

Lithic artifacts are abundant throughout (supplementary text S9, Fig. 7H, figs. S34 to S39, Table 2, and tables S38 to S41). The lithics retrieved in the 0.89 m3 of square U8 total 15.3 kg and form a representative sample for phases FB3 and FB4; the total for 0.24 m3 of FB2 excavated in the SEx trench is 1745 g. Quartz is dominant (>90% by specimen counting and >80% by weight in all three phases), features complete reduction sequences, and includes a small number of retouched pieces; most blanks were used without retouch in tasks involving the cutting and scraping of both hard (e.g., wood and bone) and soft (e.g., meat) materials.

Flint and flint-like rocks are rare: there were only 130 pieces in all units of both Entrance 3 and Area F. The inventory includes blanks, exhausted cores, and imported Levallois flakes and blades. Reduction sequences are incomplete and most cores are exhausted; their dorsal, production side features centripetal, unipolar, bipolar, and multipolar scar patterns. Flake morphology is consistent with reduction by recurrent, centripetal schemes. For both flint and quartz, most retouched tools are denticulates or sidescrapers (figs. S36 and S37); some were resharpened on-site, as shown by characteristic debris.

This raw material economy is a direct reflection of availability and remained stable through time, underpinning the absence of appreciable techno-typological change. Quartz could have been procured locally, either from marine deposits or from continental conglomerates and alluvial terraces, where the occasionally knapped cobbles of quartzite, limestone, and lydite can also be found. The flints are allochthonous. Reports that nodules exist in Paleogene formations ~5 km to the northeast could not be confirmed, and the closest sources of the assemblage’s Cenomanian varieties are on the right banks of the Tagus, in the Lisbon area, >30 km to the north (supplementary text S9).

Plant remains

Eighty-seven percent of the identified charcoal is pine. Only P. pinea, the stone pine, could be identified to species (supplementary text S4; Fig. 7, I to K; fig. S23; and table S20).

The wood was burned at hearth temperatures, indicating use as fuel, but most stone pine remains are bracts and nut shells (90, 77, and 71%, respectively, of the pine material in FB2, FB3, and FB4). Thus, the cones were not collected as fire starters because if they had been so used, they would have been thoroughly consumed. In addition, bracts often preserve anatomical shape, which implies roasting rather than burning. These patterns reflect the exploitation of the stone pine as a fruit tree.

The harvest would happen in the autumn or winter of the third year after flowering, when the nuts reach maturity but before the cones open to spread the seeds. The mature cones are always found at the very top of the canopy, which must be climbed for collection. The assemblage’s charred needles corroborate that the cones were taken directly from the tree and stored on-site; here, the nuts were extracted by low-temperature heating and then cracked open to get the kernels, whose absence from the record reflects consumption.


The marine mollusk assemblage includes specimens of Steromphala, Littorina, Bittium, Nucella, and Tritia, as well as valves of Glycymeris, Ostrea, and Pecten (supplementary text S5, figs. S26 and S27, and tables S21 to S24). Coeval, personal ornamentation–related use of these or morphologically similar taxa has been documented (2527), but no traces of anthropogenic modification could be identified in the Figueira Brava specimens.

The numbers involved, however, imply routine collection—for purposes that remain elusive to us—of the beached shells of large bivalves: in FB2, for instance, three complete and two large fragments of Glycymeris were retrieved in the 0.050-m3 excavation of unit MC5. That such shells cannot represent geologically inherited material is corroborated by the fact that none occur in the substantial beachrock remnants locally exposed across the marine abrasion platform.

The smaller gastropods may represent incidental collection. A pierced Littorina obtusata was retrieved in reworked unit IT0 (fig. S28); radiocarbon dating of this specimen yielded a mid-Holocene age (tables S1 and S2), corroborating its suspected intrusive status and the natural origin of the perforation.

The remains of edible shellfish are ubiquitous (supplementary text S5, figs. S24 and S25, and tables S21 to S24). In FB1 and FB2, the accumulations are similar in structure and density to the shell-midden deposits of the regional Holocene (fig. S40). The variation observed in the bulk samples of unconsolidated sediment collected for composition analysis closely tracks that displayed when using excavated specimen counts instead (fig. S25, Table 2, and table S21). For FB2, the values per cubic meter are 370.7 in kilograms, 2504 in number of identified specimens (NISP), and 442 in minimum number of individuals (MNI); for FB4, they are, respectively, 128.2, 2018, and 366. Mollusk shell abundance, relative to matrix and the other components, is therefore comparable in both phases. However, the preserved FB4 deposit does not display shell-midden structure; this is due to issues of site formation, which favored widespread fragmentation, as illustrated by the size class distribution of the piece-plotted and dry-sieved material used in the NISP and MNI calculations, 59% of which is <2 cm (fig. S24).

In the exposures of FB1, mussels are almost exclusive. In the excavated FB2 deposit, mussels are found alongside large numbers of the Ruditapes decussatus clams (Fig. 7B), which continues to occur afterward but in much diminished proportion. Limpets, 74% of which are Patella vulgata, dominate the mollusk assemblages in FB3 and FB4 (Fig. 7C); in decreasing order of abundance, Patella depressa, Patella ulyssiponensis, and Patella rustica are also represented. Substantial amounts of brown crab (Cancer pagurus) and spider crab (Maja squinada) remains appear for the first time at this point in the sequence; the finds include carapace fragments but are mostly composed of often burnt, cracked-open pincers with a breakage pattern that mimics present-day consume-and-discard patterns (Fig. 7D, fig. S24, and tables S25 and S26).

Carapace width, which can be estimated from the length of the pincers to average 162 mm, shows that the large crab catch is entirely made up of sexually mature animals. On average, a 16-cm brown crab weighs 800 g. Thus, the decrease in the density of shellfish remains seen when FB4 is compared with FB2 does not necessarily imply a decrease in the economic importance of marine foods; because an adult brown crab is the clean-meat equivalent of some 30 mussels, a poorer mollusk harvest would have been more than compensated by the new resource. Like the preference for limpets instead of clams, the addition of crabs to FB4’s food basket must reflect sea-level-related differences in the configuration of the shoreline and in the availability of the different species across the closer-by points of procurement.


The rocky coast where crabs and limpets were harvested during FB4 would also have been rich in fish, the bone and tooth remains of which are indeed abundant (supplementary text S6). Among those identified to family, most are eels, congers, and morays, followed by mullets, sharks, and sea breams (Fig. 7, E to G; figs. S29 and S30; and tables S27 to S29). Compared with specimens of known size, the eel bones correspond to fish that were about 30 cm long. In southern Europe, this size range is consistent with individuals in the so-called silver eel stage, which can be caught in estuaries and adjacent marine shores as they pass through on their way back to the sea for reproduction. With regard to sharks, the taxa that could be tentatively identified to species can be caught in shallow water and when trapped in large rock pools left by ebbing tides. This is not infrequently observed in the present, even in the case of the >1-m-long juvenile porbeagle specimens identified in unit IH6 of Area F.

During FB4, the sea was on average ~2 km away. Mammalian predators do not transport fish over such distances. The raptors that would be able to carry prey the size of a juvenile shark are carrion eaters, unrepresented in the faunal assemblage, or not known to nest in caves. Regardless, raptor beak or digestion marks were not observed among the fish remains, and the possibility that they represent stomachal contents can be excluded because there is no evidence that whole carcasses of hunted marine birds were being brought in. The only taxon represented at the site that is known to feed on eels is the cormorant, the remains of which, however, all come from reworked unit IT0, where eels and morays are rare (1.2% of the teleost remains) and may well derive from the Pleistocene deposit (where they represent 65.5% of the teleosts). That said, the fresh appearance of the cormorant material reflects their intrusive nature and recent Holocene age, whereas the representation of all the body parts is consistent with natural deaths in a rocky seashore such as that in existence today at the site; furthermore, no eel remains were found in a study of >400 regurgitations of the Sado estuary’s extant cormorant populations. These patterns rule out that the fish remains are incidental, and human agency is otherwise implied by the dark-brown color of several eel bones (Fig. 7E), which is indicative of low-temperature heating and thus of cooking or roasting.

Small vertebrates, marine mammals, and game

Both resident and migratory species are found among the fauna’s waterfowl component (supplementary text S7, fig. S31, and tables S30 to S34). Among them, auks, gannets, and shags are marine birds that come to shore only to breed in island or rocky cliff colonies. In contrast to the body part representation of the intrusive Holocene cormorants, only the meaty wing bones are represented in the Pleistocene assemblage (mallards and geese included). This pattern suggests human agency, as does the fact that the remains come from FB4, when the site was ~2 km inland. The same applies to the vertebrae of dolphin and the limb remains of ringed seal retrieved in Area C (fig. S33).

Land vertebrates are represented by the skeletal remains of reptiles, birds, and mammals (supplementary text S7 and S8, figs. S31 to S33, and tables S30 to S37). Excluding lagomorphs, 89% of the identified mammals are game taxa (red deer, horse, ibex, and aurochs); the remainder are carnivores, small (cat, fox, and lynx; 4%) and larger (brown bear, hyena, and wolf; 7%). Small-carnivore damage was found among the remains of lagomorphs and terrestrial birds, which may have been accumulated, at least in part, by cat or lynx. Almost all the larger carnivore bones come from the units found either side of the episode of sedimentation arrest and speleothem growth between FB3 and FB4; they reflect use of the cave during a period of human abandonment, whereas the few coprolites found in different units of the IH complex show brief hyena incursions during FB4.

The remains are commingled with abundant stone tools in a deposit rich in charcoals produced by human-lit fires. Extensive carbonate incrustation hinders observation of bone surfaces, but butchery and percussion marks could nonetheless be identified on the herbivore bones. Of these, none bears signs of gnawing or digestion by hyenas or wolves and many are burnt (fig. S32 and tables S34 and S37); likewise, the pattern of tortoise shell burning is consistent with the roasting-in-carapace technique (28).


The pine nut economy seen through the Figueira Brava sequence is well documented in the Upper Paleolithic and Mesolithic of Iberia (29). Supporting evidence for this economy to be in operation during the Middle Paleolithic comes from the LBS and SSL members of Gorham’s Cave (Gibraltar) (30). This site has been assigned to MIS 3 or MIS 4 on the basis of anchoring OSL dating results to radiocarbon chronologies, though the latter must be minimum ages only, meaning those members likely span MIS 5a to MIS 5b, overlapping in age with FB4 (supplementary text S10.3).

The tortoise, marine taxa, and ungulate remains are unquestionably anthropogenic. The hunting of aurochs, horses, and deer is an ordinary component of the Middle Paleolithic behavioral repertoire. The acquisition of small prey and marine foods would in most cases have required no more than the simplest of technologies such as low-tide handpicking in sandy bottoms, exposed rock faces, submerged crevices, and shallow waters, plus the means to bag and transport the harvest. The remains of seals and dolphins may reflect scavenging, and the waterfowl’s small numbers suggest chance acquisition not systematic procurement with netting or similarly elaborated means. These resources reflect the exploitation of all ecosystems present in the site’s catchment among mountain, estuary, and sea: rocky shores, coastal lagoons, alluvial plains, dune pinewoods, and forested slopes.

On the basis of their present behavior, the aquatic and marine birds must have been taken in autumn or winter, when mature pinecones were also harvested. Adult brown and spider crabs migrate to shallow waters in summer, so that must have been their season of collection. Thus, the change from an FB2 clam- to an FB4 limpet-plus-crab-dominated marine invertebrate assemblage possibly reflects a stronger autumn-to-winter (in the former) versus a stronger spring-to-summer (in the latter) signal. The seasonality data imply year-round presence, but given the evidence for occasional carnivore presence, in a recurrent, not continuous, manner.

The evidence from FB3 suggests sporadic use but most resources are documented in all phases. Therefore, there is no reason to think that the differences seen across the sequence imply fluctuation through time in the economic importance of food resources. When the spatial scale of the adaptive system is considered, such differences need not reflect how intensively each resource was harvested or the extent to which the product of the harvest was transported to home base as opposed to being consumed where procured. Rather, they are best explained as relating to the following: (i) where in the changing landscape the archeological record of resource exploitation formed, (ii) the way such a record formed and was (or was not) preserved in the different sites that remained active as home bases throughout, and (iii) the extent to which archeological trenches provide appropriate sampling of intrasite variation in both the vertical and the horizontal dimensions. Even though documented at a single site, the redundancy seen over the many millennia spanned by the Figueira Brava record suggests a stable settlement-subsistence system, not one-off or idiosyncratic behavior.

Based on the Holocene sites of Iberia, distance from the shore is a good predictor of the density of invertebrate shells; the Figueira Brava data fit the kilograms per cubic meter trend line (Fig. 8A), but plot below expectations in rate-of-accumulation terms (MNI per square meter per year) (fig. S42 and tables S42 to S44). This pattern suggests that site use was redundant and/or intensive in Holocene sites located at a comparable distance from the nearest shore but intermittent at Figueira Brava; here, individual occupation events nonetheless resulted in the discard of marine food remains in similar amount and manner, as revealed by the comparable structure identified in the soil micromorphology thin sections (supplementary text S10.1 and fig. S40). Carbon and nitrogen isotope analysis of the humans buried in Portuguese Holocene shell middens shows that aquatic foods may have represented up to 50% of their dietary intake (31, 32). There is no reason to think differently about Figueira Brava’s Middle Paleolithic people, the more so because their exploitation of marine birds, large crabs, and marine mammals reveals an aquatic resource base with a breadth that exceeds that of the regional Mesolithic (22).

Fig. 8 Significance.

Figueira Brava compared with Mesolithic and Last Interglacial sites in Iberia, the Maghreb, and South Africa (data from Table 2 and table S42). Note the logarithmic scale in the y-axis of (C) and the x-axis of (E). Excluding Blombos, the set of sites in (D) fits an exponential trend with R2 = 0.4635. In (E), BBC-M2 is a clear outlier and was excluded from the calculation of linear regression and coefficient of determination. In weight of shell per excavated volume versus distance to shore, Figueira Brava fits the Iberian Holocene trend (A) but falls above expectations derived from the African sites (B). If specimen counts are used instead, Figueira Brava fits the African sites’ trend, from which Blombos and Klasies River stand out due to intensive residential use (D and E), but falls well above it for fish (C). FB, Figueira Brava; HDP1, Hoedjiespunt 1; YFT1, Ysterfontein 1; BBC, Blombos; PP13B, Pinnacle Point 13B; KR, Klasies River; VC, Vanguard Cave; AV, Aviones; BJ, Bajondillo; EC, El Cuco; CBD, Contrebandiers.

When comparing against Last Interglacial marine exploitation proxies from sites in Iberia, the Maghreb, and South Africa, Figueira Brava plots above expectations (supplementary text S10.2, Fig. 8B, Table 2, and tables S44 and S45). To avoid potential biases introduced by the different approaches to the calculation of the kilograms per cubic meter index, specimen counts can be used instead; doing so changes nothing with regard to fish (Fig. 8C), but for mollusks reveals a noisier exponential trend where Blombos is the outlier (Fig. 8D). The Blombos data, however, are for the richest grid units only; they are unrepresentative of the abundance of remains across the extent of the occupied surface and, for the M2 phase, likely inflated by the differential identifiability of the Perna perna mussel. Likewise, interpretation of the high values for Bajondillo must take into consideration that they are based on Mytilus fragments counted down to the 0.5-mm fraction, whereas 2- to 10-mm mesh size ranges were used at the other Last Interglacial sites shown in the comparison plot. These examples highlight why, in addition to distance to shore, the observed variation also depends on preservation, sampling, and counting biases.

Within South Africa, Klasies River’s MNI per cubic meter values stand out as much higher than at other sites similarly located directly on the beach (e.g., Hoedjiespunt), suggesting that site function also contributes significantly to the variation in the amount of shellfish refuse (33). Indeed, the density of shellfish correlates well with that of lithic artifacts (Fig. 8E and Table 2), and the latter can be taken as a good proxy for the intensity of human occupation. The highest values for both parameters are reached at Blombos, and at Klasies River if we take the SMONE unit (34) as being representative of the MSA II phase. These two sites may have been places of long-term, intense residential activity—in agreement with their abundant fire features, human remains, and number of symbolism-related artifacts (shell beads, pigment containers, and engraved and painted items) (3539)—whereas the others may have been transiently used or only infrequently reoccupied.

Available geological descriptions suggest that some of the Klasies River levels (e.g., the SM5 midden at the base of phase MSA II) are shell supported (40); however, Pinnacle Point 5-6 is the only South African site where soil micromorphological analysis demonstrates that is the case for some levels (41) (fig. S41). On the basis of a scatterplot of piece-plotted shells, shell-midden status has been claimed for the MIS 5c accumulations in the eastern area of Pinnacle Point 13B (units SBS, URF, and LRF); taken a step further, the argument would support a difference in kind with the Middle Paleolithic of Europe because of the much lower density of the distribution of shells across an area of the same size in Vanguard Cave, Gibraltar (8). However, Pinnacle Point 13B is estimated to span a minimum of 8000 years (42), and the marine mollusk MNI represented in the scatterplot is 549 (Table 2), whereas Vanguard’s MNI is 65 and comes from a very thin deposit of high integrity corresponding to a single occupation (43, 44). Thus, the Pinnacle Point 13B record could have been produced by the equivalent of one such Vanguard event occurring at the site only once every thousand years.

When all factors that condition the density and structure of archeological deposits rich in marine food remains are duly considered, the variation seen in the Last Interglacial sites of Africa and Iberia is in the same range. Figueira Brava, with its abundance of fish, crab, and mollusk remains and presence of lenses of shell-supported matrix, is the best Iberian example so far but it is not alone. The cave sites of Aviones, Bajondillo, and Vanguard are examples that meet expectations when their Mediterranean setting is accounted for. In the Atlantic façade, the Cantabrian site of El Cuco is almost certainly another example; the shellfish-rich lower layers of the sequence (Fig. 8D, fig. S42, and tables S42 and S44) yielded finite radiocarbon ages on limpet shell in excess of 42.5 ka (45, 46), but comparable results were obtained at Aviones and Figueira Brava by the same laboratory using the same methods and the same kinds of samples—ones that were eventually shown to be of Last Interglacial age instead (supplementary text S2.1 and S10.2).


In littoral areas of Last Interglacial Iberia, Neandertals foraged much like early Holocene humans. Subsistence-wise, Neandertals were therefore geographically as diverse as might be expected, from top-level carnivores in their periglacial range (47) to fisher-hunter-gatherers in the right settings of temperate environments.

The routine harvesting of shellfish implies knowledge of tidal regimes and, along the Portuguese littoral, awareness that between late spring and autumn, the consumption of bivalves entails a significant risk of biotoxin poisoning. These cognitive aspects of the Figueira Brava subsistence data are consistent with the rapidly accumulating evidence for jewelry, cave art, and other forms of symbolic material culture in the Middle Paleolithic of Europe (26, 27, 48). The major behavioral gap once thought to separate Neandertals from modern humans would thus seem to be just another example that “absence of evidence is not evidence of absence.”

A corollary of the Iberian data is that the consumption of aquatic foods is not the differentia specifica separating anatomically modern humans in Africa from coeval Eurasians, and ultimately explaining the demise of the latter. Indeed, the possibility must now be entertained that the familiarity with marine resources and seascapes implied by the settlement of Southeast Asia, Sahul (Australia and New Guinea), and the Americas is deeply rooted in the history of our genus.

Materials and Methods

Area C, directly accessible through Entrance 1, was the target of the 1986 to 1989 paleontological investigations. Our 2010 to 2013 work concerned Entrance 3 and Area F, which can be accessed from Area C to Areas D and E through narrows of difficult speleological negotiation. In Area F, the upper part of the sedimentary fill was largely unconsolidated and amenable to normal, trowel-aided excavation but the lower part was heavily cemented and had to be excavated with power tools, as was also the case in the SEx trench opened in Entrance 3. Here, investigation of the deposit was complemented by soil micromorphological analysis of representative samples spanning the complete stratigraphic sequence. Finds were manually piece plotted against site grid and site datum. The sediment was dry sieved on-site and the residue entirely saved for subsequent wet sieving and floatation. For the sediments containing the remains of phases FB2 and FB3, the sorting of fish bones from the sieved or floated sediment has yet to be carried out; therefore, fish counts are provided for FB4 only. Following established practice in the archeology of Portuguese Mesolithic shell middens, our weight/volume shell density parameter derives from bulk samples of unconsolidated sediment. Radiocarbon dating failed because the age of the samples was beyond the method’s limit of applicability. The deposit’s Last Interglacial age is demonstrated by the uranium (U)-series and single-grain optically stimulated luminescence (OSL) results for stratigraphically associated speleothems and the sediments themselves. Additional details and an extensive description of the dating work are provided in the supplementary materials and methods. The analytical protocols used in the study of animal and plant remains and of stone tools followed standard practice and are also further explained in the supplementary materials.

Supplementary Materials

Materials and Methods

Supplementary Text S1 to S10

Figs. S1 to S42

Tables S1 to S45

References (55217)

References and notes

Acknowledgments: K. Wainer, B. Brumme, and L. Klausnitzer helped with speleothem sample preparation; P. Chistè and C. Mologni produced the high-resolution scans of soil thin sections; and J. P. Ruas helped with the photographic documentation of finds. Author contributions: J.Z. directed the research projects and the fieldwork, coordinated the postexcavation study of the finds, and wrote the paper with contributions from the other authors. D.E.A., L.J.A., E.B., M.Dem., S G., D.L.H., H.M., and M.N. conducted fieldwork and contributed analytical results. M.A.I., P.C., J.L.C., F.d.E., M.Des., C.D., P.L., A.M.M.S., P.P., and A.Q. contributed analytical results. J.D., F.R., and P.S. conducted fieldwork. Funding: Excavation and analysis of the finds were supported by research grants from Fundação para a Ciência e Tecnologia (FCT, Portugal) projects “Middle Paleolithic Archaeology of the Almonda Karstic System” (grant PTDC/HIS-ARQ/098164/2008) and “Archeology and Evolution of Early Humans in the Western Façade of Iberia” (grant PTDC/HAR-ARQ/30413/2017). U-series dating was additionally supported by the Max Planck Institute for Evolutionary Anthropology (Leipzig, Germany) and, for the Centro Nacional de Investigación de la Evolución Humana (CENIEH, Burgos, Spain) results, by Ministerio de Ciencia, Innovación y Universidades (MICINN, Spain) research grant CGL2011-27187. The study of the nonfood shells was partly supported by the Programme Investissements d’Avenir IdEx université de Bordeaux, the French Agence Nationale de la Recherche (ANR-10-LABX-52, LaScArBx Cluster of Excellence), and the Research Council of Norway (project 262618). OSL dating analyses and manuscript production were partly supported by Australian Research Council grants DE160100743 and FT130100195. The OxA radiocarbon dates numbered 19978 to 19982 were funded by the ORADS program (application 2006/1/4). The analysis of all faunal remains (except fish) was supported by a doctoral award by the London Arts and Humanities Partnership, an AHRC-funded Doctoral Training Programme, to the research project “Neanderthal subsistence in Portugal: small and large prey consumption during the Marine Isotope Stage 5 (MIS-5).” Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the manuscript or the supplementary material. The Figueira Brava finds are in storage at the Academy of Sciences (1986 to 1989 excavations) and UNIARQ (2010 to 2013 excavations) in Lisbon.

Stay Connected to Science

Navigate This Article