Introduction

Many prehistoric coastal populations have exploited the faunal resources of adjacent aquatic environments1, 2. The remains of molluscs (shells) and fish (otoliths) were often deposited in the vicinity of prehistoric settlements forming mounds known as shell middens. The composition, internal structure or the spatial and temporal distribution of these shell middens represent a valuable source of information about human dispersal, site-specific occupation pattern, subsistence strategies, associated dietary preferences or fishing and foraging seasonality3, 4. Moreover, mollusc shells or fish otoliths derived from shell middens can provide proxy records such as oxygen isotopes which enable the reconstruction of highest-resolution (sub-seasonal) palaeoenvironmental conditions of ancient coastal environments5,6,7. Despite allowing the reconstruction of foraging seasonality3, 4, past upwelling intensities8,9,10 or seasonal freshwater discharge patterns11,12,13, ontogenetic geochemical signatures of shell midden constituents have become an important tool for the reconstruction of ancient water temperature seasonality within coastal areas14,15,16,17. In particular shell midden deposits which cover time spans of up to several millennia allow the long-term reconstruction of environmental change with a sub-seasonal resolution18.

As subsistence was the primary reason for exploiting aquatic faunal resources for many prehistoric cultures, most shell midden constituents were most-likely subjected to certain processing or cooking techniques prior to deposition4, In particular molluscs, where the comparably small amount of edible flesh is protected by tightly closed valves or an operculum require sophisticated processing strategies in order to efficiently extract the small edible portion4, 6. However, evidence for certain processing or cooking techniques such as pottery and tool fragments or blackened/burned stones are often sparse in shell midden deposits or allow a variety of potential interpretations regarding specific cooking or processing strategies. This is particularly true at transitory preparation sites near fishing or shellfish gathering grounds rather than close to prehistoric settlements.

While usually unknown processing methods limit our understanding of shell midden formation, processing or cooking methods can alter the geochemical proxy records of calcareous skeletal remains, potentially introducing large uncertainties into palaeoclimate reconstructions19,20,21. Hence, the reliable reconstruction of prehistoric processing methods is crucial in order to avoid erroneous palaeoclimate reconstructions based on geochemical proxies in shell midden constituents.

In this study, we tested the potential application of carbonate clumped-isotope thermometry on aragonitic mollusc shells and fish otoliths to reconstruct specific prehistoric cooking methods. Carbonate clumped-isotope thermometry is based on the preferential formation of 13C-18O bonds (“clumping”, expressed as Δ47) in the carbonate molecule with decreasing carbonate formation temperature, independent of the isotopic composition of the precipitating fluid22,23,24 and therefore circumvents uncertainties in conventional oxygen-isotope thermometry which commonly derive from unknown ancient water chemistry23,24,25. In addition, clumped-isotope thermometry allows reconstructing the thermal histories of carbonate rocks as the relative abundance of 13C-18O bonds can be re-equilibrated during post-formational heating26, 27. The rate and magnitude of 13C-18O bond re-ordering is a function of temperature and exposure time and can occur even in closed systems without any isotope exchange with ambient media26,27,28. While re-equilibration in calcite occurs over geological time scales, aragonite shows much higher re-ordering rates resetting the clumped-isotopic signature within minutes to hours29. Thus, clumped-isotope thermometry might represent an ideal tool to detect pre-depositional heating of aragonitic shell midden constituents and may allow certain processing or cooking practices to be differentiated.

In order to test this hypothesis, we exposed the right valves of modern hard clams (Mercenaria campechiensis, Gmelin 1791) cultured in the Gulf of Mexico to a variety of prehistoric cooking methods. Our experiment comprised untreated shells (corresponding to manually opened shells and serving as controls), shells boiled in seawater (~100 °C), roasted over charcoal (174 ± 13 °C) and directly burned in charcoal (559 ± 95 °C). For all treatments, exposure times ranged from 15 minutes to 6 hours. Afterwards, we measured the bulk clumped-isotopic composition (Δ47), oxygen and carbon isotope values (δ18O and δ13C, respectively) as well as the potential conversion of the primary aragonite into secondary calcite.

In addition, we measured bulk δ18O and Δ47 values of mid-Holocene shell midden constituents excavated in the northern part of the Mauritanian coast (Banc d’Arguin), radiocarbon dated to 5,020–5,320 cal. yrs. BP. These samples comprised four catfish otoliths (Carlarius spp.) and seven bivalve shells (Senilia senilis, Linnaeus 1758). We also analysed additional samples drilled from the hinge region of each bivalve shell, as well as high-resolution ontogenetic oxygen-isotope records of three exemplary shells.

Results

Experimental shells

Depending upon the cooking temperature, variable changes in shell mineralogy, visual appearance and outer shell structure occurred within the different cooking treatments (Fig. 1). Boiling and roasting of the shells caused bleaching and blackening of the outer organic layer, respectively. While the cooked shells consistently showed a pristine outer shell structure, some roasted shells exhibited small cracks (<0.5 mm) perpendicular to the ventral margin. Bleaching and blackening of the cooked and roasted shells’ periostracum entirely disappeared after the removal of the periostracum using hydrogen peroxide. Despite the sporadically occurring cracks, overall shell integrity remained pristine, hindering a clear differentiation from uncooked control shells. Direct burning of the shells resulted in blackening of the outer surface, whitening of the inner shell layer, and was accompanied by an immediate loss of structural integrity and the conversion of the solid shell structure into brittle and chalky fragments. Boiling and roasting did not cause any conversion of initial aragonite into secondary calcite (Fig. 2a), while the burned shells showed a generally increasing conversion of primary aragonite to secondary calcite over the time of exposure.

Figure 1
figure 1

Visual comparison of untreated left valves and the corresponding right valves exposed to the different treatments prior to hydrogen peroxide cleaning. (a) Boiling for 40 minutes. (b) Roasting for 240 minutes. (c) Burning for 240 minutes. Scale bars represent 1.0 cm.

Full size image
Figure 2
figure 2

Alteration of mineralogy and geochemical proxies in modern Mercenaria campechiensis shells by simulated prehistoric cooking methods: Untreated (grey triangles and grey striped areas), boiled (blue triangles), roasted (green circles), and burned shells (orange diamonds). (a) Conversion of primary aragonite into secondary calcite with the predicted aragonite-calcite conversion based on the Arrhenius model for biogenic aragonite of Staudigel & Swart29 (solid lines). (b) Clumped-isotopic composition (Δ47) reported in the absolute reference frame (ARF) defined by Dennis et al.24 of the bivalve shells after exposure to the different cooking methods. Solid lines represent predicted Δ47 values using the Arrhenius model of Staudigel & Swart29. Error bars represent mean standard errors of the individual measurements. The grey striped area represents the range of untreated control shells (n = 5). Shaded areas represent the characteristic Δ47 windows for each cooking treatment over the entire experiment duration (light) and realistic cooking durations of <30 minutes (dark). (c) Calculated Δ47-based water temperature estimated using the equation of Dennis et al.24 and corresponding theoretical Δ47-based water temperature estimates predicted by the thermodynamic model of Staudigel & Swart29. Similar to panel (b), light and dark blue/yellow areas represent the characteristic Δ47-windows translated into water temperature. (d) Change in δ18OShell over time of exposure for the different cooking methods. Solid lines represent exponentially fitted curves highlighting the overall trend towards lower δ18OShell values. (e) Calculated δ18OShell-based water temperatures using the equation of Grossman & Ku30 and assuming a constant δ18OSeawater of 0.0‰ VSMOW.

Full size image

The exposure of bivalve shells to different cooking practices caused an alteration of their clumped-isotopic composition (Δ47). Re-ordering rates in aragonite were high enough to cause an immediate reduction in Δ47 where the magnitude is generally a function of cooking temperature and exposure time. The lowest amount of change occurred in the boiling treatment (~100 °C) where the reduction in the Δ47 values was not significantly different from the control shells. This agrees with the thermodynamic model for clumped-isotope re-ordering in aragonite at 100 °C29 which suggests that an exposure time of several hours to temperatures ~100 °C is not sufficient to cause a significant decrease in Δ47 values. In contrast, roasting and burning caused an immediate decrease in Δ47 to values ranging from 0.510 to 0.609‰, and 0.342 to 0.425‰, respectively. As shown in Fig. 2b and c, experimental lowering of shell Δ47 values within the boiling and roasting treatments are in good agreement with thermodynamic models for the relevant temperatures over the time of exposure29.

Independent of cooking temperature or exposure time, all cooking treatments resulted in a statistically significant lowering of the bulk δ18OShell values, reducing their potential for reliable water temperature reconstructions (Fig. 2d and e). The maximum decrease of δ18OShell values was −0.61‰ for boiling, −0.53‰ for roasting, and −0.90‰ for the burning treatment. Using the equation of Grossman & Ku30, the decrease in δ18OCarbonate values translates into water temperature overestimations of 2.9, 2.5 and 4.2 °C, respectively. While δ18OCarbonate values of all treatments were significantly different from the control shells, there was no significant difference between the different cooking treatments. We observed no statistically significant alteration trend with respect to shell δ13CCarbonate values among the different treatments. However, the variability in δ13CCarbonate values of the treated shells exceeds the range measured in the control shells, indicating that isotope exchange reactions also affected shell δ13CCarbonate values.

Archaeological samples

All mid-Holocene otoliths are mainly composed of aragonite (96–100%) which is in agreement with the initial mineralogy of modern Carlarius spp. otoliths (~98% aragonite, 2% calcite for Carlarius heudelotii). Likewise, all mid-Holocene S. senilis shells were composed of 100% aragonite which is the initial mineralogy also found in modern shells31. From the mineralogical perspective, the archaeological samples can therefore be considered pristine. All mineralogical data are provided in Supplementary Tab. 4.

Mid-Holocene otolith Δ47-values (0.667 to 0.716‰) largely agree with predicted values based on modern water temperature observations31, 32. In contrast, mid-Holocene bivalve shells show highly variable Δ47 values ranging from 0.601 to 0.716‰ which is considerably lower than expected for natural environments. Noteworthy, bulk samples of the shells show higher variations in Δ47 and tend towards lower Δ47 values (0.601 to 0.716‰) than corresponding samples from the hinge region of the same individuals (0.651 to 0.697‰). Subsequently, Δ47-based water temperature reconstructions using the bulk shell samples result in unreasonably high palaeo-water temperature estimates (23.9 to 51.0 °C). In contrast, the samples drilled from the hinge region of the same shells provide more realistic water temperature estimates for coastal intertidal environments (28.0 to 38.3 °C).

Mid-Holocene bulk δ18OCarbonate values are consistent amongst all otoliths and bivalve shells, ranging from −0.54 to +0.88‰ VPDB (Fig. 3). Assuming mid-Holocene δ18OSeawater values of +0.68 and +1.57‰ VSMOW for the otoliths32 and bivalve shells31, respectively, reconstructed water temperatures are slightly higher during the mid-Holocene than observed today. Measured ontogenetic bivalve δ18O records show a clear sinusoidal cyclicity that covers three to six years of growth. Measured ontogenetic δ18OCarbonate values range from −1.8 to +1.5‰ VPDB, translating into mid-Holocene water temperatures ranging from 21.1 to 36.2 °C.

Figure 3
figure 3

Clumped and oxygen isotopic composition with corresponding water temperature estimates of mid-Holocene fish otoliths and bivalve shells. (a) Clumped-isotopic composition (Δ47) reported in the absolute reference frame (ARF) defined by Dennis et al.24 of bulk otoliths (grey triangles), bulk bivalve shells (grey circles) and hinge samples from the bivalve shells (open circles). Error bars represent mean standard errors of the individual measurements. The grey area represents the theoretical Δ47-range based on modern local water temperature data31 translated into Δ47-values using the equation of Dennis et al.24. The shaded green area represent the experimentally determined Δ47-window for roasting at 174 ± 13 °C for 6 hours (light green) and realistic cooking duration of >30 minutes (dark green). (b) Δ47-based water temperature reconstruction using the equation of Dennis et al.24 with modern water temperature data31. Shaded green areas represent the experimentally determined Δ47-window of the roasting treatment translated into water temperature for 6 hours (light green) and realistic cooking duration of >30 minutes (dark green). (c) Oxygen isotopic composition of bulk fish otoliths, bulk bivalve shells and hinge samples from the bivalve shells. (d) Oxygen-isotope-based water temperature reconstructions for the otoliths using the equation of Thorrold et al.33 assuming a constant δ18OSeawater value of +0.68‰ VSMOW32 and oxygen-isotope based water temperature reconstructions for the bivalve shell samples using the equation of Grossman & Ku30 assuming a constant modern δ18OSeawater value of +1.57‰ VSMOW31 with the modern local water temperature range31.

Full size image

Discussion

Experimental data

Previous studies on the effect of prehistoric cooking on mollusc shells simulated cooking temperatures ranging from 100 to 500 °C20, 21. Our mineralogical data agrees with these studies as they also showed a conversion of primary aragonite into secondary calcite at temperatures >300 °C. This is consistent with other experimental studies that show temperatures between 300–400 °C as the minimum temperature for the conversion of biogenic aragonite into calcite29, 34, 35. However, the burned shells show a heterogeneous and considerably lower conversion of the primary aragonite into secondary calcite than predicted by the Arrhenius model of Staudigel & Swart29 for 559 °C where aragonite should entirely convert into calcite within less than a second. The lowest burning temperatures measured within our experiment (≥386 °C, see Supplemental Fig. 1) falls exactly in the window where only a partial conversion of aragonite into calcite would be expected for the experimental exposure times (see Fig. 2a)29. We therefore presume that the heterogeneous conversion among the burned shells can be related to the spatial and temporal temperature variability experienced by the individual shells throughout the experiment (see Fig. 2a and Supplemental Fig. 1).

Based on their experimental results, Milano et al.20 proposed the use of crystal microstructure for the detection of pre-depositional heating in shell midden deposits which has also been used in other previous studies on diagenetic alteration of shell midden constituents36. However, Milano et al.20 only observed changes in crystal microstructure associated with conversion of primary aragonite into secondary calcite. Since we did not find any conversion within the boiling and roasting treatments, our data indicate that mineralogical37 or shell microstructure analyses20, 36 may fail to detect pre-depositional heating if cooking temperatures are below the aragonite-calcite conversion window of 300–400 °C. Although burning causes a clear conversion of the primary aragonite into secondary calcite and would therefore be easily identifiable in shell midden deposits, the preservation potential of small burned shells in shell midden deposits is questionable considering their brittle shell structure. However, as the breakage of the burned shells occurred predominantly along the ventral margin, the thicker hinge region might still be preserved in shell midden deposits, in particular for larger shell specimens.

The observed minor overshooting of theoretical Δ47 values by the roasting treatment was most likely caused by short-term temperature spikes (>200 °C) which were also measured during the experiment (see Supplemental Fig. 1). In contrast, the observed undershooting of theoretical equilibration temperatures by the burning treatment could have been caused by several different mechanisms including: (1) disorder-to-order re-equilibration during cooling to ambient air temperature after the experiment, (2) complex interplays of re-ordering processes as a consequence of aragonite/calcite transitions during the experiments or (3) spatial and temporal temperature variability of the burning treatment including potential temperature gradients between the charcoal surface and the shells. It appears unlikely that cooling to ambient air temperature caused the observed underestimation of temperature as disorder-to-order re-equilibration occurs over considerably longer time scales28. Instead, we suspect that the conversion of aragonite into calcite causes the deviation from the thermodynamic model which has also been shown by the experiments of Stolper & Eiler35 and Piasecki38, involving at least three different mechanisms of clumped-isotope re-ordering during aragonite-calcite conversion. These mechanisms are (1) a sharp decrease in Δ47 during the initial heating of the primary aragonite (annealing) followed by a (2) partial conversion of aragonite into calcite causing a re-increase in Δ47 and (3) a second re-ordering phase of the secondary calcite resulting in a decrease in Δ47 values with considerably lower re-equilibration rates. This complex model of clumped-isotope re-ordering during the transition of aragonite into calcite is supported by our data because the measured trajectory of Δ47 alteration within the burning treatment exactly follows the pattern described by Piasecki38. Assuming a lower treatment temperature of only 386 °C increases the goodness of fit between the measured and predicted Δ47 values of the burning treatment (386 versus 559 °C models shown in Fig. 2b and c). Thus, lower burning temperatures of some shells might explain the overall lower alteration magnitude found in this study compared to the Arrhenius model of Staudigel & Swart29 for 559 °C (Fig. 2b and c). However, our experiment was designed to simulate (spatially and temporarily heterogeneous) prehistoric cooking practices and has thus only limited validity regarding the underlying thermodynamic mechanisms of clumped-isotope re-ordering during the aragonite/calcite transition.

Regardless of experimental uncertainties, the lower Δ47 values of the shells in the roasting and burning treatments are significantly different from the control as well as from each other. In particular, if only shells heated for less than 30 minutes are considered, the decreases in Δ47 values result in a confined Δ47/temperature window for each cooking treatment. Thus, our experimental data strongly support the aforementioned hypothesis that clumped-isotope thermometry applied on aragonitic skeletal structures represents a suitable diagnostic tool for detecting and differentiating between certain prehistoric cooking techniques in shell midden deposits.

The measured alterations of δ18OCarbonate values generally agree with previous studies in terms of magnitude, however, they differ in the minimum temperature required for a measurable alteration of oxygen isotopes in aragonitic skeletal structures by prehistoric cooking practices. Andrus & Crowe19 reported no alteration of δ18OOtolith values at temperatures up to ~200 °C for fish otoliths, and only detected changes in isotopic signatures of otoliths burned directly in hot (~800 °C) coals19. In contrast, Milano et al.20 showed only a clear decrease in the δ18OShell values of mollusc shells at temperatures of >300 °C. We suggest that the lower minimum temperature for alteration of oxygen isotopes in aragonitic skeletal structures found in this study as well as by Milano et al.20 compared to that of Andrus & Crowe19 is related to the direct exposure of the aragonitic mollusc shells to the cooking medium. Andrus & Crowe19 exposed the entire fish to the different cooking treatments with otoliths protected from direct heat inside the vestibular system. This hypothesis is supported by the general agreement between our data and those data of Milano et al.20 who also observed a similar decrease in the δ18O values of mollusc shells at temperatures between 300 and 700 °C. The experiment of Milano et al.20, however, did not include relevant roasting temperatures between 100 and 300 °C which most likely represent the temperature window where a decrease in the δ18OCarbonate values occurs without a conversion of primary aragonite into secondary calcite.

The different magnitudes of δ18OShell alteration between this study and Milano et al.20 at lower temperatures (<300 °C) indicates that experimental design plays a key-role in assessing the impact of prehistoric cooking practices on palaeoclimate proxy signatures of shell midden constituents. While controlled oven experiments are a suitable approach for studying the underlying processes of aragonite-calcite conversion and identifying isotope-exchange mechanisms, the realistic simulation of prehistoric cooking practices using coal/wood fire provides more reliable estimates regarding the absolute alteration of palaeoenvironmental proxies in shell midden constituents. The stronger alteration at lower temperatures found in this study is probably related to the presence of water in the boiling treatment, high-frequency temperature variability in the roasting and burning experiment, and higher CO2 and CO concentrations and fluxes compared to the closed oven setups of Milano et al.20. Such factors are likely to facilitate isotope-exchange reactions between the carbonate mineral and the ambient cooking medium and can thus explain the more extensive alteration found in this study with respect to δ18OCarbonate values at lower cooking temperatures. Together with temperature fluctuations of realistic roasting and burning experiments, such processes might also explain the observed non-linearity of δ18O alteration within the different treatments. However, experimentally determined alteration during prehistoric cooking will most likely also vary depending on the flame temperatures or different wood types used in the experiments.

Archaeological data

Although absolute mid-Holocene clumped-isotope-based water temperature reconstructions might be imprecise as a result of small but unknown deviation from theoretical equilibrium precipitation (i.e. “vital effects”), measured Δ47 values of the mid-Holocene bivalve shells are considerably lower than expected for natural environments. While our experimental data suggests that these shells were heated prior to deposition, their pristine aragonitic mineralogy suggests that they did not experience temperatures higher than 300–400 °C. The overlap of the measured Δ47 values of the mid-Holocene shell with our experimental data obtained from the roasting treatment (see Fig. 3) indicates cooking temperatures in the range of roasting or fumigation. The difference in Δ47 values between the bulk and the hinge samples of the same bivalve shells suggests that the outer shell surface of most shells experienced considerably higher temperatures than the corresponding hinge region. Therefore, we presume that the articulated shells were placed on heated surfaces to cause heat-related muscle relaxation which predominantly affected the outer shell surface and which represents a well-known processing technique for bivalve shells reported in many ethnographic observations4. This technique would imply that not all excavated valves should show heat-related Δ47 alterations (i.e. only the lower valve), which is also supported by our data (Fig. 3).

An alternative explanation for the post-mortem alteration of clumped-isotopic signatures in the mid-Holocene bivalve shells might be the long-term sub-aerial exposure with increased soil temperatures as a consequence of high solar insolation in this sub-Saharan region. However, assuming a potential re-equilibration due to surface exposure throughout the Holocene, one would also expect different bivalve shells to exhibit the same degree of alteration, similar Δ47 values within individual shells (hinge versus bulk samples), as well as similar Δ47 values for the otoliths and bivalves excavated from the same shell midden. Thus, it appears unlikely to us that the exposure to increased sediment surface temperatures caused the observed trend in the clumped-isotopic signature of the mid-Holocene samples.

Although the clumped-isotopic signatures of the bivalve shells strongly suggests pre-depositional heating, the bulk oxygen isotopic composition provides slightly higher, but still reasonable, water temperature estimates for coastal regions of NW Africa during the mid-Holocene, potentially indicating only a minor degree of pre-depositional alteration. Similarly, the high-resolution ontogenetic δ18O-based water temperature records show a clear sinusoidal oscillation but with on average ~5 °C higher water temperatures during the mid-Holocene than observed today in this region (Fig. 4)31. Whether these δ18OCarbonate values accurately identify warmer water temperatures during the mid-Holocene or if they overestimate ancient water temperatures due to pre-depositional alteration remains speculative. However, since the sinusoidal shape of the ontogenetic water temperature records seem to be preserved within all mid-Holocene shells and their water temperature amplitudes (12.4, 12.6 and 13.4 °C) agree with the modern values of this region (13.6 °C)31, we presume that even if these shells were altered by pre-depositional heating, their internal δ18O-range might have been preserved and can still provide information about water temperature seasonality and, for example, the season of capture. However, future studies on the influence of prehistoric cooking on ontogenetic oxygen isotope records in mollusc shells are necessary to further examine the potential alteration of ontogenetic oxygen-isotope records.

Figure 4
figure 4

Ontogenetic oxygen-isotope records of three archaeological mid-Holocene Senilia senilis shells and comparison with modern coastal sea surface temperature variability of the eastern Banc d’Arguin31. Water temperature reconstruction was done using the equation of Grossman & Ku30 assuming a δ18OSeawater value of +1.57‰ VSMOW31. Error bars for the measured δ18OCarbonate values and related water temperature estimates are smaller than the symbol size.

Full size image

Implications for palaeoclimate reconstructions

This study strongly suggests that pre-depositional heating of aragonitic shell midden constituents can cause considerable, but most likely unnoticeable, errors in palaeoclimate reconstructions. Depending on the cooking temperature, the alteration of oxygen isotopes in aragonitic shells is rather small (<0.9‰ ≈ <4.2 °C) and occurs even without a clear conversion of primary aragonite into secondary calcite, particularly at cooking temperatures <300 °C. This implies that such pre-depositional alterations are hardly differentiable from natural ancient water temperature variations and that conventional mineralogical analyses might fail to detect such pre-depositional heating. However, our data clearly show that clumped-isotope thermometry enables the reconstruction of the thermal history of aragonitic skeletal structures and therefore helps avoiding erroneous paleoclimate reconstructions introduced by their exposure to considerable heat prior to their deposition.

While additional experiments are needed to fully understand the underlying thermodynamic mechanisms controlling isotope-exchange mechanisms, our data show that a careful assessment of sample preservation including clumped-isotope analysis is necessary in future studies to avoid erroneous paleoclimate reconstructions using aragonitic shell midden constituents. Considering the higher stability of calcite relative to aragonite, it is likely that calcitic mollusc shells may well retain their primary oxygen-isotope ratio during exposure to prehistoric cooking practises although this also needs to be tested in future studies.

Implications for shell midden archaeology

As shown in this study, clumped-isotope thermometry represents a unique diagnostic tool for detecting but also differentiating between certain prehistoric cooking practices employing aragonitic shell midden constituents which can provide new insights into various aspects of shell midden formation. Despite evidence of human behaviour (e.g. ritualistic versus nutrition purposes of shellfish foraging or particular subsistence strategies)39,40,41 and information about the function of particular sites (e.g. processing sites, dinner-time camps or residential locales)42, the reconstruction of specific cooking methods may also help to test the validity of modern analogues of shell midden formation from ethnographic or ethnohistoric observations43. Detailed information about specific processing techniques potentially broaden our understanding of prehistoric trading activities as the use of processing methods which prevent spoiling was most likely a prerequisite for the long-term transportation of the edible portion of marine molluscs4. Moreover, potential seasonal differences in shellfish processing techniques can contribute to our understanding of fishing and foraging seasonality. For long-term shell midden assemblages deposited over several millennia, the reconstruction of specific processing techniques may enable tracing the technological advancement over time of shell midden accretion.

Given the apparent alteration of shell δ13C values, potential alteration of 14C signatures and subsequent uncertainties in measured radiocarbon ages obtained from aragonitic shell midden constituents have to be expected. This hypothesis is supported by previous experiments showing that radiocarbon dates measured from food crusts on prehistoric pottery fragments can also be subjected to alteration during cooking by e.g. a reversed old-wood effect44. Detailed future studies are therefore needed to quantify the effect of pre-depositional heating on radiocarbon signatures in shell midden constituents.

Material and Methods

Modern Mercenaria campechiensis shells

In total, 23 sub-adult M. campechiensis individuals were cultured by Southern Cross Sea Farms®, at Cedar-Key, FL, USA (N29° 08′ 32.09662″ W83° 00′ 19.66703″). All individuals had the same age (14 months), similar size (~3.52 ± 0.24 cm shell height) and were reared in the same water masses throughout ontogeny. All individuals were sampled on 30 April 2015 and shipped overnight to RSMAS and sacrificed on 01 May 2015 immediately after arrival. Shells were manually cleaned with tap water, dried at room temperature and stored at ~20 °C until the experiment.

We used natural seawater from Virginia Key, FL, USA heated in a glass jar on a hotplate for the boiling treatment. Evaporation was compensated by adding periodically deionized water. A ~10 cm thick layer of commercial charcoal briquettes (Kingsford® Hickory) were used to simulate roasting and burning. The coal briquettes were continually replenished to keep the burning and roasting temperature as constant as possible throughout the entire experiment. For the roasting treatment, individual shells were placed into perforated aluminium cups approximately 15 cm above the coal fire. For the burning treatment, individual shells were placed into small aluminium cups preventing any loss of shell material during the experiment or contamination and placed directly on/in between the burning coals. The shells were not buried in burning coal. The temporal and spatial temperature variability was determined for both treatments recurrently by measuring the temperature at different spots using an Omega® HH806AU thermocouple thermometer equipped with an Uxcell® K-type (nickel-chromium/nickel) high-temperature thermocouple probe (100–1250 °C). The reported treatment temperatures represent average values with the 1σ standard deviations. The measured treatment temperatures were boiling in seawater at ~100 °C, roasting over charcoal at ~174 ± 13 °C (maximum temperature range 126–205 °C) and burning on charcoal at 559 ± 95 °C (maximum temperature range 386–721 °C), see Supplemental Fig. 1. In each treatment, the right valves of six shells were exposed for 15 to 360 minutes while five untreated shells were used as control. Afterwards, all shells were cleaned for 24 hours with 60 ml ultrapure 10% hydrogen peroxide buffered to a pH of ~8.2 with NaOH following the protocol of Bastidas & Garcìa45 at room temperature to remove the periostracum and other organic compounds and ground using a ceramic pestle and mortar. Powdered bulk samples were stored in sealed glass vials at room temperature.

Mid-Holocene samples

Mid-Holocene specimens were excavated from a shell midden in the Ras el Sass area, Mauritania, NW Africa. For details of the study area and spatial distribution of shell midden deposits see Barusseau et al.46 and the map provided in the supplementary information (Supplementary Fig. 3). All shells and otoliths were ultrasonically cleaned in deionised water and dried at 40 °C for 48 hours. Shells were cut along the direction of maximum growth and otoliths along the transversal plane using a tap water-cooled UniPrec WOCO 50 saw with a 0.6 mm thick diamond blade. The hinge region of each shell was sampled using a U-Power® UP200 hand drill equipped with a 0.8 mm tungsten-carbide dental drill at lowest possible RPM (RPM ≈ 1000). Thick sections (~2 mm) were prepared along the direction of maximum growth and were sampled along the direction of growth in the outer shell layer using a NewWave/Merchantec Micromill system. Posterior halves of the shells and otoliths were ground using a ceramic pestle and mortar. Powdered samples were stored in sealed glass vials at room temperature until chemical analysis. The anterior halves of the otoliths were ground in a similar manner with ~50 mg aliquots used for mineralogical analyses and radiocarbon dating. Similarly, a piece of each shell was prepared for mineralogical analysis and radiocarbon dating by cutting a slice through the ventral margin of the anterior halves using a Dremel™ tool equipped with a diamond cutting wheel.

Mineralogical analyses

Shell mineralogy of the experimental shells was measured using a PANalytical X´Pert PRO diffractometer with a Cu-tube (kα 1.541 Å, 45.0 kV, 40.0 mA) at RSMAS, Miami, FL, USA. Measurements were done using a continuous scan from 23.0–35.0 2θ with a step size of 0.005 2θ and a measuring time of 0.1 seconds per step. Data collection and processing was done using the software X’Pert Data Collector and X’Pert HighScore Plus, respectively. Relative abundance of aragonite and calcite were determined using the relative peak areas of aragonite (1,1,1) peak (2θ = 26.1°) and low magnesium calcite (1,0,1) peak (2θ = 29.6°) against a calibrated standard. Archaeological XRD-samples were analysed with a Philips X´Pert Pro diffractometer equipped with a Cu-tube (kα 1.541, 45 kV, 40.0 mA) at the University of Bremen, Germany. Measurements were done using a continuous scan from 3–85° 2θ with step sizes of 0.016° 2θ. We used the Philips software X´Pert HighScore™ for data processing. Abundance of aragonite and calcite was determined using the relative peak areas of aragonite (1,1,1) peak (2θ = 26.1°) and low magnesium calcite (1,0,1) peak (2θ = 29.6°) against a calibrated standard. Mineralogical data of the experimental and mid-Holocene samples are reported in the supplementary information (Supplementary Tab. 1 and 4).

Radiocarbon dating

Radiocarbon dating of four otoliths and three bivalve shells was done at the Poznan Radiocarbon Laboratory, Poznan, Poland using a standard protocol for accelerator mass spectrometry. Calibration of conventional radiocarbon ages was done with OxCal 4.2.447 using the Marine13 calibration curve48 and assuming a local reservoir age of ΔR = −300 yrs., which represents an average estimate for the local mid-Holocene reservoir ages based on several paired radiocarbon dates along the Mauritanian coastline throughout the Holocene (J. F. Saliége, unpublished data) which is also supported by radiocarbon dates from this archaeological site. Raw as well as calibrated radiocarbon ages are reported in the supplementary information (Supplementary Tab. 3).

Clumped-isotope analysis

Clumped-isotope analysis of the powdered bulk samples was performed at the Stable Isotope Laboratory of the Rosenstiel School of Marine and Atmospheric Science (RSMAS), Miami, FL, United States, between April 17, 2015 and March 28, 2016 using the analytical procedure described in Murray et al.49. Each sample was analysed at least two times in a randomised order. All carbonate samples were prepared for analysis on the stainless-steel cryogenic vacuum extraction line at RSMAS, evacuated to <10−6 mbar using two turbo-molecular pumps (Balzer TPU 170 and Edwards 50EX). Individual sample aliquots of ~8.0 mg powder were digested for 30 minutes at 90 °C in 3.5 ml ~105% phosphoric acid (H3PO4) using a modified “Fairbanks” device50 connected to a common acid bath. Resulting CO2 was then purified on the cryogenic vacuum extraction line.

Purified CO2 was measured at RSMAS using a dual-inlet MAT-253 gas source isotope ratio mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). Samples were measured against an in-house working-gas gas standard. We analysed each sample at a signal intensity of 12 V on mass-44 over six acquisitions of 15 sample-standard measurements. Data reduction and normalisation for ∆47 and temperature calculations followed the methods of Affek and Eiler51 and Huntington et al.52. To check for potential contamination, samples were scrutinised based on their offset of δ48 and ∆48 52. Precision of clumped-isotope analysis is reported using the average standard errors. The external precision of the instrument was monitored with repeated measurement of a Carrara marble standard. The average Carrara marble ∆47 value was found to be 0.4001 ± 0.0294‰ (n = 191) which is similar to reported average values from four different laboratories of 0.395‰24 even though a different sample of the Carrara marble standard was used.

We measured δ13C and δ18O of the samples by means of masses 45/44 and 46/44, respectively using a method adapted from Craig53 modified for a multi-collector mass spectrometer. Shell and otolith δ13C and δ18O values were calibrated using NBS-19 and reported relative to Vienna Pee Dee Belemnite (VPDB). External precision of δ13C and δ18O analyses was better than ± 0.055 and ± 0.117‰, respectively. All Δ47 values defined as

$${{\rm{\Delta }}}_{47}=[(\frac{{R}^{47}}{{R}^{47}* }-1)-(\frac{{R}^{46}}{{R}^{46}* }-1)-(\frac{{R}^{45}}{{R}^{45}* }-1)]\,* \,1000$$

where R47, R46, R45 are abundance ratios of masses 47, 46 and 45 relative to the mass 4425, were calculated using the method described by Affek and Eiler51 and Huntington et al.52. Translation into the absolute reference frame (ARF) was accomplished using the method described by Dennis et al.24 with 1000, 50 and 25 °C water equilibrated gasses. All isotope data are reported in the supplementary information (Supplementary Tab. 1 and 4).

Recent publications by Schauer et al.54 and Daëron et al.55 have suggested a small difference between the correction protocols proposed by Santrock et al.56 and Brand et al.57 leading to differences in the calculated δ13C, δ18O and Δ47 values. These differences are small (~0.02‰) for δ13C and δ18O and generally within the external precision of the analytical method. For Δ47, the suggested changes were more significant and dependent upon the δ13C value of the material being analysed. These differences increased from essentially no changes for a sample with a δ13C value close to 0‰ to 0.04‰ for a sample with a δ13C value of −40‰. We compared our method of calculating δ13C and δ18O values modified from Craig53 and found the difference between our method and that proposed by Brand et al.57 to show only a 0.03‰ difference for samples with a δ13C difference of −40‰. This decreases to less than 0.01‰ for δ13C values of samples analysed in this study. Hence, based on the range of δ13C values of the samples measured in this study we conclude that the artefacts noted by Schauer et al.54 and Daeron et al.55 do not significantly impact the results presented here. We have, nonetheless, processed all results presented herein using the revised parameters from Brand et al.55.

Ontogenetic oxygen-isotope records

Sub-samples drilled from the outer shell layer (~80–100 μg) were measured on a Finnigan MAT 251 gas isotope ratio mass spectrometer connected to a Kiel III automated carbonate preparation device at the stable isotope laboratory in the Center for Marine Environmental Sciences (MARUM), University of Bremen, Germany. The long-term standard deviation of the in-house standard (Solenhofen limestone) was better than 0.06‰ for δ18O.

Statistical analysis

We used a one-way analysis of variances (ANOVA) with a Bonferroni-corrected t-test as post hoc test. The assessment of equal variances and normal distributions was done using a Levene’s test and a Shapiro-Wilk test, respectively. In case of non-normal distributed data (mineralogy), we used a Kruskal-Wallis one way ANOVA on ranks and the Dunn’s method as post hoc test. The significant levels for all statistical test were set to α = 0.05. Results of the statistical analyses are reported in the supplementary information (Supplementary Tab. 2).

Data Availability

All data generated or analysed during this study are included in this published article (and its Supplementary Information files).