'Palaeoshellomics’ reveals the use of freshwater mother-of-pearl in prehistory

Abstract

The extensive use of mollusc shell as a versatile raw material is testament to its importance in prehistoric times. The consistent choice of certain species for different purposes, including the making of ornaments, is a direct representation of how humans viewed and exploited their environment. The necessary taxonomic information, however, is often impossible to obtain from objects that are small, heavily worked or degraded. Here we propose a novel biogeochemical approach to track the biological origin of prehistoric mollusc shell. We conducted an in-depth study of archaeological ornaments using microstructural, geochemical and biomolecular analyses, including ‘palaeoshellomics’, the first application of palaeoproteomics to mollusc shells (and indeed to any invertebrate calcified tissue). We reveal the consistent use of locally-sourced freshwater mother-of-pearl for the standardized manufacture of ‘double-buttons’. This craft is found throughout Europe between 4200–3800 BCE, highlighting the ornament-makers’ profound knowledge of the biogeosphere and the existence of cross-cultural traditions.

Introduction

The selection of shell as a raw material by prehistoric populations implies that it possesses an inherent attractiveness that makes it suitable for displaying social connections, wealth and prestige (Bar-Yosef Mayer et al., 2009; d'Errico et al., 2005; Giacobini, 2007; Kuhn et al., 2001; Saunders, 1999; Zilhão et al., 2010). Unravelling these nuances is especially important in times of lifestyle transformation (e.g. from mobile hunting and gathering to sedentism), when the way people perceived their ‘homeland’ may also have shifted: personal adornments thus help track cultural continuity or discontinuity (Rigaud et al., 2018; Rigaud et al., 2015; Stiner, 2014; Stiner et al., 2013; Taborin, 1974; Vanhaeren and d'Errico, 2006). Exotic marine shells, transported to inland sites, are thought to have acquired special value and are interpreted as a marker of status and a proxy for long-distance exchange and trade (Alarashi et al., 2018; Bajnóczi et al., 2013; Borrello and Micheli, 2011; Taborin, 1974; Trubitt, 2003). The exploitation of local shells is instead usually viewed in utilitarian terms (e.g. Colonese et al., 2011), as an easily accessible, convenient, but inherently less prestigious resource. However, the collection of certain local marine shells may have a deeper meaning; for example, creating a feeling of familiarity with a new environment for seafarers and colonisers (Bar-Yosef Mayer, 2018). The use of freshwater shells as raw material, although often documented in archaeological sites, has been somewhat overlooked (Borrello and Girod, 2008), and the relative importance of freshwater over marine shells has never been systematically addressed, introducing a bias in archaeological interpretations.

A typical example of such bias is the interpretation as exotica of the findings of prehistoric mother-of-pearl (shell) miniature double-buttons (doppelknöpfe), worked in a way to look like ‘true’ pearls (Figure 1). Experimental archaeological work has shown that they are excellent as ornaments pressed into thin leather, for example armbands or belts (Kannegaard, 2013). The raw materials actually vary from shell to (possibly) copies made of stone, teeth, bone and ceramic, but they are all consistently white, and their appearance standardized (Heumüller, 2009). Double-buttons have been reported primarily from a range of Neolithic Central European sites on the Danube Valley, but occasional findings occur from the Ligurian coast to Northern and Southeastern Europe (Girod, 2010a; Heumüller, 2012). Among these sites, the shell double-buttons from the Danish Ertebølle/Early Funnel Beaker shell midden at Havnø (Andersen, 2008) and the hundreds of examples from the Neolithic submerged pile dwelling settlement of Hornstaad-Hörnle IA on Lake Constance (Heumüller, 2009) are especially significant, as they could represent an instance of exchange of materials or ideas between hunter-gatherers (Havnø) and farmers (Hornstaad-Hörnle IA; Figures 1 and 2, Table 1). Furthermore, if the double-buttons were made of marine shell (in particular of oyster shells, abundant at Havnø [Andersen, 2008]), then a case could be made for exotic materials from the coastal Ertebølle site of Havnø being imported (because highly desirable) to the inland Neolithic settlement of Hornstaad-Hörnle IA (Rowley-Conwy, 2014). However, the edible flat oyster (Ostrea edulis), so abundant at Havnø, can hardly be the raw material of the double-buttons: visual inspection (later confirmed by our analytical data) revealed that these are made of mother-of-pearl (nacre) and O. edulis does not form this microstructure. Furthermore, the ‘true’ pearl oysters (marine genus Pinctada) are not found in North Atlantic cold waters.

10.7554/eLife.45644.003

Figure 1.

The double-buttons analysed in this study.

Generate image description

10.7554/eLife.45644.004

Figure 2.

Map displaying the location of Havnø, Hornstaad-Hörnle IA and Peştera Ungurească, together with other archaeological sites from which double-buttons (Heumüller, 2012) and a variety of ornaments made with Unio sp. shells (Rigaud et al., 2015) have been reported.

Generate image description

10.7554/eLife.45644.005

Table 1.

Summary of the materials analysed in this study (archaeological ornaments and reference shells) and information on their context, chronology and taxonomic determination.

Sample type	Site	Cultural group	Time span	Taxonomic determination before this study	Other molluscan fauna present at the site
Double-buttons	Havnø	Ertebølle	4200–4000 cal BCE (radiocarbon dating of the Late Mesolithic horizon; Andersen, 2000; Andersen, 2008)	Unknown, presumed marine shell	Abundant edible marine shells: Ostrea edulis, Littorina sp., Mytilus edulis, Cerastoderma edule (Andersen, 2000; Andersen, 2008)
	Hornstaad-Hörnle IA	Hornstaad group (early phase of the regional Late Neolithic)	3917–3902 BCE (dendrochronology; Billamboz, 2006)	Debated: marine (Ostrea edulis) vs freshwater (Margaritifera margaritifera; Heumüller, 2010; Borrello and Girod, 2008)	Mediterranean marine shells (exotic, non edible): Columbella rustica (Borrello and Girod, 2008), Callista chione, Astarte borealis, Dentalium vulgare (Heumüller, 2010)
	Peştera Ungurească	Toarte Pastilate and transition to Coţofeni	4260–3820 cal BCE (range of radiocarbon dates, at 1σ, of layers 2B, 2A3 and 2A, Toarte Pastilate)(Biagi and Voytek, 2006)	Unio cf. crassus (Girod, 2010a, Figure 6B)	Abundant terrestrial taxa (naturally occurring). Occasional presence of freshwater species (Anisus spirorbis, Pisidium milium, Lithoglyphus naticoides, Lymnaea truncatula, Planorbis cf. carinatus; Girod, 2010a)
Reference shell	Limfjord (Northern Jutland)		Modern	Ostrea edulis
	Limfjord (Northern Jutland)		Modern	Modiolus modiolus
	Limfjord (Northern Jutland)		Modern	Margaritifera margaritifera (determined by F.M.)
	France (Izeure)		Modern	Unio pictorum
	Peştera Ungurească	Toarte Pastilate and transition to Coţofeni	4260–3820 cal BCE (range of radiocarbon dates, at 1σ, of layers 2B, 2A3 and 2A, Toarte Pastilate) (Biagi and Voytek, 2006)	Unio cf. crassus (Girod, 2010a)	Abundant terrestrial taxa (environmental signal). Occasional occurrence of freshwater species (Anisus spirorbis, Pisidium milium, Lithoglyphus naticoides, Lymnaea truncatula, Planorbis cf. carinatus). Not suitable as raw material for the double-buttons (Girod, 2010a)
	Isorella	Vhò	5226–5023 cal BCE at 2σ (Starnini et al., 2018)	Pseudunio auricularius (Biddittu and Girod, 2003; Girod, 2010b)	Marine taxa (typically used as ornaments): C. rustica, Spondylus (fragment of a bracelet; Girod, 2010b)

Despite being the basis for archaeological inference, the knowledge of the biological origin of molluscan mother-of-pearl in prehistoric Europe is only vague (Taborin, 1974). This is usually based on macroscopic and microscopic observations of heavily worked and degraded objects, which have thus lost any diagnostic feature that might have been useful for identification (Demarchi et al., 2014). Using analytical methodologies that are the basis of shell biomineralization studies (Marin et al., 2016; Marin et al., 2013), including the first application of proteomics to archaeological shells (‘palaeoshellomics’), we investigated seven archaeological double-buttons, dated between ~4200 and ~3800 BCE and recovered from a wide geographic area (Figure 2): three (HorA, HorB, HorC) from Hornstaad-Hörnle IA (Borrello and Girod, 2008; Heumüller, 2012), three (HavA, HavB, HavC) from Havnø (Andersen, 2008), and one (PesB) from Peştera Ungurească, a cave site in Transylvania (Biagi and Voytek, 2006; Girod, 2010a). The richest record comes from Hornstaad-Hörnle IA, where more than 564 double-buttons were recovered, although many were partially fragmented and affected by fire. The origin of the raw material had been debated (Table 1): either edible oyster (marine) or freshwater pearl mussel (Borrello and Girod, 2008; Heumüller, 2012). Around 40 double-buttons were excavated from the shell midden in Havnø - the raw material used to make these was undetermined. A handful of double-buttons were found at the cave site of Peştera Ungurească, together with Unio crassus valves, and therefore this freshwater mollusc was assumed to be the raw material used to make the double-buttons at Peştera Ungurească (Girod, 2010a).

Our work had two main aims:

to develop ‘palaeoshellomics’, a new molecular approach to characterize ancient proteomes preserved in mollusc shells. The recovery and identification of these proteomes is challenging, because shell contains only a small fraction of proteins embedded in the mineral skeleton (approximately 0.1–1% vs 30% in bone) and there is a lack of molecular sequence data, which are needed for comparing taxa within the same clade. When studying ancient materials, these issues are further compounded by the effects of diagenesis and by the small sample sizes available. Therefore, developing palaeoshellomics has an impact on understanding the fundamental mechanisms of biomineralization (protein-mineral interactions) in molluscan shells and other invertebrates, the extent of proteome variability within the same molluscan clade, as well as the diagenetic pathways of degradation and preservation.

to integrate ‘palaeoshellomics’ within a set of well-established analytical techniques, in order to determine the origin of the shell ornaments from Hornstaad-Hörnle IA and Havnø, and to confirm that of the double-buttons from Peştera Ungurească. This is archaeologically significant, as similar ornaments were recovered from three geographically distant sites belonging to three different and broadly contemporary cultural groups (Table 1): Late Mesolithic (Ertebølle), early Late Neolithic (Hornstaad Group), and Copper Age (Toarte Pastilate/Coţofeni).

To achieve both of these goals, the archaeological double-buttons were studied alongside a selected set of reference shells (Table 1). These include two marine species (Ostrea edulis, Modiolus modiolus), which commonly occur in northwestern European seas, and four freshwater mussels, which are characterized by a thick layer of mother-of-pearl, that is Unio pictorum, Unio crassus, Margaritifera margaritifera, Pseudunio auricularius.

Sample selection

Two modern marine mollusc shells, O. edulis and M. modiolus, were collected in northern Jutland (Denmark) by Søren H. Andersen and were selected for the following reasons: O. edulis shells had been suggested as the potential raw material for the Hornstaad-Hörnle IA assemblage (Heumüller, 2010) and are very abundant at the shell midden site of Havnø; M. modiolus is a thick-shelled mussel with a nacreous layer, therefore a suitable raw material for the Havnø ornaments (Appendix 1, section 2). Furthermore, close relatives of both species are present in public sequence databases, which is important for palaeoproteomics: O. edulis belongs to family Ostreidae (genomes available for Crassostrea gigas and C. virginica) and M. modiolus to family Mytilidae (genome available for Mytilus galloprovincialis).

With regard to the freshwater species (order Unionoida), U. pictorum and U. crassus belong to family Unionidae, P. auricularius and M. margaritifera to family Margaritiferidae. Modern U. pictorum shells were collected in a stream close to Izeure (Burgundy) by Frédéric Marin and modern M. margaritifera was collected in northern Jutland by Søren H. Andersen. The morphological determination of both taxa was carried out by Frédéric Marin. U. crassus and P. auricularius are archaeological shell specimens from the sites of Peştera Ungurească and Isorella (Neolithic, Po Plain, Italy [Starnini et al., 2018]). The determination of U. cf. crassus had been carried out by Alberto Girod on the basis of morphological observations of the whole shell valves and comparison to extant specimens from the area (Girod, 2010a). Archaeological P. auricularius was used as this species is critically endangered (Altaba, 1990) and extant populations rare (Appendix 1—figure 1). The morphological determination of the species had been carried out by Alberto Girod using comparative specimens from museum collections (Biddittu and Girod, 2003). An advantage of including archaeological shells as reference materials is that we were able to assess the extent of molecular preservation in shells that are contemporary to the double-buttons. Furthermore, whole and fragmented valves of U. cf. crassus had been recovered from all the archaeological layers that yielded the double-buttons at Peştera Ungurească (Girod, 2010a). Therefore, in this case, potential raw material and finished product have experienced the same post-depositional conditions.

None of the Unionoida species is well-represented in public sequence databases, especially with regard to proteins related to shell biomineralization (see Methods section).

Results and discussion

Morphological analysis

The double-buttons described here are circular, with a groove in the middle of the body and no perforation. The main body is shiny but all of them have a thin matte layer on one of the two surfaces (Figures 1 and 3). These two layers, both aragonitic (as shown by infrared spectroscopy, see Appendix 1, section 3.2), are the nacre and prisms of a mollusc shell. Scanning electron microscopy showed the presence of the ‘brickwall’ microstructure of nacre (sheet nacre) juxtaposed with the thin layer of prisms, the latter having elongation axes perpendicular to the nacre plane (Figure 3). No secondary calcite was observed and there was no sign of the occurrence of diagenetic recrystallization. The overall nacre appearance is typical of bivalves and not of gastropods. The combination of nacroprismatic microstructure and aragonitic mineralogy is observed in freshwater unionoid mussels but also in marine trigonioids and anomalodesmatans (Appendix 1, section 2).

10.7554/eLife.45644.006

Figure 3.

Microstructure (SEM) and mineralogy (FTIR-ATR) of double-buttons, showing shiny nacreous (a) and matte prismatic (c) layers, both aragonitic (b, d).

Generate image description

Stable isotopes of carbon and oxygen

Stable isotope analyses for all of the samples yielded average δ¹⁸O and δ¹³C values of −5.3 ± 0.4 and −11.1 ± 0.6 ‰ for Havnø, −6.1 ± 1.0 and −11.9 ± 1.7 ‰ for Peştera Ungurească and −9.3 ± 0.5 and −10.6 ± 1.6 ‰ for Hornstaad, respectively (Appendix 1, section 3.3). The consistently low δ¹⁸O and δ¹³C values of shells from Peştera Ungurească and Hornstaad indicate a local freshwater origin for the shells (Keith et al., 1964; Leng and Lewis, 2016), whereas the δ¹⁸O values at Havnø suggest some mixing of marine water or changes in the atmospheric circulation, with precipitations slightly enriched in ¹⁸O compared to present day over the region. Our interpretations are broadly supported by the average annual δ¹⁸O values of modern local precipitations for the sites (WaterIsotopes.org, 2018). The isotope data therefore suggest that the shells were locally sourced (Keith et al., 1964).

Amino acid analysis

The absence of recrystallization observed by SEM was consistent with the concentration, composition and relatively low D/L values for all the amino acids analysed (e.g. alanine D/L ~0.1 for Peştera Ungueraşca, ~0.2 for Havnø and Hornstaad), except for sample HorA. This supports a non-fossil origin for the shells used to make the double-buttons, that is the makers used ‘fresh’ or recently dead mollusc shells. However, the extent of degradation for HorA was significantly higher and both D/L and concentration values showed a clear ‘burning’ signal (Crisp, 2013; Demarchi et al., 2011). The amino acid composition was similar to that of freshwater bivalves (Unio, Margaritifera) present in the reference database of Demarchi et al. (2014), although HorA and PesB appeared to be rather different from the other double-buttons (Appendix 1, section 3.4).

Palaeoproteomics

We characterized the proteomes preserved within the seven double-buttons and performed bioinformatic searches (PEAKS 8.5, Bioinformatics Solutions Inc [Ma et al., 2003]) of the product ion spectra against both a Protein sequences and an Expressed Sequence Tags (ESTs) database, restricting the taxonomy to Mollusca (see Methods section). This resulted in the identification of 1973 and 3233 peptide sequences, respectively, which represent the 3.5% and 5.1% of the total number of sequences generated by the de novo algorithm of the software (excluding contaminant sequences). For comparison purposes, we also performed shotgun proteomics on the shell matrices extracted from the six reference mollusc species (Appendix 1, section 3.5) and analysed the data using the same databases and parametres (see Table 2—source data 1 for full results of the palaeoproteomic analyses).

Table 2 shows the top-scoring proteins from the seven double-buttons: the numbers indicate the peptides supporting each protein identification, while protein coverage (i.e. the percentage of sequence for which we could detect peptides) is represented by different colours. Additionally, on the right hand side of the table we indicate if each of the double-button proteins also occurred in the reference shell proteomes (the list of all shell proteins identified in both the double-buttons and the reference shells can be found in Table 2—source data 1).

10.7554/eLife.45644.007

Table 2.

Main protein sequences identified in the double-buttons from Havnø, Hornstaad-Hörnle IA and Peştera Ungurească and their presence/absence in the analysed set of reference freshwater and marine shells (black dots).

Numbers indicate total number of peptide sequences identified and the cell colour is proportional to the coverage of the sequence itself. Threshold values for peptide and protein identification: false discovery rate (protein FDR) = 0.5%, protein score −10lgP ≥ 40, unique peptides ≥ 2, de novo sequences scores (ALC%) ≥ 50. Asterisks (*) indicate proteins identified only when using less stringent parametres: protein score −10lgP ≥ 20; unique peptides ≥ 1. Note that molecular sequence databases for molluscan species are incomplete and biased towards well-studied model organisms. The peptide sequences recovered in our study were identified using sequence homologies with proteins originally described from Hyriopsis cumingii, Crassostrea sp., Pinctada sp., Mytilus sp. and several others. As a result of database insufficiency, the bioinformatic search of these ‘shellomes’ could not identify the exact taxon of our samples, but provided a strong indication of the fact that the closest taxon to that of the ornaments (and of the freshwater reference shells) is the pearl-producing triangle sail mussel Hyriopsis cumingii (Unionoida).

Proteins present in database from		Identified proteins	Double-buttons							Freshwater				Marine
Order	Genus		HavA	HavB	HavC	HorA	HorB	HorC	PesB	U.p	U.c	M.m	P.a	Mo.M	O.e
Unionoida	Hyriopsis	Hic74 [Hyriopsis cumingii]	132	158	260		6	11	21	●	●	●	●
		Hic52 nacreous layer matrix protein [Hyriopsis cumingii]	1*	1*	2*					●	●	●	●
		Silkmapin (isoforms: nasilin 1 and nasilin 2) [Hyriopsis cumingii]	1*	3	3		5			●	●	●*	●*
Ostreida	Pinctada	MSI60-related protein [Pinctada fucata]	6		27			12		●	●		●	●
		Insoluble matrix protein [Pinctada fucata]	4		33					●	●		●
	Crassostrea	Glycine-rich cell wall structural protein-like [Crassostrea virginica/gigas]	17	11	12				14		●
		Glycine-rich protein 23-like [Crassostrea virginica]	8	11	6					●		●
		Antifreeze protein Maxi-like [Crassostrea virginica]		4	4								●
Mytilida	Bathymodiolus	MSI60-related protein partial [Bathymodiolus platifrons]		6	11					●			●	●
	Mytilus	Precollagen D [Mytilus edulis]	16	26	23				9	●	●	●
		Nongradient byssal precursor [Mytilus edulis]	10	10					25		●
Other	Other	Predicted: transcription factor hamlet-like partial [Octopus bimaculoides]	5	6	11					●		●
		Hypothetical protein OCBIM_22008720 mg partial [Octopus bimaculoides]	6	11
							Coverage		≥55%	≥35%	≥15%	≥10%	≥1%	Presence ●

Shellomes: significant protein hits

The main protein sequences from the double-buttons were identified as belonging to the pearl-producing triangle sail mussel Hyriopsis cumingii (Bai et al., 2013).

Protein Hic74 (GenBank: ARG42316.1) was found in all of the archaeological samples, except HorA. The percentage coverage for the Hic74 sequence was highest for the Havnø beads (35–55%), where it was supported by 132, 153, 255 unique peptides in HavA, HavB and HavC, respectively (Appendix 1, section 3.5). This protein was also securely identified in all of the freshwater unionoid reference shells (coverage varying from 34% in M. margaritifera, supported by 67 peptides, to 50% in U. crassus, supported by 203 peptides). Hic74 is an acidic, Ala- and Gly-rich shell matrix protein (Liu et al., 2017a). Consisting of 19 poly-A blocks, GA repeats, short acidic motifs (that probably bind to the mineral) and a GS-rich domain at the C-terminus (which resembles that of lustrin-A), this silk fibroin-like protein is likely to play a structural role in nacre formation and in enhancing its mechanical properties (Liu et al., 2017a).

Protein Hic52 (GenBank: ARH52598.1) was identified in all the reference unionoid shells and in the Havnø samples, but only when less stringent parameters were used for the identification (i.e. number of unique peptides ≥ 1 (instead of 2) and protein score −10lgP ≥ 20 (instead of 40)). Hic52 is a very basic (theoretical pI > 10), Gly- and Gln-rich protein, with few poly-Q and poly-G blocks and several degenerate G-rich repeats of different lengths along the sequence. It possesses a collagen-like structure which suggests a structural role in nacre formation (Liu et al., 2017b). Silkmapin (GenBank: AIZ03589.1, and its isoforms nasilin 1 and 2) are Gly-rich non-acidic proteins with a structural function, probably related to the formation of both nacreous and prismatic layers (Liu et al., 2015; Marie et al., 2017). Present in the shell matrix of all the unionoids, these proteins were also detected in the Havnø samples and in one of the Hornstaad beads (HorB). Finally, we also identified protein sequences from marine mollusc genera (mainly Pinctada, Crassostrea and Mytilus), but all these ‘marine’ sequences only displayed repeated low-complexity (RLC) domains (typically consisting of Ala and Gly-rich repeats and/or poly-Ala blocks). RLC-containing peptides are not sufficient for distinguishing between freshwater and marine shells. On the contrary, in double-buttons and in unionoid reference shells, the top-scoring protein Hic74 was supported by remarkably high (for shell proteins) coverages, and, together with Hic52 and silkmapin/nasilin, showed a number of specific peptides that do not exhibit RLC domains. These proteins showed no homologues with any other shell proteins of marine origin currently present in the NCBI database (BLASTp search), being unique to H. cumingii and suggesting their specificity to freshwater unionoid shells. We argue that their presence (where identified as the major shell matrix proteins, supported not only by RLC domains) is specific to Unionoida, freshwater mother-of-pearl shells, which in combination with the isotopic data, and supported by the microstructural and amino acid results, excludes a marine origin for the raw material used to make the double-buttons.

Comparison with reference shells

We performed a search of the product ion spectra from the double-buttons and the six reference shells against the redundant EST database, so that we could recover complementary information from non-annotated sequences. For example, a search of Hyriopsis cumingii on NCBI will retrieve 246 protein sequences but 10156 EST sequences. The dataset was used to explore the similarities between double-buttons and molluscan shell proteomes, presented as a circular plot in Figure 4. This output was derived from an adjacency matrix, showing which proteins (EST sequence identifiers) occurred in two or more samples (R code and data files can be found in Figure 4—source data 1). From this, the subset of unique identifiers, present both in the double-buttons and in any of the reference shells, was associated with its sequence coverage (%) in the archaeological samples. This information is represented in the right-hand side of Figure 4: the length of the circular segment for each molluscan species is proportional to the number of sequences that the shell shares with the archaeological samples, scored on the basis of the coverage. On the left-hand side of the graph, each double-button is represented by a circular segment, which is proportional to the number of unique peptides that supports each shared protein sequence. The similarity between double-buttons and mollusc taxa can be visualised through the thickness of the connecting bands. Overall, the data showed that the EST sequences shared between the ornaments and the shells were mainly from the unionoids, consistent with the results obtained by searching the annotated protein database (Table 2). From all of the archaeological samples, the Havnø set showed the best match to the freshwater unionoids, owing to better-preserved proteins, that is with high coverage and number of supporting peptides. The LC-MS/MS analysis of PesB produced a number of tandem mass spectra comparable to the other samples, but a lower number of sequences were identified. Sample HorA (burnt), from which no proteins had been identified using the annotated protein database, yielded some matches to EST sequences, most of which shared with U. pictorum.

10.7554/eLife.45644.009

Figure 4.

Circular diagram representing the similarity between the proteomes of seven double-buttons (left) and six mollusc shell taxa (right).

Generate image description

Database-independent comparison

In order to provide further, independent, evidence for the origin of the raw material, we developed an in-house tool (in C language, available in Figure 5—source code 1) for ‘proteome comparison’, using all the peptide sequences generated by the de novo algorithm of the software PEAKS, that is before performing any database search (Appendix 1, section 3.6). The tool was able to provide a score for the sequence similarity between two lists of peptides and to generate a similarity matrix from all pairwise comparisons, which was then converted to a distance matrix. Multidimensional scaling (MDS; Gower, 1966) was used to visualise the similarity of observations (Figure 5) and confirmed that the Havnø set and the freshwater reference unionoids display the higher degree of proteome similarity, while the samples from Hornstaad and PesB fall in a different area of the plot from each other and from the marine reference shells. The results were also in accordance with those obtained from another database-independent approach Appendix 1—figure 25, based on direct product ion spectra comparison (Rieder et al., 2017), which was adapted for this study (Appendix 1, section 3.6). Overall, our study, which represents one of the few that attempts to compare molluscan proteomes within the same clade (genus, family or order), shows that unionoid shells exhibit very similar proteome profiles, sharing many sequences between species Appendix 1—figure 23. This may suggest that this group has a rather conserved, homogeneous and recognisable proteomic signature, a conclusion that is completely congruent with earlier findings (Marie et al., 2017). Furthermore, all analyses showed that the three sets of archaeological ornaments have similar proteome profiles (Table 2, Figures 4 and 5, Appendix 1—figure 24, but do not exhibit a simple correspondence to a molluscan species, at least among the Unionoida considered here, further highlighting the complexity of molluscan shell proteomes.

10.7554/eLife.45644.011

Figure 5.

Proteome comparison based on peptide sequence similarity, represented by multi-dimensional scaling (MDS).

Generate image description

Analysis of the Hic74 sequence

We examined the sequence of the top-scoring protein, Hic74, recovered from the reference shells and ornaments, with the aim of assessing the presence and frequency of any amino acid substitutions, which could potentially yield taxonomic resolution within Unionoida. Figure 6 shows the alignment (performed using the software Geneious Prime 2019.1.1) of these incomplete sequences to the reference (Hic74 from Hyriopsis cumingii). The sequence coverage of each sample was obtained from the ‘Spider’ output of PEAKS 8.5. The Spider algorithm takes into account potential amino acid substitutions, as well as a large number of in vivo, laboratory-induced (e.g. carbamidomethylation), and diagenetically-relevant (e.g. deamidation) modifications, therefore it is especially useful in highlighting possible mutation sites. In our sequence reconstruction we only considered peptides displaying typical sample preparation-induced or diagenesis-induced modifications (Figure 6—figure supplement 1, Figure 6—figure supplement 2, Figure 6—figure supplement 3, Figure 6—figure supplement 4, Figure 6—figure supplement 5, Figure 6—figure supplement 6, Figure 6—figure supplement 7, Figure 6—figure supplement 8, Figure 6—figure supplement 9, Figure 6—figure supplement 10). The potential amino acid substitutions and their positions are summarised in Table 3 (supporting product ion spectra can be found in Figure 6—source data 1).

10.7554/eLife.45644.013

Figure 6.

Alignment of Hic74 sequences recovered from the Unionoida reference shells and the ornaments. 

The reference Hic74 [Hyriopsis cumingii] is shown at the top of the alignment and is highlighted in yellow. Dashes indicate where the sequence was not covered in the samples analysed in this study; amino acid residues highlighted in colour show all disagreements with the reference Hic74 [Hyriopsis cumingii].

Generate image description

10.7554/eLife.45644.014

Figure 6—figure supplement 1.

Unio crassus - Hic74 coverage.

Generate image description

10.7554/eLife.45644.015

Figure 6—figure supplement 2.

Unio pictorum - Hic74 coverage.

Generate image description

10.7554/eLife.45644.016

Figure 6—figure supplement 3.

Margaritifera margaritifera - Hic74 coverage.

Generate image description

10.7554/eLife.45644.017

Figure 6—figure supplement 4.

Pseudunio auricularius - Hic74 coverage.

Generate image description

10.7554/eLife.45644.018

Figure 6—figure supplement 5.

PesB - Hic74 coverage.

Generate image description

10.7554/eLife.45644.019

Figure 6—figure supplement 6.

HavA - Hic74 coverage.

Generate image description

10.7554/eLife.45644.020

Figure 6—figure supplement 7.

HavB - Hic74 coverage.

Generate image description

10.7554/eLife.45644.021

Figure 6—figure supplement 8.

HavC - Hic74 coverage.

Generate image description

10.7554/eLife.45644.022

Figure 6—figure supplement 9.

HorB - Hic74 coverage.

Generate image description

10.7554/eLife.45644.023

Figure 6—figure supplement 10.

HorC - Hic74 coverage.

Generate image description

10.7554/eLife.45644.025

Table 3.

Potential amino acid substitutions detected in the samples analysed in this study, compared to the reference Hic74 sequence [Hyriopsis cumingii].

Positions are derived from the sequence alignment shown in Figure 6. Dashes indicate that the position was not covered for that sample; question mark symbols indicate ambiguous substitutions. Hic74 coverages for each sample and supporting product ion spectra are presented in Figure 6—figure supplement 1, Figure 6—figure supplement 2, Figure 6—figure supplement 3, Figure 6—figure supplement 4, Figure 6—figure supplement 5, Figure 6—figure supplement 6, Figure 6—figure supplement 7, Figure 6—figure supplement 8, Figure 6—figure supplement 9, Figure 6—figure supplement 10, as well as in Figure 6—source data 1.

Hic74 [Hyriopsis cumingii]	A	A	A	A	A	A	A	G	D	G	S	E	G	A	A	L	V	G	L	I	A	G	A	Q	R	E
AA position	83	85	91	93	106	108	110	111	151	152	163	172	175	282	283	284	289	292	306	310	403	801	804	821	822	827
U. crassus	-	-	A	A	A	A	A	G	D	G	-	Q	G	A	?	L	?	G	L	F	A	-	-	Q	R	E
U. pictorum	-	-	?	?	?	-	-	-	-	-	-	Q	G	-	-	-	-	-	L	F	?	S	S	Q	R	E
M. margaritifera	-	-	-	A	-	-	-	-	-	-	-	Q	G	-	A	L	L	G	L	F	A	-	-	Q	R	E
P. auricularius	-	A	A	A	A	A	A	-	?	?	D	E	L	-	A	F	V	E	F	I	-	-	-	H	H	D
HavA	A	S	A	A	A	?	?	A	-	-	-	Q	G	A	A	L	L	G	L	F	-	-	-	Q	G	E
HavB	?	S	A	A	A	V	-	-	-	-	-	-	-	-	A	L	L	G	L	F	-	-	-	Q	G	E
HavC	?	S	A	A	A	V	?	A	D	G	G	-	G	?	A	L	L	G	L	F	-	G	S	Q	G	E

In the best-case scenario, only around half of the Hic74 (Hyriopsis cummingii) reference sequence was covered (Table 2, Figure 6). This may be due to: genuine sequence differences between Hyriopsis and all the other Unionoida examined here; low susceptibility of low-complexity domains to enzymatic cleavage; selective post mortem (or laboratory-induced) degradation of half of the sequence; errors in the transcriptome assembly of the protein. It is likely that a combination of all of these factors is responsible for this, particularly as the Hic74 regions not covered in the samples are mainly low-complexity domains, sometimes highly polar and thus prone to hydrolysis (Ser and Asp-rich). Within the limits due to the incomplete coverage of the mutation sites in reference shells, it is interesting to note that the Hic74 sequence of P. auricularius diverges significantly from those of U. pictorum, U. crassus and M. margaritifera, which appear to have a higher degree of sequence similarity (Table 3). While this is in contrast with a recent taxonomy reassessment of Unionida based on mitochondrial DNA (Lopes-Lima et al., 2018), it has been noted before that shell proteomes do not follow a simple phylogenetic signal (Jackson et al., 2010). The sequence coverage was insufficient to attempt any further consideration for the Hornstaad and PesB samples, but the better-preserved Havnø double-buttons shared the same amino acid substitutions, supporting the hypothesis that the same species was used to make these three ornaments (Table 3). Furthermore, Table 3 shows that this taxon was unlikely to be P. auricularius, and more likely to be Unio or Margaritifera sp.

Raw material identification

The raw material used to manufacture the seven double-buttons can be firmly and consistently identified as Unionoida, freshwater shells with a thick mother-of-pearl layer, on the basis of morphological, microstructural, mineralogical, geochemical and biomolecular data. From a microstructural viewpoint alone, the combination of aragonitic prisms and sheet nacre (‘brickwall type’) structures is restricted to three bivalvian orders: the Unionoida, the Trigonioida and the Anomalodesmata, the first two belonging to sub-class Palaeoheterodonta (Taylor et al., 1973). The Trigonioida relic order could be ruled out, since it is represented nowadays by a single genus, Neotrigonia, with very small shells and living exclusively on the Australian and Tasmanian coasts. Geochemical data, that is stable isotope values of carbon and oxygen, overall indicated that all biominerals studied here were formed in freshwater environments, and δ¹⁸O values in double-buttons tracked the average annual δ¹⁸O values of local precipitations. This excluded Anomalodesmata as potential candidates, since this order of enigmatic, rare and specialised bivalves are strictly marine (Taylor et al., 1973). This finally left only one possibility, Unionoida, the representatives of which are all freshwater bivalves.

Biomolecular analyses showed that the proteome similarity is highest between the double-buttons and the unionoid reference shells. The identification of proteins Hic74, Hic52 and silkmapin in almost all of the archaeological samples confirms the freshwater nacre (Unionoida) origins of the double-buttons. With our current knowledge on shell proteins, these sequences probably represent taxon-specific adaptations for the biomineralization of nacroprismatic structures in unionoid shells: they do not bear any homologues with other shell proteins and are not found in the proteomes of other non-nacreous shell structures characterized here (O. edulis). Furthermore, the analysis of the amino acid substitutions on the Hic74 sequence, recovered from both the Unionoida shells and the ornaments, indicates that Unio or M. margaritifera (not Pseudunio) had been used for making the Havnø ornaments.

Sources of bias in ‘palaeoshellomics’

Except technical bias, inherent to standard proteomics per se and discussed elsewhere (Marin et al., 2016), we identified three potential sources of bias that may hamper, limit or confound the current use of ‘palaeoshellomics’: 1) the intrinsic peculiarities of several shell matrix proteins; 2) the completeness of the dataset used for identification searches; 3) the diagenetic degradation of shell proteins.

Sequence analysis of shell matrix proteins has revealed that a large proportion exhibits unconventional primary structures, with abundant long stretches of repeated residues, the RLCDs/LCDs (repetitive low-complexity domains; Marie et al., 2013). The sequences of such domains are neither taxon-specific nor do they carry phylogenetic information. Because they play a structural role, they are ubiquitous and may be detected in phylogenetically-distant lineages. For example, our proteomic analysis identified many sequences known in marine bivalves such as glycine-rich, insoluble matrix and MSI60-related proteins, but these findings are only supported by RLC domains that can be detected in very different molluscs. More generally, silk fibroin-like domains (poly-Ala), acidic (D-rich or poly-D) or collagen-like repeats (G-X-A triplets) are widespread, ubiquitous and cannot be simply assigned to one given shell protein and/or mollusc genus.

As stated before, protein identification is database-dependent and the quality of the interpretation is proportional to the size of the data set. The list of known shell proteins has dramatically expanded in the past few years, with the use of high-throughput screening (Marin et al., 2016). However, the ‘shellomes’ of only about thirty mollusc genera have been identified, in a phylum that comprises between 80000 and 100000 species. We do not know yet whether the set of sequences at our disposal is a representative sample of the sequence diversity for the whole phylum. In other words, the taxonomic origin of a given archaeological shell sample (such as the double-buttons analysed here) may be indeterminable because extant representatives of the corresponding genus (or of closely related genera) are not yet registered in molecular databases via their genomes, mantle transcriptomes or shellomes. As a consequence, palaeoshellomics, for the time being, will provide information mainly at intermediate taxonomic levels (order, family, sub-family). Nevertheless, the continuous expansion of the dataset of shell protein sequences will increase the power of palaeoshellomics: archaeological shell samples, which are currently unidentified, can be revisited and re-interpreted in the future via in silico investigations alone (i.e. without any further proteomic analysis). Furthermore, the detection of possible amino acid mutations in the sequence of different reference shells, such as Hic74 in this study (Figure 6, Table 3), can be helpful in narrowing the range of candidates and in obtaining more precise taxonomic information.

Diagenesis, that is the slow transformation of sediments and their contents with time, also affects shell biominerals, whether from continental, lacustrine or marine environments. As shell biominerals are organo-mineral composites, diagenesis alters shell proteins too. The diagenetic stability of one given shell protein is a complex phenomenon that depends on several parameters, including its primary structure, its conformation, its localization within the biomineral, the availability of water molecules at the vicinity of the protein, the presence of saccharide moieties. Some shell proteins that could be important for taxonomic determination may be diagenetically unstable, that is easily degraded. As a result, the information they carry would be lost. In the present case, we were fortunate to obtain a consistent set of peptides that do not correspond to LC/RLC domains, i.e. peptides that carry relevant information for protein/clade identification. These unambiguously target Hic74, Hic52 and silkmapin proteins that correspond to freshwater unionoid bivalves. Beside taxonomic determination per se, our data suggest that these proteins (especially Hic74) are diagenetically stable and that they may constitute accurate markers for further studies.

While diagenesis may represent a true source of bias, we were able to accurately evaluate its effects, and we found that the extent of protein degradation (racemization, deamidation) was consistent with the age and burial history of the samples. More specifically, we observed that samples from Havnø were the best preserved - the coverage of the main proteins was high, especially for Hic74 (up to 55%). Surprisingly, we found that the coverage of this protein in Havnø samples is similar to that of modern U. pictorum and Neolithic U. crassus and P. auricularius - indeed, the number of Hic74 peptides in Havnø even surpasses that of the reference unionoid shells, despite the fact that the sample size for the archaeological double-buttons was at least 100 times smaller (Appendix 1, section 3). We assume that this effect is due to early diagenetic changes (such as protein unfolding, loss of linked sugars) that render the protein backbone more accessible to proteolytic enzymes, thus increasing the chance of releasing and identifying peptides. Interestingly, we observed a similar phenomenon in other mineralised systems, for example ostrich eggshell (Demarchi et al., 2016).

Samples from Hornstaad yielded significantly lower coverages (7% and 12% respectively for HorB and HorC) with fewer supporting peptides (6 and 11; Table 2). In one instance (sample HorA), no proteins were identified. This is consistent with the results of the chiral amino acid analysis, which had flagged this sample as compromised and probably burnt, as well as with archaeological evidence for widespread fire destruction of the settlement (Heumüller, 2012). However, the same sample yielded a high number of unidentified peptide sequences (~6000), some of which appeared to be highly acidic and reminiscent of biogenic carbonate-associated proteins (Marin and Luquet, 2008). Furthermore, neither the microstructure nor the mineralogy of the double-buttons from Hornstaad showed any apparent sign of recrystallization to secondary calcite. We therefore hypothesize that the exposure to high temperatures had been relatively moderate, sufficient for inducing protein degradation and/or modification (Asp and Ser decomposition, amino acid racemization, Appendix 1—figure 9) but not high enough to induce mineral conversion, which starts to occur around 300 °C (Yoshioka and Kitano, 1985). Double-button PesB yielded low D/L values but only a modest number of peptides were identified (~100, much less compared to the Havnø samples, where at least 200–400 peptides were matched to known shell protein sequences); this suggests that the sample had not been diagenetically compromised (also supported by the amino acid data, Appendix 1—figure 7, Appendix 1—figure 9).

Freshwater mother-of-pearl: archaeological significance

We found that mother-of-pearl of freshwater origin (Unionoida) was used in three European sites over a wide geographical range but relatively short time span (~4200–3800 BCE). Crucially, the crafters manufacturing such highly-standardized ornaments belong to different cultural groups: Late Mesolithic, Neolithic and Copper Age. Our results settle the ‘marine vs freshwater’ debate (Heumüller, 2012) for the double-buttons from Hornstaad (Borrello and Girod, 2008), and confirm previous identifications for the Peştera Ungurească examples (Girod, 2010a). The use of freshwater nacre (Unio or Margaritifera) comes as a surprise for the Havnø material, a coastal shell midden with a dominance of marine resource exploitation and rich in marine shells perfectly suitable for the purpose of making beads, including the horsemussel M. modiolus. Therefore, this finding suggests that the importance of freshwater mother-of-pearl be re-evaluated.

Unio sp. shells were probably selected to make ‘disc beads’ in the Epipaleolithic of the Levant, at Eynan (Natufian, 10,000–8,000 BCE; Bar-Yosef Mayer, 2013), and in Europe the presence of Unio sp. beads has been recorded 259 times according to the dataset gathered from the literature by Rigaud et al. (2015), mainly from Neolithic sites. Despite this relative frequency of (presumed) freshwater mollusc ornaments in prehistoric Europe, a systematic study of their exploitation as raw materials is almost completely lacking. This is especially surprising since it is known that unionoid shells were exploited for mother-of-pearl until the Middle Ages (Bertin, 2015). Indeed, North American freshwater mussels were the basis for the ‘pearl rush’ during the 19th century, and their overexploitation for pearl harvesting, for making nuclei to be inserted in Pinctada pearl oysters as well as button-making on an industrial scale, almost drove a high number of species to extinction (Haag, 2012).

The lack of comprehensive archaeological studies on freshwater molluscs can be explained by two main factors. The first is methodological: the typical chaîne opératoire of bead-making involves several steps that obliterate most of the anatomical features (e.g. hinge apophysis) that are usable for taxonomic identification. These include: cutting and abrading small pieces until they take a circular shape; perforating the disc (Gurova and Bonsall, 2017) or, in the case of the double-buttons, working the side (with an abrasive wire?) to shape the central groove (Bertin, 2015; Borrello and Girod, 2008). Our work provides a series of analytical tools for overcoming this issue and for determining the biological origin of the raw material.

A second, perhaps more relevant, factor is the long-standing perception that freshwater molluscs are inherently less ‘prestigious’ than marine species, because of their presumed local origin. However, marine and freshwater molluscs are used side-by-side in a number of instances, for example the high-status burials at Mulhouse-Est (Bonnardin, 2009), or complex parures from the Swiss Early Bronze Age (Borrello and Girod, 2008). This clearly demonstrates that both were held in the same ‘esteem’ by the craftsman and that her/his choice was dictated by reasons other than the ‘exoticism’ of the material. The use of freshwater mother-of-pearl at Hornstaad and Peştera Ungurească, two sites with large procurement networks of exotic raw materials and at which there are clear signs of specialised production of ornaments (including gold at Peştera Ungurească; Biagi and Voytek, 2006; Heumüller, 2012), also confirms that freshwater pearl mussels were seen as prized materials, locally available. Furthermore, the use of freshwater molluscs for the manufacture of the doppelknöpfe recovered from Havnø (together with unworked fragments of the shells, Appendix 1—figure 2) shows that the manufacture of these ornaments was consistently associated with the use of freshwater mother-of-pearl, even in marine settings. Therefore, the Late Mesolithic people of Jutland and the Neolithic people of central Europe were either exchanging the finished products/raw materials, or the knowledge that the manufacture of the double-buttons required the use of unionoid shells.

Why freshwater nacre?

It is clear that mother-of-pearl (nacre) from freshwater molluscs was a prime material of choice for the manufacture of shell double-buttons. Further investigation of other types of shell ornaments may reveal that this raw material was more frequently selected than previously thought, but in the meanwhile it is necessary to consider the reasons behind this choice.

Unionoids inhabit clean flowing waters (they are occasionally also found in lakes) and are dependent upon the presence of sufficient salmonid fish to carry the larval glochidial stage of the pearl mussel life cycle (IUCN: International Union for Conservation of Nature, 2019 ). It is highly likely that freshwater mussels were collected near the site (as supported by the ‘local’ isotope signatures in this study), and that the procurement of the mussels was not especially difficult (for example, M. margaritifera lives at depths of up to two meters) nor too time-consuming. Therefore, the choice of this material must have been linked to reasons other than its long-distance provenance, the skills involved in procurement, or its rarity; rather, it is more likely a result of the characteristics of the raw material per se (mechanical properties and aesthetic qualities) and its connection to other things, be these in the sensory world (the river and its water, the landscape) or in the symbolic.

Mother-of-pearl is exceptionally hard - a thousand times more resistant to fractures than its mineral alone (Currey, 1977; Jackson et al., 1990) - and unionoid shells (Unio and Margaritifera sp., but not Anodonta) typically have a rather thick layer of nacre, while in some of the marine molluscs (particularly those occurring in European waters, such as Modiolus sp.), the ratio between nacre and prisms shifts, favouring the latter, where the nacre only partly covers the inner surface of the shell. The preservation of the prismatic matte layer may indicate that the coloured periostracum, which can give an appealing effect of chromatic contrast, was deliberately kept, for aesthetic reasons. Alternatively, if the periostracum was removed by mild abrasion, this would have resulted in fully white ornaments, showing both the brilliance of nacre and the dullness of prisms. The white colour of the ornaments may have been associated to wellbeing, peace and fertility (Trubitt, 2003). White was certainly a sought-after effect, so much so that red-purple Spondylus shells were often worked in order to remove the striking hue and reveal the white underneath (Borrello and Micheli, 2011). At the same time, the gloss of mother-of-pearl has been linked, in historical periods, with spirituality, life, royalty, and pearl fishing is a tradition that dates back to the same period considered here, around or a few centuries before 5500 BCE, in the Arabian Peninsula (Charpentier et al., 2012).

The choice of the raw material could also be a reflection of the role of freshwater environments: the Neolithic is the period in which water, together with plants and animals, is ‘domesticated’ (Garfinkel et al., 2006; Mithen, 2010). Rivers provided fast access routes to Central, Western and Northern Europe for hunter-gatherers during the Palaeolithic and Mesolithic and, later on, for agriculturalists coming from the East (Rowley-Conwy, 2011). Despite their ‘fluidity’, rivers and lakes were meaningful and persistent places in the prehistoric landscape.

In summary, the streams, rivers and lakes near occupation sites were inhabited by organisms that provided the crafters with exceptional-quality raw material, easy to procure and which could be worked following a well-established chaîne opératoire in order to obtain a standardized result. The small white double-buttons could then be threaded using the central groove or pressed into the fabric or leather (Kannegaard, 2013). Our work thus highlights an interpretative bias whereby exoticism is considered the primary reason for choice of raw materials, and suggest that local environments held an equally important place in the mind of prehistoric people.

Conclusions

The first application of ‘palaeoshellomics’ has demonstrated that it is possible to recover and identify ancient proteins sequences from mollusc shell, despite significant analytical challenges due to the combined effects of several factors, including low protein concentrations, small samples sizes, diagenesis and database insufficiency (Table 2).

Our molecular data showed that molluscan proteins are similar across the four freshwater taxa we examined (U. pictorum, U. crassus, M. margaritifera, P. auricularius) and differ significantly from the two marine species (O. edulis, M. modiolus; Table 2, Figures 4 and 5). We confirmed that freshwater molluscan matrix proteins are characterized by highly repetitive low complexity domains (RLCs). This is consistent with results obtained on other shell taxa, and improves our understanding of the biomineralization mechanisms within these invertebrate systems.

The archaeological double-buttons examined here were all confidently identified as Unionoida, freshwater shells with a thick layer of mother-of-pearl, using a combination of mineralogical, geochemical and biochemical techniques (SEM, FTIR-ATR, oxygen and carbon isotopes, chiral amino acid analyses, palaeoshellomics). The analysis of the sequence of the shell matrix protein Hic74 supports the use of Unio or Margaritifera as the raw material for the three Havnø ornaments (excluding Pseudunio), but lack of coverage of most amino acid modification sites in the reference samples hampered identification to a lower taxonomic level (Figure 6, Table 3).

The high degree of standardization of the ornaments (Figures 1 and 2), as well as the consistent choice of freshwater mother-of-pearl as raw material indicate that, in Europe, between ~ 4200 and~3800 BCE, there was a common notion of the manufacture of the doppelknöpfe, which was shared by different cultural groups: Late Mesolithic (Ertebølle), early Late Neolithic (Hornstaad Group), and Copper Age (Toarte Pastilate/Coţofeni). Our in-depth study therefore puts into question the most commonly accepted interpretations, which privilege the preponderant use of exotic marine shells as prestigious raw materials for the manufacture of prehistoric shell ornaments.

Methods

Non-destructive characterization

Whole beads and fragments of the reference shells were observed using an environmental Scanning Electron Microscope (Hitachi TM1000 Tabletop Microscope). The mineralogy of the beads was identified by infrared spectroscopy in attenuated total reflectance (ATR) mode (FTIR-ATR) (Appendix 1, sections 3.1 and 3.2).

Biogenic carbonate isotopic analyses

Isotopic analysis was carried out on biogenic carbonate to obtain bulk δ¹³C and δ¹⁸O values for the double-buttons. Small amounts of cleaned samples (bleached using concentrated NaOCl (12% w/v) for 48 hr) were analysed using a Delta V Plus mass spectrometer coupled with a Kiel IV carbonate device (ThermoFisher). All steps are detailed in Appendix 1, section 3.3.

Protein analysis

All reference shells and beads were powdered using a clean mortar and pestle and accurately weighed.

Amino acid racemization

~2 mg of powder were selected for each double-button and were bleached for 48 hr using concentrated NaOCl (12% w/v) in order to isolate the intra-crystalline amino acids. The analysis of total hydrolysable amino acids (THAA) was carried out as detailed in Demarchi et al. (2014) and Appendix 1 (section 3.4).

Proteomics (Appendix 1, sections 3.5, 3.6)

Powdered reference shells and double-buttons were bleached using diluted NaOCl (2.6%) for 48 and 3 hr respectively. Demineralization was carried out using cold diluted acetic acid, the resulting solutions were thoroughly desalted, concentrated and lyophilized. All samples were digested using two proteolytic enzymes (trypsin and elastase) in order to maximise sequence coverage. LC-MS/MS analyses were carried out using a nanoflow HPLC instrument (U3000 RSLC Thermo Fisher Scientific) coupled to a Q Exactive Plus mass spectrometer (Thermo Fisher Scientific) for M. modiolus, O. edulis, U. pictorum, M. margaritifera and the Havnø and Hornstaad-Hörnle IA ornaments (MSAP CNRS laboratory, University of Lille); an Ultimate 3000 Dionex nanoHPLC instrument coupled with an Orbitrap Fusion (Thermo Fisher Scientific) mass analyzer was used for the analysis of U. crassus, P. auricularius and for sample PesB (Mass Spectrometry Biomolecules core facility, University of Turin). No systematic difference was detected between the data obtained at the two facilities, which could have potentially affected the identification of the samples.

Bioinformatic analysis was carried out using PEAKS Studio 8.5 (Bioinformatics Solutions Inc, Ma et al., 2003). The thresholds for peptide and protein identification were set as follows: protein false discovery rate (FDR) = 0.5%, protein score −10lgP ≥ 40, unique peptides ≥ 2, de novo sequences scores (ALC%) ≥ 50. The FDR is calculated by the software PEAKS using an approach called decoy fusion (Zhang et al., 2012), whereby target and decoy are concatenated for each protein, rather than searching a target database and a decoy database separately (which can result in FDR underestimation); the effect is that of improving accuracy without impacting on sensitivity. The de novo algorithm derives the peptide sequences from the tandem mass spectra without using a database, and it is therefore suitable for the study of organisms where molecular reference sequences are scarce, such as in the present case. Each amino acid residue in the sequence is given a score (0–99%), indicating how confident the software is for that local identification. The overall confidence of the de novo sequence is calculated as the Average of Local Confidence (ALC) score.

The Molluscan Protein Database used in this study comprised 633061 protein sequences, that is all sequences available on the National Centre for Biotechnology Information (NCBI) repository restricting the taxonomy to Mollusca (fasta database downloaded on 15/02/2018), excluding all common contaminants (cRAP; common Repository of Adventitious Proteins: http://www.thegpm.org/crap/). The ESTs database included 1149,723 expressed sequence tags, also restricting the taxonomy to Mollusca (fasta database downloaded on 15/02/2018) and including cRAP sequences. Unionoida molluscs are poorly represented in these databases: a search for Unionoidae on the NCBI Identical Protein Database retrieves 4562 entries, almost exclusively belonging to soft tissue proteins, for example cytochrome oxidase, NADH dehydrogenase and ATP synthase. Shell matrix-related sequences are few (<35), almost all from the transcriptome of the triangle sail mussel, Hyriopsis cumingii (first released in 2013 [Bai et al., 2013]), including proteins related to shell biomineralization (i.e. upsalin, Pif, calmodulin, hicsilin, hichin, Hic74, Hic52, nasilin, Ca-binding P-glycoprotein, perlucin). The same search for Margaritiferidae yields 207 entries, none of which related to shell matrix proteins.

Data deposition

All the mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository (Vizcaíno et al., 2013) with the data set identifier PXD011985.