Archaeometric perspective on the emergence of brass north of the Alps around the turn of the Era

Ancient brass (aurichalcum) was a valued commodity in the Antiquity, notably because of its gold-like appearance. After mastering brass fabrication using the cementation procedure in the first century BC in the Mediterranean, this material became widely used by the Romans for coins, jewellery and other artefacts. Because of its visual qualities, it is believed that since this period, brass played an important role in diplomatic and economic contacts with indigenous communities, notably Celtic and Germanic tribes north of Danube and west of Rhine. To test this hypothesis, we performed for the first time the advanced statistical multivariate analysis based on chemical composition and lead isotope systematics, coupled with informed typo-chronological categorisation, of a suite of late Iron Age and Early Roman period (first century BC – first century AD) brass and other copper-alloy artefacts from the territory of Bohemia. In order to to discuss their provenance, the results were compared to known contemporary sources of material. The new results for brass artefacts from this early phase of the massive occurrence of Roman aurichalcum in the Barbarian territories point to the ore deposits in the western Mediterranean or the Massif Central area in Gaul, consistent with historical events. These new findings underscore the great economic and political importance of the new and rich mineral resources in the Transalpine Gaul acquired due to Caesar's military campaigns.

Brass is undoubtedly one of the most valued materials in Antiquity. Its high appreciation in the contemporaneous society is also underscored by the written sources, particularly in Pliny the Elder's and Cicero's works 1,2 . Since the discovery of Zn-rich alloys in the material culture of the Early Roman period, several studies have summarised its origins, the technological process of its fabrication-including the cementation-by the Romans and its importance during that period [3][4][5][6][7] . It appears that the widespread distribution of brass is connected with the period of the reign of Augustus and his coinage reform in 23 BC 8 .
The original brass produced by Roman workshops in the first half of the first century BC with very distinctive composition and material properties has been referred to as the aurichalcum 1,4,7 . Using modern analytical tools of chemical composition (XRF, EPMA, AAS, PIXE, ICP-MS), there has been some progress in identifying brass manufacture (possibly from the Roman imports) in the broader area of Europe among the artefacts dated already to the early 60s BC; however, the geological provenance of used ores remained mostly unrevealed 2,[9][10][11][12] . More recently, studies systematically dealing with the Pb isotope compositions of selected materials and artefacts frequently used in the society and for constructions, such as copper [13][14][15][16] , lead 17,18 , and brass 2 , have become available. It has been noted that the provenance analyses of ancient Cu from the Iron Age and later periods might be challenging due to the complexity of the interpretations imposed by numerous and often unknown resources, widespread material mixing, recycling, depletion, or other reasons. However, even if the determination of the exact origin of the artefacts in question proves to be difficult, the provenance studies remain to be a great source of information for the understanding of contemporary socio-economic networks that often are key to understanding the historical events 18-21 . Lead isotope systematics. Lead isotope analysis shows similar results as the PCA (Fig. 5). The most significant variability is observed in the category of local items ('L'), followed by a slightly more homogeneous Norican ('N') group and the imports ('I'). Samples in the 'Western' category ('W') form the tightest cluster, which is also coherent chronologically (phase R B1; Fig. 5a). Their linear trend and its spatial overlap are most similar with the Pb isotope systematics of the ores from the Massif Central (Fig. 6). Almost all samples from the 'LT-brass' category are mutually close and show a significant consistency with the ore deposits in the western Mediterranean (i.e. Iberia) and the Massif Central in France. Sample TRS 003 is offset from the rest of the suite but still plots in the Pb isotope space of the Spanish or French deposits.
All brass artefacts in the 'I' category (imports) are from the phase R A, and all of these samples show a tendency towards more radiogenic 206 Pb/ 204 Pb ratios. A tendency towards less radiogenic 207 Pb/ 204 Pb and 206 Pb/ 204 Pb ratios is observed for items from the chronologically youngest period (R B2). A costume pin sample RIM 001 made of brass with high Zn content (> 15 wt.%) represents an ' outlier' with the lowest 207 Pb/ 204 Pb. A clear tendency of samples with low Zn content (< 5 wt.%), dated either in the earliest or in the latest phase (R A or R B2), towards less radiogenic 207 Pb/ 204 Pb values is apparent (Fig. 5b). The samples from the phase R B1 with high Zn contents appear to be dispersed around 206 Pb/ 204 Pb value of 15.68 and are relatively homogeneous. Lead-rich (~ 5 wt.%) imported vessel has the same Pb isotope systematics as the high-Zn low-Pb items from the phase R B1. A drinking horn fitting with the highest Pb content of 9.5 wt. % plots separately with 207 Pb/ 204 Pb and 206 Pb/ 204 Pb ratios of 18.55 and 15.65, respectively.

Discussion
The original aurichalcum, i.e. the brass produced in Rome, must have contained at least 22-28 wt.% Zn 4 . However, the content of Zn in aurichalcum started to decrease already in the first century AD and later produced aurichalcum further continued to lose its original qualities 62 . It was shown that alloy with the Zn content between 10 and 15 wt.%, that could be produced via a simple dilution process of equal quantities of bronze and the aurichalcum, already yielded its typical golden colour that was in demand among the indigenous communities 1,4,63 . Table 1. Summary of the chronological system employed in the sample categorisation. The generally accepted chronological framework for the late La Tène 12 and the Early Roman period 22,44,54,55 , respectively, were followed. www.nature.com/scientificreports/ www.nature.com/scientificreports/ Those were especially the visual qualities of brass that were favoured for producing costume parts such as brooches, rings, pins and belts made both in Roman and Barbarian cultural environments. However, geochemical data for metal finds from the Barbaricum are still sparse, particularly those of the local (i.e. Germanic) provenance. In parallel to such compositional evolution on the Roman side, lower Zn contents in Barbarian artefacts measured here can be attributed to local mixing and recycling of the imported objects. In addition, a repeating pattern of small contents of Sn was observed randomly occurring in high Zn-brass artefacts throughout the entire period except for the earliest and youngest samples (Table 3).
In order to reveal further details about the manufacturing of brass in the indigenous territories, the assemblage analysed in this study was compared with chronologically and typologically compatible data from published reports. Given the overall purity of the analysed alloys in terms of chemical composition, we can exclude mixing of brass with Sn bronzes and leaded brass/bronze alloys known from the Roman Cu-alloy production for most of our samples with high Zn content 56 . A scenario of a 'melting pot of all Cu-alloys' may thus be definitively excluded. However, mixing of brass with a close chemical fingerprint is still plausible but, in such case, the Zn content would be lowered 4 . In case of artefacts with significantly lower Zn contents (2-9 wt. %), we may assume (i) technological experimenting given the inexperience with new material in the earliest Roman period (R A; Fig. 3; Suppl. Figure 2b), and/or (ii) repeated recycling of various Cu-alloys, including brass, tin-brass, leaded brass etc. The later process leads to a gradual depleting of alloying components with a lower evaporation point, such as Zn and Sn. In our group of samples this seem to occur more towards the end-period (R B2; Fig. 3; Suppl. Figure 2b) and correspond with the contemporary findings from elsewhere within the territories with imported Roman brass 64 . This chemical pattern can also indicate gradual lowering of the quality of the used materials that was also reported in the case of Roman coins from the same period 65 .
Comparison with contemporary assemblages. The comparative dataset of the chemical compositions includes early Roman finds from Bohemia (NAA method 28 ), and Cambodunum (AAS method 66 ), both analysed in the 1990s, brass brooches from the territory of Slovenia (PIXE method 9,10 ), and brass staters from Gaul (FNAA 67 ). Due to the currently leading hypothesis about the Roman brass used for the fabrication of the costume parts in the Barbaricum, chemical data from the brass coinage 65 were used for a more detailed comparison. The latest data for the Roman brass coinage were obtained by PIXE 68 and, unfortunately, do not provide sufficiently accurate results for the comparison. When comparing the earliest brass artefacts from the first century BC, the main alloying components (Zn, Sn, Pb) reveal a cluster of late Iron Age brass staters of VERCA and Vercingetorix CAS series because of their lower mean Zn contents (ca. 12.2 wt.%; Fig. 7 Figure 3. Ternary Pb-Zn-Sn plot for the Bohemian brass samples from this study categorised according to their dating. Zinc content is divided by a factor of ten, whereas the Pb content is multiplied by factor of ten. For categorisation according to the cultural groups see Suppl. www.nature.com/scientificreports/ coins and other brass artefacts of the assemblages mentioned above is probably due to chronology because the 'pure' brass, with high Zn contents (> 15 wt.%), and low contents of Sn and Pb (sum below 0.5 wt.%) appeared no earlier than around the Augustan period. Besides, there is evidence of Zn-rich brass in the second half of the first century BC (phase R A; Suppl. Figure 2, 3). A second PCA with only selected trace elements only (Co, Ni, Sb, Ag) was carried out including the comparative datasets. To avoid inconsistency, the dataset was reduced solely to brooches. This step further enhanced the chronological compatibility among the typological groups of the artefacts. Also, in the archaeological categorisation of the groups, several trends were revealed (Suppl. Figure 4a). Similarly to 'N' samples from Bohemia, also Norican brooches tend to contain more Ag, which is paralleled by higher contents of Sb, thus indicating fahlore copper 69,70 used for their fabrication. The most significant variability in the trace element composition was observed for the local items ('L'). Chronologically speaking, a notable heterogeneity in the chemical composition may be observed in the phase R B1a compared to the following phase R B1b (Suppl. Figure 4b). A specific group of the artefacts of the Norican tradition from the phase R B2 forms a tight cluster in both plots. These findings indicate rather heterogeneous supply patterns of brass in the beginnings of the trade contacts between the early Empire and the Germanic communities, compared to the late Republic and Celtic agglomerations of the second and first century BC on the one hand and late first century AD on the other.
When these results are compared with the Roman copper AES coinage, a weak correlation between the part of the Bohemian samples with elevated Sb and Ag levels (Sb 0.01-0.2 wt. %; Ag 0.001-1.5 wt. %.) and the AES coinage elemental pattern group III (EPG III), characterised also by increased levels of Sb (0.02-0.1 wt. %) and www.nature.com/scientificreports/ Ag (0.03-0.7 wt. %), is apparent 13 . Similar Sb-and Ag-rich copper AES coinage is also specific for the Lyon altar series I (LAS I) AES coinage 71 . The comparison of brass artefacts with > 5 wt.% Zn from Bohemia and Cambodunum 66 with the Roman brass coinage 65 is based on the systematics of Sb and Ag (Fig. 8). It clearly shows the incompatibility between the local brass artefacts and the Roman sestertii. Due to the different analytical approaches to obtain the compositional data, the results must be treated with caution. Also, a significant variability of the Roman metal supply for the coin production [cf. 13 ] should be considered. Therefore, for successful future provenance studies, a targeted archaeometric analysis of the Roman coins is vital.
A brooch with an unusually high Ag content (20.8 wt. %) is, generally speaking, uncommon-even in the context of the broad spectrum of Early Roman finds from Central Europe 29,66 . Nonetheless, there are artefacts known from the contemporary cemeteries to had been manufactured from pure Ag, and particularly this type of brooch (Almgren 24) is the one most frequently fabricated artefacts from precious metals 48,72 . Considering the technologically advanced metallurgy -both Roman and Barbarian -unintentional contamination caused by the accidental use of Ag-rich ore is considered rather implausible. Local manufacturing of Ag-rich brooches by deliberate alloying with Ag thus remains a possible explanation.
Provenance analysis -mineral exploitation and raw resources in the Early Roman period. Lead isotope analysis has become a conventional method for tracing the archaeological artefacts containing Pb to their possible geological origins, i.e. the ores the artefacts were fabricated from 73,74 . The provenance analysis testing the consistency between the samples and the known ore deposits was carried out using a combination of the conventional biplots and the Euclidean (ED) and Mahalanobis (MD) distance algorithms 21 . The same approach was then applied to compare the contemporary bronze and brass assemblages of various cultural backgrounds. The ED algorithm has initially been suggested by Stos 75 as a simple metric to compare how far the point distributions are from one another in a multivariate space defined by individual Pb isotope signals. While the ED algorithm is currently widely used, it is advised to be complemented with MD in which the effects of the shape, scale and trend of the distribution of the data are accounted for 76 . Therefore, the metric can measure the distance from a data point to distribution in the multivariate space and can account for the distance of points as well as for the linear trends in the data and distributions of the data clouds. Plots derived using such an approach ( Fig. 9 www.nature.com/scientificreports/ Suppl. Figure 5) are then used to predict the allocation of the analysed brass artefacts from Bohemia in the comparative datasets. Due to considerable overlaps in the data distributions of different ore deposits, we note that the predictive value of ED + MD can vary among distinctive sources. Because 92% (n = 50) of our samples have a low Pb content (< 1 wt.%), we do not consider the sampled alloys to have been deliberately leaded 56,64,77 . The Pb isotope systematics thus appear to reflect the intrinsic mineralderived lead than can be used for reliable prediction of their provenance. The best level of consistency for most of our samples is observed with polymetallic deposits from the Massif Central ( Fig. 6; Suppl. Figures 5, 6). These results were verified by both the ED and MD algorithms; however, the outliers in the Massif Central ore dataset provided a less pronounced consistency than the standard biplot. There is also a possibility of mixing the sources from various deposits, namely the Mediterranean (Iberia, Sardinia, Macedonia, or Attica) or the Alpine (namely the south-eastern Alps and the Inn Valley; Suppl. Figures 5, 6). The Alpine signal is the most apparent in the 'N' category. British source ores did not come into consideration until AD 43 when Britain came under the Roman control and data were thus omitted for historical reasons.
Additional chemical data from the early Roman imperial Pb artefacts were included in the comparative analysis: the Augustan Pb water pipes from Pompeii [code 'Pb pipes'; 78  Because no Pb isotope data for the artefacts of the 'Barbarian' provenance are available, only the early Roman Imperial datasets could be included in the comparative analysis. Regarding the provenance of Cu, there is an extensive corpus of comparative data from the Roman AES coinage 14,71 and Cu ingots of the Sud-Lavezzi 2 Bonifacio wreck from the beginning of the first century AD 15 . Finally, since we aimed to reveal a possible consistency between the Germanic finds and the preceding late Iron Age artefacts, bronze artefacts mainly from the first century BC ('Oppida' set) were also included in the comparative dataset 20,21,35 to detect potential looting of the abandoned Celtic oppida by the newly incoming Germanic populations.
The results show that the Pb isotope compositions of most of the Early Roman samples in this study are generally inconsistent with late Iron Age finds (cf. results of ED and MD, Fig. 9; Suppl. Figure 7). A subset of samples, consisting mostly of artefacts dated to the La Tène period or late first and/or second century AD with less radiogenic 207 Pb/ 204 Pb and 206 Pb/ 204 Pb ratios are closer to ore deposits in Germany (Suppl. Figure 6), and are consistent with Roman Cu coins, Cu-Zn, Pb ingots and part of the copper AES coinage from the LAS I. There appears to  www.nature.com/scientificreports/ Figure 9. Euclidean (ED) and Mahalanobis (MD) distances between Pb isotope ratios of various sets of the analysed artefacts from the Early Roman period (red labels) and the comparative datasets of bronze and brass artefacts (coinage, ingots, Iron Age objects from the oppida, Pompeii objects; for references, see main text). The graph shows the density distributions ('density violin plots') according to their distance. Unlike ED, the MD distributions also take into consideration the shape of the data clouds and their trends. The closer to zero the higher the probability of the analytical match. Sources: see description in the main text + this study. www.nature.com/scientificreports/ be a consistency with the preliminary findings of Roman copper/brass alloys from Elsfleth-Hogenkamp dated to the second -third century AD as well, but we note that the report of Merkel lacks analytical data and we cannot make any further conjectures 77 . No analytical match with the Cu bars from Sud-Lavezzi 2 Bonifacio was observed, although the Cu bars might be a very convenient and contemporary source of copper. The samples with low Zn contents ( Fig. 3; Suppl. Figure 2) from the R A and R B2 phases, respectively, combined with their tendency towards less radiogenic 207 Pb/ 204 Pb ratios (Fig. 5b), are still within the range of the ores from the Massif Central. However, three out of four samples from the early phase (R A) show some proximity towards the south-eastern Iberian zone, most compatible with the late La Tène samples 20 . Sample RIM017 with a lower 206 Pb/ 204 Pb ratio contradicting this chronological explanation could equally be dated into the phase R B1 (Tables 2, 3, 4). The LAS I is consistent with a subset of the Bohemian samples in their Pb isotope ratios and Ag contents 71 . A tendency towards less radiogenic 207 Pb/ 204 Pb and 206 Pb/ 204 Pb ratios may result from the influence of the Iberian Massif ( Fig. 6; Suppl. Figure 5, 6). The southern Spanish mines are thought to be most important in the organisation of the Roman Cu supply 14 , which is also evidenced by the most chronologically compatible dataset-the imperial AES copper coinage. Data from the AES coinage are partly inconsistent with Cu ingots, but the variability of Cu sources corresponds well with the suggested complexity of the Cu industry of the Roman Empire 15 . In the case of samples from the phase R B2, i.e. after 43 AD, Pb-Zn deposits from Great Britain may also come into consideration [cf. 2 ]. Since data from this late phase is rather subordinate in number, further historical analysis, such as the trend comparisons with discussed ore deposits, cannot be performed. Collectively, there are three most distinctive analytical matches in terms of possible ore resources. All these scenarios are historically plausible and can be thus discussed further: I. The mixing of Mediterranean sources has been thoroughly discussed for the Pb pipes from Pompeii, which had Pb isotope signature close to the samples from this study ( Fig. 9; Suppl. Figure 7) 78 . The authors interpreted lead from Pompeii as a mixture of Sardinian, Iberian and Laurion ores; however, as recently pointed out 60 , the possibility of the involvement of the Massif Central ores was initially omitted from the discussion. Considering the original 'mixing scenario' , a more satisfying explanation for Pb in the Pompeiian pipes would favour the Cartago Nova deposits with a minor influence from Sardinian ores 60 . II. The Alpine origin of brass is unsupported because of the lack of clear historical evidence of Roman copper or lead mining in this region. There is a partial Pb isotope overlap with the deposits from the Central Alps (Valais) that may be associated with the 'Sallustian' copper, mentioned by Pliny the Elder, and linked to the Haute Savoie (Suppl. Figure 5, 6), which was discussed in the context of the chemical composition of Lyon altar series AES coinage. This explanation, however, was abandoned because of the inconsistency of the LAS coinage with the Pb isotope ratios of given deposits 71 . A subordinate correlation of the chemical composition and Pb isotope compositions of the 'N' category of samples can be considered for geographical reasons, but this consistency is far from being proven. III. According to Leblanc's 79 map, the Pb-Zn mineralisation in the Massif Central is spread from Les Malines to Lyon, where the production of brass is dated from the middle of the first century AD to the beginning of the second century AD 80,81 . The south-eastern part of the Massif Central is rich in various Cu-bearing ore bodies with specific combinations of the trace elements, for example, ophiolites (Ni, Co, Ag), pyrite ores (Ag, Au), Permo-Triassic (As, Pb, Ag), and Hercynian veins (Sb-Ag-Pb) with the Salsigne type mineralisation (As, Bi, Au) 79 . The connection of the polymetallic deposits in the Massif Central 61 with the Roman lead metallurgy has been suggested earlier 59 . To support this argumentation, six samples in this study that are made of Sn-bronze, Pb-bronze or Ag-rich bronze, i.e. without any cementation process possibly taking place, still have their Pb isotope compositions consistent with the Massif Central ores, and we may thus assume that even Cu was extracted in the same region. A specific mining site, consistent with the Pb isotopic data from this study, cannot be assigned because the available data cover the entire Pb isotope diversity of the Massive Central ore deposits 60 . At present, this dataset appears to bear similarity with Pb isotope values from the Les Malines Pb-Zn deposit 61 . Whether the deposits in the Cevénnes area also served as a Cu source remains uncertain 2 .

Consideration of possible contamination.
A particular methodological risk should be considered when comparing brass artefacts with possible Cu ores because Zn ores, as an essential constituent of the produced brass, may also contain trace amounts of Pb 2,6,82,83 . The inclusion of such Pb may then disturb or obscure the Pb isotope signature of the intrinsic Cu source during the cementation process 84 . There is also the uncertainty on how exactly and how much the cementation medium had impacted the trace element composition, which is crucial for the correct interpretation of the chemical composition of the analysed brass artefacts. So far, it is known that at least Fe and As can enter Cu metal during the cementation process 84 . In Roman Imperial workshops, where a very pure Cu was manufactured due to the advanced refining 13 , the risk of contamination could be exceptionally high. Therefore, it must be acknowledged that the Pb isotope signal from the samples may point to the Pb-Zn source ore instead of the Cu ore [cf. 2 ]. Furthermore, the hypothetical contamination during the cementation process could strongly influence the comparison of Roman Cu coins and brass artefacts based on trace elements such as Sb and Ag. These notions, however, require carefully controlled metallurgical experiments. We assume that the cementation process was carried out using Zn in Pb-Zn ore rather than Zn in the form of ZnO, typically developed in furnaces during the pyrotechnological process [cf. 6 ]. The Pb contents in brass samples from this study are significantly higher than those in the LAS I coinage 71 , representing at present the purest available copper from the Massif Central. The possibility of Zn source for Roman brass production was recently further discussed by S. Merkel  www.nature.com/scientificreports/ isotope compositions of the Dossena Zn ores with the oldest (LT) and the youngest artefacts (R B2) in our sample suite with the tendency towards less radiogenic 207 Pb/ 204 Pb and 206 Pb/ 204 Pb ratios can be observed ( Fig. 5a; Suppl. Figure 5, 6). However, the majority of new data from this study are inconsistent with the Zn ores presented in this study. Collectively, we posit that the Massif Central ores still represent the most plausible source of Zn. www.nature.com/scientificreports/ Roman brass production in the Massif Central area. Archaeological evidence for copper mining in the southern Massif Central during the Roman period is still rare; however, two sites in the Cavénes area served as Cu mines during the late first century BC and the early Imperial period 85 . One deposit around Carcassone, exploited during the later Roman Republic (second and first century BC), has also been documented 71,86 . In general, there always is a possibility of missing archaeological evidence of past extraction activities due to medieval and later mining that may have obscured or eradicated the traces of earlier exploitation. Therefore, clear evidence of the Gallic metal supplies is still missing 17 . The best evidence for the early Roman (i.e. Augustan period) mining in the Massif Central is the Pb isotope analysis of the AES coinage of the so-called Lyon altar series I 71 , supposedly originating in the Cévennes part of the Massif Central. Despite the fact that a part of the Roman brass coin production took place in the Lugdunum (Lyon) mint 71 , similarly to Pb ingots of Santa Maria and partially also to Pb pipes from Pompeii, an Iberian origin was initially expected. However, as the chemical analysis of further Lyon altar series coins (the LAS II collection of the AES coinage) indicates, the Gallic production alone might not have been sufficient for the great demand for Cu during the reign of Augustus, and another Cu source (possibly of the Iberian origin) was thus used for this other series of ases 71 . From the historical perspective, there has been a suggestion that copper used for coins of the LAS I may have been the 'Livian' Gallic copper, mentioned by Pliny the Elder, as one that was quickly depleted 71 .
It should be noted that also other Pb objects have been assumed to originate in the Massif Central, including the artefacts found in Germania 87 . Recently, another assemblage of Roman and Byzantine Cu and Cu-alloy coins nummi minimi from the fourth to eighth century AD was found to be consistent with the deposits in the Massif Central, suggesting a long-term mining tradition of local mineral resources 88 . On the other hand, critical notes have also cast some doubt on the Lyon crucibles, pointing out their lack of technical properties 6 . Nevertheless, based on the reasons presented above, we are confident that our data, in fact, indicate the early Roman cementation in the south-eastern part of the Massif Central, and the increasing evidence for the Imperial exploitation of Cu, Zn and Pb in general 59,71 supports the original interpretation of the Lyon crucibles 80,81 .
The origin of most of the Bohemian brass in the Massif Central, and possibly its fabrication directly in Lyon, is not entirely impossible, as it is in accordance with the recent research, regularly pointing out the Gallic production 2,59,71,87 . The consistency of these deposits with the data from an entirely different cultural tradition may appear surprising at first; however, they only underscore the complexity of the socio-economic networks and the organisation of the metal supplies taking place already in the Early Roman period. Given the presence of the mint in Lugdunum (Lyon), a hypothetical origin of brass for the imperial coinage in the Massif Central appears to be very likely and should be verified by further analyses. The proximity of numerous rich ore deposits to Lugdunum was undoubtedly crucial for its economic importance. These indices could have been underestimated before the publication of chemical data from this area that supported the ancient exploitation of local resources. Furthermore, the consistency of the Pb isotope compositions with samples from the late Iron Age may indicate a long-distance distribution of these mineral resources as soon as around the middle of the first century BC, i.e. the time directly around Caesar's military campaigns in Gaul.

Conclusions
The majority of samples from this study were made of high-quality brass, arguably of the Roman origin. The Pb isotope data show a clear consistency with ore deposits in the Massif Central, especially with Pb-Zn deposits near Les Malines. In addition, a high degree of homogeneity of the analysed samples in terms of their Pb isotope ratios probably excludes recycling using significantly different resources. Whether the Massif Central connection is provided by Pb contained in the Cu source (metallic or geological), or is a result of the cementation process, cannot be unambiguously distinguished. Given the high purity of the Roman Cu -known from the Cu ingots and Cu-based coinage -with low Pb levels compared to slightly Pb-enriched brass samples, a Pb isotope signal linked to the Pb-Zn ores is more likely. Nevertheless, as indicated by non-brass samples from our assemblage and local provenance of the AES coins of the first altar series from Lyon 71 , even Cu could come from the same territory.
The archaeological cultural groups used to categorise samples appear to have only a moderate significance in the pattern of trace element composition; even the Pb isotope ratios were not influenced significantly compared to categorisation of the samples according to dating. If we accept the possibility that the Pb isotope compositions refer to lead originating in the cementation medium, the variation in trace element patterns may point to various Cu sources, or it may be a result of some further admixtures. Such fact does not contradict a possible different origin of a given artefact suggested by its typological classifications and refers solely to the material used for its fabrication.
An important message provided by the combined chemical and Pb isotope analysis is the inconsistency of brass artefacts with contemporary brass coinage, the Roman sestertii. However, this finding requires verification by future analysis of a more varied selection of brass coins using the state-of-the-art analytical methods.
The best level of consistency is found among samples with high Zn content from the phase R B1, i.e. a period around the turn of the Era and the following five decades of the first century AD. Already the artefacts from the late Iron Age do not fall outside the range of Pb isotope ratios of the Massif Central deposits. Therefore, it can be assumed that the brass production might have started in the Massif Central as early as around the middle of the first century BC. The existence of Gallic brass coins from the time of Caesar's military campaigns in Gaul supports this hypothesis 67 . Also, there is evidence of large-scale exploitation of Au in the Massif Central that took place already prior to the Roman conquest 89,90 . It is generally accepted that the Romans benefited from the developed tradition of local Gallic mining 91 . The importance of natural resources in the Massif Central for the expanding Roman Empire is underlined by the intensive Fe production around the Montagne Noire area that became significant in the first century AD. According to archaeometric analyses, local Fe was distributed widely via long-distance trade and served as a vital source of material for the Roman army 92 www.nature.com/scientificreports/ The influx of brass to the territories north of the Alps occurred as early as in southern Europe and the Gaul, thus indicating the instant popularity of the new and attractive material. The earliest evidence of brass used in diplomatic contacts with the indigenous populations can be already seen in the late Iron Age. The nature of its distribution mechanisms is hard to evaluate, but in that period, brass was still a rare commodity. Massive-scale and, perhaps more importantly, a regular occurrence is dated no earlier than the Augustan and Tiberian Era. Brass became a ubiquitous yet still highly valued commodity in Germanic society. Its special social status was derived from its distinctive visual qualities and, initially, its exclusive Roman provenance. The level of dependency of the Barbarian society on an external material supply from the Romans appears to be very high. Based on the current scientific evidence, the importance of brass in political relations between the Romans and Barbarians, possibly similar to the role of the silver coins in northern Britain 94 can be assumed. This material with the connotations of prestige and luxury could serve as an effective medium in determining the quality of relations among different Barbarian groups. The 'value to cost-effectiveness ratio' for the material such as brass seems to have played in the Roman favour. The archaeological evidence from other regions beyond Bohemia suggests that a similar strategy of diplomatic contacts may also apply to other territories where early Roman brass artefacts occur (i.e. Slovakia, Poland, Germany) 95 .
Given the sufficient influx of the Roman brass into the Barbarian territories, the recycling has not affected the geochemical properties attributed to the original Roman aurichalcum as much as is observed for materials from the second century AD 66,96 . Only the samples from the earliest phase of the Early Roman period (R A) may have had their Pb isotope ratios influenced by Iberian Cu sources. A specificity of data from the latest phase (R B2) could be explained at this point by exploitation of different deposits in the Massif Central than in the early stages of brass production. Such a hypothesis is also supported by the second century AD brass ingots from the shipwreck of Aléria that share the Pb isotope signature with our samples and is also thought to be produced in the Massif Central. Another explanation brings the recently analysed Zn ores from Northern Italy into consideration as well 77 . Naturally, mixing of various resources in these later stages of Roman brass production is always an important issue for consideration and hopefully will be addressed in future studies on this topic.
The volume of material entering the Germania Magna in the Early Roman period is hard to estimate and represents a research topic on its own. There have been some rough estimations in the work of Becker 97 for the Barbarian territory of the late Roman Germania, which led to an estimated 2.5 tons of material just for brooches. Given the larger dimensions of the Early Roman brooches compared to Late Roman types, plus the overall abundance of the metallic goods in the Early Roman graves, the quantity of consumed material must have been probably higher than that estimated.

Methods
Selected artefacts were drilled to the metal core to avoid the corrosion layers and collect the minimum sufficient amount of material for the chemical and Pb isotope analyses. Due to the small sizes of the artefacts and the high corrosion stage of some, the sample weight varied between 0.01 and 0.05 g. Because of the sample preparation methodology, As contents were not determined.
Samples of drilled-out bronze/brass materials were carefully weighed into pre-cleaned Savillex beakers, dissolved in a mixture of 6 M HCl-7 M HNO 3 (3:1 v/v) with several drops of 23 M HF and placed on a hotplate for 24 h at 50 °C. For the measurements of element abundances, freshly prepared solutions were dried down and re-dissolved in 2% HNO 3 . The abundances of selected elements were determined using an Agilent 7900x inductively coupled plasma mass spectrometer (ICP-MS), housed at the Czech Geological Survey.
The chemical procedures for Pb isolation and purification employed two chromatographic columns. The first step was modified from Pin et al. 98 and used pre-cleaned and pre-conditioned Sr.Spec resin (50-100 mesh; Triskem, France) packed in 0.2 mL columns. Samples were dried down and re-dissolved in 2 M HCl. Lead was eluted with 6 M HCl. The second step employed anion-exchange resin BioRad AG 1 × 8 (100-200 mesh) combined with HCl and HBr as elution media, following the methodology outlined in Romer et al. 99 . The eluted Pb fraction was then dried down and repeatedly re-dissolved with 50 μl 14 M HNO 3 to remove any residual organic material.
Prior to Pb isotope measurements, the dried Pb fractions were re-dissolved in 1 mL 2% HNO 3 and doped with Tl solution (NIST SRM 997; 205 Tl/ 203 Tl = 2.3871). Lead isotope compositions were determined using a Neptune multi-collector ICPMS (ThermoFisher) coupled to an Aridus 2 desolvating unit (Cetac), housed at the Czech Geological Survey, in static mode. Sample analysis followed a conventional standard-sample-standard bracketing protocol in which the SRM-981 reference material solution was run after every unknown sample. Potential 204 Hg isobaric interference on 204 Pb was monitored at mass 202 Hg and corrected by assuming natural Hg isotope ratios ( 202 Hg/ 204 Hg = 4.35). Correction of the measured Pb isotope ratios for mass discrimination utilised a generalised power law and natural isotope composition of Tl 100 . The results were then normalised off-line to the certified values for SRM 981, and the combined statistics for three measurements of each unknown sample were calculated. Data represent the uncertainty-weighted mean of three replicate measurements. Repeat measurements of NBS 981 yielded mean 206 Pb/ 204 Pb, 207 Pb/ 204 Pb and 208 Pb/ 204 Pb ratios of 16.942 ± 0.003, 15.4998 ± 0.0030 and 36.725 ± 0.007 (2SEM, n = 66), respectively.
Samples dated into the La Tène period and comparative Iron Age samples were processed with a slightly different methodology detailed elsewhere 21 .
For consistency in the data evaluation, the ore deposits data were prepared for the comparative analysis by removing multivariate outliers (i.e. those that would significantly affect the quality of the analysis) detected using the Mahalanobis distance 76 . This step is a prerequisite to fitting linear models to the data or using any other method to visualise the trends. By outliers, we understand data points with extreme values regarding the www.nature.com/scientificreports/