Results of sulfur isotope analysis

Human sulfur isotope results are presented in Figure 2 alongside their respective δ¹³C and δ¹⁵N values, and the 1 s.d. range of the average δ³⁴S for terrestrial (coastal and inland herbivores) and aquatic (marine and freshwater) organisms sampled from similar contexts. Full results along with quality indicators can be found in Supplementary Information S2. Majority of samples complied with quality criteria established by Nehlich and Richards (Reference Nehlich and Richards2009), with C:S ratios of 600±300, N:S ratios of 200±100 and %S between 0.15% and 0.35%. Eight samples have %S≤0.14% (minimum of 0.10%) but these were included in the analysis as the other collagen quality indicators were acceptable.

Figure 2. Sulfur stable isotope ratios compared with their respective nitrogen and carbon isotopic ratios by site. St Barbara and Kaberla data are from Aguraiuja-Lätti and Lõugas (Reference Aguraiuja-Lätti and Lõugas2019), carbon and nitrogen isotope ratios of the remaining samples are from Aguraiuja-Lätti and Malve (Reference Aguraiuja-Lätti and Malve2023). Blue samples are from coastal sites, red samples from inland sites. Full symbols are Late Iron Age sites, open symbols Medieval sites and patterned symbols Early Modern sites. Color-shaded areas represent the 1 s.d. range of the group average for coastal and inland herbivores, freshwater fish, and marine organisms (from Aguraiuja-Lätti et al. Reference Aguraiuja-Lätti, Tõrv, Sayle, Lõugas, Rannamäe, Ehrlich, Nuut, Peeters, Oras and Kriiska2022).

Temporal variations in human δ³⁴S values are very moderate, thus we will not discuss differences between periods further but rather focus on spatial variations, i.e., between coastal and inland sites. The results demonstrate that similarly to terrestrial faunal reference values, coastal humans display the lowest δ³⁴S values (ca. +5.6±2.6‰ 1 s.d.) but also the largest range (‒1.9‒14.0‰). Humans from inland sites have average values of 9.2±1.3‰, ranging from +3.7‒11.1‰. There is a statistically significant difference between the two groups (Mann‒Whitney U test, U=6412, p≤0.0001). Figure 2 shows that inland populations are also visually well distinguished from coastal communities, characterized by a combination of low(er) carbon isotopic values and high sulfur isotopic values. This is also indicated by the small yet statistically significant negative correlation between δ¹³C and δ³⁴S values (Pearson’s r = ‒0.428, p<0.001) of all sampled humans, i.e., as δ¹³C decreases, δ³⁴S increases, following a coastal‒inland trajectory.

Terrestrial baselines are generally in agreement with human δ³⁴S values for both regions, although many individuals have values that are considerably higher than the 1 s.d. range of their local terrestrial average (Figure 2). For inland humans, consumption of aquatic resources with higher δ³⁴S values could be considered as a possible explanation, although it is admittedly difficult to differentiate between marine and freshwater fish due to their overlapping δ³⁴S ranges. While inland humans tend to have lower δ¹³C and higher δ³⁴S values—similar to some freshwater fish—their relatively low δ¹⁵N values suggest that their diet was instead predominantly terrestrial.

For coastal populations, high human δ³⁴S values (and the observed much larger range) may be influenced by several factors. For example, the sea-spray effect has likely affected local terrestrial baselines to a certain degree as evidenced by some coastal herbivores from Estonia having δ³⁴S values of up to +10‰ (Aguraiuja-Lätti et al. Reference Aguraiuja-Lätti, Tõrv, Sayle, Lõugas, Rannamäe, Ehrlich, Nuut, Peeters, Oras and Kriiska2022). Humans consuming various terrestrial animals, some affected by the sea-spray and others not, could thus display a range of δ³⁴S values somewhere between the two.

Additionally, consumption of omnivorous animals (e.g., pigs) who feed on aquatic resources could also result in high human δ³⁴S values, although coastal omnivores have only somewhat higher average values (+3.4‰) compared to herbivores from similar contexts (+2.3‰) (Aguraiuja-Lätti et al. Reference Aguraiuja-Lätti, Tõrv, Sayle, Lõugas, Rannamäe, Ehrlich, Nuut, Peeters, Oras and Kriiska2022), indicating that both groups had a predominantly terrestrial diet. Considering there exists a statistically significant positive correlation between nitrogen and sulfur isotopic ratios of coastal humans (r=0.316, p<0.001), elevated δ³⁴S values are more likely caused by consumption of aquatic resources, which have both high δ¹⁵N and δ³⁴S values. The slightly elevated δ¹³C values of coastal populations compared to inlanders suggests their sulfur isotope ratios have been influenced by marine and not freshwater resources. No similar (statistically significant) distinction in coastal vs. inland δ¹³C (or δ¹⁵N) values is evident in terrestrial animal (including omnivore) values from coeval sites (Aguraiuja-Lätti et al. Reference Aguraiuja-Lätti, Tõrv, Sayle, Lõugas, Rannamäe, Ehrlich, Nuut, Peeters, Oras and Kriiska2022), implying that diet (most likely marine resource consumption), rather than the isotopic baseline, is the primary factor behind the observed differences between coastal and inland human populations.

These results demonstrate that at least in coastal regions, a combination of high δ³⁴S and high δ¹⁵N can be used (with some reservation) to detect aquatic resource consumption. This creates the assumption that there could be notable reliance on aquatic resources for some coastal individuals, and consequently this needs to be taken into account when dealing with radiocarbon dates of individuals from this temporo-spatial context.

Quantitative dietary reconstruction

The Bayesian statistical program FRUITS (Food Reconstruction Using Transferred Isotopic Signals, beta 3.0; Fernandes et al. Reference Fernandes, Millard, Brabec, Nadeau and Grootes2014) was employed to provide quantitative dietary reconstruction for a sub-sample of 29 individuals (all from coastal sites) that were radiocarbon dated (Table 1). FRUITS calculates probability estimates of the relative contribution of multiple food sources both to diet as a whole and to individual dietary proxies of δ¹³C, δ¹⁵N, and δ³⁴S, considering consumer’s isotope values and those of the different food groups, in addition to factors such as differences in protein content between dietary resources and uncertainties in trophic level offsets and food-source isotope values (see Supplementary Information S3 for further details).

Table 1. List of individuals radiocarbon dated as part of this study alongside their ¹⁴C ages, dietary isotope ratios, FRUITS estimates, and calibrated age ranges. Stable isotope ratios of St Barbara (SB) and Kaberla (KB) individuals are previously published in Aguraiuja-Lätti and Lõugas (Reference Aguraiuja-Lätti and Lõugas2019), δ¹³C and δ¹⁵N values of the rest of the individuals published in Aguraiuja-Lätti and Malve (Reference Aguraiuja-Lätti and Malve2023). “Fish (%) from ¹³C” shows the average estimated contribution (with ±1 s.d.) of fish to the ¹³C content of the sample as modeled by FRUITS; the second scenario was run excluding δ³⁴S as a dietary proxy (“w/o ³⁴S”). Three calibrated age ranges are compared: uncorrected dates using only IntCal20, RE corrected dates using FRUITS estimates from the first scenario, and RE corrected dates using FRUITS estimates from the second scenario without δ³⁴S. Calibrated age ranges are shown at 95.4% probability. See Supplementary File S3 for more details on RE correction.

Three isotopically distinct food groups are employed in the model (plants, animals, fish) to reflect the main staples of the medieval diet (i.e., bread, meat, fish) (Põltsam-Jürjo Reference Põltsam-Jürjo2013). Reference values for Animals include the main species consumed, e.g., cattle, goat, sheep, pig and chicken. Although herbivores and omnivores differ in their δ¹⁵N values, we have included them as one group to simplify the model (see also S3). For Fish, reference values for marine and brackish water (i.e., freshwater species living in the brackish coastal waters of the Baltic Sea) fish are used from Aguraiuja-Lätti et al. (Reference Aguraiuja-Lätti, Tõrv, Sayle, Lõugas, Rannamäe, Ehrlich, Nuut, Peeters, Oras and Kriiska2022). Although freshwater species are numerous in zooarchaeological deposits of this period (outnumbering those of marine species in inland sites, c.f., Lõugas and Aguraiuja-Lätti Reference Lõugas and Aguraiuja-Lätti2023), they are not included as a separate food source (i.e., with very low δ¹³C values indicative of origin in “inland” rivers and lakes). This is in line with criticism brought up by Schulting et al. (Reference Schulting, MacDonald and Richards2023), who assert that if a food group is included in FRUITS, the model will never estimate its importance as zero, even if the consumption is highly improbable, for example, in the case of “inland” fish being consumed at rural coastal villages.

Figure 3 reflects the average estimates produced by FRUITS on the contribution of the Fish food group to the ¹³C composition of the sample. FRUITS generally provides estimates on the caloric contribution of a food group to diet as a whole, but to assess radiocarbon reservoir effects we are specifically interested in where the dietary carbon (¹³C) is coming from. To evaluate the advantage of using δ³⁴S as a proxy for aquatic resource consumption, the model was run twice, the second time omitting δ³⁴S from analysis (i.e., using only δ¹³C and δ¹⁵N). In most cases, the average estimated fish contribution between the two scenarios only differs by a few percentages (on average 2.8%), implying some redundancy of information, especially considering the correlation between δ³⁴S and δ¹⁵N. The estimates differ the most for those individuals whose δ³⁴S values stray from the average regional terrestrial baseline (e.g., δ³⁴S of <1‰ and >7‰). Unfortunately, we cannot typically rule out whether these “outliers” were locals eating different diets or recent migrants originating from areas with divergent baseline values (but see discussion further below).

Figure 3. Estimated average contributions of fish to the ¹³C composition of each radiocarbon dated individual modeled by FRUITS. A comparison between the two scenarios is shown: utilizing all three dietary proxies (δ¹³C, δ¹⁵N, δ³⁴S; black columns) vs. only two (δ¹³C, δ¹⁵N; diagonally striped columns). Sample numbers correspond to those shown in Table 1.

Radiocarbon measurements and chronological modeling

Radiocarbon measurements have been calibrated with OxCal using the “Mix_Curves” function with a defined mix of terrestrial and aquatic components corresponding to the range of estimated Fish contribution to ¹³C and a reservoir equivalent to the estimated local DRE for the aquatic component. The appropriate DRE in aquatic resources cannot be directly estimated from available data. According to Lougheed et al. (Reference Lougheed, Filipsson and Snowball2013, their Fig 6), the waters of the Baltic Sea around Estonia have an estimated reservoir age of about 225±75 ¹⁴C years. But Lougheed and colleagues also documented that some coastal regions are influenced by hard-water from land areas dominated by limestone bedrock or aquifers (which is the case for most of the northern and western coastal regions in Estonia), which can contribute much older carbon, with a reservoir age of up to 1000 ¹⁴C years. Large reservoir ages have also been estimated for some Estonian inland bodies of water (see Tõrv Reference Tõrv2018). As such, we may need to consider a much greater reservoir age (about 500±100 ¹⁴C years) for local Baltic Sea fish that takes into account possible hard-water effects in coastal areas. While we tested both the lower and the upper range of the estimated local DRE (Supplementary File S3), in the final analysis only an average of 300±100 is used for chronological modeling (Table 1). This estimate is also similar to that proposed by Pesonen et al. (Reference Pesonen, Oinonen, Carpelan and Onkamo2012) as the average Baltic Sea RE. Since testing and refining the DRE is not the topic of this paper, the presented results should be taken as approximations.

Niguliste Street, Tallinn

At the end of the Iron Age (12‒13th century AD), Tallinn was a proto-settlement centered around a hill-fort, until 1219 the Danish led a crusade against northern Estonia, conquering the site and rebuilding it as a freshly converted Christian settlement. The church of St Nicholas (Niguliste in Estonian) was one of the two earliest and most important churches in the Old Town of Tallinn. During earthwork, a triple grave was found underneath the street next to the church, with burial inventory characteristic of native (i.e., pre-conquest) Late Iron Age inhumations (Talvar Reference Talvar2000). The grave consisted of a non-adult, an adult female (NIG2) and an adult male (NIG3).

While the woman and the child have very similar isotope values (ave. ‒19.8‰, +9.3‰ and +3.9‰ for δ¹³C, δ¹⁵N and δ³⁴S) consistent with a local, terrestrial-based diet, the man showed very different results for δ¹⁵N and δ³⁴S (+12.7‰ and +14.0‰), but not for δ¹³C (‒19.8‰). Individual estimates provided by FRUITS suggest the woman’s diet was 90% terrestrial, whereas the man had a diet very high in animal protein, including an average estimated contribution of about a quarter from fish. Considering the isotope data of the male, it was assumed that he may also display a significant reservoir effect, thus both the male (NIG3) and female (NIG2) from the same grave were radiocarbon dated (Table 1). However, their uncalibrated ¹⁴C ages are statistically the same, as tested by the “R_Combine” function in OxCal (1039‒1206 cal AD; χ²-test: df=1, T=1.4[5% 3.8]).

The woman’s stable isotope ratios do not allow us to assume any significant reservoir effect, thus we can consider her radiocarbon age to be an accurate estimation of the true age of the triple grave. Since the male essentially had the same age, his stable isotope ratios (which were measured from the same collagen extract as was used for ¹⁴C dating) are also unlikely to have been influenced by consumption of resources with a notable reservoir effect. Although there is some evidence concerning freshwater fish with no measurable reservoir effects, e.g., from fishponds or other stagnant water bodies (Dury et al. Reference Dury, Eriksson, Fjellström, Wallerström and Lidén2018; van der Plicht et al. Reference van der Plicht, Kaupová, Velemínský, Smolík, Kučera, Kameník, Havránek, Brůžek, Vellev and Rasmussen2020), we are not aware of any such features to be present around Tallinn at this time. Neither can the male’s very high δ³⁴S value (which is the highest measured δ³⁴S of any human in this whole dataset by a margin of +3.4‰ over the second-highest measurement) be reasonably explained by the sea-spray effect since coastal terrestrial animals sampled from Estonia have not shown δ³⁴S values above +10‰ (Aguraiuja-Lätti et al. Reference Aguraiuja-Lätti, Tõrv, Sayle, Lõugas, Rannamäe, Ehrlich, Nuut, Peeters, Oras and Kriiska2022).

As such, the high δ¹⁵N and δ³⁴S values of the male can be alternatively interpreted as being caused by feeding on omnivore protein and originating from a region with much higher terrestrial sulfur isotope ratios. Since the bone sample was taken from the rib, which has an estimated collagen turnover period of ∼5 years (Manolagos Reference Manolagos2000; Ubelaker et al. Reference Ubelaker, Buchholz and Stewart2006), it can be inferred that the man could have moved to Tallinn shortly before his death—i.e., before his rib collagen had adjusted to the local, much lower δ³⁴S terrestrial baseline. In this instance, it seems that δ³⁴S as a dietary proxy is misleading, although including it in the model has not resulted in conflicting RE corrected calibrated dates (Table 1). Excluding δ³⁴S from the FRUITS model and using a more similar average estimated fish contribution to ¹³C for both NIG2 and NIG3 (9% and 12%, respectively) results in very good agreement among the DRE corrected calibrated dates (1040‒1214 for NIG2 and 1047‒1275 for NIG3). While this range is quite large, it overlaps nicely with the one estimated based on burial inventory (ca. 1150‒1250 AD).

St Catherine’s Church, Tallinn

The St Catherine’s church of the Dominican Friary was situated inside the medieval town walls of Tallinn Old Town. Burials recovered from the ruins of the church nave have been previously the subject of a multi-proxy analysis which demonstrated that they belonged to either monks or foreign merchants, most likely of high social status and having spent their childhood outside Estonia (Lightfoot et al. Reference Lightfoot, Naum, Kadakas and Russow2016). One of the burials (KAT2) was suggested by FRUITS as having consumed a diet typical of the medieval elite, with a strong focus on animal protein, including an average contribution of 34% from fish. This elite burial had the highest δ¹³C and δ¹⁵N values of the whole dataset (‒18.5‰ and +14.1‰, respectively), and a δ³⁴S value which would be consistent with consumption of aquatic resources (+10.6‰), thus it is not surprising that it produced the highest estimated contribution of fish to dietary carbon. While excluding δ³⁴S from FRUITS analysis resulted in a reduction in the average fish contribution, it was still quite high at 27%. The uncorrected age of 1299‒1401 cal AD has thus likely been affected by reservoir effects, but death is unlikely to have occurred later than AD 1531 when the church was burnt down and never rebuilt (Lightfoot et al. Reference Lightfoot, Naum, Kadakas and Russow2016). Using the estimated average fish contribution and a terminus ante quem of 1531, the calibrated RE corrected date for KAT2 is 1308‒1475 AD. Excluding δ³⁴S and using a lower estimate for fish contribution results in a more or less similar corrected age range (1306‒1452 AD).

St Barbara cemetery, Tallinn

The suburban cemetery of St Barbara was situated right outside the Medieval town walls of Tallinn and was used to bury low social status and ethnic Estonians throughout the later Medieval and Early Modern Period, up until 1772 when burying inside towns was officially forbidden and the urban cemeteries closed. The site likely included individuals with diverse backgrounds, and the wide variation in protein consumption of St Barbara burials has been already discussed elsewhere (Aguraiuja-Lätti and Lõugas Reference Aguraiuja-Lätti and Lõugas2019). Four individuals were radiocarbon dated from St Barbara, with varying levels of estimated fish consumption (SB524, SB798, SB969, SB1366). A terminus ante quem of 1772 was set in the OxCal model as the known end date of the cemetery, and the calibrated RE corrected dates fall mostly between 1460 and 1660 AD. Although previously it has been thought that this part of the cemetery may have been used already from the second half of the 14th century (Sokolovski Reference Sokolovski1996), these data indicate a later usage of the cemetery. The inclusion of δ³⁴S as a dietary proxy did not change the outcome of the chronological modeling since the average FRUITS estimates usually differed only by a few percent between the two scenarios.

St John’s Hospital, Tallinn

The St John’s Hospital was a leprosarium in the outskirts of medieval Tallinn, which later became an almshouse for the poor in the Early Modern Period. The associated cemetery was used to bury both the hospital patients and later also the local suburban population up until 1772 (Sokolovski Reference Sokolovski2002). Two individuals with notable pathological conditions were radiocarbon dated: one with venereal syphilis (JS6) and the other with advanced leprosy (JS8); both were likely residents of the hospital. JS6 was calibrated to 1520‒1751 AD and JS8 to 1503‒1667 AD. These ranges are very similar to the ones produced by the scenarios without δ³⁴S. The results for JS6 in particular are compatible with syphilis starting to spread in Estonia at the end of the 15th century (Malve Reference Malve2023).

Kaberla village cemetery

A rural cemetery situated ∼30 km east of Tallinn was in use from the end of the Iron Age up until the 17th century (Aguraiuja-Lätti and Lõugas Reference Aguraiuja-Lätti and Lõugas2019). The site is thought to have been originally a burial place of a local 12th century wealthy family, which later grew into a Christian village cemetery (Selirand Reference Selirand1962). Four burials were selected for radiocarbon dating: KB173 was buried in a wooden coffin but with a rich burial inventory including copper alloy and silver jewelry; KB179 was buried with multiple dress ornaments and decorations typical of the 12‒13th century; KB95 had no grave goods but was buried under limestone slabs; KB180 also had no grave goods. All four had similar carbon and nitrogen stable isotope ratios, with FRUITS estimating on average around 10% contribution of fish to their dietary carbon. The inclusion of δ³⁴S produced more variable estimates for fish contribution, but the differences in the RE corrected date ranges between the two scenarios were negligible for all four.

KB173 and KB179, both with rich burial inventories, and KB95, buried under limestone slabs, all date to between 1040‒1277 cal AD, reflecting the earliest phase of the cemetery. The fourth burial (KB180) is clearly younger, with a RE corrected date between 1330 and 1475 AD, consistent with the Medieval Period.

Valjala churchyard

The burials from Valjala churchyard are associated with one of the oldest stone churches in Estonia, on the largest island of Saaremaa. Based on archaeological context, the burials represent some of the earliest graves at this site, originating from the 12‒13th century (1300±20 AD was set as the terminus ante quem in the OxCal model). Twelve individuals were dated, all producing RE corrected calibrated ages between ca. 1190‒1305 AD, consistent with the erection of the stone church in the 13th century (Mägi et al. Reference Mägi, Malve and Toome2019). FRUITS estimated on average a 12% contribution of fish to dietary carbon; the inclusion of δ³⁴S as a proxy has not significantly affected the results of dietary, nor chronological modeling.

St Michael’s churchyard and probable Gallows’ site, Rakvere

Rakvere is a small town in northern Estonia, situated ca. 20 km from the modern coastline. Once the location of an Iron Age hill-fort, it became an urban settlement following the Livonian Crusades in the 13th century. St Michael’s Church was erected next to the town’s main road in the 15th century, but arguably as early as the 13th century (Malve et al. Reference Malve, Viljat, Rannamäe, Vilumets and Ehrlich2020). Two individuals buried at the churchyard have been radiocarbon dated: a male with signs of healed trauma (RK21) produced a RE calibrated date of 1046‒1264 AD, and another male who likely died in a violent battle (RK47) was dated to 1221‒1300 AD. Both have estimated average fish contribution of around 10% and are otherwise isotopically similar to other Late Iron Age coastal populations such as those from Kaberla. The inclusion of δ³⁴S has not greatly affected the results of dietary or chronological modeling. The RE corrected dates for St Michael’s churchyard seem to confirm that the church was already present in the 13th century.

Nearby St Michael’s Church, outside the borders of the medieval town, a group of burials was also discovered, thought to belong to victims of execution (Malve and Vilumets Reference Malve and Vilumets2020). No grave goods were found with the burials, which were unearthed right underneath the modern road, so dating them through archaeological context was impossible. RV2 was an adult male, who had died through complete decapitation. The RE corrected date is 1305‒1427 AD (or 1304‒1422 AD if δ³⁴S is excluded as a dietary proxy), suggesting that an execution site may have been present here in the Medieval Period.

Lihula burial site

The burial from Lihula (LIH2) in coastal western Estonia is somewhat of an outlier. LIH2 along with a few others from Lihula was included in the original analysis due to the underrepresentation of samples from this region of Estonia. The burials were recovered without any grave goods and from a location not previously associated with a known burial site, so contextual evidence about their chronological age is unavailable. The radiocarbon date for LIH2 is calibrated to the end of the 17th century up until the end of the calibration curve in 1943. It is the youngest of the dated individuals in our dataset, and all three Lihula individuals have some of the lowest reported human δ³⁴S values (ca. ‒0.6‰), whereas terrestrial fauna from western Estonia has thus far shown more elevated values (Aguraiuja-Lätti et al. Reference Aguraiuja-Lätti, Tõrv, Sayle, Lõugas, Rannamäe, Ehrlich, Nuut, Peeters, Oras and Kriiska2022). However, excluding δ³⁴S from FRUITS analysis did not significantly alter the calibrated date. They may have been non-locals, their diet could have been exclusively terrestrial, or this region could have been unaffected by the sea-spray effect in the same way that other coastal sites were. Further work on samples from mainland western Estonia are necessary to provide more context for interpreting the Lihula samples.