Tooth oxygen isotopes reveal Late Bronze Age origin of Mediterranean fish aquaculture and trade

Past fish provenance, exploitation and trade patterns were studied by analyzing phosphate oxygen isotope compositions (δ18OPO4) of gilthead seabream (Sparus aurata) tooth enameloid from archaeological sites across the southern Levant, spanning the entire Holocene. We report the earliest evidence for extensive fish exploitation from the hypersaline Bardawil lagoon on Egypt’s northern Sinai coast, as indicated by distinctively high δ18OPO4 values, which became abundant in the southern Levant, both along the coast and further inland, at least from the Late Bronze Age (3,550–3,200 BP). A period of global, postglacial sea-level stabilization triggered the formation of the Bardawil lagoon, which was intensively exploited and supported a widespread fish trade. This represents the earliest roots of marine proto-aquaculture in Late Holocene coastal domains of the Mediterranean. We demonstrate the potential of large-scale δ18OPO4 analysis of fish teeth to reveal cultural phenomena in antiquity, providing unprecedented insights into past trade patterns.

freshwater and marine fish, probably due to bone diagenesis 20 . Hence other isotopic proxies are required to assess fish provenance.
Longinelli and Nuti's 21 pioneering study on the distribution of the phosphate oxygen isotopes (δ 18 O PO4 ) in fish bioapatite demonstrated that δ 18 O PO4 values are controlled by the water temperature and oxygen isotope composition of the ambient water (δ 18 O Water ) 22,23 . Enameloid of fish teeth is highly resistant to diagenetic alterations 24 , and thus in many cases preserve information regarding the original salinity and temperature 25,26 of their past aquatic habitat (i.e. marine, rivers, lakes, and lagoons) [27][28][29][30] . In closed or semi-closed water bodies with a high degree of evaporation, the δ 18 O Water are enriched in 18 O relative to the seawater that feeds them 23,31 (Fig. S2). Enameloid of fish teeth from such water bodies carry distinctively high δ 18 O PO4 values, which reflect hypersaline habitats of the fish 23,28 . The Bardawil lagoon (Sabkhat al Bardawil; along the northern coast of Sinai, Egypt (Fig. 1, 31°09\N, 33°08\E) is such a water body 23,31 .
This study builds on Sisma-Ventura et al. 28 , where phosphate oxygen isotopes of tooth enameloid of S. aurata were used as a new proxy to identify the provenance of archaeological fish remains from the Iron Age in the south-east Mediterranean. Our aim is to provide a first long-range assessment of the sources, exploitation and trade of S. aurata in the context of Egypto-Levantine inter-regional interaction and commercialism.
We assess the provenance of ancient S. aurata from archaeological layers in the southern Levant by using phosphate oxygen isotope analysis of fish teeth enameloid as a proxy for fish habitat salinity. We analysed the oxygen isotope ratio ( 18 O/ 16 O in the PO 4 group, expressed as δ 18 O PO4 value; see Methods) of enameloid phosphate of the first molariform teeth (n = 100; Table S1) and jawbones (n = 24, Table S1) of this species from a broad range of 12 coastal and inland archaeological sites spanning the Pre-Pottery Neolithic to the Islamic period: ~9,700 BCE to 600 CE (~11,700-1,400 years BP; Fig. 2).
During this time span, southern Levantine societies evolved from hunting-gathering, to sedentary ways of lives, to complex societies, to territorial states, and were intermittently subsumed under the aegis of vast or lesser  Map of archaeological sites in the southern Levant from the beginning of the Pre-Pottery Neolithic (PPN: ~9,700 years BCE) until the Byzantine period (BYZ: 300 to 600 CE), from which S. aurata remains were analysed for their phosphate oxygen isotope composition. Also indicated is the location of the hypersaline Bardawil lagoon along the north coast of Sinai, Egypt, which is the proposed source of S. aurata with high δ 18 32,33 . The fortunes of the Mediterranean southern Levant have always been intertwined with that of its Saharan neighbour in Egypt. Those two different agro-ecological regions were economically interdependent and from the first consolidation of centralized power in Egypt in the 3 rd millennium BCE (the Egyptian Old Kingdom), Egypt episodically controlled the southern Levant, whence it could extract the Mediterranean products required for its subsistence and spiritual life. Interconnections between the two regions, however, were not only dictated by cultural and political factors but by climatic fluctuations in both 34,35 . Traffic between Egypt and the Levant was conducted through marine or terrestrial routes through northern Sinai, which was both affected by it in antiquity and provides archaeological proxies for its intensity today. The main question we asked is whether the distribution of these fish to the Levant was a historically-unique and context-dependant phenomenon or rather of more sustainable nature.

Oxygen Isotopes as Proxy of Fish Aquatic Environment
Bioapatite of fish teeth forms in oxygen isotopic equilibrium with the body fluid at ambient water temperature and δ 18 23,31 , which are recorded in the hard tissues (i.e. teeth) of migratory fish that exploit these habitats 23,[28][29][30] .
The temperatures of the Eastern Mediterranean littoral generally range from 15 °C in late winter (February-March) to 30 °C in summer (July-August) 37 . Variations in East Mediterranean δ 18 O Water are small [36][37][38] , varying between 1.4‰ (February-March) and 1.8‰ (July-August). Therefore, the calculated δ 18 O PO4 values (range: 21.2-24.2‰) for bioapatite forming in isotope equilibrium with seawater 22 (see Methods) in the southeast Mediterranean reflect mostly the seasonal changes in water temperature (Fig. 3). This agrees well with the δ 18 O PO4 range from 21.5‰ to 23.4‰ (n = 18) in teeth of modern S. aurata caught in the southeast Mediterranean littoral zone (Fig. 3). In contrast, S. aurata from the Bardawil lagoon along the Southeast Mediterranean (Sinai) coast display significantly higher δ 18 O PO4 values, between 23.5‰ and 25.4‰ 23 (Fig. 3). Bardawil lagoon Sparidae are adapted to a salinity value as high as 60‰ 23 . The Bardawil instrumental salinity range typically varies between 36.9‰ and 74.5‰, but could temporarily reach higher values between 70‰ and 90‰, at times when the inlets to the sea were closed artificially 39 .
The Bardawil lagoon (Fig. 1) is a large (30 km long, 14 km max. width), shallow (0.3-3 m deep) hypersaline coastal lagoon, separated from the Mediterranean Sea by a narrow sandbar 39 . The Bardawil is connected to the sea via two small natural inlets (Boughaz Zaranik). Water exchange in the lagoon is controlled by Mediterranean Sea tides with a mean height of 50 cm. As a result, it has an elevated salinity level and δ 18 O Water values around 3.7‰ (range: 1.8‰ near the Mediterranean inlet, reflecting inflowing seawater, up to 7.2‰ 23,31 (Fig. S2). The unique environmental conditions of the Bardawil lagoon: shallow, warm (17.3-28.3 °C) and hypersaline water (39.0-74.5‰) provides optimal growth conditions for several species of fish, including S. aurata 11 . Today, juveniles of S. aurata enter the lagoon seeking shelter and food. At about two years of age 11,13 , most fish reach sexual maturity and leave the nursery, migrating back into the open sea where they live in a variety of habitats such as sea grass beds and sandy or rocky bottoms 13,40 .
Fish teeth evolve from the epidermal eruption in the skin of the jaw and are continuously replaced throughout the fish's life cycle 41 . The δ 18 O PO4 values in teeth of modern S. aurata cover nearly the entire seasonal range  23 ; data for southeast Mediterranean fish: Sisma-Ventura et al. 28 (this study), reflecting the season of tooth formation. Note that molariform tooth crown mineralisation seems to have occurred year around. This reference frame was used to infer the habitat of ancient S. aurata.
SCIenTIfIC RepORTs | (2018) 8:14086 | DOI:10.1038/s41598-018-32468-1 of predicted δ 18 O PO4 values for teeth formed in isotopic equilibrium with the southeast Mediterranean and the Bardawil lagoon water in the according temperature and salinity range, respectively (Fig. 3). We, therefore, assume that tooth formation and replacement occurs on a seasonal basis.
The δ 18 O PO4 of fish teeth from the archaeological sites in the southern Levant indicate that S. aurata were caught in two distinct habitats: the southeast Mediterranean littoral characterised by low δ 18  Notably, records from sediment cores of the Levant coast, dated to the Holocene, show no evidence for any hypersaline lagoon, similar to the Bardawil, neither in the dimension, nor in the unique environmental setting of this habitat 42,43 . Therefore, as is the case today, the Bardawil lagoon was the only known source for archaeological S. aurata with hypersaline δ 18 O PO4 values consumed in the Levant. A Bardawil provenance for these specimens from the Late Bronze Age (LBA; ~3,200 BP) to the Byzantine period is further supported by the calculated water temperatures, ranging between 16 and 28 °C (using the mean Bardawil δ 18 O Water of 3.7‰), which agree well with the annual water-temperature range of the present-day Bardawil lagoon 39 . We note that the few hypersaline fish dated to the PPN-EBA period (early to mid-Holocene), were likely caught off the southern Levantine shore, and were not exported from the Bardawil, as proposed for the LBA and onward (Late Holocene). This suggests that short-lived, hypersaline lagoons may have formed along the Levant coast when the rapidly rising sea level flooded low-lying coastal areas during the Early Holocene 43,44 .  (Table S1). Note, different width of the box plots reflects the age ranges of each site (see Table S1 for details). Changes of Eastern Mediterranean Sea level are taken from 44,45 . The sea level stabilised at the present level (±1 m) about 1,600 years BCE 46

Identification of Fishing Grounds/Habitats and Formation of Coastal Lagoons in the Southeast Mediterranean
A key factor in the formation of coastal lagoons in the southeast Mediterranean was the post-glacial stabilisation of the sea level. Over the last 4,000 years sea level stabilised close to its present-day level 44,45 , reaching the current level (±1m) at around 3,620 ± 160 years BP 46 (1,620 BCE; based on optically stimulated luminescence dating of marine sand deposits, overlain by aeolian sand) (Fig. 4A). Sea-level stabilisation allowed the formation of the perennial shallow hypersaline Bardawil lagoon along the northern Sinai coast, due to the establishment of long-shore currents that transported Nile sands which built up blocking sandbars 42,43 .
During the Early Holocene (Pre-Pottery Neolithic; PPN): PPNA; ~9,700 years BCE; 11,700 years BP) δ 18 O PO4 values indicate that S. aurata was captured mainly from southeast Mediterranean waters and to a lesser extent from hypersaline lagoons (Fig. 4A). The latter results are the first proof of the past existence and exploitation of hypersaline coastal lagoons along the eastern Mediterranean coast during the Early Holocene. Due to the sharp rise in sea level at the onset of the post-glacial period the nature of these lagoons remains unknown. Nevertheless, local fishing in Mediterranean littoral waters was previously assumed based on fish remains from the now-submerged PPNC site of Atlit-Yam in northern Israel 47 .
Mid-Holocene (Chalcolithic period and the Early Bronze Age: 4,500-2,500 years BCE; 6,500-4,500 years BP), δ 18 O PO4 values indicate that S. aurata was primarily captured 'locally' , namely along the southeast Mediterranean coast (Table S1) (Figs 3 and 4A). In addition, from the late LBA onwards, both dentary bone and teeth of the same specimens of S. aurata displayed hypersaline isotopic signatures, suggesting that those fish may have spent their entire life cycle in the Bardawil lagoon (Fig. 6).
The δ 18 O PO4 values of tooth-jaw pairs of ancient S. aurata specimens from PPN to Byzantine are positively correlated (n = 24, r 2 = 0.68), yet δ 18 O PO4 values of jawbones are consistently lower than those of tooth enameloid. An average offset (Δ 18 O tooth-jaw ) of 1.8 ± 0.95‰ may reflect systematic differences of bioapatite biomineralisation in bones and enameloid. These bone-tooth δ 18 O PO4 differences suggest that the bone δ 18 O signatures likely represents a mean of the entire lifespan of S. aurata. This further supports that S. aurata spent a significant part of their adult life in the open sea, but continued to exploit hypersaline coastal lagoons as part of their trophic migration, as is the case with modern S. aurata [11][12][13] .
Specimens from the Byzantine period have oxygen isotope signatures typical of hypersaline water both in their teeth and jawbone (Fig. 6). This suggests that these fish lived their full life cycle in the hypersaline Bardawil lagoon. Jawbones from other archaeological periods have lower δ 18 O PO4 values than the teeth from the same individual/jaw, reflecting a marine origin and indicating that these fish lived predominantly in southeast Mediterranean seawater. A few jawbones have values lower than those expected for bioapatite forming in isotope equilibrium with modern southeast Mediterranean seawater. Some alteration of bone, which is known to be more prone to diagenesis than enameloid 24

Fish Body Mass as Indicator for Fishing Intensity
The body size of S. aurata (i.e. total length-TL (cm) and body mass-BM (kg)), can be calculated from the maximum length of the molariform tooth crown 48 . Body size of S. aurata reflects ontogenetic age and can thus be used as a proxy of fish exploitation (Fig. S3). Smaller average fish body size in both archaeological and modern contexts is associated with higher intensity exploitation of their nursery, i.e. the Bardawil lagoon 11,13,29 . During the Holocene, the size pattern of S. aurata clearly changes, exhibiting a decrease and lower range in fish size (i.e. harvesting of younger individuals) for specimens from the hypersaline lagoonal waters of the Bardawil (Table S1;  This trend of reduction in fish body size accords chronologically well with cultural changes and spreads over a millennial time-scale, starting in the LBA, continuing to the Iron Age and to the Byzantine period. Interestingly, we witness the appearance of larger fish only at Tel Dor, where we identified in the past S. aurata captured from the Mediterranean littoral zone 28 . Overall, the reduction in S. aurata mean and maximum body size demonstrates similarity with present day fishery data from the Bardawil lagoon (age group 1-3 years), where S. aurata are intensively exploited due to their abundance in this unique hypersaline nursery 11,13 . The decrease in S. aurata body size, to a range similar to present-day fish from the lagoon 11,13,40 is in line with observations from traditional extensive aquaculture 49,50 .
From the LBA onwards we find evidence for extensive and likely year-round exploitation of S. aurata in the Bardawil lagoon. This unique exploitation pattern continues until the present 10,13 , and agrees well with the definition of traditional extensive aquaculture 49,50 . It is similar, for example, to the Italian 'vallicoltura' or the Egyptian 'hosha' -both representing traditional fish exploitation systems which utilise natural fish traps by taking advantage of the trophic migration of juveniles from the sea into coastal lagoons. These systems therefore capitalise on naturally occurring foods from this highly productive lagoon [11][12][13] . Sparus aurata were traditionally exploited extensively in coastal lagoons, until intensive rearing systems for this fish were developed during the 1980s 51 . Therefore, the cradle of marine aquaculture may be rooted in the Bardawil hypersaline lagoon. For more than 2,000 years it functioned as the major source for Sparus aurata for the Levant. This started not later than the LBA (~1,200 years BCE; 3200 years BP) as extensive harvesting and still continues today as intensive aquafarming 11,13 . A Diachronic Summary: Fish Exploitation in The Context of Egypto-Levantine

Exchanges
In the southern Levant, for the last 50,000 years, S. aurata were exploited by local coastal fishing communities 4,5 . The first evidence for fishing in hypersaline lagoons appears during the PPNB ca. 9,500 years BP, but these remains are scarce and are insufficient to indicate systematic exploitation of hypersaline lagoons during this period. However, the occurrence of fish remains in Neolithic inland sites of the Judean Mountains (central Israel) indicates that the transportation of dry fish from the Mediterranean coast was already established in the Early Holocene 4,52 .
From the LBA period onwards, non-local ("exotic") S. aurata with hypersaline δ 18 O PO4 signatures were imported from the Bardawil lagoon, almost entirely replacing S. aurata caught locally in coastal waters. These results contradict the conventional null hypothesis assuming that in coastal sites of the Levant S. aurata remains will represent local fishing, exhibiting that regardless of the site location (coastal or inland) most of the S. aurata were "exotic" (nonlocal) with δ 18 O PO4 signatures of the Bardawil lagoon.
In archaeological sites across the southern Levant, this fundamental change in fish provenance coincides with a sharp increase in the abundance of exotic Nilotic fish, such as Nile perch (Lates niloticus) and the Nile catfish 6,7 (Table S3). Our results, therefore, provide new evidence for the intensity of Egypto-Canaanite long-distance, inter-regional trade connections in this period, which seem to have even included commercialization of fish from Bardawil lagoon. This pattern, which further intensified in the Iron Age and lasted at least until the Byzantine period, most likely comprised dried S. aurata, as depicted in Egyptian reliefs 53 and as observed from the fragmentation patterns on some of the S. aurata remains recovered in Jerusalem (inland) 54 .

Contextualising the Results Historically
Prior to Classical times, the densest network of archaeological sites in northern Sinai is documented for the LBA, the imperial epoch of Egypt's New Kingdom. In this extremely arid, inhospitable region, meaningful population and economic infrastructure could only be sustained when backed by a centralised power. During the LBA Egypt controlled Canaan 55 and well-documented terrestrial and maritime routes across northern Sinai and along the coast served as the main military and commercial artery between the two regions. Intensive traffic between them is recorded both textually and archaeologically, the latter including dozens of waystations, granaries, reservoirs, etc. 56,57 . The Egyptians called this route The Ways of Horus-the southernmost leg of the famed international trunk road linking Egypt with the Fertile Crescent, better known today as the Via Maris.
It is easy to envisage how the Bardawil fish industry and export emerged and functioned in this context. Although our earliest substantial evidence of S. aurata teeth with Bardawil-like, hypersaline δ 18 O PO4 signatures in Canaan dates to the early 12 th century BCE (at Lachish, an Egyptian administrative centre), the single 14 th century BCE hypersaline specimen found at Tel Rehov may hint at an earlier beginning of this trade during the LBA. Moreover, although we lack data from the MBA (~2,000-1,500 BCE), it is possible that this phenomenon started even earlier. Throughout the MBA, contact between Canaan and Egypt was close, although at the end of this period administrative control was inversed: Canaanites controlled parts of Egypt at this time. Egypto-Canaanite commerce flourished in this period too, though most of it appears to have been conducted via the sea 56 . Dozens of MBA settlements have been surveyed in northern Sinai, especially south of the Bardawil lagoon.
Importantly, our findings demonstrate that despite climatic changes and frequent socio-political, economic and demographic upheavals in both regions, once industry and marketing were in motion, they lasted at least until Byzantine times (i.e. minimally for two millennia), providing a paradigmatic example of a Mediterranean exchange network driven by the diversity and interdependence between ecological micro-regions 58,59 .
As if to support our main claim in this paper, Rabbi Abbahu, a Jewish sage living in 4 th -century CE Caesarea Maritima-a major Southeast Mediterranean harbour city (just south of Tel Taninim, Fig. 2)-declared that: "any fish [brought to the city] must come either from Apamea [in Syria] or from Pelusium [Bardawil's harbour town from the 6 th century BCE until the drying up of the eastern arm of the Nile 60 ]". Our results support Safrai's concomitant assumption that even in coastal markets most of the fish in Roman Palestine were imported and demonstrate that this state of affairs had already been in place many centuries earlier.

Methods
A sample of 100 S. aurata molariform teeth recovered from 12 archaeological sites in the southern Levant was analysed in this study. Identification to species level is based on modern Mediterranean ichthyofauna housed at the University of Haifa reference collection and on OL's personal research collection. Modern S. aurata teeth for this study were obtained from specimens captured offshore in Haifa Bay, Israel 28 . Their phosphate oxygen isotope signatures as well as those of extant S. aurata from the Bardawil lagoon 23 were compared with those of teeth from archaeological S. aurata to assess their past habitats, specifically whether they derive from southeast Mediterranean marine or hypersaline lagoonal waters.
Analytical Methods. The enamel cap (~0.2-0.4 mm thickness) of each tooth was separated from the dentine using a diamond-head micro-dental drill, washed three times with distilled water, and dried overnight at 50 °C. Each sample was crushed and ground to powder using an agate mortar and pestle. Organic matter was removed from the samples soaking the samples in 2% NaOCl overnight. The phosphate fraction of the samples was separated using a method modified after Dettmann et al. 61 and described in detail by Gehler et al. 62 . In summary, approximately 5 mg of pretreated sample powder was dissolved in 0.8 ml 2 M HF and placed on a vibrating table for ca. 12 h. After centrifuging, the supernatant sample solution was separated from the CaF 2 precipitate and transferred to new centrifuge tubes. After neutralising the HF solution with NH 4 OH (25%) in the presence of bromothymol blue as a pH indicator, Ag 3 PO 4 was rapidly precipitated by adding 0.8 ml of 2 M AgNO 3 . Following settling of the Ag 3 PO 4 crystals, the samples were centrifuged and the supernatant solution was removed using a pipette. The Ag 3 PO 4 was then rinsed five times with MilliQ water and dried overnight in an oven at 50 °C. Ag 3 PO 4 aliquots of 0.5 mg were placed into silver capsules and analysed in triplicate by means of high temperature reduction using a Finnigan TC-EA coupled via a Conflo III to a Micromass 100 GC-IRMS at the University of Mainz, or to a Finnigan Delta Plus XL GC-IRMS at the Universities of Tübingen and Lausanne, following Vennemann et al. 63  i.e., as the deviation in per mil (‰) relative to Vienna Standard Mean Ocean Water (VSMOW), the international reference material. The δ 18 O PO4 values were measured with an external precision of ±0.3‰ (1 SD). The raw δ 18 O PO4 values were normalised to an Ag 3 PO 4 standard produced by Elemental Microanalysis with a certified value of 21.7‰ (silver phosphate P/N IVA33802207, batch no. 180097, distributed by IVA Analysentechnik, Germany). The analytical precision for this standard was better than ±0.3‰ (1σ). For untreated NIST SRM 120c Florida phosphate rock standard reference material, we obtained a δ 18 O PO4 value of 21.9 ± 0.3‰ (n = 9). This value compares well with the values around 21.7‰ initially measured by Lécuyer et al. 64 and currently reported by most other laboratories as compiled in Chenery et al. 65 .
The δ 18 O PO4 theoretical range of Sparidae bioapatite. We calculated the equilibrium range of Body mass estimation of S. aurata. Body mass of ancient fish can be estimated from species-specific regressions with bone or molariform tooth size. In this study, we used the tooth length measurements recommended for the first molariform tooth (Fig. S1), and regression equations to estimated fish total length (cm) and body mass (kg) 48 : Linear regression calculated from S. aurata first molariform tooth maximum length (FMTL) to fish total length (TL).