Diet of the prehistoric population of Rapa Nui (Easter Island, Chile) shows environmental adaptation and resilience

Abstract Objectives The Rapa Nui “ecocide” narrative questions whether the prehistoric population caused an avoidable ecological disaster through rapid deforestation and over‐exploitation of natural resources. The objective of this study was to characterize prehistoric human diets to shed light on human adaptability and land use in an island environment with limited resources. Materials and methods Materials for this study included human, faunal, and botanical remains from the archaeological sites Anakena and Ahu Tepeu on Rapa Nui, dating from c. 1400 AD to the historic period, and modern reference material. We used bulk carbon and nitrogen isotope analyses and amino acid compound specific isotope analyses (AA‐CSIA) of collagen isolated from prehistoric human and faunal bone, to assess the use of marine versus terrestrial resources and to investigate the underlying baseline values. Similar isotope analyses of archaeological and modern botanical and marine samples were used to characterize the local environment. Results Results of carbon and nitrogen AA‐CSIA independently show that around half the protein in diets from the humans measured came from marine sources; markedly higher than previous estimates. We also observed higher δ15N values in human collagen than could be expected from the local environment. Discussion Our results suggest highly elevated δ15N values could only have come from consumption of crops grown in substantially manipulated soils. These findings strongly suggest that the prehistoric population adapted and exhibited astute environmental awareness in a harsh environment with nutrient poor soils. Our results also have implications for evaluating marine reservoir corrections of radiocarbon dates.


| I N T R O D U C T I O N
Rapa Nui (Easter Island, Chile) is frequently used as an exemplar of human social competition and an avoidable ecological disaster, in which rapid destruction of the native palm forest had devastating consequences for the island's environment and human population (e.g., Diamond, 2005). Recent archaeological research has brought such Malthusian claims into question: the arrival of the Pacific rat (Rattus exulans) shortly after the island's colonization may have extensively contributed to the palm forest's demise (Hunt, 2007) and with the use of fire the island was transformed into an agricultural landscape (Hunt & Lipo, 2006).
Revised chronologies indicate settlement of Rapa Nui centuries later than previously supposed, with evidence for a more balanced use of the environment and a greater degree of human adaptability to a changing ecosystem than the "ecocide" model purports (Hunt & Lipo, 2006;Stevenson et al., 2015). Knowing past diets is crucial for understanding the impacts of human occupation on Rapa Nui. Despite the intrinsic connection between human behavior and the utilization of natural resources in this small and ecologically-constrained island environment, prehistoric diets of the native islanders are still debated and poorly understood. Dietary evidence is spatially and temporally scattered, and its quality is affected by unfavorable preservation and taphonomic transformations; thus diet has been inferred from stable isotopic compositions.
Rapa Nui is a small (171 km 2 ), remote volcanic island located in the south-eastern Pacific initially colonized in the early 13 th century AD (Hunt & Lipo, 2006). Polynesian settlers introduced chicken (Gallus gallus) and the Pacific rat, but the island has no endemic mammals or land birds (Klemmer & Zizka, 1993). Introduced plants for subsistence purposes are mainly taro, sweet potato, and yam, but also ti, bananas, and sugarcane, and these are thought to have been cultivated in prehistoric times. Evidence for this comes from ethnohistorical accounts (e.g., M etraux, 1940;Roggeveen, 1908), surveys of the current flora (Flenley, 1993) and from microfossils in bioarchaeological and paleoecological studies (e.g., Dudgeon & Tromp, 2014;Horrocks et al., 2012aHorrocks et al., , 2012bHorrocks & Wozniak, 2008). Archaeological evidence documents extensive use of lithic mulch in small-scale rock gardens and planting enclosures (manavai) that served to increase soil nutrients, regulate soil conditions, and offset the effects of aridity and strong winds (Ayala-Bradford, Lipo, & Hunt, 2005;Ladefoged et al., 2010). The importance of seafood in prehistoric Rapa Nui diet has been subject to debate. Evidence for marine resource use is typically elusive in the archaeological record on Rapa Nui due to the acidic volcanic soils, yet is demonstrated through abundant fish and marine mammal remains wherever deposits that preserve these kinds of materials are found (Hunt, 2007;Martinsson-Wallin & Wallin, 1999). The presence of bone and stone fishhooks also attest to the creation of tools specifically for acquiring fish (M etraux, 1940). In addition, it has been suggested that microwear patterns on prehistoric human dental enamel are consistent with seafood consumption (Polet, 2011) and petroglyphs could suggest the cultural importance of fishing (Arana, 2014). However, estimates of the relative contribution of seafood to Rapa Nui diets vary considerably; for example while M etraux (1940) considers it to have been less important, Thomson (1891) described seafood as a principal dietary staple.
Apparent variation in the archaeozoological record has been thought to relate to temporal differences and/or social stratification (e.g., Martinsson-Wallin & Wallin, 1999). Existing studies of carbon and nitrogen stable isotope ratios in Rapa Nui prehistoric human collagen (Commendador, Dudgeon, Finney, Fuller, & Esh, 2013;Polet & Bocherens, 2015) suggest that seafood only contributed a minor part of dietary protein, and that the focus of subsistence was on consumption of chickens and rats. Similar patterns have been suggested elsewhere in prehistoric Pacific contexts (Richards, West, Rolett, & Dobney, 2009).
Unlike on Rapa Nui, however, populations in those locations benefited from large introduced domesticates such as pigs and dogs, which could perhaps better satisfy nutritional demands than birds and small rodents.
If marine resources were not utilized to a great extent on Rapa Nui, human subsistence could place substantial strain on other terrestrial resources.
Up to now, all stable isotope studies from Rapa Nui have used bulk collagen d 13 C and d 15 N values related to a reconstructed food web based on measured or hypothesized plant and faunal isotopic end members (Commendador et al., 2013;Fogel, Tuross, Johnson, & Miller, 1997;Polet & Bocherens, 2015). A disadvantage with this method is that it relies heavily on the appropriateness of selected reference values for the local environment. In their study of Rapa Nui bulk collagen isotopic composition, Commendador et al. (2013) used terrestrial end member d 13 C (-20.5&) and d 15 N (6.5&) values derived from published modern data for C 3 plants from Pacific islands other than Rapa Nui, concluding that only a minor proportion of prehistoric diets came from marine resources. Although using modern bulk data is common practice in archaeology, the approach is problematic for Rapa Nui for several reasons. Firstly, significant environmental and land-use changes have taken place on Rapa Nui over the past millennium (Flenley, 1993;Hunt & Lipo, 2011), so most plants grown under current conditions there are unlikely to be representative of prehistoric environments. Only 40% of the current flora on Rapa Nui is indigenous and an estimated 90% of Rapa Nui is now covered by grasslands, with post-contact introduced grasses making up 80% of this Poaceae flora (Finot, Marticorena, Marticorena, Rojas, & Barrera, 2015). This is of particular relevance in isotopic studies as most invasive grasses on Rapa Nui are C 4 plants in the subfamilies Chloridoideae and Panicoidae, which make up 60% of the grass species, whereas indigenous grasses were dominantly C 3 plants (Finot et al., 2015). Secondly, Commendador et al. (2013) showed that the Rapa Nui humans exhibited particularly high d 15 N values, an observation also noted by Fogel et al. (1997). Both studies suggested environmental factors, including the effect of seabird guano, as possible influences on this raised baseline. Szpak (2014) showed that d 15 N values in particular are very sensitive to environmental variation, which demonstrates the importance of using reference values specific to a particular location if possible. Elsewhere archaeologically preserved cultivars have been used as proxies for evidence of past agricultural manipulation (Styring et al., 2015), but no such preserved macroremains suitable for isotope analysis have been found on Rapa Nui.
Here, we use unique carbon and nitrogen compound-specific isotope analysis of amino acids (AA-CSIA) in human and faunal remains to explore the extent of marine resource use in prehistoric Rapa Nui diets and to investigate the drivers of the elevated bulk d 15 N values observed in previous studies. We also reevaluate prehistoric terrestrial end member d 13 C values used to model bulk collagen isotopic compositions to account for possible variations in the local environment.
d 13 C AA-CSIA has been shown to be particularly helpful for distinguishing between marine and terrestrial diets (Corr, Sealy, Horton, & Evershed, 2005). Experiments have shown that dietary essential amino acids (EAAs) are routed to bone collagen with minimal carbon isotope fractionation, and must therefore reflect a consumer's diet (Howland et al., 2003;Jim, Jones, Ambrose, & Evershed, 2006). In ecological contexts, multivariate statistical analyses have been used to separate marine and terrestrial consumer groups using EAA d 13 C "fingerprints" specific to dietary food sources (Larsen et al., 2013). Based on relative patterns between EAA d 13 C (d 13 C EAA ) values, this fingerprinting approach provides an estimate for dietary protein resources that is independent from mass-balance approaches, which must assume absolute d 13 C values for terrestrial and marine end-members. are determined in analysis of consumers. In samples of consumer tissues, "source" amino acids (e.g., phenylalanine) retain the isotopic composition of nitrogen sources at the base of the food web, whereas "trophic" amino acids (e.g., glutamic acid) become 15 N enriched with each trophic transfer (Chikaraishi et al., 2009;Hannides, Popp, Landry, & Graham, 2009;McClelland & Montoya, 2002;McClelland, Holl, & Montoya, 2003;Popp et al., 2007). The magnitude of 15 N enrichment in trophic amino acids is most likely related to kinetic isotope effects associated with C-N bond breakage, and the relative flux of the deamination products for each trophic amino acid (Chikaraishi, Kashiyama, Ogawa, Kitazato, & Ohkouchi, 2007). In contrast, the metabolic route for source amino acids, like phenylalanine, involves little formation or breakage of bonds to N atoms (e.g., Bender, 2007). Thus source amino acid d 15 N values change little during trophic transfer (Chikaraishi et al., 2007(Chikaraishi et al., , 2009. For these reasons, the trophic position of an organism can be determined using the difference in the d 15 N values of glutamic acid (d 15 N glu ) and phenylalanine (d 15 N phe ). The major benefit of this technique is that it allows for direct paleodietary comparison in different temporal and geographical contexts, offering a refined estimate of TPs without requiring a full independent characterization of nitrogen isotopic composition at the base of the food web in the ancient local environment like that required for bulk collagen isotope analyses (Naito, Chikaraishi, Ohkouchi, Drucker, & Bocherens, 2013). This method is particularly relevant for Rapa Nui as it also allows for investigation of the d 15 N baseline.
Our results indicate that marine protein formed a substantial component of the islanders' diet and do not support subsistence based on protein consumption predominantly from rats. The isotopic results further suggest that the crops consumed by the prehistoric population resulted from deliberate and sustained manipulation of infertile agricultural soils, indicating considerable adaptation to a barren island environment. These conclusions not only affect our understanding of resource use on the island, but also have implications for radiometric dating on human and faunal remains. Radiocarbon dating of such materials depends on reliable reconstructions of marine dietary input, to calibrate against marine reservoir effects (MRE) caused by incorporation of marine carbon with considerably older reservoir ages than terrestrial counterparts (Arneborg et al., 1999).

| Archaeological human, faunal, and botanical material
The majority of human (n 5 10) and faunal (n 5 25; rats, birds, fish, marine mammals) bone samples used in this study originate from 1986 to 1988 archaeological investigations at Anakena (Figure 1), carried out as a collaboration between the Kon Tiki Museum (Oslo) and the Museum in Hanga Roa (Rapa Nui), with a further two samples of human bone from excavations directed by Thor Heyerdahl in 1956 (see Skjølsvold, Wallin, & Martinsson-Wallin, 1994). All human bone was sampled as rib fragments or phalanges from adult individuals, and with the exception of the 1956 material, no further information about age or sex was available. In addition to the human and faunal samples from Anakena, one sample of unburnt, preserved totora reed (Schoenoplectus californicus) associated with the burials in Grave 2 from Ahu Tepeu (Heyerdahl, Ferdon, Mulloy, Skj€ olsvold, & Smith, 1962) was analyzed along with a modern Totora sample collected as a part of Thor Heyerdahl's expedition to Rapa Nui in 1956. We also analyzed a preserved palm nut endocarp from Anakena. A summary is given in Table 1, and full list of materials analyzed is given in Supporting Information Table S1.
Permission for destructive analysis of the human remains was granted by the Kon Tiki Museum, in line with original permissions to export and analyze excavated material granted by Consejo de Monumentos Nacionales, Chile, for the excavations in the 1950s and 1980s.
Further ethical approval for the study was granted by the University of Bristol Committee for Research Ethics.
Additional rat bones (n 5 10) were obtained from excavations at Anakena Beach directed by Terry Hunt, Carl Lipo, and Sergio Rapu in 2004to 2005(Hunt & Lipo, 2006. These field studies focused on early deposits located in an area adjacent to Trench K, which formed part of the Kon-Tiki Museum excavations in the 1980s. Excavated material from Anakena Trench K has previously been 14 C dated to between AD1259 and AD1580 on the basis of two charcoal samples from different contexts (Skjølsvold et al., 1994). Dates from the nearby area excavated by Hunt and Lipo (2006) support this range of ages for the deposit. The sample from Ahu Tepeu (RN026) originates from a multiple grave in which one of the individuals has previously been 14 C dated to 330 6 150 BP (Heyerdahl et al., 1962).

| Soils and modern reference materials
Forty one modern soil samples were chosen for bulk carbon and nitrogen isotopic analysis. Twenty nine soil samples were associated with manavai and twelve were associated with rock gardens. Samples were collected along north-south transects across two manavai at Maitaki Te Moa and two at Anakena (for details see Hunt and Lipo, 2011). Thirteen samples were collected within the rock walls of manavai and the remainder outside these structures. Twelve soil samples from the Anamarama region were collected by Vitousek, Chadwick, Hotchkiss, Ladefoged, and Stevenson (2014) along an intensively sampled transect across two rock gardens (n 5 8) separated by a tephra-covered area with few surface rocks (n 5 4). Details of sampling at Anamarama and the soil properties are given in Vitousek et al. (2014). Eight archaeological soil samples were analyzed; five from beneath fallen Moai statues and three from likely habitation contexts at Anakena and Vinapu.
We relied on both modern and archaeological training data for constructing the d 13 C EAA fingerprinting model. These training data comprised of 12 plant food samples (modern), three marine fish  samples (modern), five marine mussels (modern), and six archaeological bone collagen samples from fish, marine mammals, and birds (Table 1 and Supporting Information Table S1).

| Collagen extraction and bulk isotopic analysis
Collagen for stable isotope analysis was extracted from bone samples following a method similar to of M€ uldner and Richards (2005 and are reported in standard d-notation relative to V-PDB and atmospheric N 2, respectively and are reported in Supporting Information Table S2. Accuracy and precision were <0.2&, as determined from multiple laboratory reference materials extensively calibrated using National Institute of Science and Technology reference materials and analyzed every 10 samples. Molar C:N ratios were determined to assess collagen preservation according to the commonly accepted range of 2.9-3.6 (DeNiro, 1985) and samples outside this range were excluded from further analysis.

| Radiocarbon dating
Three samples of human bone from the current study (RN035, RN036, and RN037) were submitted for 14 C AMS dating at the Center for Applied Isotope Studies at the University of Georgia. Collagen for 14 C analysis was extracted from bone samples using 1N HCl, the residue filtered and rinsed with deionized water and collagen dissolved at 808C for 6 hr. Lyophilized isolated collagen was graphitized in evacuated and sealed ampoules containing CuO at 5758C and the cleaned charcoal converted to CO 2 by a similar procedure at 9008C in the presence of CuO. CO 2 was cryogenically purified and then reduced to graphite (Vogel et al., 1984). Graphite 14 C/ 13 C ratios were measured using a CAIS 0.5 MeV accelerator mass spectrometer and normalized to measured values of Oxalic Acid I (NBS SRM 4990) as well as sample 13 C/ 12 C ratios. Measured radiocarbon dates were calibrated using OxCal 4.2 (Ramsey, 2009), with the SHCal13 atmospheric curve (Hogg et al., 2013) and Marine13 marine calibration curve (Reimer et al., 2013).

| Preparation of samples for amino acid isotope analysis
A subset of 19 samples used for bulk collagen isotope analysis was chosen for compound-specific amino acid (AA) d 13 C and d 15 N analysis (AA-CSIA). This included all human samples with adequate collagen preservation, and a representative selection of possible food sources.
In addition, the d 15 N values of amino acids in the archaeological totora reed were also measured.
Samples for AA-CSIA were hydrolyzed and derivatized according to the methods of Popp et al. (2007) and Hannides et al. (2009) for trifluoroacetyl/isopropyl ester derivatives and Larsen et al. (2015) for Nacetylmethyl ester (NACME) derivatives. For preparation of trifluoroacetyl/isopropyl ester derivatives, 5-10 mg of collagen were hydrolyzed using trace-metal grade 6M HCl and the hydrolysate purified using low protein-binding filters and cation exchange chromatography. Purified samples were esterified using 4:1 isopropanol:acetyl chloride and derivatized using 3:1 methylene chloride:trifluoroacetyl anhydride. Finally, trifluoroacetyl/isopropyl ester derivatives were purified using solvent extraction (Ueda et al., 1989) and stored at 2208C for up to two weeks before analysis. Samples were prepared in batches of eight with an additional vial containing a mixture of 15 pure AAs purchased commercially (Sigma Scientific). For the NACME derivatives, 9-15 mg of dried material was hydrolyzed in 6M HCl (analytical grade HCl, Merck, Darmstadt, Germany). The samples were purified using cation exchange chromatography, dried and then methylated with acidified methanol and subsequently acetylated with a mixture of acetic anhydride, triethylamine, and acetone, forming N-acetyl methyl ester derivatives. The esters were stored in ethyl acetate at 2208C until analysis.

| Nitrogen isotope analysis of amino acids
The d 15 N values of trifluoroacetyl/isopropyl ester derivatives of AAs were determined using gas chromatography combustion isotope ratio mass spectrometry (GC-C-IRMS, Hayes, Freeman, Hoham, & Popp, 1990). Specifically, we used an isotope ratio mass spectrometer (IRMS; Thermo Scientific Delta V or MAT 253) interfaced to a gas chromatograph (GC; Thermo Scientific Trace) fitted with a 60 m BPX5 forte column (0.32 mm internal diameter with 1.0 lm film thickness; SGE, Inc.) through a GC-C III combustion furnace (9808C), reduction furnace (6508C), and liquid nitrogen cold trap. Helium (11.4 mL min 21 ) was used as the carrier gas. Immediately before analysis, samples were dried and redissolved in an appropriate volume of ethyl acetate. The analysis consisted of at least 6 injections for each sample, with norleucine and aminoadipic acid internal reference compounds co-injected in each run. The suite of 15 pure amino acids was also analyzed every 3 injections to provide an additional measure of instrument accuracy.
The isotopic composition of all pure amino acid reference compounds were previously measured using the bulk isotope technique described For our samples the correction equation was: d 13 C AA 5 (d 13 C CSIA - (1 -X) 3 d 13 C ISO )/X, where d 13 C AA is the corrected isotope value for the AA of interest, d 13 C CSIA is the isotope value initially determined by GC-C-IRMS, X is the mole fraction of amino acid C in each AA (vs. derivative C), and d 13 C ISO is the isotope value of the isopropanol added to each AA during esterification (the correction factor). Additionally, norleucine and aminoadipic acid reference compounds prepared with each sample batch were co-injected with all samples and the reference suite of compounds analyzed, their d 13 C values corrected using the above equation, and the results analyzed to establish instrument accuracy. For replicate collagen injections, average d 13 C AA standard deviations were 0.38& and ranged from 0.02&-1.37& (Supporting Information  (Howland et al., 2003;Jim et al., 2006).
We based our analysis on d 13 C values of Leu, Lys, Phe and Val because they have previously been identified as the most informative EAA for distinguishing between marine and terrestrial food sources (Arthur et al., 2014;Larsen et al., 2013). We used samples that clustered nearby one another to define potential food groups. These food groups were used to calculate relative contributions of marine and terrestrial proteins to consumers with the software FRUITS version 2.0 (Fernandes, Millard, Brabec, Nadeau, & Grootes, 2014). For the groups with plant-based proteins, we mostly included the most calorie dense plant foods; bananas, starchy vegetables and palm seed (Supporting Information Table S1). FRUITS is executed with BUGS, which is a software package for performing "Bayesian inference Using Gibbs Sampling" that includes an expert system for determining an appropriate Markov chain Monte Carlo scheme based on the Gibbs sampling.

| Results of d 15 N AA-CSIA
The nitrogen isotopic composition of amino acids was analyzed in 20 samples (Supporting Information Table S3). We targeted analysis of 13 amino acids, which we consistently measured in many samples.  and terrestrial plants and birds (-24.2&).
We applied d 13 C EAA fingerprinting to determine proportional contribution of terrestrial vs. marine EAA sources. We relied on both modern and archaeological training data for constructing the d 13 C EAA fingerprinting model. These training data comprised ten plant food samples (modern), three marine fish samples (modern), five marine mussels (modern), and six archaeological bone collagen samples from fish, marine mammals, and birds (Supporting Information Tables S1 and   S4). Our PCA with mean-centered d 13 C values of Leu, Lys, Phe and Val revealed that the training data clustered in four distinct groups; two marine and two terrestrial groups. Marine-I samples comprised of fish, marine mammal, and seabird, Marine-II of mussels, Plant-I of yam, common purslane, palm seed, and leaves and tubers of sweet potatoes, and   Commendador et al., 2013Commendador et al., , 2014Polet & Bocherens, 2015). In order to compare our compound-specific techniques described below to this approach, we first estimated the proportion of marine protein in human diets using bulk d 13 C data in a simple linear marine-terrestrial two-component mixing model (Phillips & Gregg, 2001), with a re-evaluation of the most appropriate end mem-  Table   S2. 213.1 6 2.1&, n 5 9; Commendador et al. (2013) 212.2 6 1.3&, n 5 43 data digitized from their Figure 4).
Establishing a representative terrestrial end member d 13 C value for Rapa Nui is more problematic. Richards et al. (2009) (Farquhar, Ehleringer, & Hubick, 1989). Increased soil moisture enhances leaf conductance that increases the partial pressure of CO 2 within a leaf resulting in lower plant d 13 C values (Farquhar et al., 1989).  Tables S1 and S2) suggest at least some contribution from C4 plants regardless where they were collected, most likely due to a high proportion of modern invasive C4 grasses. Edwards and Still (2008) found that the distribution of C4 grasses on Hawaii correlated with areas of lower mean annual precipitation compared to their C3 counterparts. The lower proportion of C4 plants in modern soils within manavai and rock garden soils inferred from d 13 C values therefore is consistent with higher soil moisture contents in agricultural soils that affect C3/C4 plant ratios (Edwards & Still, 2008;Ehleringer, Cerling, & Helliker, 1997). Historical land use changes, such as extensive sheep farming from ca. 1888 to 1953 (Flenley, 1993;Hunt & Lipo, 2011), has extensively altered the flora on Rapa Nui (Finot et al., 2015). Nevertheless, our data from modern soils suggest that plants grown in nonagricultural soils can yield different isotopic results from those grown in agricultural soils and that these differences are driven by soil moisture content.
No macroremains of archaeologically preserved cultivars suitable for isotopic analysis have been found on Rapa Nui, and the only preserved archaeobotanical samples available for this study were totora reeds and palm nut endocarps. The totora is a perennial sedge known to have been of cultural importance on Rapa Nui and in South America.
Totora typically grow in muddy soils in freshwater marsh and riparian areas but is also found in cismontane and occasionally in desert environments (Munz, 1974

| 351
All archaeological soils and the prehistoric palm nut endocarp are enriched in 13 C relative to the totora reeds analyzed, which is the relationship expected from nonagricultural soils or plants growing in welldrained soil (palm) compared to plants grown in moist agricultural soil (totora, and presumably taro, sweet potato and yam). Importantly, using the mean d 13 C values for archaeological soils corrected by 5& for the diet-to-collagen isotope fractionation in our calculations results in f marine < 0 for all ancient Rapa Nui humans, which violates mass balance. A similar calculation using the d 13 C values for the palm nut endocarp as a terrestrial end member results in f marine < 0 for 95% of ancient Rapa Nui humans. We therefore use the archaeological totora reed d 13 C value as our best estimate for our terrestrial end member, assuming a conservative uncertainty of 62& and the commonly accepted value of 5& as the 13 C enrichment of collagen relative to plant materials (Ambrose et al., 1997; but see also below).
Our two-component linear bulk isotope mixing model is similar to previous approaches in the literature where only bulk isotope data has been available. Using our marine (-12.4 6 1.5&) and terrestrial (-26.6 6 2.0&; 221.6& when corrected for diet-to-collagen isotope fractionation) end members to calculate f marine , suggests that the prehistoric Rapa Nui humans obtained 35% (range 7-67%) of their food from marine sources with a propagated uncertainty of 5%. It should be noted that this estimate is not only dependent upon our chosen appropriate marine and terrestrial end member d 13 C values, but is also dependent upon assumptions about the magnitude of diet-collagen carbon isotope fractionation. The commonly used 5& diet-to-collagen fractionation value we assumed has been shown to vary by as much as 12.3& depending on diet type and quality (Ambrose et al., 1997;Howland et al., 2003).
However, we have no evidence for variation in the diet-to-collagen fractionation value on Rapa Nui. Additionally, bulk collagen may be biased towards the protein part of the diet, which could also affect estimates of marine consumption. For example, in a study of prehistoric diets from the southern Mariana Islands, ƒ marine estimates based on collagen d 13 C values were considerably lower than those determined from carbon isotope analysis of bone apatite, thought to represent whole diet d 13 C values, in two out of the three islands studied (Ambrose et al., 1997). A number of controlled feeding experiments have yielded similar results.
Recently, Webb et al. (2017) showed that in pigs fed on diets with different amounts of marine protein, there was considerable variation in the relative difference between the whole diet and tissue d 13 C values with even just a small change in marine input. Significantly, it was shown that this variation occurred on the individual amino acid level through differences in synthesis and routing of e.g., glycine. As a result, the authors concluded that bulk isotope values in e.g., collagen may significantly mask or even misrepresent the actual resources consumed in mixed marine/terrestrial diets.
In summary, even with the caveats outlined above, the results of bulk collagen d 13 C values suggest that seafood comprised approxi-  EAAs. It is therefore independent from the bulk collagen d 13 C mass balance described above and results in different estimates of marine resource usage. Both our d 13 C EAA and bulk collagen isotope results support the interpretation that marine resources were more important for the subsistence of prehistoric Rapa Nui people than previously assumed (Commendador et al., 2013;Polet & Bocherens, 2015 where d 15 N glu and d 15 N phe are the isotopic composition of glutamic acid and phenylalanine in a consumer, respectively, and D glu-phe is the extent of 15 N enrichment in glutamic acid relative to phenylalanine in the consumer with each trophic transfer, assumed here to be 7.6& for mammals (Naito et al., 2010), although recent work has shown that D gluphe values of marine organisms may be up to 1.5& lower (Bradley et al., 2015;Nielsen et al., 2015). b marine defines the 15 N enrichment in glutamic acid relative to phenylalanine in aquatic photoautotrophs, and is taken here as 23.4& (Bradley et al., 2015;Chikaraishi et al., 2009;Nielsen et al., 2015).
A similar quantification (Equation 2) can be used to determine the trophic position of organisms feeding on C 3 vascular plants (TP terrestrial , Chikaraishi, Ogawa, Doi, & Ohkouchi, 2011).
In this equation, the value for b terrestrial is 8.4& (Chikaraishi et al., 2011), where b terrestrial defines the 15 N content in glutamic acid relative to phenylalanine in terrestrial (i.e., vascular C3) plants. d 15 N glu , d 15 N phe and D glu-phe are as defined above. For C 4 plants, the b value is significantly lower than 8.4& (Chikaraishi et al., 2011). However, as noted historical and archaeological evidence indicates that the main crops grown on prehistoric Rapa Nui were C 3 (Hunt & Lipo, 2006 HTL 5 2.14; Solomon Islands HTL 5 2.10; Vanuatu HTL 5 2.14).
Although HTL in many island populations in Oceania showed increases up to 2.3-2.5 over the next 3-4 decades (see Supporting Information Table S2 in Bonhommeau et al., 2013), we speculate that these increases may be due to effects of globalization of the food supply and may not be characteristic of subsistence living in prehistoric contexts.
When using established wholly marine or terrestrial TP equations

| Dietary protein estimated from AA-CSIA d 15 N values and mixed terrestrial/marine food sources
In order to estimate the fraction of marine protein (f marine ) in human diets, we used an amino acid nitrogen isotope mass balance model that combines Equations 1 and 2 (see also Hebert et al., 2016;Styring, Fraser, Bogaard, & Evershed, 2014). This approach uses a b mixed value, which multiplies the estimated fraction of marine and terrestrial foods (f marine and 1-f marine , respectively) with their corresponding known marine or terrestrial b values.
TP mixed 5 d 15 N glu 2d 15 N phe 1ð12f marine Þðb terrestrial Þ1ðf marine Þðb marine Þ D glu2phe 11 where d 15 N glu , d 15 N phe , b marine , b terrestrial and D glu-phe are as defined in Equations 1 and 2. Hence, calculation of TP mixed requires knowledge of the proportions of marine and terrestrial protein in the prehistoric human Rapa Nui diets.
As a first approximation we assume that the HTL estimated from modern and historic human diets are representative of prehistoric humans on Rapa Nui and can therefore estimate the fraction of marine protein (f marine (HTL) ) by rearranging Equation 3 to solve for ƒ marine : The estimate of ƒ marine (HTL) for prehistoric Rapa Nui derived from Nui population. We test this assumption below using an isotope mass balance of measured terrestrial and marine d 15 N phe values.
Our use of d 15 N AA-CSIA resulted in new estimates of marine input into human diets, substantially higher than previous estimates based on bulk collagen isotope analysis (Commendador et al., 2013;Polet & Bocherens, 2015) but in agreement with our independent estimates based both our bulk collagen d 13 C and d 13 C EAA results. Using Here, we used our average d 15 N phe value determined for prehistoric Rapa Nui humans (d 15 N human-phe 5 11.6&), the average measured marine fish value (d 15 N marine-phe 5 1.0&), and the average ƒ marine (HTL) range from 27& to 20.1& (Bradley et al., 2015;Naito et al., 2010Naito et al., , 2013Nielsen et al., 2015;Styring et al., 2015;Styring, Sealy, & Evershed, 2010). Only a small proportion (n 5 14) of the latter are above the highest d 15 N phe value of humans among our dataset (RN035, With an independent approximation of a d 15 N terrestrial-phe end member based on plants grown in lithic mulch gardens and manavai soils, we rearrange Equation 6 to obtain an estimate of fraction of marine protein in human diets (ƒ marine (phe) ) that does not depend upon an assumed HTL.
Here we assume that the d 15 N phe measured in the archaeological totora Using this approach, we obtained an ƒ marine (phe) average of 0.50 and a range from 0.39 6 0.03 to 0.56 6 0.03. Returning to our initial TP calculations, we can use this average ƒ marine (phe) value in Equation 3 to achieve an average TP mixed for humans of 2.19 ( Figure 4c). This again supports both the feasibility of calculating TPs from d 15 N phe end member values, and ƒ marine for humans using the global average human HTL.
We believe that the archaeological Totora reed is a well-preserved end-member for this calculation and that its d 15 N phe value has not been diagenetically altered. Tremblay and Benner (2006) Hunt and Lipo (2006) and Hunt (2007) show a significant fraction (28-29%) of all preserved faunal remains to be from fish. Similarly, in a recent summary Arana (2014) details ethnohistorical sources that describe a range of boats, canoes, and totora reed floats used by the Rapa Nui population for fishing purposes. These, and extensive representation of fish and marine species, can also be seen on petroglyphs and as stone statues (e.g., Heyerdahl, 1958;Heyerdahl et al., 1962;Lee, 1993). It has been argued by some (e.g., Polet & Bocherens, 2015) that Anakena may have represented a more specialized or higher status site than other locations on the island resulting in systematically higher nitrogen and carbon isotope values. The notion that status is related to food consumption is difficult to justify given that Anakena is also situated in a location with easy access to marine resources. Given the lack of resource access restriction and the very localized structure of prehistoric Rapa Nui communities (Dudgeon, 2008;Lipo, Hunt, & Hundtoft, 2010;Morrison, 2012), the differences in fish consumption likely reflect relative proximity to these resources and not status.

| Implications of results for human consumption of rats
Our study does not support the hypothesis proposed elsewhere that rats formed a major dietary source (Commendador et al., 2013). An independent estimate of fraction of marine protein in prehistoric Rapa Nui human and rat diets (ƒ marine (phe) ; i.e., not assuming TP of the consumer) can be made using end member d 15 N phe values of the archaeological totora reed (22.7 6 0.9&) and the lowest measured marine fish (0.4 6 0.6&) using Equation 7. These values of ƒ marine (phe) can then be substituted for ƒ marine in Equation 3 to evaluate the relative difference in calculated TP mixed for humans and rats (Table 3), and thus allow us to discern the probability of a direct dietary connection between rats and humans.
The modeled TP mixed values in Figure 5 allow us to evaluate the suggestion that rats were a major dietary source of protein for humans (e.g., Commendador et al., 2013). Gut contents of the Pacific We propose that the elevated terrestrial d 15 N values are a result of substantially different soil N cycling processes in anthropogenically altered archaeological soils in the lithic mulch gardens and manavai that favored denitrification or ammonia volatilization (Szpak, 2014). Soil moisture levels are higher in modern lithic mulch gardens and manavai compared to natural soils (Hunt & Lipo, 2011), which would favor denitrification (e.g., Houlton et al., 2006). Furthermore, numerous studies have demonstrated that soil nutrients are higher in Rapa Nui lithic mulch gardens and manavai compared to natural soils, in part as a result of human interventions (Hunt & Lipo, 2011;Ladefoged et al., 2010;Louwagie, Stevenson, & Langohr, 2006;Vitousek et al., 2014). In addition, modern manavai and rock garden soils have significantly higher d 15 N values compared to adjacent nonagricultural soils despite the predominance of introduced plants. Manuring can markedly increase the 15 N content of agricultural soils, and this 15 N enrichment is passed on through the food chain to humans (Fraser et al., 2011).
The use of seabird guano for manure has been proposed as a mechanism for the elevated d 15 N values observed in Rapa Nui human collagen (Commendador et al., 2013;Fogel et al., 1997), and it has been demonstrated that guano as a fertilizer can have a substantial effect on d 15 N values in plants (Szpak, 2014;Szpak, Longstaffe, Millaire, & White, 2012). The manavai were likely used as deposits for household waste (Hunt & Lipo, 2011), and chicken husbandry may also have contributed to the elevated d 15 N values observed.

| C O NC LU S I O N S
Our results of carbon and nitrogen isotopic analysis of individual amino acids show that in our samples, seafood made up about half of the protein in human diets, which is considerably higher than previous estimates based on bulk data with similar isotopic compositions. Our estimates are consistent across four independent modelling approaches.
Additionally, we show that rats are unlikely to have made up a significant source of human dietary protein. These results may demonstrate a more balanced subsistence strategy, which is less likely to have placed unnecessary strain on natural terrestrial resources. Furthermore, the more accurate estimation of marine input in human and faunal diets allows for improved MRE corrections of 14 C dates. On Rapa Nui, the dating of the initial colonization of the island has been subject to extensive debate (Hunt & Lipo, 2006), and a detailed chronology is essential to understanding the impact of human settlement on the environment.
The 14 C dates from human and faunal remains have typically been considered less reliable due to MREs (Lipo & Hunt, 2016) despite their potential as more direct evidence for human occupation.
Significantly, our nitrogen isotopic results also suggest cultivation of agricultural crops in lithic mulch gardens and manavai, as documented in the archaeological record, was the source of the high d 15 N values observed in prehistoric human remains. This is further supported by our analysis of ancient and modern soils from both agricultural and noncultivated contexts. We do not know if biogeochemical conditions in these agricultural plots favored denitrification or ammonia volatilization, or if manuring through bird waste produced the high cultivated plant d 15 N values that were inherited by prehistoric humans. Regardless, these gardens required considerable effort in transporting the stones required to construct and maintain manavai and mulched areas (Ayala-Bradford et al., 2005;Bork, Mieth, & Tschochner, 2004), attesting to the effort invested in cultivating terrestrial resources. Burning of the native forest would have temporarily increased soil fertility on Rapa Nui, but over time the soils would have lost fertility (Hunt & Lipo, 2011). Our results point to concerted efforts to manipulate agricultural soils, and suggest the prehistoric Rapa Nui population had extensive knowledge of how to overcome poor soil fertility, improve environmental conditions, and create a sustainable food supply. These activities demonstrate considerable adaptation and resilience to environmental challenges -a finding that is inconsistent with an "ecocide" narrative.

ACKNOWLEDGMENTS
We thank Jim Ehleringer, Thure Cerling, John M. Hayes and Alex Bentley for comments on an early draft of the paper, and to Janet Becker for assistance with error propagation calculations. We also thank Natasha Vokhshoori for her help analyzing the food samples