The face-to-face survey, broken up into 40 rounds of collection, was implemented over a period of 10-months (May 2019–March 2020) in order to capture seasonal fluctuations in consumption, production and income. In addition to staff from KNSO, a total of 10 interviewers and 5 supervisors—divided into 5 teams—were employed to implement the survey. The survey questionnaire was developed in English using the World Bank Survey Solutions Computer Assisted Personal Interview (CAPI) software. The HIES questionnaire was divided into a series of modules (Supplementary Table S1).

Beyond the household-level food recall, there was also a module including a semi-quantitative food frequency questionnaire, where individual respondents within households were asked about the frequency of their consumption of 79 foods. They were also asked an open-ended question about whether they consumed any additional foods from five categories: cereals, seafood, fruit, vegetables, and snacks. These data were collected from all individuals aged 12 and older residing in households selected for the clinical health component of the research. Further information on the HIES, including the survey document, is publicly available (50).

The Village Resource Survey comprised a maximum of 164 questions that provided context for the resources available within the village, recent environmental history, access to fishery benefits, and formal and informal rules around fishing and gleaning activities. The survey had seven sections. The first section detailed the characteristics of the interviewees, who were selected based on key informant criteria. The second section focused on property ownership and transfers. The third section included a series of questions about the proximity of the village to a range of public services, such as the nearest hospital, bank, and market. It also collected information about available transportation options. The fourth section asked about migratory work opportunities. The fifth section focused on fisheries assets, commonly targeted and consumed aquatic resources, recent changes in fisheries, including identification of recent natural hazards and coral bleaching events, and documentation of rules and customs related to fisheries. The sixth section detailed the availability of physical infrastructure, such as sewage and waste disposal. The final section focused on broad changes experienced in the village within the past 10 years.

Of the 147 villages in Kiribati, the HIES was implemented in 111 and it was intended that one VRS would be conducted in each village (i.e., VRS target was one interview to be conducted in 111 different villages). In total, 109 villages were included in the VRS with 111 interviews being completed (i.e., some villages were interviewed more than once). The VRS questionnaire was administered to village authorities. Respondents included elected village leaders (51), unimane/unaine (elders; 41), teachers (1), pastors (2), and others (5). The data were collected via interview of the HIES team supervisors using Survey Solutions Software.

The clinical health survey was conducted on a tablet using software from the Dharma Platform of BAO Systems. These questions were meant to complement the socio-economic, demographic, and health questions that were included—as standard—in the HIES. These questions focused on food taste preferences and health and vaccination histories. Targeted questions for reproductive-aged women included information on pregnancy, lactation, menarche, menopause, and birth control. The survey itself lasted only 10 min per individual. The survey, all physical health measurements, and all biological sample collection was conducted by one of five licensed nurses, who were employed by MHMS, overseen by HM and ET, and trained by CG, WK, and JA on specific research protocols. Because of internet and data syncing issues with the Dharma Platform, we lost a substantial amount of survey data.

A team of ecologists (JGE, KKB, and ATeki) conducted marine community surveys to assess the species assemblage and functional health of I-Kiribati coral reefs. All diurnal fish species, non-cryptic invertebrates and the associated benthic habitats were quantified across the FR, LR, and BR sites by free-divers between 08:00 and 17:00 h. Observers were kept consistent to limit observer bias. For each site, three replicate 50 m transects were laid out using the belt transect method (n = 18 transects per habitat, per island; method described further in (24)) with >10 m spacing between transects. Transect placement was determined by habitat structure, depth (2–4 m), and heading.

Underwater visual census (UVC) was used to characterize the diurnal fish community assemblage. In order to avoid preferentially recording highly mobile and more bolder fish species and to limit diver effects (54), the fish observer (JGE) laid out the transect tape during the first “swim out” while simultaneously recording, identifying, sizing (nearest cm), and sexing all fish with a total length >10 cm across a survey width of 4 m (50 × 4 m benthic area per transect). On the second pass, less mobile, cryptic, small-bodied, and site-attached species were targeted with a more detailed examination of crevices across a swath of 2 m (50 × 2 m benthic area per transect). All individuals with a total length ≤ 10 cm were recorded, identified to the lowest recognizable taxon, sized (nearest cm), and sexed (when possible) as the observer more permanently fixed the transect. All fishes were recorded, including individuals in the water column within the defined benthic area (i.e., three dimensional surveys). When large schools or shoals of fish were present, species were binned by size class.

After the completion of the UVC of fishes, a second observer (ATeki) entered the water to quantify the invertebrate composition and density along the fixed transects. For the assessment of all mobile and sessile invertebrates, a swath of 2 m was used (50 × 2 m benthic area per transect). All species were recorded to the lowest recognizable taxon if >50% of the individuals' body was inside the swath. To limit observer bias, “percent effort” was standardized. This included time spent searching, team assistance, conditions (i.e., visibility, surge, and swell), and the lack of additional visual tools (i.e., no underwater flashlight). In the instance of an overabundance of a particular organism a sub-sampling method was used in order to maintain percent effort. Sub-sampling requirements were met when >50 individuals of a species were recorded within one of the two 25 × 2 m transect sections (0–25, 25–50 m). When this occurred, the observer recorded the distance surveyed (in cm) and total count to extrapolate the density out for the remaining distance within that section.

A benthic habitat assessment was conducted using a uniform point contact (UPC) method along the fixed transect (50 m). At each meter mark the observer recorded the identity of the primary space-holding organism, excluding epiphytes, or substrate beneath each point (n = 50). UPC cover categories consisted of macroalgae, turf, seagrass, sessile invertebrates (if a mobile invertebrate was landed on an adjacent point was used), non-biological substrate (e.g., sand coarseness, rubble, boulder), and corals [using definitions from (55)]. Low lying and unidentifiable masses of tightly packed turf-forming filamentous algae were recorded as turf algae [defined by Hay (56)]. All macroalgae and corals were recorded to species or lowest recognizable taxon by a consistent observer [KKB; using Kelley (57)]. For crustose coralline algae (CCA), encrusting calcareous algae, turf, or hard corals, growth form was classified as: branching, tabulate, digitate, plating, massive, submassive, and encrusting. When a live coral was landed on, the health of each colony was additionally assessed by recording percent bleached, presence and type of disease, and/or percent mortality when present. Bleaching scores were based on the 6-point color saturation scale on the CoralWatch Coral Health Chart in situ to minimize subjective assessment of bleaching state (58, 59). If bleaching was present, the coral was additionally characterized as either stressed or bleached and percent mortality of total coral structure was recorded.

In addition to point cover, the benthic observer recorded relief (n = 50) and structural complexity (n = 50) for each point. The change in vertical height between the shallowest and deepest substrate point within a 25 cm² box was measured and recorded as relief. Relief was binned by depth: (1) 0–10 cm, (2) 10 cm−0.5 m, (3) 0.5–1.0 m, (4) 1.0–2.0 m, (5) 2.0–3.0 m, and (6) >3.0 m. Complexity characterized the porosity of the reef structure. For this study, an exponential scale of 1–>10 was used: (1) 1, (2) 2–3, (3) 4–6, (4) 7–10, and (5) >10. Similar to relief, complexity measurements were considered by vertical height, however, unlike relief measurements, complexity measurements were confined to the vertical plane of that specific point (i.e., not within a 25 cm² box). For example, a point on the absolute bottom was scored a (1) where a highly complex coral, such as Acropora with >10 branches that crosses the vertical plane, was scored a (5).

Fine-resolution reef community surveys using the belt transect method can occasionally underestimate rare species, large-bodied individuals, and highly mobile or flighty species (54). The manta tow technique, where a diver on breath-hold is towed behind a small boat, is a method that has been adapted to census these species and supplement belt transects. A primary advantage of the technique is that it enables an observer to census species of interest over large areas of reef benthos at speeds far greater than a free-swimming diver. We conducted surveys at 3 manta tow (MT) sites per island across a deeper aspect of the fore reef to account for underestimation of rare, large, and flighty species. Each MT site was established adjacent to a FR site, but was conducted at a deeper depth (6 m) and on a different day to limit observer bias (including boating disturbance). MT sites were additionally selected to have consistent reef substrate (via satellite imagery) and large channels or sand patches were avoided. At each site 6,300 m transects with a swath width of 10 m were surveyed (18 km² observed per site) with >30 m spacing between transects. The tows were conducted by a consistent observer (JGE) at a constant heading and depth (±1 m due to tidal swings), in visibility >10 m, and at 3–4 km/h. A GPS was used to track speed, distance, and heading. Fish and invertebrate species of interest were recorded to the lowest taxonomic resolution, sized, and sexed.

While large-bodied reef-based invertebrates account for most of the I-Kiribati subsistence invertebrate catch (e.g., Tridacna spp.), a portion of fishing effort occurs on soft sediment habitat adjacent to the communities (7). Shellfish gathering, or gleaning, can fulfill nutritional needs when faced with fluctuations and seasonal inequalities in the availability of other resources. In order to quantitatively assess the abundance and diversity of species targeted by gleaning (animals that burrow and live beneath the substrate; e.g., the ark shell Anadara maculosa or te bun clams and Trochus spp.) we conducted soft infaunal quadrats (SIQ) on the soft sediment back reefs in areas exposed during low-tide [see Pakoa et al. (60) for more detailed methods]. SIQ sites were established 100–300 m inshore of the BR sites below the mean-high tide mark. Community assessments were made across a 40 m transect, laid parallel to shore, and replicated 3 times with >10 m spacing between transects. Observers randomly placed 4 25 cm² quadrats within a 2 m swath along the transect for each 5 m segment (n = 32 per transect) and recorded the relief, complexity, substrate (mud, sand, rubble, etc.), and cover (seagrass, algae, sponges, epiphytes, etc.), when present. The observer then excavated, using traditional I-Kiribati hand and spoon harvesting methods (14), all sediment and organisms down to approximately 10 cm. The composition of the sediment and all visually identifiable species with >50% of the individuals' body inside the quadrat was recorded. During sediment sorting, “percent effort” (as described above) was not kept constant due to inconsistencies in the sediment. Instead, effort was considered complete when observers were 100% confident that all infaunal species within the sampling area were quantified.

To supplement the HIES and VRS fishing pressure estimates, in situ fishing effort surveys were conducted for all 10 focal islands. While the ecological field research was being conducted, observers (JGE, KKB, ATeki, and I-Kiribati boat drivers) visually quantified the amount of (1) motorboats, (2) paddle canoes (waa n oo), (3) sail canoes (waa n ieie), (4) shore-based net fishers, (5) shore-based spear fishers, (6) shore-based hook and line fishers, (7) gleaners, and (8) other efforts that were actively fishing and/or transiting to, between or after a confirmed fishing effort. The surveys were conducted using the maximum number (MaxN) framework to estimate relative abundance and were recorded twice daily; a.m. 8:00–12:59, and p.m. 13:00–18:00. Within the recording periods, all observers actively observed fishing effort within line of sight. MaxN is a commonly used method of estimating maximum relative abundance using video recordings or line of sight estimates when absolute observations are difficult to obtain (61). Thus, MaxN is a conservative method where the maximum number of samples observed at any one given time throughout the entirety of a distinct trial period is recorded (62). The method was designed to avoid the recurring counting of individuals that enter a field of view within a trial (63).

When a potential fishing effort was observed by line of sight during a trial period the observers watched for fishing behavior, fishing evidence and discussed the boats intentions with the local boat driver. A effort was not recorded when the observers were not certain, with reasonable doubt, that the effort was fishing based. When the method was not possible to decipher it was recorded as fishing other. However, most fishing effort was obvious, routine and verbally confirmed with the fishers when catch was observed. Influencing factors, such as the day of the week (limited fishing on Sunday), weather, wind, and tide, were recorded. Lastly, observer distance was noted as: a shore observation, lagoon near (associated village from departure/arrival location), lagoon far (outside the study sites for that village), leeward fore reef, windward fore reef or equal distance, and other to account for spatial differences in effort.

To complement the ecological survey data and assess differences in oceanographic variables across the biogeographical and anthropogenic gradients, we collected 8 oceanographic measurements at 3 FR, 3 LR, 3 BR, and 3 SIQ sites per island. At each water sampling site, we measured (1) salinity, (2) conductivity, and (3) temperature using a handheld conductivity meter (YSI Model 30), salinity tester (Hach Pocket Pro) and a refractometer (Agriculture Solutions). All devices were calibrated prior to each measurement following the calibration specifications sheet provided by the manufacturer. Additionally, (4) chlorophyll A was sampled following the EPA Method 445.0 (64). Following EPA's guidelines, ocean water was filtered using a glass microfiber filter paper (Whatman 1823-025, 25 mm diameter) and then stored in a petri dish, wrapped in aluminum foil to prevent light damage, and frozen at −80°C. Lastly, (5) nitrogen, (6) alkalinity (KH), (7) calcium, and (8) phosphorus were tested in situ using a non-laboratory grade chemical kit for approximation and preliminary qualitative analysis.

Three ocean water samples were collected at each site to assess site-level microbiology. For the first sample, ocean water was filtered through a glass micron membrane (Whatman 934-AH) to a volume of 150 mL of ocean water (~100 mL of cells) to identify E. coli and giardia and assess the general bacteriology and microbiology of the ocean water. The filter was transferred to a sterile tube and DNA/RNA Shield (Zymo Research) was added to ensure a 9:1 ratio (shield to ocean water cell solution). The filter was submerged and agitated to preserve all cells present on the filter surface. The above sampling and preservation process was repeated for the second sample, but with 9 mL of ocean water filtered through a glass micron membrane and preserved with 1 mL of DNA/RNA Shield. The second sample of higher ocean water volume supplements the first sample (of standardized volume) to ensure sufficient cells for analysis were captured if a site had a naturally low concentration of microbes. The third ocean water sample was collected to preserve and identify diatoms, dinoflagellates, and other primary producers. Ocean water was filtered again using a glass micron membrane, preserved in a 1% solution of 50% concentrated glutaraldehyde and shaken aggressively. All tubes were stored at −80°C.

Quantifying concentrations of heavy metals in marine environments is challenging because trace metals (e.g., insoluble sulfur) can interfere with salt matrix refraction (51). Standard methods, such as inductively coupled plasma mass spectrometry (ICP-MS) with ISO 17294-2 protocol measurements can be used, but satisfactory performance results range from 41 to 86% with various metals (51). Instead, previous studies have used macroalgae as a bioindicator of heavy metals in saltwater (67–69). In the present study, Halimeda opuntia, a species of green macroalgae, was selected as a bioindicator of metals to establish a proxy of saltwater concentrations (70). Samples were collected underwater (>50 g per sample) along the ecological community survey transects with 2–5 m distance between samples, dry-weighed in the field (Hochoice milligram scale), and stored at −80°C.

In order to explore the presence of ciguatera in microalgal communities, via benthic dinoflagellates of the genus Gambierdiscus, we sampled turf algae from recently degraded branching hard corals from the genus Acropora. Coral growth form, turf morphology, height and density [following (71)], temporal persistence and sediment load and composition (e.g., limited sand or silt) was standardized to the best of our ability in situ when selecting coral colonies for sampling. Turf algae was scraped, using a scalpel underwater, from the branch tips (>1 cm) of each colony along the ecological community survey transects with 2–5 m distance between coral colonies. Upon surfacing, the sample was divided into two tubes for later processing. DNA/RNA Shield was added to the first sample to ensure a 9:1 ratio (shield to turf algae) depending on the amount of turf collected. A 1% glutaraldehyde solution was added to the second sample to allow for microscopic imaging of the cells present. Turf algae samples were stored at −80°C.

Reef fish were collected from both northern and southern sectors for each island to assess concentrations of heavy metals and ciguatoxin across geographical scales. Survey locations were selected during the initial capacity-building meetings with the Island Council as the most common fishing grounds. Discussions revolved around target catch (finfish) and reef habitat. To standardize fishing grounds across islands, only leeward fore reefs were sampled due to differences in island geomorphology (particularly lagoon reefs). To engage the community and ensure appropriate locations were sampled, we employed local fishers to catch reef fish for the study. Species were selected based on previously known I-Kiribati target species (14, 40). For each sector, we sampled 3 fish from each of the 4 families (family; example target species): groupers (Serranidae; Cephalopholis argus, Epinephelus merra), snappers (Lutjanidae; Lutjanus bohar, L. gibbus), parrotfish (Scaridae; Chlorurus sordidus, Scarus frenatus), and surgeonfish (Acanthuridae; Ctenochaetus striatus, Acanthurus nigricans). A total of at least 24 fish were sampled per island. While specific target species were requested and represented the majority of samples, other species from the same genera were occasionally substituted. Additionally, we opportunistically sampled other supplemental species, when present. If a ciguatera hotspot was mentioned by the Island Council (e.g., certain locations in Tabiteuea South), the area was sampled in addition to the two fishing sectors. For all reef fish, a geographic location within the sector was distinguished by having the fisher select a location using satellite maps. Fish were kept on ice or in a −17°C freezer, where possible, after catch, during transit, and for storage prior to dissections.

For each fish, liver and muscle tissue samples were taken for analysis of ciguatera and heavy metals, including mercury. After standard and total length were measured, the entire liver from the fish was removed and dissected into two 1 cm³ pieces. One sample was submerged in DNA/RNA Shield and stored for bacteriology, while the other standardized piece and remaining liver, if any, were stored in separate zip lock bags for metal analysis and future isotopic analysis, respectively. Scalpels were sterilized between tissue dissections with 90% isopropyl alcohol to avoid cross-contamination. The same procedure was followed for the white muscle tissue samples resulting in 6 samples per fish. Tissue samples were stored at −80°C. The animal study was reviewed and approved by University of California Santa Barbara IACUC.

Food Consumption. The monetary value of food consumption was attributed to the quantity consumed over the past 7 days for each food consumed by each food acquisition source (cash, own production, gifts, and exchange). The market survey was used to convert quantities of consumption expenditure that were reported in non-standard units of measurement to grams. Whole food acquisition was converted into edible quantities and nutrients using the Pacific Nutrient Database (72). Caloric availability was derived using the Atwater equation where 1 gram of protein = 4 calories, 1 gram of carbohydrate = 4 calories, 1 gram of fat = 9 calories, and 1 gram of pure alcohol = 7 calories (72). Only food consumed by the household was included, whether it was sourced from cash transactions, own-account production, gifting, or through exchange.

Non-durable (and Non-food) and Semi-durable Consumption. The consumption of non-food non-durable and semi-durable items was calculated as the annualized value of reported transactions for individual and household expenditures. Outliers were detected by product type and area in log-space using +/−2.5 × IQR and, where detected, locational medians were imputed.

Durables. Durables are items that are infrequently purchased by the household and have a long life (>1-year). Consumption of durables was estimated based on their use value.

Intermediate Expenditure. All expenditure related to productive activities and business were treated as intermediate expenditure and were not included in the consumption aggregate.

Transfers. All transfers (e.g., cash gifts to other households, or payments for fines) were excluded from the consumption aggregate.

Imputed Rent. The imputed rent component of the consumption aggregate was computed for owner-occupied housing using a predictive hedonic model, which was based on a range of variables including tenure, physical dwelling characteristics (number of rooms, building materials for walls, floor, roofing, water connection, flush toilet, electricity grid connection, fuel for cooking, and fuel for lighting) and location characteristics (province, urban/rural) characteristics. The model was based on rental expectations from the owner-occupying households because only 5 of the 2,182 households were renting, a sample deemed too small for an imputation model. For consistency across renting and non-renting households, the imputed rent from the model was used for all households, and actual rents were not used in the consumption aggregate. Deductions were made from the imputed rent for maintenance costs of owner occupiers.

Dietary Intake Food Frequency. Each of the 79 food groups were coded into their respective frequencies of consumption based on a provided serving size. All foods reported in “other” categories were translated, coded, and incorporated during the data cleaning and analysis stages. For each of the food groups, participants reported the frequency with which they consumed one serving based on the following options: (1) never/almost never; (2) once per month; (3) once per week; (4) three times per week; (5) daily; (6) more than daily; (7) more than three times per day.

Anthropometric data were analyzed using standard WHO cutoffs to assess nutritional status. Children under 5 were assessed as underweight (weight-for-age z-score < -2), being stunted [height-for-age z-score (HAZ) < -2], wasted (weight-for-height z-score < -2) and overweight (weight-for-height z-score > 2) based on the WHO Child Growth Standards (52). Mid-upper arm circumference (MUAC) was used to assess severe acute malnutrition in children based on a MUAC <11.5 cm in children under 5, and moderate acute malnutrition based for a MUAC ≥ 11.5 cm and <12.5 cm. For children <24 months, a head-circumference-for-age z-score < -2 was classified as a case of microcephaly. In children ages 5–19, overweight was classified based on a body-mass-index-for-age z-score (BMIZ) > 1, and obesity was classified as a BMIZ > 2. In adults ages 20 and above, body mass index (BMI) was used to classify nutritional status in the following way: underweight, BMI <18.5; 18.5–24.9, normal weight; 25.0–29.9, overweight; 30.0–34.9, obesity class I; 35.0–39.9, obesity class II; and >40, obesity class III. Observations were considered to be outliers if their WAZ or HAZ were > |6| or their head circumference z-score or WLZ were > |5|.

Anemia status was classified into non-anemia or mild, moderate, or severe anemia based on hemoglobin levels (g/dL) with respect to age and pregnancy status per the WHO recommendations (73). In children 6–59 months of age, anemia was classified as follows: mild, 10.0–10.9 g/dL; moderate, 7.0–9.9; and severe, <7.0. In children 5–11 years of age: mild, 11.0–11.4 g/dL; moderate, 8.0–10.9; and severe, <8.0. In children ages 12–14 and non-pregnant women over age 15: mild, 11.0–11.9 mg/dL; moderate, 8.0–10.9; and severe, <8.0. In pregnant women: mild anemia, 10.0–10.9 g/dL; moderate, 7.0–9.9; and severe, <7.0. Lastly, men ages 15 and above were classified as having anemia as follows: mild, 11.0–12.9 mg/dL; moderate, 8.0–10.9; and severe, <8.0. Non-anemia was designated in the absence of anemia per the classifications of each group.

Blood pressure was considered to be: normal if systolic <120 mmHg and diastolic <80 mmHg; elevated if systolic was ≥120 and ≤ 129 mmHg and diastolic <80 mmHg; Stage 1 hypertension if systolic was ≥130 and ≤ 139 mmHg or diastolic was ≥80 and ≤ 89 mmHg; Stage 2 if systolic ≥140 and ≤ 180 mmHg or diastolic ≥90 mmHg; and Hypertensive Crisis if systolic > 180 mmHg and/or diastolic >120 mmHg (74).

Metabolic syndrome, and its constituent measures, all followed from American Heart Association guidance (75). In individuals over age 12, fasting plasma glucose levels ≥180 mg/dL were considered hyperglycemic; 70–180 mg/dL normal, and ≤ 70 mg/dL hypoglycemic. An individual was considered to be pre-diabetic if their hemoglobin A1c (HbA1c) levels were between 5.7 and 6.5%; diabetic if their HbA1c ≥ 6.5%; and non-diabetic if their HbA1c <5.7%. Cholesterol levels were considered normal if total cholesterol <200 mg/dL, HDL > 40 mg/dL, and LDL <130 mg/dL. Cholesterol levels were considered at-risk if total cholesterol ≥200 or HDL ≤ 40 or LDL ≥ 130. Triglycerides were assessed as follows: <150 mg/dL as low-normal; ≥100 and <150 mg/dL as high-normal; ≥150 and <200 mg/dL as borderline hypertriglyceridemia; ≥ 200 and <500 as moderate hypertriglyceridemia; and ≥500 mg/dL as severe hypertriglyceridemia. Note that the CardioChek Plus cannot read triglyceride values below 50 mg/dL, therefore neither triglyceride nor LDL cholesterol values were recorded for these individuals. LDL cholesterol is estimated as a function of triglyceride, HDL cholesterol, and total cholesterol levels using the Friedewald Equation (LDL cholesterol = total cholesterol—HDL cholesterol—triglycerides/5) (76).

All OmegaQuant filter papers, treated with HUFASave^TM, were sent to OmegaQuant laboratories for fatty acid analysis following established protocols (53). Their analyses produce 24 fatty acid profiles, including eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), and an overall omega-3 index, which has been associated with various forms of morbidity and mortality (53).

DNA was extracted from dried blood spots using Zymo Quick-DNA extraction Kits. Participants will eventually be genotyped using Illumina's Infinium Global Screening Array v1.0 (GSA) from Illumina (77), which includes 642,824 SNPs. Genotypic calls and CNV analysis will be performed using the Genome Studio 2.0 (Illumina Inc. SD, California). We will exclude participants if (i) genotyping fails at more than 5% of SNPs, and/or (ii) if their heterozygosity rate is >3 standard deviations from the cohort mean. We will also exclude rare alleles (minor allele frequency (MAF) <0.01) and those that are not in Hardy–Weinberg Equilibrium (p < 1 × 10–6). LD will be calculated using r² implemented in PLINK (78) and removing one SNP of a pair each time r² > 0.6.

Two blood smears (Peripheral Smear) were collected for each participant to calculate blood cell counts. Following fixation, each slide was evaluated for differential white blood cell count (WBC). This count was done manually using a microscope and a cell counter for neutrophils, lymphocytes, monocytes, eosinophils, and basophils. WBC percentages were standardized for each WBC. Normal ranges in adults are: neutrophils (55–70%), lymphocytes (20–40%), monocytes (2–8%), eosinophils (1–4%), and basophils (0.5–1%) of total white blood cells (79). We also evaluated each participant for RBC irregularities (e.g., anisocytosis—indicative of anemia; poikilocytosis and anisopoikilocytosis), the presence of abnormal cells (e.g., presence of immature white blood cells such as blasts, indicative of leukemia, or serious bone marrow disease) and infectious agents. Note that a variety of diseases and conditions can affect the relative number of WBC, and we primarily intend to use variation in WBC as a covariate in downstream analysis (e.g., blood gene expression profiling).

To prepare samples from Halimeda opuntia and fish for analysis, they were removed from a deep freezer (−80°C), lyophilized (MechaTech Systems, LyoDry Compact), and transferred to sterile tubes (VWR Centrifuge Tube, 76204-404, Batch: 190717060 and 190609058). Samples were ground using a metal spatula sterilized using 10% Hydrochloric acid (J.T. Baker, Hydrochloric Acid 36.5–38.0%, Baker Instra-Analyzed Reagent, 9530-33, Batch No: 000024173 diluted with deionized water). Tissues were weighed for dry-weight before Thermo Quant'X EDXRF (XRF) and mercury analysis (Mettler-Toledo AG XPE205DR).

Ground samples were analyzed at Harvard University for total mercury concentration (T-Hg) on a direct mercury analyzer (DMA) (NIC MA-3000 Mercury Analyzer and MA3 Win software) following US EPA Method 7473 (US EPA 1998). This process involves measurement by thermal decomposition, reduction, amalgamation, and atomic absorption spectrometry. A quality control and assurance series was run every 12 samples. This series included the following: blanks (2), deionized water purge (1), dry purge (1), Apple Leaves 1515 (1) (NIST Standard Reference Material (SRM), US Department of Commerce National Institute of Standards and Technology, Gaither, MD 20899), MESS-4 SRM (1) (Marine Sediment SRM, National Research Council Canada, Lot G 4169010, Serial CC 567765) to verify recovery rates and sample integrity. Calibration curves were set using R² > 0.99 and SRM recovery was 87–102% confirming the accuracy of measurements. Amalgamator and catalyst tubes were replaced as needed due to excessive calcium carbonate deposition from the tissues.

To prepare human fingernail samples, they were cleaned with a 1% triton deionized (DI) water solution, sonicated for 30 min, vortexed, and soaked for an additional 2 days (89). This process of sonification and rinsing was repeated three more times, including an acetone rinse, and nails were then freeze-dried overnight. Total mercury (Hg) concentrations were measured with a Nippon MA-3000 mercury analyzer (U.S. EPA 1998) along with a certified reference material (European Reference Materials (ERM-DB001), human hair).

To test the concentration of heavy metals present, samples were analyzed using inductively coupled plasma mass spectrometry (Thermo Scientific, iCAP Q) following EPA Method 6020B (U.S. EPA 2014). This process involves ~100 mg of dried samples digested at 95°C for 6 h using a 4 mL nitric acid and hydrogen peroxide solution (3:1, respectively, J.T. Baker, Instra-Analyzed Reagent, Batch No: 0000221802). Post digestion, samples were diluted to 30 mL using deionized water and measured on ICP-MS. Calibration of ICP-MS was conducted using Multi-Element Solution 2 (Spex CertiPrep, USA). Average recovery rates for calibration standards and trace metals in water SRM (NIST 1643f) analyzed as samples were generally in the range of 90–110%, supporting measurement accuracy. These results allowed for comparison of mercury values between two machines. All samples were then tested for heavy metals using X-ray Fluorescence, which is a faster analysis process.

We used a Thermo ARL Quant'X EDXRF (XRF; Thermo Fisher Billerica, MA) for non-destructive analysis of a suite of elements. The system was run for 30 min live time for each sample with x-ray settings of 50 kVp and 1 mA using a silver filter and anode. Samples were measured in 30 mm diameter, polyethylene XRF sample cups with a 2.5 um mylar support film. XRF was optimized for sampling over a 20 mm² area to a depth of approximately 0.5 cm. Samples would rotate during measurement to ensure a full homogenous measure of the cup and sample. XRF spectra was analyzed using in-house peak fitting methods calibrated against standards with known composition (89).

Water samples will be analyzed to explore the presence of various human relevant diseases. DNA/RNA solutions can be later used for extraction and DNA sequencing to screen for bacteriology analysis which can detect cryptosporidium, fecal coliforms, giardia, and general bacterial presence. We will use a combination of 16S rRNA and shotgun metagenomics sequencing to identify variations in community composition that may correlate with adverse health outcomes.

Reef fish samples (liver and white muscle tissue) and turf algal samples were shipped to the National Oceanic and Atmospheric Administration (NOAA) to analyze the presence of ciguatoxin (CTX) using their fluorescent receptor binding assay protocol (90). This process involves a fluorescence-based receptor binding assay [RBA_(F)], based on competition binding between CTX and fluorescent-labeled brevetoxin-2 with voltage-gated sodium channel receptors. The analysis detects the presence or absence of ciguatoxin, toxicity levels, and the percent binding equivalents.