Dietary transitions among three contemporary hunter-gatherers across the tropics

The diets of contemporary hunter-gatherers are diverse and highly nutritious, but are rapidly changing as these societies integrate into the market economy. Here, we analyse empirical data on the dietary patterns and sources of foods of three contemporary hunter-gatherer societies: the Baka of Cameroon (n = 160), the Tsimane’ of Bolivia (n = 124) and the Punan Tubu of Indonesia (n = 109). We focus on differences among villages with different levels of integration into the market economy and explore potential pathways through which two key elements of the food environment (food availability and food accessibility) might alter the diets of contemporary hunter-gatherers. Results suggest that people living in isolated villages have more diverse diets than those living in villages closer to markets. Our results also suggest that availability of nutritionally important foods (i.e., fruits, vegetables and animal foods) decreases with increasing market integration, while availability of fats and sweets increases. The differences found seem to relate to changes in the wider food environment (e.g., village level access to wild and/or market foods and seasonality), rather than to individual-level factors (e.g., time allocation or individual income), probably because food sharing reduces the impact of individual level differences in food consumption. These results highlight the need to better understand the impact of changes in the wider food environment on dietary choice, and the role of the food environment in driving dietary transitions.


Introduction 30
The quality of diets in traditional hunter-gatherers has been a topic of heated 31 debate, to the point where a "paleo" diet has been promoted as a healthy alterative to In this article we address these two questions. In the first part of the article, we 65 analyse empirical data on the dietary patterns and sources of foods of three 66 contemporary hunter-gatherer societies. We specifically focus on differences between 67 communities with diverse levels of integration into the market economy. In the second 68 part of the article, we explore potential pathways through which changes in the food 69 environment associated with integration into the market economy might alter the diets 70 of contemporary hunter-gatherers. The analysis is based on the assumption that nutrition 71 transitions result from changes in elements of the food environment that are likely to 72 impact dietary choice. 73 Work on the role of food environments in dietary choice has, to date, largely 74 focused on the food environment in markets, with little attention to cultivated and wild 75 food sources (Powell et al. 2015, Ahmed and Herforth 2017). Powell et al. (2015) argue 76 that "In areas where market access is difficult or where markets do not function well, 77 economic factors and market food environments may not be the strongest determinants 78 of food choice: in these settings, we need to understand how the landscape (or natural 79 food environment) affects diets." Herforth and Ahmed (2015)  is also likely to change as hunter-gatherer communities become more integrated into the 96 market economy. Livelihood transitions often imply that foods that were once obtained 97 from the wild or from subsistence agriculture must be purchased, oftentimes because 98 people no longer have the time to collect, hunt, or grow them. Finally, desirability or 99 food preferences are learnt and highly bound by socio-cultural factors (Serrasolses et al. 100 2016;Bowles 1998;Fischler 1988) and are also likely to change as hunter-gatherer 101 communities become more market integrated.
In our analysis, we examine differences in importance of different aspects of the 103 food environment in communities with more and less market integration. Because no 104 detailed market survey data were collected with the study, we use seasonal differences 105 to examine the importance of food availability and livelihood strategy (assessed by the 106 activity an individual spends most time pursuing) and income as to assess the 107 importance of food accessibility (including both monetary access and access in terms of 108  the steep slopes, irregular rain patterns, and lack of agricultural inputs still make 152 harvests highly variable. Palm sago has been replaced by easier to prepare, cassava 153 sago, which is an alternative to rice, particularly before rice harvest when the stored rice 154 has been consumed (Table 1) (Table 1). They rely on slash-and-burn farming of cassava, 170 plantains, maize, rice, and chickens, supplemented by hunting, fishing, gathering wild 171 fruits. Game and fish are generally more abundant in more remote villages (Díaz-172 Reviriego 2016). Some Tsimane' men, mostly in villages close to town, increasingly 173 engage as wage laborers in logging camps, cattle ranches, and in the homesteads of 174 colonist farmers. The commercialization of forest products (e.g. thatch palm) also provides a primary source of income for many households, often through barter (Vadez 176 et al. 2008). Partly due to these shifts in livelihoods, Tsimane' diets are undergoing 177 rapid change including the introduction of market foods and beverages, such as dried 178 and salted meat, sugar, noodles, lard, vegetable oil, white flour/ bread, and soda 179 In each society, we worked in two villages that differed in their distance to the 203 main market town (i.e., isolated and close villages) (Table 2) foods products according to the 12 following food groups: 1) starch (i.e., cereals, white 213 tubers and roots); 2) dark green leafy vegetables; 3) other vitamin-A rich fruits and 214 vegetables; 4) other fruits and vegetables; 5) meat and fish foods (including insects); 6) 215 organ meat; 7) eggs; 8) milk and milk products; 9) legumes and nuts; 10) fats (including 216 oils); 11) sweets; and 12) spices (including condiments and beverages) (Supplementary 217 Material 1). We asked informants to list all the foods and drinks they had consumed 218 during the previous 24-hours, inside and outside the house, and each food item was 219 noted in the corresponding food group. Probing was used to help ensure informants did 220 not omit added foods (e.g., sugar) or food items consumed outside the house (e.g., in the 221 forest). The questionnaire was administered in the morning, avoiding holidays, 222 celebrations and/or fasting periods. We also recorded the source of each food item 223 differentiating between items that were cultivated, obtained from the forest, or bought 224 from the market. To capture seasonal variation in food consumption (Table 1) aforementioned food groups, except fats, sweets and spices. We then calculated the 263 mean for each society and differentiating between people living in the isolated and the 264 close village. We used a Pearson chi2 test to assess whether there were statistically 265 significant differences between villages (Table 3). 266 To analyze sources of foods, we created three new variables for food groups 267 (crop, wild, and market) which were coded as 1 if at least one of the food items in a 268 group came from that source and 0 otherwise. We then calculated the percentage of 269 diets which included at least one food item of each of the food groups obtained from 270 agricultural fields, the wild, and the market ( Table 4). As with dietary patterns, we differentiate between the isolated and the close village in each society and assessed 272 differences using a Pearson chi2 test. 273 To analyze if and how changes in food availability and food accessibility might 274 be responsible for the differences in diet associated with market integration of 275 contemporary hunter-gatherers we used bivariate and multivariate analysis. We first 276 look at potential variations in sources of food according to temporal availability 277 (seasonality). We did so by grouping information on food consumption differentiating 278 between questionnaires that were conducted during the "Rainy" and "Other" seasons 279 and then analyzing sources of food for each season (Table 5). To explore how 280 accessibility might alter dietary patterns, we aggregated data on time allocation into four 281 categories: subsistence agriculture, foraging, wage labor, and other (e.g., leisure, 282 cooking, household work). We calculated the share of times an individual was mainly 283 devoted to each activity and classified individuals as predominantly: a) agriculturalists, 284 b) foragers, or c) wage workers if >50% observations were in the corresponding 285 category, and as d) diversifiers, if they did not fit in any of the previous groups. We then 286 explored if there were differences in the food consumed between people in these 287 categories (Table 6). 288 In the last part of the analysis, we assess the relative weight of these various 289 factors in modeling WDDS by using multivariate analysis. Specifically, we analyze 290 how variation in WDDS relates to food availability, food accessibility, and village and 291 individual level of integration into the market economy using expression fixed-effects, we included a set of dummies for the societies of study (S). And εihv is the 305 error term, that basically captures the information that the model cannot explain (Table  306 7). In additional analyses we replaced binary variables for societies by binary variables 307 for villages. To control for the fact that observations may be correlated within 308 individuals, but independent between them, in all regression models we used clusters by 309 individual. The statistical analysis was done using STATA for Windows, version 13. 310 We report p-values < 0.10 as indicator of statistical significance, 311 Results 312

Dietary patterns, food sources, and market access 313
In each society, food consumption differed substantially between villages with 314 more and less market access, diets being generally more diverse in the isolated villages 315 (Table 3). Thus, Baka living in the isolated village had a WDDS about 0.3 food groups 316 higher than their peers in the close village (p<0.001). The Baka living in the isolated 317 village consumed starchy staple, meat and fish, and legumes and nuts more frequently 318 than the Baka living in the close village. Conversely, the Baka living in the isolated 319 village consumed less sweets but more spices, condiments and beverages than their peers. Similarly, the WDDS of the Tsimane' living in the isolated village was about 0.5 321 food groups higher than the WDDS score of Tsimane' living closer to the market 322 (p<0.001). Food items from the categories of 'other fruits and vegetables', organ meat, 323 meat and fish, legumes and nuts, and milk and milk products were consumed more 324 frequently by Tsimane' living in the isolated village. Unexpectedly, Tsimane' living in 325 the isolated village also consumed fats and sweets more frequently than their peers. 326 INSERT TABLE 3  327 The pattern is somewhat different among the Punan Tubu, as we did not find 328 statistically significant differences in their WDDS score. Thus, although Punan Tubu in 329 the isolated village consumed 'other fruits and vegetables' more often than Punan Tubu 330 in the close village, they consumed meat and fish and other vitamin A rich fruits and 331 vegetables less often. The Punan Tubu in the isolated village also consumed oils and 332 fats more often than people in the close village. 333 We also found variation in food sources associated to market access, although 334 there was less variation between near and far villages in terms of the sources of food 335 than in the frequency of consumption of different food groups (Table 4). Our data show 336 a difference in the source of meat and fish, with more of these food items coming from 337 the wild in the isolated village and more from domestic animals and the market in the 338 close village. Among the Baka, those living in the close village obtained less of their 339 staples from cultivated crops but more from the wild and less of their dark green leafy 340 vegetables from markets than Baka living in the isolated village. Among the Punan 341 Tubu, those living in close village obtained less starch from cultivated crops but more 342 from the market, than people the more isolated one. They also obtain less of their fruits and vegetables from cultivated sources and more from the wild than those living more 344 isolated. 345 Oils and sweets were the food groups with more differences in source between 346 the close and the isolated villages (Table 4). The Baka in the isolated village obtained 347 all their sweets from the wild (i.e., honey), whereas those in the close village obtained 348 them from the market. For both the Baka and the Tsimane', isolated villages obtained 349 more of their fat and oils from the market, while those in the close communities 350 obtained them on farm (e.g., from cultivated oil palm or domestic animals). 351

Food availability, food accessibility, and market access 353
Overall, there is a strong seasonal variation in diets, reflected in a lower WDDS 354 during the rainy season than during the rest of the year (Baka and Tsimane' p<0.001, 355 Punan Tubu p=0.06; Table 5). During the rainy season, the Baka consumed starchy 356 staples, dark green leafy vegetables, other fruits and vegetables, and meat and fish less 357 frequently, and obtained a lower percentage of these from the market and a greater 358 percentage from the wild. In the rainy season, the Baka also consumed oils and fats, 359 sweets, and spices and condiments less often and obtained a lower percentage of these 360 from the market. Similarly, during the rainy season the Tsimane' consumed foods in the 361 categories of staples, organ meat, and meat and fish less frequently, but other fruits and 362 vegetables more frequently (with a greater percentage of these obtained from the wild). 363 The lower consumption of meat and fish in the rainy season is concurrent with a greater 364 percentage of fish and meat obtained from the market at that time. They also consumed 365 oils and fats, sweets and spices and condiments less frequently in the rainy season, 366 obtaining less of them from the market (Table 5).

INSERT TABLE 5 368
The diets of the Punan Tubu showed less seasonal variation. The most important 369 patterns of seasonal variation for the Punan Tubu refers to a greater dependence on 370 purchased staples in the rainy season, and a less frequent consumption of meat and fish 371 and organ meat during the rainy season (Table 5). While oil and fat consumption did 372 exhibit seasonal variation, a larger share of the fat and oil consumed in the rainy season 373 came from the market. Consumption of sweets was less frequent in the rainy season and 374 a lower share of them came from the market (Table 5). 375 Time allocation did not seem to relate to diet (and presumably food 376 accessibility), as the individual consumption of different food groups varied little 377 according to individual time allocation, and did not seem to have an overall impact in 378 dietary diversity (Table 6). Thus, despite some specific differences in the three societies 379 (e.g., Baka foragers consume more oils and fats, Punan Tubu agriculturalists consume 380 more fish and meat, and Tsimane' wage workers consume more organ meat than people 381 in other groups), there are no statistically significant differences in the WDDS across 382 time allocation groups. 383

The correlates of WDDS 385
When considering the three societies together (Table 7, Model 4), the distance of 386 the village to the market and season were associated in a statistically significant way 387 with WDSS. Overall, people living in the isolated villages had a higher WDSS than 388 their peers living in the close villages. Additionally, the WDDS was generally lower 389 during the rainy season than during the rest of the year. We find the same pattern when 390 analysing data for the Baka (Model 1) and the Tsimane' (Model 3) samples, but not among the Punan Tubu (Model 2). When looking at the pooled sample, we also found a 392 weak and negative association between people who allocate more time to foraging and 393 WDDS. However, the association is not found in any of the regressions with separate 394 samples. None of the variables that proxy for individual level of integration into the 395 market economy are consistently associated to WDDS across the three case studies. 396

INSERT TABLE 7 397
Discussion 398 In this work we used data from three contemporary hunter-gatherer societies 1) 399 to assess variations in dietary patterns and food sources associated to market integration, 400 and 2) to explore the role of two key elements of the food environment, food 401 availability and food accessibility in explaining such variability. We organize the 402 discussion around the main findings for these two goals. 403 In the three studied societies, we found variations in diets and food sources 404 associated with integration into the market economy. Although the diets of the three 405 societies were different from one another, we found a similar pattern in that there was 406 higher dietary diversity in isolated villages, a trend that was corroborated through 407 multivariate analyses. Moreover, the difference found was relatively large when 408 compared with results from previous studies (see Jones (2017) for a review of the 409 literature). Importantly, the higher dietary diversity in the isolated villages was due to 410 the more frequent consumption of nutritionally important food groups (e.g., fruits, 411 vegetables, meat, fish). Our results, however, also point at some counterintuitive 412 findings, such as that people living in the isolated villages also consumed more Hadiwijaya, L. Napitupulu, and D. Suan in Indonesia. We thank them all. We also thank 505 CIFOR for logistical assistance during field-work, A. Pyhälä for database management, 506 and C. Vadez-Reyes for research assistance. This work contributes to the "María de 507 Maeztu Unit of Excellence" (MdM-2015-0552). 508

Conflict of Interests 509
The authors declare they have no conflict of interests. 510    Note: For each food group we calculated the percent of food items obtained as crops, from the wild, or from the market. Since food groups can have items from more than one source, percentages do not necessarily add to 100.   Standard errors in parentheses * p< 0.10, ** p< 0.05, *** p< 0.01