A test of four evolutionary hypotheses of pregnancy food cravings: evidence for the social bargaining model

The onset of cravings for items not typically desired is often considered a hallmark of pregnancy. Given the ubiquity of cravings, this phenomenon remains surprisingly understudied. The current study tested four hypotheses of pregnancy food cravings: behavioural immune system, nutrient seeking, resource scarcity and social bargaining. The research took place in Tamil Nadu, South India, with pregnant women residing in rural villages (N = 94). Methods included structured interviews and anthropometric measures. Findings revealed that unripe mango and unripe tamarind were the two most frequently mentioned food cravings among this population, but were not sufficiently supported by the a priori models. Results confirmed that the social bargaining model was the best explanation for the etic category of toxic/pathogenic food items, suggesting that pregnant women crave dangerous foods when experiencing heightened social pressures. Finally, toxicity/pathogenicity was a confounding factor for the nutrient seeking and resource scarcity models, calling into question the validity of these models in adverse environments. Overall, these findings present important implications for research on pregnancy food cravings, such that in resource-scarce and pathogen-dense environments, cravings might target teratogenic items that signal a need for increased social support.


Introduction
Across cultures, a craving for items not typically desired is often considered a hallmark of pregnancy. Women are known for craving sweets, fruits, calorie-dense foods, odd combinations such as 'pickles and ice cream', or pica substances, such as clay and chalk [1][2][3]. Despite the widespread occurrence of cravings, this phenomenon remains relatively understudied [1]. Evolutionary theories proposed to explain food cravings include a need to seek foods to either satisfy the energetic 2017 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited.

Methodology
This study relied on Agar's [44] 'ethnographic funnel' method that starts with a general question (e.g. 'What do women crave during pregnancy?') and then hones in on a more specific set of questions that lead to hypothesis testing. Early stages of this project consisted of informal interviews with women in the community (n = 10) who told stories of common pregnancy cravings. Information gathered at this stage led to a semi-structured questionnaire that asked community women (N = 54) specific questions about pregnancy cravings and the consequences of consuming the mentioned items. Results for these stages are reported in [13] and [36]. In the final step of data collection, N = 95 pregnant women underwent structured interviews that assessed their aversions, cravings and consumption patterns in pregnancy. Women also addressed questions focusing on resource scarcity and pathogen exposure. Finally, anthropometric measurements were collected. Participants were recruited from primary health centres (PHCs) located within each village. Women in this region are given incentives for registering with PHCs, therefore, this sample is probably inclusive of all women who were pregnant in this region at the

Analysis
Data were analysed in R v. 3.3.3 (3 June 2017) for Macintosh and can be found publicly on Figshare [45].
The following hypotheses were tested: behavioural immune system, nutrient seeking, resource scarcity and social bargaining. Univariate analyses and multivariate modelling were used to test the hypotheses. First, free-list data for food cravings were graphically represented to determine the most frequently craved items. Additional outcome variables were created from free-list data to represent the different model predictions. Next, models were generated to test each hypothesis individually. Since outcome measures differed for each model, statistical tests were chosen by examining the distribution of variables and appropriate tests were determined by comparing model options based on Akaike information criterion (AIC) [46]. For example, tests for negatively skewed ordinal variables were selected based on the lowest AIC score between Poisson and negative binomial regressions. Overall model strength was determined by using McFadden's R 2 -values for Poisson and negative binomial models, and Nagelkerke R 2 -values for logistic regression models. The strength of individual predictors was determined by coefficient estimates (Poisson and negative binomial models), odds ratios (logistic regression models), 95% CIs and p-values. Since these models are not mutually exclusive, some predictor variables are included in more than one model. Table 1 presents each model including outcome and predictor variables, and specific analytic details for each model are described below.

Behavioural immune system
The behavioural immune system predicts that pregnant women experience cravings for micronutrients that are lost due to increased nausea and vomiting that occur during the early months of pregnancy. According to this hypothesis, cravings will focus on foods rich in antioxidants, such as unripe and regular fruits [5]. To test this hypothesis, logistic regression models were fitted to predict cravings for ripe and unripe fruits. A dichotomous variable was created that measured cravings for ripe fruits, and the most frequently craved food items, unripe mango and unripe tamarind, were also transformed into dichotomous outcome variables (presence = 1, absence = 0). Predictor variables included the presence/absence of nausea and vomiting (yes = 1, no = 0), and months pregnant.

Nutrient seeking
This model predicts that women will crave calorically dense or nutritionally rich foods, and these will be positively associated with months pregnant, and indicators of nutritional need, such as triceps thickness and BMI. Calories per food item were computed using the Nutritive Value of Indian Foods measure by Gopalan et al. and is accessible as freeware (http://bit.ly/ncalculator). Foods not listed in this source were located on the South Indian foods calorie chart (http://southindianfoods.in/south_indian_food_ caloriechart.html). Next, a calorie-density composite score was computed for each woman's free-listed cravings. An outcome variable for nutritious foods was created from the free-listed cravings. Foods that were considered nutritious were situated within a South Indian context of what is considered healthy to consume in pregnancy: vegetables, ripe fruits, meat, eggs, fish and dairy products [34]. Unripe fruits were not considered healthy, because they are perceived to lead to spontaneous abortion if consumed in high quantities [13], and were thus excluded. Predictor variables included months pregnant, triceps thickness and BMI. To control for toxicity and pathogenicity of each food item, a control variable was created and included in the models. To create this variable, each food was given a '1' if it was potentially toxic or pathogenic (e.g. vegetables or meat), and a '0' if not.

Resource scarcity
The resource scarcity model indicates that women should experience cravings for nutritious foods because their diets are limited to a few food categories [6]. The measure of nutritious food consumption was selected as the outcome variable. Predictor variables included total food insecurity, dietary diversity, triceps thickness and BMI. Months pregnant was included as a control measure and was predicted to have no association with the outcome. Food insecurity was measured with the six-item short-form measure of food security that has demonstrated adequate validity and reliability on studies conducted in India [47]. Dietary diversity was measured using a version of the Food Frequency Questionnaire that asked women how often they consume a particular food item. Responses were coded on a 7-point Likert scale (0 = never, 6 = daily). There were a total of 28 food items (see electronic supplementary material, S1). Frequencies were added together to create a total score for dietary diversity.

Social bargaining
The social bargaining model predicts that women will crave foods that are harmful to the developing fetus. The outcome variables for this model, therefore, included both an etic and emic measure of 'dangerous' foods. The dichotomous toxicity/pathogenicity variable was used as the etic measure. A study by Placek & Hagen [36] found that 'hot' and 'black' foods are considered dangerous to eat during pregnancy. In this study, hot foods included sour items, unripe mango, mango, papaya, pineapple, palmyra sprouts, eggplant, chicken and fish. Black items included black grapes and jamun. A 'hot' and 'black' variable was computed from cravings to serve as the emic outcome variables. Finally, unripe mango and unripe tamarind were modelled separately because these two items were craved in high frequency.
Predictor variables included the presence or absence of pressure to have a son (yes = 1, no = 0), the total number of living children, food insecurity and psychological distress. An interaction term was added for pressure to have a son and total number of living children because these variables are often related [48]. The total score from the Kessler-6 was included as an indicator of general psychological distress. The K-6 is a six-item measure that assesses serious mental illness in World Health Organization surveys [49]. Measures tested in India have demonstrated adequate internal consistency [50] and have proven to be reliable among pregnant women residing in South India [51]. The initial item pools for the K-6 include both depression (e.g. Beck Depression Inventory) and anxiety measures (e.g. Self-Rating Anxiety Scale; [52]), and subsequent studies have used the K-6 as both a measure for depression and anxiety, but most commonly for psychological distress (e.g. [52]). Questions include symptoms of depression, fatigue, motor agitation, worthless guilt and anxiety experienced within a 30-day time frame; e.g. 'In the last 30 days, how often did you feel worthless?' [52]. Responses were based on a 4-point scale ranging from none of the time to most of the time.

Results
Pregnant women had an average age of 23.29 years (range: 19.00-35.00) and had completed an average of 9.20 years of school (range: 0.00-18.00). Nearly half of the women were primigravid (n = 44, 46.81%), whereas 44 (46.81%) had one child, four (4.2%) had two children and two (2.1%) had three children. At the time of data collection, only six participants were in their first trimester of pregnancy, 41 were in the second trimester, and 47 were in the third trimester. Table 2 presents the descriptive statistics for all study variables. According to thresholds set by Blumberg et al. [47], a large proportion of women were food insecure (42.6%). Fifty-one women (54%) reported a craving for at least one item. The average number of cravings was 0.77 (range = 0-3). Four women (67%) in the first trimester reported at least one craving, followed by 22 women (54%) in the second trimester and 25 (61%) in the third trimester. Total cravings did not differ based on trimester (χ 2 = 0.40, p = 0.82). The most common cravings were for unripe mango (82.0%) and unripe tamarind (26.6%). Figure 1 presents results for women's food cravings.

Behavioural immune system
Logistic regression was used to model the presence and absence of cravings for fruit, unripe mango and unripe tamarind. Fruit cravings were reported by 13.83% of women and were not significantly associated with the predicted measures of behavioural immunity. Unripe mango cravings were also not significantly associated with the study variables, and neither were unripe tamarind cravings. See table 3 for results, including the Nagelkerke R 2 values for each model.

Nutrient seeking
To test the nutrient seeking hypothesis, each food item was ranked according to calories per serving (see electronic supplementary material, S2). These results are also presented in figure 1  that unripe mango, the most frequently craved item, has an average of 100 calories per serving. Although unripe tamarind, the second most frequently craved item, has higher caloric density, the items with the highest calories (e.g. fish curry and biriyani) were mentioned less frequently. Results from the negative binomial regression revealed that the total score for calories of craved foods was not significantly predicted by months pregnant, triceps thickness or BMI. Interestingly, the presence of toxins or pathogens in each craved item was positively and significantly associated with the total calories composite score, indicating that cravings for calorie-rich items were confounded by possible toxicity and/or pathogenicity (Est. = 1.06, p < 3.09 × 10 −9 , 95% CI = 2.05, 4.14). McFadden's R 2 indicated that little variance was explained by this model (R 2 = 0.060).
Poisson's regression was used to model cravings for nutritious foods. Nutritional food cravings were not significantly associated with months pregnant, triceps thickness, or BMI; however, the presence of toxins or pathogens was significant (Est. = 1.69, p < 2.08 × 10 −7 , 95% CI = 2.93, 10.6). This model predicted 20.7% variance in cravings for nutritional foods based on McFadden's R 2 (table 3).

Resource scarcity
Poisson regression was used to test the resource scarcity hypothesis. Results indicated that none of the main effects were significant: food insecurity, dietary diversity, triceps or BMI. As predicted, months pregnant was also not significant. Toxicity/pathogenicity was a significant confound (Est. = 1.63, p < 7.7 × 10 −7 , 95% CI = 2.74, 10.13). The McFadden's R 2 value was 0.21.

Social bargaining
Thirty-four per cent of women reported craving at least one pathogenic or toxic food item, 0% reported cravings for 'black' foods and 90% of women reported a craving for at least one 'hot' item. Logistic regression was used to analyse etic toxic/pathogenic food cravings. As predicted, food insecurity was positively and significantly associated with cravings for dangerous foods (OR = 1.29, p < 0.02; 95% CI = 1.06, 1.6), and the main effects for pressure to have a son and number of children predicted lower odds of craving potentially toxic or pathogenic foods (OR = 0.07, p < 0.01; 95% CI = 0.01, 0.41; OR = 0.31, p < 0.02; 95% CI = 0.11, 0.77, respectively). Psychological distress was not a significant            Figure 2 displays the interaction effect of pressure to have a son and number of children, and as predicted, as the number of children increases, the pressure to have a son also increases. The Nagelkerke R 2 was 0.25 (table 3). Poisson's regression was used to model cravings for 'hot' food items. None of the variables were significant. McFadden's R 2 value was low at 0.01. Unripe mangoes were also not significantly predicted by any of the study variables. This model only predicted 0.07% of the variance in unripe mango consumption according to the Nagelkerke R 2 .
Number of living children predicted greater odds of consuming unripe tamarind (OR = 0.31, p < 0.04; 95% CI = 0.09, 0.9). The remaining predictors were not significant: psychological distress, food insecurity, pressure to have a son, and the interaction effect of pressure to have a son and number of living children. The Nagelkerke R 2 for unripe tamarind consumption was 0.11.

Discussion
This study investigated four hypotheses for food cravings in pregnancy: behavioural immunity, nutrient seeking, resource scarcity and bargaining. Similar to existing studies on pregnancy cravings, women were most likely to experience cravings for fruits [2,18,21]. Interestingly, cravings focused on unripe mango and unripe tamarind, two items that are high in antioxidants [53,54], yet culturally perceived as dangerous for the fetus if consumed in large quantities [13]. Although these cravings seem reflective of a behavioural immune strategy, the remaining analyses did not support this hypothesis, perhaps because few women were in their first trimester of pregnancy, or more importantly-these cravings likely serve an alternative purpose. The bargaining model also failed to provide strong support for unripe fruit cravings. Cravings for unripe fruits in pregnancy are common across several states in India [55]. In Karnataka, cravings for unripe mango are locally perceived to reflect a need to reduce bodily heat, or pitta according to Ayurvedic theory [56], whereas in Tamil Nadu, cravings for unripe tamarinds are considered to be heat-causing and, therefore, harmful to consume in pregnancy [57]. The sour taste of fruits like these are sometimes appealing to pregnant women due to gustatory changes [58] and could signal an increased need for vitamin C [59]. Fruits rich in vitamin C aid in the absorption of iron, and iron requirements increase during pregnancy due to the maternal-fetal transfer of iron and changes in haemodilution [60]. Iron deficiency anaemia is a major public health concern for Indian women of reproductive age and higher among pregnant women who have had multiple pregnancies [61]. These fruits are rich in antioxidants, help stabilize red blood cells [53,54,62], and have highest vitamin C content during the unripe state [63]. Therefore, perhaps these cravings enable women to increase iron bioavailability, particularly among those who have higher fertility, as this study found for unripe tamarind cravings. The social bargaining model predicted the etic classification of potentially toxic and/or pathogenic items, suggesting that women crave these foods in pregnancy when social and material resources are unstable. Psychological distress was not a significant predictor, however, which is surprising given the evidence that depressed mood states commonly reflect bargaining strategies [64,65], particularly among individuals who lack the physicality for aggressive manipulation [66]. Furthermore, this study found that the presence of toxins and/or pathogens in foods was a confounding factor for calorie-dense food and nutritional food cravings. This finding raises questions about the validity of the nutrient seeking and resource scarcity hypotheses because instead of seeking nutritious foods to protect the pregnancy, women appear to be seeking potentially teratogenic foods when resources are inadequate.
This study was an observational design and, therefore, cannot assume causality. Furthermore, only a small proportion of the participants were in their first trimester of pregnancy, which limits the ability to sufficiently test the behavioural immune hypothesis. Per the social bargaining model, more research is needed to see if the food cravings are successful in eliciting investment. The inclusion of a more precise measure of food consumption to see how nutrition influences pregnancy desires would also enhance this study. Finally, these findings are specific to pregnant women in Tamil Nadu and might not generalize to other populations of pregnant women in India and other locations. Regardless of these limitations, this study contributes to a growing literature on South Indian women that investigates how current environmental and social factors influence hypothesized dietary adaptations in pregnancy. These studies are revealing that dietary habits do not serve a single evolutionary function to prevent the ingestion of toxins and pathogens, but nonetheless, probably aim to protect the mother and/or developing fetus [13,36,67].
To conclude, although pregnancy cravings are a common phenomenon that occurs across populations, our understanding of the causes remains limited [1]. Cravings are likely to vary according to resource availability, psychosocial distress, pathogen exposure, familial support and nutritional need. This study, for instance, demonstrated that pregnancy cravings for toxic and pathogenic items most likely reflect a bargaining strategy among women who face increased social pressure to conform to cultural standards of the 'ideal' family. Another study conducted among these women in South India found that cravings for soils and other pica substances were a response to psychological distress and resource scarcity [13]. Further, women in the United States experience cravings for calorie-rich items, which may or may not be due to resource scarcity, but has implications for gestational weight gain that is difficult to lose during the postpartum period [1]. Collectively, these studies demonstrate that future research should consider the cultural and environmental niches that frame pregnancy cravings. Rather than focusing on biological causes of women's dietary abnormalities, followed with biomedical treatments, for instance, iron tablets to treat iron deficiency, public health efforts will be improved through systematic investigations of the evolutionary processes and cultural factors that shape consumption patterns.
Ethics. The Washington State University Institutional Review Board approved the research reported here. All participants provided informed consent before participating in the study. Disclaimer. The content is solely the responsibility of the author and does not necessarily represent the official views of the National Institutes of Health.