点击显示 收起
1 From the Nutritional Epidemiology Group (JEC, JAM, BO, and VJB) and the Biostatistics Group (DCG), Centre for Epidemiology and Biostatistics, University of Leeds, Leeds, United Kingdom; the Genetic Epidemiology Division, Cancer Research UK Clinical Centre, St James’s University Hospital, Leeds, United Kingdom (JM, KK, and DTB); and the University of Wales College of Medicine, Cardiff, United Kingdom (MW)
2 Supported by Food Standards Agency grant NO5 023. The UK Women’s Cohort Study was supported by the World Cancer Research Fund. Two summer-placement students were supported by Rank Prize scholarships. 3 Reprints not available. Address correspondence to JE Cade, Nutritional Epidemiology Group, Centre for Epidemiology and Biostatistics, University of Leeds, 30-32 Hyde Terrace, Leeds LS2 9PL, United Kingdom. E-mail: j.e.cade{at}leeds.ac.uk.
ABSTRACT
Background: Gene mutations associated with iron overload have been identified. How food and nutrient intakes affect iron status in persons who may be at risk of iron overload because their genetic status is unknown.
Objective: The objective was to determine the relation between food and nutrient intakes, HFE genotype, and iron status. Foods and nutrients associated with iron stores, with adjustment for gene mutations associated with hemochromatosis, were explored.
Design: A prospective cohort of women aged 35–69 y (the UK Women’s Cohort Study) provided information on diet through a questionnaire and food diary; 6779 women in the cohort provided cheek cell samples, blood samples, or both, which were genotyped for C282Y and H63D mutations, and 2489 women also had their iron status assessed. Relations between serum ferritin and iron intake were investigated by using multiple linear regression, with adjustment for potential confounders.
Results: The strongest dietary association with serum ferritin concentration was a positive association with heme iron and not with nonheme or total iron. Weaker positive associations were seen with red and white meat, and negative associations were seen with total energy and white and brown whole-meal bread, independent of genotype and other potential confounders. The effect of genotype on ferritin concentrations primarily occurred after menopause, at which time a strong interaction between genotype and heme iron intake was observed. Other factors associated with serum ferritin concentrations were age, body mass index, blood donation, menopausal status, and HFE genotype.
Conclusions: Postmenopausal women eating a diet rich in heme iron and who were C282Y homozygotes had the highest serum ferritin concentrations.
Key Words: Iron • genotype • ferritin • UK Women’s Cohort Study • heme • diet
INTRODUCTION
Public health policy regarding iron is to try to prevent the occurrence of iron deficiency or overload. In general, the emphasis has been on preventing deficiency through public education and provision of foods with adequately bioavailable iron. However, high levels of iron storage, even within the normal range, may predispose individuals to many chronic diseases, including heart disease, diabetes, and some cancers (1).
Recent advances in genetic research have made it possible to identify genetic mutations that are associated with hemochromatosis. Hemochromatosis is an autosomal recessive disease that is characterized by progressive iron overload. Two common mutations of the HFE gene have been linked to hereditary hemochromatosis: C282Y (cysteine-to-tyrosine mutation at codon 282) and H63D (cysteine-to-glycine mutation in codon 63) (2, 3).
Homozygosity for the C282Y mutation was associated with the highest risk of iron overload from a pooled analysis of 14 studies (4). In the United Kingdom, >90% of patients with genetic hemochromatosis are homozygous for the C282Y mutation (5), although it appears that the penetrance of this mutation is very low (6, 7). This mutation is confined to populations of European origin; 12–20% of northern European populations are heterozygous for the mutation and 0.4–1% are homozygous (8). Twenty-five percent of the UK population carries the H63D mutation. This alone does not appear to be important in raising iron concentrations; however, in combination with the C282Y mutation, persons with this genotype may accumulate iron, although to a lesser extent than those homozygous for C282Y. Three percent of the population is heterozygous for both mutations (9).
The most important sources of dietary iron are foods that are rich in iron and eaten in reasonable quantities and from which iron is reasonably absorbed. In an elderly cohort, intakes of highly bioavailable forms of iron (supplemental iron and red meat) and fruit promoted high iron stores, whereas whole grains decreased stores (10). No studies to date have considered the influence of diet in relation to gene mutations and levels of iron storage. This study provides the opportunity to explore these issues in detail.
SUBJECTS AND METHODS
Study population
The UK Women’s Cohort Study has recruited >35 000 women aged 35–69 y at recruitment, which began in 1994. The women live across England, Wales, and Scotland and were selected to represent a wide range of dietary intakes with roughly one-third being vegetarian, one-third eating fish but not meat, and one-third being meat eaters. The study is described in detail elsewhere (11). Baseline questionnaire data were collected by mail from all subjects between 1995 and 1998. The study was approved by the Multiple Research Center Ethics Committee in North Yorkshire and, because the study involved a national sample, 202 local research ethics committees.
Fieldwork
Cheek cell samples
The cohort was contacted for a second time between 1999 and 2002 to acquire more detail on diet via a 4-d food diary, a 1-d activity diary, and a questionnaire. A subgroup of 15 000 subjects was also sent 2 cytology brushes and instructed to provide cheek cell samples that would be used to assess the 2 mutations of interest. These cheek cell samples were returned by mail and refrigerated until DNA extraction could be undertaken.
Blood samples
All subjects who were found to be homozygous or heterozygous for the C282Y gene mutation were asked to provide a blood sample, which would be used to confirm the result from the cheek cell DNA samples and also for the measurement of markers of iron storage. In addition, an additional 3000 women from the cohort were asked to provide a blood sample for both measurement of iron storage markers and DNA. These were randomly selected from women in the cohort who had not been asked to provide a cheek cell sample but who had previously stated that they would be prepared to provide a blood sample if it were requested.
Gene mutation analysis
A protocol for high throughput screening was developed for the 2 most common HFE mutations associated with hemochromatosis—C282Y and H63D. This involved a simple DNA extraction method that uses sodium hydroxide cell lysis for cheek cells, adapted from the method of Rudbeck and Dissing (12). This was followed by a highly sensitive fluorescent Amplification Refractory Mutation System (ARMS) technique.
DNA from the buccal cells was of insufficient quality to use a traditional restriction digest for detection of the 2 mutations, C282Y and H63D (5). The ARMS technique is a sensitive analysis that can be used for any point mutation (13, 14).
Blood iron measurements
Ferritin was measured with a 2-site chemiluminometric (sandwich) immunoassay (Bayer, Newbury, United Kingdom) performed in the Department of Clinical Biochemistry and Immunology. Methods were validated by participation in recognized quality-assurance schemes. Serum iron and unbound-iron-binding capacity were assessed with a Hitachi automated analyzer (Boehringer Mannheim, Mannheim, UK). Hemoglobin was measured with a Coulter STKR analyzer (Coulter Electronics, Hialeah, FL).
Dietary data
Dietary intake was assessed initially with a 217-item food-frequency questionnaire (FFQ) which was developed specifically to include vegetarian items and was based on the FFQ being used by the UK EPIC study (15). The FFQ data were entered twice, and the nutrient analysis was carried out by using SPSS syntax (SPSS Inc, Chicago, IL), which included nutrient values for food items on the questionnaire from standard UK food-composition tables (16). Heme iron was estimated by using specific values obtained from the literature for percentage of heme iron in meats, fish, and poultry rather than using the standard assumption of 40% (17-19) Detailed food intake and supplement information was available from a 4-d food diary for the women who had provided a blood sample. The blood sample was provided in the same year as the food diary was completed, whereas the FFQ was collected earlier. It is therefore possible that the diary may have been more likely to show associations with blood measures.
Statistical analyses
Statistical analyses were carried out by using STATA version 8 (20) and SPSS version 10.1 for WINDOWS. Serum ferritin was log transformed as the dependent variable in t tests and analysis of variance. Tukey’s honestly significant differences were used after analysis of variance to allow for multiple comparisons.
Multiple linear regression was used to investigate the relation between serum ferritin concentrations and iron intakes. Variables to be included were based on univariate analysis or had previously been shown to be associated with serum ferritin. Nondietary factors were included to control for their potential confounding effect on ferritin concentration. Two separate models were created to examine the effects of nutrients or foods and beverages on iron stores. The numbers in each model varied because of missing data for certain variables. Ferritin was included in the model as a natural log value to address the lack of normality of this variable. The antilog of regression coefficients gives the multiplicative change in serum ferritin for every unit difference in predictor. CIs for the coefficients are computed as the antilog of the geometric mean ± 1.96 x SE. The influence of menopausal status on the relation between iron intake and serum ferritin concentrations was assessed by adding the appropriate interaction term to the original model. Likelihood ratio tests were used to test the null hypothesis of no interaction.
RESULTS
Description of the sample
In total, 5349 (36%) of the 15000 women contacted returned used cytology brushes. Blood samples were returned by 1877 (63%) of the 3000 women contacted solely for blood. Combining the information from cheek cell and blood samples, C282Y and H63D genotypes were measured for 6726 (93%) and 6755 (94%), respectively, of the subjects who returned samples; 6779 women had either or both genotypes available. Blood iron storage levels were available for 2528 women, of whom 2489 also had genotype data. This subsample included 1745 C282Y wild types, 713 heterozygotes, and 31 homozygotes. Characteristics of the sample included in this article are presented in Table 1: 22% described themselves as vegetarian (although 50% of these women reported eating some meat or fish), 1% (n = 65) described themselves as vegans, 59% said that they used dietary supplements, 26% had at least a degree-level education, 64% were classified as being in the professional and managerial social class according to the NS-SEC (21), 54% had not had any menstrual periods in the past 12 mo, and 1.6% (n = 106) and 7.5% (n = 507) said that they currently or previously had diabetes and cancer, respectively.
View this table:
TABLE 1. Characteristics of the sample of the UK Women’s Cohort Study with either or both C282Y and H63D mutations analyzed
Genotyping
Final scores for the C282Y and H63D genotypes were available for 6724 (93%) and 6755 (94%) of the subjects, respectively. Of these subjects, 31 (0.5%) were homozygous for the C282Y mutation, 171 (2.5%) were compound heterozygotes (C282Y and H63D mutations), 888 (13.2%) were heterozygotes for C282Y only, 5805 (86.3%) were wild type, and 167 (2.5%) were homozygous for the H63D mutation (Table 1).
Blood iron stores
Of the 2528 women who had provided blood, results were weighted to represent results from the whole cohort. The weighted geometric mean serum ferritin concentration was 45.9 µg/L (95% CI 44.5, 47.3 µg/L) (Table 1). In the subsample that had provided blood, 78 (3%) women had serum ferritin values that were considered to be high for women (>200 µg/L) (22); 13 of these women were homozygous for the C282Y mutation (42% of homozygotes). The C282Y homozygotes had a mean (±SD) transferrin saturation of 72.7 ± 16.4%.
Nutrient intakes
The association between ferritin concentrations and nutrient intakes from the FFQ and supplements and alcohol from the food diary was explored univariately by using correlation. There were weak inverse correlations of ferritin concentrations with intakes of nonheme iron, total iron, nonstarch polysaccharide (NSP) fiber, calcium, and total energy; a stronger positive correlation with heme iron intake; and a weak positive correlation with units of alcohol (Table 2). The subsample of women who were included in the regression model had a median heme iron intake of 0.17 mg/d (range: 0–3.61 mg/d).
View this table:
TABLE 2. Nutrient intakes of the whole sample
Food intake from 4-d diary
The consumption of bread, white or brown, on the 4 d of recording was not strongly associated with serum ferritin concentrations (P for trend = 0.02 for both white bread and the brown and whole-meal categories). Increasing consumption of beans and pulses and of nuts and seeds showed a significant inverse association with ferritin concentrations (P for trend = 0.001 for both beans and pulses and nuts and seeds). Consumption of red and white meat and fish showed a strong positive association with ferritin (P for trend < 0.001 for all 3 food categories). Consumption of iron-rich foods (eg, liver, pté) showed a positive association with ferritin (P for trend = 0.003). However, the top 2 categories contained only small numbers of subjects. Consumption of iron-fortified breakfast cereal was associated with ferritin concentration, although the highest concentrations were seen with those who did not consume any or who ate it daily (P for trend = 0.01). Consumption of polyphenol-rich beverages with meals was not associated with ferritin concentrations. Information on fruit and vegetable, fruit juice, and alcohol intakes was also captured in more detail from the food diary. There was no correlation between grams of fruit, vegetables, or fruit juice eaten daily and ferritin concentrations. In contrast, there was a weak positive association between the number of units of alcohol recorded in the diary and ferritin concentrations (r = 0.06, P = 0.001).
Use of drugs and supplements
Women who said that they were blood donors had statistically significantly lower median (10th, 90th centiles) serum ferritin concentrations than did women who were not blood donors: 31 (14, 80) and 58 (23, 143) µg/L, respectively. However, women who reported having an infection or illness in the 2 wk before they provided blood samples did not have altered ferritin concentrations. We also explored whether the use of certain drugs or supplements, either on the day of blood letting or on the previous day, had any association with ferritin concentrations. Use of aspirin, other nonsteroidal antiinflammatory drugs, or warfarin was not associated with ferritin concentrations; however, the category "other medicine" was associated with ferritin; users had significantly higher ferritin concentrations than did nonusers. Consumption of dietary iron or vitamin C supplements on the day of blood letting or the previous day was not associated with differences in ferritin concentrations. Dietary supplements were recorded in detail in the food diary. The average intakes of iron, calcium, or vitamin C as dietary supplements were not strongly correlated with ferritin concentrations.
Other lifestyle factors
Age and body mass index (BMI; in kg/m2) were highly positively correlated with ferritin concentrations (age: r = 0.37, P < 0.001; BMI: r = 0.12, P < 0.001). Postmenopausal women had ferritin concentrations that were more than twice those of women who were premenopausal (postmenopausal status: 59 µg/L; premenopausal status: 27 µg/L). Ever having had hormone replacement therapy was associated with higher ferritin concentrations (yes: 57 µg/L; no: 44 µg/L). Smoking habit did not appear to be related in this cohort to ferritin concentrations, although the numbers of regular smokers (ie, every day) was rather low (n = 143).
Nutrient model
All of the factors in Table 1 were included in the multiple linear regression model, which together explained 33% of the variance in ferritin concentration (Table 3).
View this table:
TABLE 3. Multiple linear regression model for associations of nutrients and other characteristics with ferritin concentrations1
Heme iron intake from the FFQ was strongly positively associated with ferritin, such that an increase in heme iron intake of 1 mg/d was associated with a 41% increase in serum ferritin concentration (P < 0.001). Nonheme iron intake and iron intake from supplements were not associated with ferritin concentration, although they were positively associated with percentage transferrin saturation.
Alcohol intake from the food diary was also strongly positively associated with ferritin. Ferritin concentrations increased by 1.4% (P < 0.001) with each 1-unit increase in alcohol intake. Vitamin C from the diet (FFQ) and calcium supplements (food diary) were also positively associated with ferritin (P = 0.006 and 0.005, respectively). Total energy intake (FFQ) was inversely associated with ferritin; for every 100-kcal increase in energy intake, ferritin concentrations decreased by 1% (P = 0.003).
Of the nondietary factors, age and BMI were positively associated with ferritin concentrations (P < 0.001 and 0.03, respectively). Ferritin concentrations increased by 1.3% with each 1-y increase in age. Being a blood donor was strongly linked with lower ferritin concentrations; donors had 33% lower concentrations than did nondonors (P < 0.001). Being postmenopausal was associated with 68% higher ferritin concentrations than was being premenopausal (P < 0.001). The only contrast that was statistically significant for HFE genotype was C282Y homozygote compared with wild type; the subjects with C282Y homozygote status had 141% (95% CI: 90, 206%) higher concentrations of ferritin (P < 0.001).
Exploration of the interactions between HFE genotype and menopausal status showed that heme iron affected ferritin concentrations in postmenopausal C282Y homozygotes rather than in premenopausal C282Y homozygotes (Figure 1). There was a stronger relation between heme iron intake and predicted serum ferritin concentrations in the homozygotes, but this was not observed in heterozygotes or wild types and narrowly missed being statistically significant in compound heterozygotes. In addition, the C282Y homozygotes had higher ferritin concentrations overall. Predicted serum ferritin concentrations, based on the linear regression model for nutrients (including interaction terms for genetic status and menopausal status), are shown in Figure 1. The homozygotes appear to be in 2 groups, 1 with higher predicted serum ferritin concentrations than the other; these groups separate pre- and postmenopausal homozygotes. The premenopausal homozygotes had ferritin concentrations similar to those of the heterozygotes and wild types.
FIGURE 1.. Plot of predicted serum ferritin concentrations from a regression model including interactions between genotype and menopausal status. n = 29 C282Y homozygotes, n = 631 C282Y heterozygotes, and n = 1541 C282Y wild type. The wild-type group included all subjects who were either H63D wild type or heterozygotes.
Food and beverage model
All of the variables included in the food model accounted for 34% of the variance in ferritin concentrations (Table 4). As with the nutrient model, age and BMI were positively associated with ferritin (P < 0.001 and 0.003, respectively). Being a blood donor was negatively associated and being postmenopausal was positively associated with ferritin. Ferritin concentrations were 32% lower in blood donors than in nondonors (P < 0.001) and were 66% higher in postmenopausal women than in premenopausal women (P < 0.001). In this model, 2 contrasts were statistically significant with regard to HFE genotype, being either a C282Y homozygote or a compound heterozygote compared with a wild type. Both of these genetic mutations were associated with higher concentrations of ferritin. The C282Y homozygote group had ferritin concentrations that were 151% higher than those in wild-type women (P < 0.001), and the compound heterozygotes had ferritin concentrations that were 18% higher than those in wild-type women (P = 0.008). The C282Y heterozygote group had a ferritin concentration that was 7% higher than that of the wild-type women; this difference was nearly significant (P = 0.07).
View this table:
TABLE 4. Multiple linear regression model for associations of foods and other characteristics with ferritin concentrations1
Despite white bread being fortified with iron, both white and brown whole-meal bread consumption were associated with lower ferritin concentrations if they were consumed daily compared with no consumption at all. Daily white bread consumption led to 24% lower ferritin concentrations, and daily brown whole-meal bread consumption led to 21% lower ferritin concentrations than did no consumption of these foods (P < 0.001 for both). Red and white meats were positively associated with ferritin concentrations; daily red meat eaters had 36% higher ferritin concentrations than did nonconsumers (P < 0.001). However, fish, fortified breakfast cereals, beans and pulses, iron-rich food, and consumption of polyphenol-rich beverages with meals were not associated with ferritin. Alcohol intake was again positively associated with ferritin concentrations (P < 0.001). Fruit juice consumption (g/d) was negatively associated with ferritin concentrations (P = 0.01), which possibly reflected the large number of vegetarians in the cohort. The amount of calcium supplements consumed was positively associated with ferritin concentrations, albeit weakly (P = 0.007).
DISCUSSION
This is the first large epidemiologic study to explore the relation between dietary iron intake, HFE genotype, and iron status. The estimated prevalence of the C282Y and H63D mutations in this population was similar to that reported previously in the United Kingdom (8, 9) and the United States (6).
Total iron intake and iron stores
There is no clear picture from the literature of the relation between dietary iron intake and body iron stores. The use of serum ferritin as a measure of iron stores should be made with caution (23) because there may be considerable biological variation resulting from, for example, the presence of chronic disease or infection. Inconsistencies may be due to methodologic problems, which we attempted to address. To avoid an inadequate or short-term assessment of diet, we used an FFQ that covered the previous 12 mo to assess nutrient intake. We studied a large population-based sample. Previous studies did not adjust for all of the potential confounders (of which there are many) and iron-absorption modifiers. We adjusted for various nondietary factors, including genotype, taking into account absorption modifiers such as dietary supplements of calcium and vitamin C and polyphenol consumption.
Most dietary iron is present as nonheme iron. The absorption of nonheme iron is affected by many dietary factors. Dietary factors do not tend to influence heme-iron absorption; only calcium is believed to inhibit heme-iron absorption (24). Our results showed lower intakes of heme iron than found in a national survey of British women of a similar age (1.0 mg/d) (25). This may have been due to the large numbers of vegetarians and low meat eaters in the cohort or differences in methodology.
Heme iron
Our study showed that heme iron—and not nonheme or supplementary iron—was strongly positively associated with serum ferritin concentration. Specific values for the heme-iron content of individual food items have been used to calculate total heme iron intake (19). Other studies have used a standard formula of 40% total iron in meat, fish, and poultry for the assessment of heme iron in the diet (18). Heme iron as a percentage of total iron varies from 50% to 80% in red meat and from 25% to 40% in white meat (26). Use of the standard formula may explain the inconsistency with the Framingham Study, which showed that red meat but not heme iron intake was independently predictive of high iron stores in the elderly (10). Alternatively, red meat could be a better predictor of serum ferritin than heme iron because meat also contains nonheme iron, which could be available if consumed with enhancers of nonheme iron. A previous analysis of the same data showed a positive association of heme iron but not of nonheme iron with serum ferritin concentrations (27). This analysis did not account for genotype. Our results showed that the predominant effects of the C282Y mutation are likely to be on the absorption and metabolism of heme iron.
Genetics of iron metabolism
Our linear regression models including nutrients, lifestyle factors, and HFE genotype explained 33% of the variance in serum ferritin, and a similar model replacing nutrients with foods explained 34% of the variance in serum ferritin. Within each model, there were many highly significant predictive factors that together explained a considerable amount of the variance.
This study has shown interaction effects between genetic status and menopausal status on serum ferritin. The postmenopausal C282Y homozygotes showed a much stronger influence of heme iron on iron status than did wild-type women. In addition, only the postmenopausal C282Y homozygote women had higher ferritin concentrations than did the wild-type women.
Predictors of serum ferritin
Heme iron (nutrient model) and red meat and white meat (food model) were significant positive predictors of serum ferritin. Nonheme iron was not a predictor of serum ferritin. Heme iron, but not nonheme iron, was also found to be associated with serum ferritin concentrations in elderly subjects from the Framingham Heart Study (27). This study found that supplemental iron was positively associated with serum ferritin. Iron supplements were not associated with serum ferritin in our study, although they were significantly positively associated with transferrin saturation. These differences may have been due, in part, to the fact that the Framingham Study did not account for genotype.
Dietary factors known to enhance iron absorption (23) and found in our study to be positively associated with serum ferritin were vitamin C and alcohol (28). However, there was no effect of vitamin C supplements on serum ferritin. A previous study using very large doses of vitamin C showed no significant effect on iron reserves (29) and concluded that altering the availability of nonheme dietary iron has little effect on iron status when the diet contains substantial amounts of meat. Alcohol has been shown to be positively associated with serum ferritin (27, 30). Although the precise mechanism for this effect is not clear, it may in part be related to alcohol increasing the intestinal mucosa permeability to iron and the relatively high amount of iron present in certain wines and beers (31).
Nuts and seeds are known inhibitors of iron absorption, and these foods were negatively associated with serum ferritin. There was a lack of effect on serum ferritin of some other potential inhibitors of iron absorption: calcium, polyphenols, and fiber. Calcium has been shown to be negatively related to serum ferritin concentrations (32). However, calcium may be only a weak inhibitor of iron absorption when studied in the context of the whole diet (23, 33). In fact, our study found that calcium supplements were positively associated with serum ferritin concentrations.
Increasing energy intakes and increasing BMIs were associated with higher serum ferritin concentrations, independent of iron intake and other factors. BMI in an elderly American population (27) and in postmenopausal Danish women (30) has also been shown to be positively associated with serum ferritin.
There was no effect of fortified breakfast cereals on serum ferritin concentrations, and consumption of both white and brown breads and fruit juice were significantly negatively associated with serum ferritin. Since 1953, all flour in the United Kingdom, except whole meal, has been fortified. It has been suggested that this is no longer necessary because there is evidence that little if any of this iron is absorbed (31). These results support this suggestion. A clear understanding of factors that influence iron absorption is important when considering fortification strategies (33).
As expected, blood donation was associated with reduced serum ferritin concentrations. We also found a positive relation with age and serum ferritin. In general, serum ferritin remains low in women until the menopause, after which time it increases rapidly (34).
Our results may have been influenced by the use of an FFQ to assess nutrient intakes. However, fully coded 4-d food diaries were available on a subgroup of 430 persons. Comparison between the 2 methods showed strong correlations, particularly for heme iron (data not shown).
Conclusion
Women with the highest serum ferritin concentrations were postmenopausal, were C282Y homozygotes, and consumed a diet rich in heme iron. In premenopausal women, poor regulation of iron absorption may be masked by menstrual blood loss, and iron overload may still be possible later in life.
ACKNOWLEDGMENTS
We thank James Thomas for designing and managing the database; the nutrition students Breige McNulty, Caroline Owen, Emma Elliott, Michelle Spence, Gavin McArt, Sinead Boylan, Aine McConnon, Alison Long, Nicola Leadbetter, Leyla Okhai, Laura Henderson, Jenny Lui, and Tina Hawke for coding the food diaries and mailing the cheek cell samples; Zoe Kennedy for assisting with the DNA analysis and extraction; Rupert Gaut for technical advice and DNA extraction; and Jenny Barrett for PhD student support.
JEC had the original idea for the research, and DTB, VJB, DCG, JM, and MW contributed to the study concept and design, interpreted the results, and critically reviewed the manuscript. JAM and KK helped to develop the laboratory techniques and undertook the laboratory analyses. BO developed the nutrient analysis methods. All authors commented on the manuscript. There were no conflicts of interest.
REFERENCES