Applied Geochemistry, Vol. 7, pp.

145-158, 1992
Printed in Great Britain

0883-2927/92 $5.00+ .00 Pergamon Press Ltd

Reconstruction of human diet from 813C and ~lSN in contemporary Japanese hair: a stochastic method for estimating multi-source contribution by double isotopic tracers
MASAO MINAGAWA Mitsubishi Kasei Institute of Life Sciences, 11 Minami-Oya, Machida, Tokyo 194, Japan (Received 19 November 1990; accepted in revised form 15 November 1991)
Abstract--A stochastic method has been proposed for estimating contributions of multi-sources based on

a simultaneous measurement of C and N stable isotope compositions of a mixture. Double isotope tracer analyses essentially allow a determination of the mixingproportion of up to three sources by mass balance calculations. For more than three sources, a stochastic approach may provide a possible range of mixing proportions. In this work, a stochastic method using the Monte Carlo simulation was employed for reconstructing the dietary consumption of contemporary Japanese. The mean 6 13C and 6 15~ N values of contemporary Japanese scalp hair were found to be - 18.2 _ 0.4 and 10.3 + 0.4%orelative to PDB and the atmospheric N2, respectively. A dietary model was constructed by using mean isotope ratios of five major food groups and hair for modern Japanese. Based on this model, a stochastic method was applied to simulate human feeding. A range of dietary patterns, consistent with reasonable energy/protein uptake ratio, was estimated for fitting C and N isotope distributions of hair. The estimated mean dietary pattern has protein contributions of 35, 9, 16, 14, 27% from C3 plants, legumes, C4 plants, land animal products, and fish products, respectively, in good agreement with the observed food consumption in the National Statistics Report. The contribution of C4 type plant was estimated to range from 10 to 20% in protein, indicating much higher consumption than the statistical estimation. Thus, this stochastic method is useful in dietary analysis based on C and N isotope ratios. If the isotopic compositions of source materials are sufficientlydifferent and each source is not isotopicallyreproducible by mixing of others, the stochastic simulation may indicate accurately the different source contributions. This method is applicable also to many geochemical problems where mixing occurs.

INTRODUCTION NATURALLYoccurring stable isotopes of C and N have been often measured for tissues of humans and wild animals (ScHOENINGERe t al., 1983; DENIRO and EPSTEIN,1978). The purpose of such analyses is to estimate relative proportions of food materials in diets, as well as to search for unknown pathways of nutrient transport. The isotope composition is conventionally represented by the delta values, and is governed by the law of conservation of mass. The isotope ratio of mixed materials can be calculated by a mass balance equation based on the proportions of source materials and their isotope compositions. Hence, isotope ratio measurements have been successfully used to estimate the proportional contributions of source materials. Recent isotope studies on natural organisms have depicted a variety of C and N isotope distributions in various ecosystems. Carbon and N isotope compositions of living organisms are ultimately controlled by the isotope ratios of primary producers that are initial suppliers of organic C and N in an ecosystem (PETERSON and FRY, 1987). There are three different pathways in photosynthesis; the most abundant Calvin and Benson system (called the C3 type), Hatch and Slack system (C4 type), and crassulasian acid metabolism (CAM type) (SMrrH and EPSTEIN, 1971;

WHELANet al., 1973). Because the C isotope fractionations associated with these three modes of photosynthesis are different, these plant groups possess significantly different isotope ratios even though they all obtain their initial C from the same source, atmospheric CO 2. In contrast, the N isotope ratios of plants are mostly controlled by the isotope ratios of their substrates, such as NH 3 and NO3 in soil and water, and atmospheric di-nitrogen for leguminous plants (HOBNER, 1986). In addition to the isotopic variations in flora, feeding processes of animals modify both C and N isotope ratios of their body tissues. Previous studies on many kinds of animals have revealed that the isotope fractionation during feeding and metabolism falls in a limited range for both C and N. The N isotope ratio of many animals is normally enriched 3.5%o over that of dietary proteins (MINAGAWA and WADA, 1984). The C isotope fractionation seems to be less than that of N (RooNlC~ and WtNXERBOORN, 1986). This evidence indicates that the isotope fractionation during the human feeding and metabolism processes is approximately constant for most food sources (MINAGAWAet al., 1986). These biological processes in plants and animals finally determine the C and N isotope compositions of a given consumer in an ecosystem. Because human food has been produced from 145

146

M. Minagawa directly transferred to human hair and other macromolecules do not participate. This assumption was investigated in relation to the contributions of fat and carbohydrates. The C and N isotope ratios of human tissues differ in organs (Minagawa unpublished data; LYON and BAXTER, 1978, for C isotopes), but the isotopic trends among organs appear to be similar in other mammals. Therefore, the isotope ratios of hair may be used as representative of the human body for a human feeding model.

biological resources in natural and agricultural ecosystems, the isotope composition of the human body is also controlled by the same rules that apply to other animals. Therefore, the hypothesis that the C and N isotope compositions of animal tissues reflect those of diets is also the case for human food webs. This was partially shown by a C isotope study of modern human hair analysis (NAKAMURAet al., 1982). Further use of C and N isotope ratios for interpreting modern American human food web was reported by SCHOELLER et al. (1986). As human hair lengthens a few centimeters each month, the isotope ratios of hair reflect the food digested through the last several months. In spite of this time lag, it was shown that the C and N isotope ratios of hair can be explained fairly well by the national food consumption statistics and the mean isotope ratios of major foods in U.S.A. The apparent discriminations between diet and human hair were estimated to be - 2 and 4%0 for C and N, respectively. These studies support the above hypothesis in human food webs. On the basis of these results, the next question is how to reconstruct the dietary pattern of humans from isotope compositions of food web samples. Dietary analysis has been used in the field of archaeology and anthropology, and many reconstructions of ancient human diets have been reported. In such applications, the number of dietary sources was usually restricted to two or three sources (e.g., CHISHOLM et al., 1983; FA~NSWORTHet al., 1985). For contemporary humans, however, the food web is usually more complex than that of prehistoric humans who depended solely on local food. This is partly ascribed to the expansion of food trading in the modern world. Accordingly it is difficult to limit major foods to just two or three groups. From C and N isotope compositions, modern human food sources are characterized in at least four or five representative groups (MINACAWAet al., 1986). In this study, the C and N isotope compositions were analyzed for scalp hair and some food samples from contemporary Japanese food webs, and a new data analysis method was proposed to estimate a dietary pattern. Emphasis was placed on an evaluation of a stochastic method that enables several food sources to be reconstructed by C and N isotope analysis.

Analytic feeding model Suppose a person is living on an unchanging diet of known isotope ratios, for either C or N. The isotope ratio of the body tissues should be controlled by the isotope ratios of the food, and by the isotope fractionation between food and human tissue. In most animals, including humans, the isotopic fractionation of C and N appears to be nearly constant. If we know the isotope discrimination values (Ahuma,_diet) for a certain animal-diet system, the isotope composition of the diet can be estimated by
6m = 6h . . . . -- mh . . . . . diel'

(1)

where 6 m and 6huma n are delta values of the diet and human tissue, and Ahuman_diet is an offset due to isotope fractionation between diet and human tissue. In the case of a mixed diet composed of two different food sources, the proportion of each source can be calculated from the mass balance equation:
6m = f l 61 + (1 - - f l ) 62, fl -- 6m -- 62 (2) (3) (4)

61 - 62'
f2 = 1 - f l ,

DIETARY RECONSTRUCTION MODEL

The biological behaviors of C and N isotopes have not been studied closely in metabolic processes. In particular, it is still ambiguous as to how these isotopes are distributed in animal tissues after the macromolecules in diet are metabolized. These processes are retained as a black box. However, a relatively constant isotope fractionation apparently can be seen between diets and animal tissues. In the present model, it is assumed that because the major component of hair is protein, protein in diet is

where fl and f2 are proportions of source 1 and source 2, respectively, and b 1, d2 and 6m indicate delta values of C or N, for source 1, source 2 and a mixed diet, respectively. In this case, C and N isotope ratios of source 1, source 2 and the mixed diet give an estimation of the proportions of source 1 and source 2 in the mixed diet. In the case of three sources, simultaneous measurements of both C and N are needed to estimate the contribution of each source. Using the mass balance equation, the proportion of one source is determined analytically as follows: fl= (d m - 63)(6 ~ - d~) - (d 2 - 63)(d m - d~) (5) (61 63)(6~ d~) ( 6 ~ - - 6 - ~ " ' " (6)

f2 = 6m - d~ - fl(6] - 6~) and

f3 = 1 --f2 --f3,

(7)

Reconstruction of human diet from d13C and 615N in contemporary hair where 61, 62, 6 3 and 6 m represent C delta values of food sources 1, 2, 3 and mixed food, which is composed of food from each source in proportions )q, f2 and f3, respectively. The 6~, 6~ and 6~ indicate N delta values in the same manner. Thus, when the number of food sources is limited to three or less, the proportions of the food sources are unequivocally determined from the isotope compositions of the sources and of a mixed diet.

147

Stochastic feeding model

This analysis can be extended to those cases with more than three food sources. Suppose that a person eats food from many sources which have significantly different isotope ratios, and that the dietary intake proportions are kept constant in the period sufficiently longer than the turn-over time of hair. The isotope ratios of body tissues should be controlled by the isotope ratios of each diet alternative and their mixing proportions, as given by the following mass balance equation:
dh . . . . = Ah. . . . . diet + fl d l ~" f2

62+...+fnd

n,

(8)

where,f1 ,f2 . . . . f, and 61, d2 . .. d n are mass fraction of the diet alternatives and each isotope ratio, respectively, and Ahuman_diet is the isotope fractionation due to the feeding process. Then, the isotope ratios of the averaged diets from the isotope data for human tissue and the isotope discrimination factor can be estimated. In this case, the dietary mixing proportions cannot be calculated exactly, as was done in the analytical model. If C and N isotope compositions of every diet alternative and their consumption proportion can be obtained, it is possible to reconstruct the isotope ratios of the total food intake, i.e., the average diet eaten. The isotope ratios of the total diet do not suggest the proportion of food consumption directly, but it is useful to judge whether certain possible mixing proportions produce the observed isotope ratios. To carry this out, the following stochastic approach was developed; numerous hypothetical people are served a variety of meals that are composed of varying amounts of defined food sources. The isotope ratios of C and N of such diets were calculated. This operation was programmed in a computer simulation, and run many times, using randomly generated food proportions. Where the calculated isotope ratios were close to the observed data, the food proportions yielding that result were recorded as possible cases of dietary mixtures. After sufficient numbers of trial and error, the possible ranges of proportions were determined for each food source. With this program, it is possible to determine combinations of multiple food sources which satisfy

the observed isotope values of human tissues. If this process is repeated the above trial and error operation for all proportions with equal possibilities, the distribution of possible proportions that can satisfy the isotopic conditions can be obtained. For this purpose, a special function that generates random proportions having equivalent probabilities of occurrence is necessary, because the sum of proportions in each combination must be unity. For such a function, the Sweeping-out method is generally appropriate and is capable of generating such homogeneous proportion frequencies. Mixing proportions are generated in series from 0 to 1 for each food source using a constant interval for the contribution proportion of each source, then each combination is tested against the observed isotopic condition. This method is suitable to find all proportion ranges with equivalent probabilities, but the frequency distribution is discontinuous and it generally takes a very long time computing until all proportion ranges are covered. Another technique is a Monte Carlo method (MC), that generates random mixing proportions, then checks if each case satisfies the isotopic condition. This method is convenient for an actual operation, because computing time is saved. However, it requires an appropriate function to generate homogeneous random proportions. In this work, a program which generates random numbers under restricted conditions using the algorism of WAKIMOXO (1976) was used. The program was verified by running problems of which results are known, and by comparing the result obtained by the analytical method. The result is initially given as a frequency distribution, then illustrated by a histogram or a box-hinge plot using the accumulated proportion of each source. The most likely contribution pattern of each food source can be suggested by optimizing additional factors in the human feeding process, such as protein and energy demand.

EXPERIMENTAL METHOD

The C and N isotope ratios of human hair were analyzed for 42 contemporary Japanese (23 males and 19 females), randomly chosen from volunteers of wide age range, living in Tokyo, Okinawa, and Akita areas during 1984 and 1985. Another hair sample was the NIES No. 5 reference sample, which was powdered hair sample mixed from many contemporary Japanese males in Tokyo, combined, and distributed from the National Institute of Environmental Sciences, Tsukuba, Japan, especially for the purpose of an intercalibration of chemical analyses. Scalp hair specimens were cleaned by distilled water, methanol, chloroform and acetone. The solvents were evaporated at room temperature, and then freeze dried. Hair was cut from the scalp by stainless steel scissors. The position on the scalp and the length of hair from the skin were not specified. Normal colored hair and gray from the same old person did not result in significant differences in both dl3C and blSN isotope ratios nor the content of C and

148

M. Minagawa
MUP.A et al., 1982). These similarities suggest that the isotope ratios of the examined specimens are likely to represent the mean isotope ratios of contemporary Japanese hair. SCHOELLER et al. (1986), NAKAMURAet al. (1982) and MINAGAWAet al. (1986) reported that 613C value of human hair was enriched about 1.4-2.0%o compared to that of the diets. For N, an enrichment in 15N of -4.3%o from diet to hair was found ( M I N A G A W A et al., 1986; SCHOELLERet al., 1986). According to these observations, it appears that the 613C and 615N values of th e average Japanese diet may be - 19.6 to -20.2%0 and 6%o, respectively. This estimate will be discussed later.

N. Therefore, gray and partially gray hairs from older people were analyzed in a similar manner. The food materials for isotope analysis were obtained from several public markets in Tokyo during 1985. They were homogenized, treated with methanol-chloroform (1:2) to remove the solvent extractable lipid fraction, and then were freeze dried. The food samples used in this work, therefore comprise both protein and carbohydrate fractions. Dried hair and food samples of -10 mg were combusted and converted into COE and N2 gas by the conventional quartz combustion method described by MINAGAWAet al. (1984). Amounts of N and C in the samples were determined volumetrically using a mercury manometer prior to collection of the gases. The isotope ratios of carbon (13C/12C) and nitrogen (15N/14N) were analyzed by a FINNIGAN MAT 251 isotope ratio mass spectrometer. Isotope ratios were converted to delta values relative to the international standards; PDB and atmospheric N2, respectively (CRAIG,1957; MARIOTTI,1983), by using the isotope ratios of several laboratories working standards, such as reagents of proline, histidine and valine. The overall analytical precision of the measurement was _+0.08 for both 6]3C and dilSN in five replicate analyses of powdered human hair.

The 6 i3C and 6 X SN o f recent Japanese diet

Carbon and N isotope data of food available in the Tokyo area are listed in Table 2, and plotted in Fig. 3. The results show that contemporary Japanese foods fall into at least five groups based on C and N isotope compositions: C3 plants including rice and most vegetables; C4 plants such as corn and millet; N2-fixing plants (mostly leguminous plants); land animal products including meat, eggs and dairy products; and fish products (both marine and freshwater). The 613C of plants vary from - 3 5 to -10%o due to the differences in photosynthesis. Consequently the 613C of herbivorous animals vary depending on their feeding patterns. As the animal feeding ecology could change seasonally and locally, the isotope composition of food for humans may not be constant even for the same items. In this work, the sample size and

RESULTS AND DISCUSSION

The 613C and 6 I5N o f contemporary Japanese hair

There were no clear differences in isotope compositions between males and females, nor dependence on the age of individuals (Fig. 1, Fig. 2 and Table 1). The mean 613C and 615N of 42 individuals were consistent with those of the NIES composite hair sample. The 613C of the examined hairs also agreed well with the previous report for different Japanese groups collected at least two years previously (NAKA12

11
0 o O

_m

0 0

U)
0 ~eO O 0

8 g ~o
OOQ 0 0

10

-20

I -19

I -18 613C PDB

i -17

-16

FIG. 1. Carbon and N isotope compositions of modern Japanese hair. Open and closed circles are data from female and male samples, respectively.

Reconstruction of human diet from 613C and 615N in contemporary hair

149

-17
O

0
0

-18
CO v-

o,~OO
0

o
O

~o
-19

male 12

o female

e 0

o-~ 11

8 O
o

Z
~ 10
0

OD

_%oOo
~'~ IP"~

O0 e

2toO

4'0 Age ( a )

610

80

FUG.2. The relation between the C and N isotope compositions (613C) of hair and the age of Japanese, Open and closed circles indicate data from female and male Japanese, respectively.

selection may be still insufficient to determine the representative isotope composition of Japanese foods. However it is believed that the major food groups are represented. The average isotope compositions of these food groups are probably not very different from the results reported here. The measured isotope compositions for Japanese foods overlap the distribution of American food data reported by SCHOELLERet al. (1986) as shown in Fig. 3. This is partly because the carbon and nitrogen isotope compositions of contemporary food resources are mainly controlled by general factors, such as C3 and C4 photosynthesis in terrestrial food chains and marine food chains. In other words, the isotope data on major food groups presented here may apply to wide areas of the world. In addition, many food resources consumed in modern Japan are imports from the U.S.A., e.g. most of the soy beans and corn are imported from the U.S.A. The C isotope data on meat (beef and poultry) also indicate that they were fed forage composed of corn imported from the

U.S.A. However, dairy products in Japan, such as fresh milk, have quite negative 613C values suggesting that cows in Japan are fed dominantly C3 rich forage. The isotope analysis of foods should be continued to obtain a more complete picture of isotope values. In particular, the range in isotope composition of fish is still not well understood. This is important because modern commercial fish in Japanese markets are collected from the world oceans. The habitat and the trophic state of these new resources are not well known, so that their C and N isotope ratios may not be in the range of fish customarily used in Japan.

Mean isotope compositions of C and N for Japanese food

Because the food items which are used in contemporary Japanese life are very diverse, it is difficult to prepare average dietary mixtures for chemical analy-

Table 1. The C and N isotope compositions of Japanese scalp hair 613C S.D. (No.) Japanese Female Male Total Nies hair Japanese -18.35 + 0.33 -18.04 + 0.45 -18.19 _+0.43 - 18.00 + 0.08 -18.2 _+0.8 (19) (23) (42) (5*) (15) 615N S.D. (No.) 10.31 + 10.36 + 10.31 + 10.54 + 0.41 0.48 0.45 0.08 -(19) (23) (42) (5*) Mean age (range) 34 (3-83) 43 (5-78) 39 (3-83) Composite of male hairs NAKAMURA et al. (1982)

*Number of measurements of powdered hair.

150

M. Minagawa t h a t p r o t e i n c o n s u m p t i o n rates should be a m a j o r factor in u n d e r s t a n d i n g the N c o n t r i b u t i o n of t h e diet alternatives, in principal, because N from food is primarily f o u n d in p r o t e i n s fractions a n d m o s t N in h u m a n tissues is also f o u n d in proteins. H o w e v e r , for C, p r o t e i n is not the only p o t e n t i a l source in a n i m a l tissues. T h u s , in o r d e r to find r e a s o n a b l e factors to average the isotope c o n t r i b u t i o n of foods, it m a y be necessary to consider two c o m p o n e n t s , the c o n s u m p t i o n rate a n d the n u t r i t i o n a l factor of each food group. T h r e e weighting factors were c o n s i d e r e d in the nutritional factor: the u p t a k e rates indicated by food weight, energy a n d protein. T a k i n g these c o m p o n e n t s as weighting factors, the m e a n isotope c o m p o s i t i o n of C and N for a typical J a p a n e s e can b e estimated. This is e v a l u a t e d by c o m p a r i n g t h e m with the m e a n isotope c o m p o s i t i o n of J a p a n e s e diets e s t i m a t e d f r o m hair isotope o b s e r v a t i o n s (Table 4). A s n o t e d before, the p r o t e i n c o n s u m p t i o n rate

sis. In place of this, the dl3c a n d 615N values of the r e p r e s e n t a t i v e food groups were e s t i m a t e d from the isotope data o n individual foods. T h e food c o n s u m p tion statistics r e p r e s e n t average dietary p a t t e r n s of m a j o r foods for typical J a p a n e s e . T h e 613C a n d 615N values of m a j o r foods can b e m e a s u r e d individually. T h e 613C a n d 615N values of an average diet for a typical J a p a n e s e can t h e n b e e s t i m a t e d by use of a mass b a l a n c e equation. T a b l e 3 shows the n u t r i t i o n a l statistics of J a p a n e s e food c o n s u m p t i o n . To calculate the average isotope ratios of food groups, the arithmetic m e a n is i n a d e q u a t e , because the C or N e l e m e n t a l c o n t r i b u t i o n f r o m diet to h u m a n tissue is not u n i f o r m a m o n g food groups. F o r example, the N c o n t e n t of oil a n d sugar is negligible, whereas t h e i r c o n s u m p t i o n is significant in b o t h weight a n d calories. T h e c o n s u m p t i o n rate of each food is i m p o r t a n t to o b t a i n the relative weighting of isotopes. M e a n w h i l e , the efficiency of diet r e t e n t i o n in h u m a n tissues is not yet clear. It m a y be correct

Table 2. The C and N isotope compositions of Japanese food available in Tokyo Food Cereal Rice Wheat Millet Buckwheat foxtail millet Corn Legume Cowpeas Soy bean Vegetable Welsh onion Sweet pepper Lettuce Udo Potato Yam Fruit Banana Apple Meat Beef Beef Pork Chicken Chicken Egg Yolk White Whole Dairy Milk Fish Skipjack Tuna Puffer Carp Salmon Shellfish Clam Shortneeked clam Scalp Corb shell (n) 2 2 1 4 613C -24.7 _+2.1 -26.8 -22.6 -10.8 _+0.4 - 11.2 -10.4 -25.5 + 0.8 -26.2 -24.7 -25.8 _+ 1.6 -27.0 -29.0 -25.4 -25.0 -25.4 -25.4 -26.9 _+0.4 -27.3 -26.5 -16.3 _+ 1.7 -15.8 -19.5 - 15.9 - 16.1 - 14.3 -14.2 + 0.6 -13.5 - 14.9 - 14.3 -21.1 -21.1 -17.6 _+ 1.1 - 18.2 -16.8 - 16.1 -18.9 - 17.8 -18.0 _+ 3.1 - 17.3 - 15.2 -16.7 -23.3 615N 4.4 _+ 0.3 4.7 4.1 1.7 _+ 2.8 4.5 -1.1 2.0 4,- 0.5 2.4 1.5 3.5_+ 1.1 4.6 2.4 0.2 0.2 2.3 +_ 1.6 3.9 0.7 6.4 _+ 1.0 7.7 7.0 6.6 6.0 4.9 6.1 _+ 1.2 7.7 5.7 4.9 6.6 6.6 12.4 _+ 3.8 10.3 19.0 12. l 9.4 11.4 8.1 _+ 1.1 7.2 9.6 7.8 9.7 C3 plant C3 plant Land animal Isotope category C3 plant C4 plant
N 2 fixer

C3 plant

1 2 5

Land animal

1 4

Land animal Marine animal

Marine animal

(Brackish-water animal)

Reconstruction of human diet from (~13C and ~15N in contemporary hair

20

151

F 15

10 C3 PLANT LEGUME

MEAT,EGG & DAIRY

PLANT

-3!

-30

-25

61"aC */,,

- 20

-15

-10

-5

Fie. 3. Carbon and N isotope compositions of contemporary human food. Open circles show the Japanese food obtained in this work; closed circles were the data of U.S.A. food from SCHOELLERet al. (1986).

Table 3. Nutritional data of contemporary Japanese food groups Consumption rater (/d/person) Food Cereal Rice Barley Wheat Millet Legume Vegetable Green vegetable Oil Sugar Potato Fruit Meat Beef Pork Poultry Whale Other Egg Dairy Milk Other Fish Type* Weight (g) 309 216 0.6 91.3 0.9 66.6 73.9 178.1 17.7 11.2 63.2 140.6 71.7 16.2 27.5 16.9 0.7 10.5 40.3 116.7 108.0 8.7 90.0 Calorie (cal) 986 763 2.1 217.1 3.7 94.1 18.7 42.7 144.2 42.4 51.0 71.8 167.6 33.5 73.5 34.9 0.9 24.9 65.2 81.8 63.7 18.1 141.4 Protein (g) 21 15 0.1 6.3 0.1 6.5 1.5 2.1 0.1 0.0 1.1 0.7 12.7 3.1 4.7 3.2 0.2 1.5 5.0 3.8 3.1 3.7 18.7 Content (/g) Calorie (cal) 3.2 3.53 3.5 2.38 4.1 1.41 0.25 0.24 8.15 3.78 0.81 0.51 2.34 2.07 2.67 2.07 1.29 2.37 1.62 3.1 0.59 2.08 1.57 Protein (g) 0.068 0.068 0.167 0.069 0.111 0.098 0.02 0.012 0.006 0.000 0.017 0.005 0.177 0.191 0.171 0.189 0.286 0.143 0.124 0.033 0.029 0.425 0.208 Ratio Calorie/protein (cal/g) 47.0 52.3 21.0 34.5 37 14.5 12.5 20.3 46.4 102.6 13.2 10.8 15.6 10.9 4.5 16.6 13.0 21.5 20.3 4.9 7.6

C3 C3 C3 C4 N2 C3 C3 C3/C4 C3 C3 LA LA LA LA MA LA LA LA LA LA MA

*C3, C3 plant; CA, C4 plant; N2, di-nitrogen fixing plant; LM, terrestrial animal; MA, marine animal. tData from MlNIs~v OF HEALTHAND WELFARE, JAPAN(1987).

152

M. Minagawa

Table 4. Carbon and N isotope contributions of average Japanese foods Contribution (%)* Weighted 13C Weight -6.5 -0.0 -1.5 -5.7 -1.4 -0.4 -0.i -3.2 -1.0 -0.5 -2.1 -1.4 -23.6 Calorie -12.7 -0.0 -1.2 -0.9 -0.7 -2.0 -0.3 -1.0 -1.4 -0.5 -0.9 -1.3 -23.0 Protein -7.1 -0.0 -2.3 -1.3 -0.4 ---0.3 -2.8 -1.0 -1.1 -4.5 -20.7

Weighted 15N
Weight 1.2 0.0 0.1 0.7 0.0 0.0 0.0 0.3 0.4 0.2 0.7 0.8 4.4 Calorie 2.4 0.0 0.1 0.1 0.0 0.0 0.0 0.1 0.6 0.2 0.3 0.8 4.6 Protein 1.3 0.0 0.2 0.2 0.0 --0.0 1.1 0.4 0.3 2.7 6.3

13C
Cereal Millet Legume Vegetable Potato Oil+ Sugar Fruit Meat Egg Dairy Fish Total -24.7 -10.8 -25.5 -26.6 -25.4 -26.9 -12.0 -26.9 -16.3 -14.2 -21.1 -17.8

6 1 5 N Weight 4.7 1.7 2.0 3.5 0.2 --2.3 6.4 6.1 6.6 10.7 26.2 0.1 5.7 21.4 5.4 1.5 0.9 11.9 6.1 3.4 9.9 7.6 100

Calorie 51.6 0.2 4.9 3.2 2.7 7.5 2.2 3.8 8.8 3.4 4.3 7.4 100

Protein 28.6 0.1 8.9 4.9 1.5 0.1 0.0 1.0 17.3 6.8 5.2 25.5 100

*Calculated from data in Table 2. tlsotope data were quoted from NAKAMURA et al. (1982). from each food group should be used as the weighting factor for the isotope contributions of N. As shown in Table 4, the 6~3C estimate based on the protein fraction is 6.3%0 for the gross Japanese diet. This value is supported for the mean diet, because the 15N enrichment from diet to human hair has been reported to be -4.3%o which will give a human hair 615N near the observed hair isotope data (10.3 + 0.5%0). Other possible factors such as food weight and calorie consumption rate were also examined as shown in Table 4, and gave 4.4 and 4.6%0 as the mean isotope ratio for gross N uptake, respectively. This range is too small to interpret the human 61SN as due to the isotope enrichment on feeding process. H e n c e it would be more reliable to consider the protein fraction as the main factor controlling 6 ~sN in human food webs, and not calorie content or food weight. The C sources for human tissues are not as simple as is the case for N. The C chains of some amino acids may have formed through the T C A cycle, and consequently produce amino acids of different isotope compositions than found in protein (GALIMOV, 1985). H e n c e dietary protein may not be the only source that can alter the 6J3C of animal tissues, although the contribution of C from carbohydrate and lipid to the protein c o m p o n e n t is unknown. Food weight indicates the amount of food in the particular diets, including every macronutrient. Caloric content can be an indicator of uptake of energy that should be directly related to respiration in the T C A cycle. Results obtained with these different factors are shown in Table 4. The 613C values of the proposed Japanese diet, estimated using these different factors were - 2 3 . 6 , - 2 3 . 0 and -20.7%o, based on weight, caloric and protein content, respectively. In previous studies, the apparent enrichment factors for 613C in modern humans were reported in the range from 1.4 to 2.0%o (ScHOELLER et al., 1986; MINAGAWA et al., 1986). Taking these biasing factors into account, the estimation based on the protein contribution seems to be the most reasonable, particularly when compared with observed hair 613C values of -18.4%o. Thus, the food isotope data suggest that the isotope composition of the Japanese diet should be - 2 0 . 7 and 6.3%o for 6~3C and 615N, respectively. These estimates were consistent with the isotope composition of human hair, when the isotope fractionation between diet and hair were taken into account.

M C simulation to reconstruct Japanese dietary pattern from human isotope data

The stochastic model was used to estimate dietary patterns from the 613C and 615N observations of Japanese hair. As described before, the 613C and 615N values of the diet can be estimated by subtracting the isotope fractionation factors due to feeding. Five food sources were selected on the basis of food isotope data and the 613C and 615N values of each food group were compiled from measured data as shown in Table 5. Bars drawn as light tone in Fig. 4 show the result of the MC simulation using the isotope data from modern Japanese. In the simulation, 2000 possible combinations of five food groups are recorded, from >232,000 trials. The frequency pattern of proportions which can reproduce the same isotope composition as the diet is shown by open bars in the figure. Because the isotope data on diet was originally estimated from isotopic data on human hair (=protein), this dietary pattern shows the protein contribution from food groups. The bar height in the graph shows the number of cases that were accepted at each protein dependence level (presented in per cent). The histogram presented by the light-toned bars shows the frequency pattern that satisfies the

Reconstruction of human diet from blJC and bISN in contemporary hair

100

153

l-Z

50
o9 (.3

50

100

100

LU (.9 iii

50

0 100

isotope condition of the hair. The probable contribution of fish was restricted to a narrow range from 12.1 to 42,0% of the total protein uptake, and the contributions of C4 plants and leguminous plants were estimated to be <27 and <42%, respectively. The results show a rather wide range of possibilities, less than 50% are other food sources (meat and C3 plants). Thus, the results indicate that it is impossible to include >50% of any one source in any mixed combination. Any mixture within the above ranges can be used to produce a diet with the required isotope ratios. For example, the mean value proportions of each source is one possible solution. where the protein uptake proportions for C3. N2 fixing legumes, C4 plants, meat and fish are 23.6. 18.4, 11.0, 19.3 and 27.7%, respectively.

Nutritional analysis
50 100

Z J

~_
(.3

50

0 100

50

100

Given a food mixture, an uptake pattern of macronutrients (e.g., protein, carbohydrate and lipids) is consequently defined. These nutrient conditions also indirectly define an energy uptake rate. Hence, when a mixture of foods is specified, the energy/protein uptake ratio is simultaneously fixed. This parameter is useful to evaluate food quality. Because the energy and protein content of food groups are given in the food analysis literature (RESOURCECOUNCILSCIENCE AND TECHNOLOGYAGENCY, 1982), the energy/protein ratio of a composite diet can be obtained by the following equation:

Rm=flRi+f2Rz+...+f,~xR,,,

(9)

Z <~ C~ Z

50

50

100

160

T 120 CO L~ LU 80 Z .<

~;

40

50

100

PROTEIN D E P E N D E N C E ( % )

FIG. 4. Frequency distribution of dietary proportion by the MC stochastic analysis of Japanese diet. The lightly shaded bars indicate the frequency distribution satisfying only isotope conditions, while the dark shaded bars indicate the range which is reasonable also for the calorie/protein uptake rate (see text).

where ft J2 . . . . . J~l and R I , R 2 . . . . R, are dependencies estimated by the MC simulation and energy/ protein ratio of each food source, respectively. Thus, the energy/protein ratio can be calculated for each dietary proportion. Because the mixing proportions in the previous MC simulation model satisfy only the isotope condition of the particular diet, the frequency distribution includes inappropriate cases, from criteria such as the energy/protein ratio needed for human survival. In fact. the calculated energy/ protein uptake ratio in the previously cited diet mixture is 18.5 kcal/g on average which is much less than the observed value of 26.4 kcal/g for Japan in 1985. The energy/protein ratios of food sources vary from 5 kcal/g (meat) to 103 kcal/g (plant) in contemporary Japanese food (Table 3). On the other hand. recent national nutritional observations show that the average energy and protein uptakes for Japanese are 2000 kcal/d per person and 80 g/d per person. respectively. This means that average energy/protein uptake for Japanese is near 26.4 kcal/g. The actual uptake rate of energy and protein for >90% of the Japanese ranged from 1800 to 3600 kcal/d and from 50 to 150 g/d, respectively, in 1985. This means that the possible energy/protein ratio may vary from 12 to

154

M. Minagawa
Table 5. Isotope and nutritional data used for the MC simulation for the contemporary Japanese food Group C3-plant N2-fixer (leguminous plant) C4-plant Land animal (meat, egg etc.) Marine animal 613C (%0) -25 -25 - 11 - 15 - 18 615N (%0) 3.5 1.0 3.0 6.0 13.0 Protein content* (g/g) 0.11 0.1 0.11 0.14 0.21 Energy content* (kcal/100 g) 3.4 1.41 4.11 1.76 1.65

(~sh)
*Mean values weighted by real consumption statistics (see text). 72 kcal/g. The mixing proportions estimated in the previous MC simulation include some cases outside this range. It is therefore reasonable to omit these improbable combinations from the initial results of the MC simulation. The protein and energy content of contemporary foods have been reported in the Standard Tables of Food Composition in Japan (RESOURCECOUNCIL SCIENCE AND TECHNOLOGY AGENCY, 1982). Using these data, the protein and energy content was calculated for five food groups (Table 5). The protein or caloric content of food varies within the same food category. Therefore, to calculate the representative content of these nutritional components for each group, only the edible parts were used by weighting the actual consumption rate of major foods in 1985. For example, milk and beef are classified in the same animal food group, but their protein content is different; the protein content is 19 g/100 g for beef, but 2.9 g/100 g for raw milk. On the other hand, based on current statistics, the protein dependence on milk and beef is only 15% each of all animal products ingested. To estimate the average protein content of all animal food, the protein content of beef and milk was weighted by 0.15. By calculating other animal foods, using similar weighting corrections, the mean protein content for land animal food is estimated to be 13.9 g/100 g. In the same manner, the mean protein content and mean caloric content of all other food groups were calculated, based on food nutrition data and the actual consumption rate of food items (Table 5). The energy/protein ratio of each food group was calculated using these results. Applying these energy/protein ratios for food groups to the mixing proportion obtained by the MC simulation, the energy/protein ratios for each combination were calculated for all recorded proportions. The densely toned bars in Fig. 5 show the frequency distribution of proportions whose energy/protein ratio are in the range for current Japanese consumption (12-72 kcal/g). Hence, the distribution indicates the possible range of protein dependence which can satisfy both isotope conditions (~3C and 15N content) and the nutritional condition (protein and energy

Protein 0
I

contribution 40
---.r~ m

(*,4,) 60 80

20

C3 plant Legume C4 plant Meat Fish

i
m

~['--~
t

A i I I t

: M H W report (1988) F A O report (1984)

F1G. 5. Box hinge plot of dietary pattern as protein contribution obtained by the MC simulation for Japanese hair.

Reconstruction of human diet from 613C and 615N in contemporary hair Table 6. Classification of the isotope category of food groups and statistic results of food consumption rate as a protein source for the average Japanese Annual supply of protein (%) FAO (1983) 24.2 0.6 1.0 9.9 0.9 4.0 10.5 5.2 5.5 25.2 6.1 Annual consumption of protein (%) (MINISTRY O F HEALTH ANn WELFARE,JAPAN1987) 21.0 0.1 1.1 6.5 0.7 3.6 12.7 5.0 3.8 18.7 --

155

Food Cereal Millet Potato Legume Fruit Vegetable Meat Eggs Milk Fish Other

Isotope category C3 plant C4 plant C3 plant C3 N2 fixer C3 plant C3 plant Land animal Land animal Land animal Marine animal

demand) for mixed diets. This should provide more reliable reconstructions of the contemporary Japanese dietary pattern.

Comparison between MC simulation result and food consumption statistics

The average food consumption for Japanese is reported in two different sources: a nationwide observation based on the questionnaires administered by the MINISTRY OF HEALTH AND WELFARE, JAPAN (1987) (MHW), and the food balance sheets reported by the FOOD AND AGRICULTUREORGANIZATIONOF THE UNITED NATIONS (1984) (FAD). Table 6 shows protein consumption reported by MHW and protein supply reported by FAD of 11 food groups based on the food statistic report space (1979-1981). Because the FAD data were estimated from food statistics based on the annual balance of consumption, stock, import and export, the supply rate does not completely indicate the real consumption of food. On the other hand, the food consumption report from the MHW is based on a questionnaire given to about 20,000 people from 300 areas throughout Japan and represents an actual food consumption. However, the data were gathered during three days in November, every two years. This may cause errors due to seasonal differences in food consumption. Thus, both estimates may not accurately report the actual consumption, but only reflect the general pattern. Further, these reports give poor information about the rate of C4 plant use, because only millet and corn are distinguished as being C4. It is not known if the vegetables and fruit category includes other C4 plants. This also applies to the category of "other foods" which includes cooked foods (Table 6). Hence, the food statistics tend to give an underestimation for the C4 plant contribution in the contemporary diet. To compare these observational data with the estimation by the isotope-MC method, an isotopic category was classified by the major food item in each group.

The food consumption patterns obtained from the MC simulation are shown in a box-hinge graph (Fig. 5). The box and the vertical line in the figure indicate the range from the 25th to the 75th percentile of accepted cases and the median of the protein dependence, respectively. The solid triangles represent the mean value from the MHW observations, and the protein supply rate by the FAO report were figured by open triangles. These results clearly show that all statistical data (MHW and FAO data) fall in the range indicated by the MC simulation. Further, the consumption rate of all food groups, except C4 plants, are distributed between the 25th and the 75th percentile. In particular, the MC simulation result agreed well with FAO and MHW data for fish, C3 plants, and legumes. The estimate based on the modern food model suggested a protein contribution of C4 plant possibly ranging from 0 to 25% for contemporary Japanese, whereas the MHW data suggests a contribution of 1% or less. As noted before, the MHW data on C4 plants may give a minimum estimate of the actual C4 contribution. Estimates of food group consumption from different methods are listed in Table 7. For the MC method, the median values optmized by the energy/ protein ratio agree well with the observations of the MHW and FAO, except for C4 plants. In addition, the estimate of the amount of animal food in the diet was also consistent with the MHW and FAO results for both protein and energy. Thus, it has been demonstrated that the MC simulation has produced quite reasonable estimates of the proportional protein consumption rate in the modern Japanese food web.

Implications of frequency pattern in a stochastic simulation

The frequency pattern obtained by the MC method represents the range of the proportion of each food

156

M. Minagawa
in 1984 (MINISTRY OF HEALTH AND WELFARE, JAPAN,

source in the total diet. The frequency patterns of each food group show different distributions; some are symmetric with a single peak, others not symmetric (Fig. 4). Are these distributions related to the actual frequency pattern of human feeding? To obtain the frequency pattern it was postulated that thousands of people ate mixed diets with random proportions of food groups. Hence, it is unlikely that the generated frequency pattern reflects the real frequency pattern of feeding, because normally a human does not choose his diet randomly. Various factors, such as economic, religious and cultural reasons may influence dietary selection and may constrain it in a particular direction. If the examined human population does not have strong preferences in food choice, a stochastic model will provide a frequency pattern indicating the distribution of individual feeding patterns in the population. Such a case may be expected for populations facing food shortage, for instance, for wild animals in natural field and probably for prehistoric human populations who depended on natural food. Figure 5 shows that the mean or median value of the contributions of C3 plants, legumes and fish is close to each observed contribution reported in FAO and MHW statistics, but it is different for meat (land animal food). One of the possible reasons for this disagreement may be non-random food selection. A reason for the differing MC result is that the increasing palatability of meat products might affect current Japanese feeding habits. This tendency agrees with the fact that the consumption rate of meat and milk increased annually following World War II (MINISTRY OF HEALTH AND WELFARE, JAPAN, 1987).

Evaluation o f result obtained in the model The average uptake rate of protein for the Japanese was - 7 9 g/d based on the national observation

1987). Assuming that the dietary pattern shown in Fig. 5 is true, this protein uptake rate introduces 1706 kcal/d per person, because the mean energy/ protein ratio is to be 21.6 based on the nutrient condition of the mean Japanese diet presented in Table 3. This energy uptake rate is - 3 0 0 kcal higher than the actual observed daily energy uptake. Still not considered is the contribution of oil fat and sugar in the dietary model, because their protein content is effectively zero. These foods are normally utilized as energy sources, but do not participate in the C and N metabolism of tissues composed of protein. Consequently, these carbohydrate and lipid compounds should not affect the isotope compositions of human tissues, such as hair, muscle and bone. Further, to obtain the isotope composition of protein, the lipid fraction in hair and food samples was extracted prior to isotope analysis. Thus the dietary proportion estimated by the MC simulation represents the proportion based on protein. Reconstruction of the energy balance was based mainly on the contribution of protein. The contribution of carbohydrate is not completely included and fat, oil and sugar were excluded in the calculation. Thus, the present model should estimate the minimum amount of energy. The consumption rate of such C sources for average Japanese were reported to be 144.2 and 42.4 kcal/d, respectively. Adding these to the total energy uptake estimated from the protein intake results in an energy uptake for average Japanese of 1900 kcal. This is still - 1 0 0 kcal less than the observed average energy consumption rate. The disagreement between these is partly due to the mismatch between the actual feeding pattern that simulated the model, which was based on the random food selection. Real food selection by contemporary Japanese may tend toward food yielding greater energy when compared to a diet made up through random selection.

Table 7. Comparison of Japanese food consumption (% in protein) between the MC model and the statistical reports MC method (mean) Food group C3 plant (%) Legume (%) C4 plant (%) Land animal (%) Fish (%) Other Calorie/protein ratio Animal protein (%) Animal energy (%) Initial 23.6 18.4 11.0 19.3 27.7 -After nutrition consideration 32.0 11.3 13.8 16.2 26.7 -21.6 (18.0-31.4) 40.5 17.8 MHW report* 33.4 8.5 0.1 27.9 23.7 7.0 26.4 50.8 22.3 FAO report ~33.2 12.0 0.8 23.9 28.4 1.7 32.8 51.1 19.3

*National nutrition observation report, 1984from the M|NISTRY OFHEALTH ANDWELFARE,JAPAN(1987). t 1979--81Food balance sheets report by FOODANDAGRICULTURE ORGANIZATIONOF THE UNITED NATIONS(1984).

Reconstruction of human diet from ~13Cand 615Nin contemporary hair

157

Limitation of reconstruction
The isotopic conditions of all source materials can influence the limitation of reconstrucion of a diet mixture. If the isotope compositions of source materials are sufficiently different, it is expected that the stochastic method can specify the probable distribution of the diet components within a reliable range. The isotope differences among source groups should be larger than the isotope variations within each food group. In addition, if a source material has the same isotope composition as can arise with mixtures of other food sources, the frequency distributions can be affected. This might be happening to the estimates for land animal use in the Japanese food web model presented above. Recent cattle and poultry in Japan are fed on a mixed forage of corn and millet. Because of this C4 plant contribution and the 15N enrichment associated with feeding, a mixture of marine food, C4 plants and C3 plants can produce a quasi food mixture which has the same isotope composition as land animals. This quasi mixing value might fall in the frequency distribution obtained by the MC simulation. This may be a reason for the disagreement on the contribution of land animals in the diet as estimated by the MC method and the MHW and FAO stastical data. The number of sources chosen for a model is also an important factor affecting the reconstruction. If the model includes an improbable source, erroneous results will be obtained. In order to obtain accurate estimations, the number of sources should be minimized. On the other hand, it is more difficult to determine the mean value of one group when it is combined with other food groups to minimize sources, because it requires knowledge of the proportion of these combined food sources in advance. In most cases, this information is what is needed. Thus, the determination of source groups should be done carefully. Isotope data on diet are still limited. Further analysis of diet should be continued to revise the isotope composition of diet in future. Further, the dietary model involves several hypothetical concepts on the behavior of isotopes, which are partly related to the biochemical and physiological problems with C and N isotopes. In particular, the isotope fractionation of C and N must be confirmed. Regardless of what is revealed in future work, this approach can be applied as a general method to find the probable range of mixing proportions of known sources.

SUMMARY

A new data analysis method for estimating mixing proportions of multiple food sources from a double isotope tracer analysis has been proposed. If the mixing is governed by stochastic processes and the

law of conservation of mass holds for stable isotopes in the system, the present method is useful for analyzing the range of possible mixing proportions of known diet components. The tracer used for analysis is not limited to only stable isotopes, but other chemical components including radioactive tracers may be applicable. The number of sources must be identified prior to the stochastic analysis, and the concentration of tracers in each source must be measured. The lack of any of these data makes reasonable estimates unlikely. Additional information characterizing the system, such as energy balance, can be used to increase the degree of certainty of the results. If the proportion of any source can be estimated by other methods, it helps to decrease the number of unknown sources. Because the number of sources included in a model is critical to determining the range of possible proportions accurately, it is important to minimize the number of unknown sources which are analyzed. Unnecessary choices Of sources makes the estimate vague. The human food web model may not be the most appropriate application of this physical mixing model, because the food selection by humans may not always be performed in a stochastic manner. Moreover, the consistency of the isotope fractionation for each food source must be ascertained in more detail, whereas it was assumed to be constant for the foods examined in this study. In spite of these disadvantages, the human food web is still unique because the food consumption data have been recorded in detail. The present estimate showed good consistency in comparison to the published data. Based on the hypothesis that the C and N isotope compositions of human hair are essentially controlled by the mixing proportions of diets, and of their isotope compositions, it will be possible to reproduce the human isotope composition. If this mixing process can be described in an experimental model, it may be possible to judge if a proportion of mixed foods can be shown to agree with observed human isotope compositions. In this work, a Monte Carlo simulation was designed and employed for performing such an experiment. The C and N isotope compositions of contemporary Japanese scalp hair were measured, and gave values of - 18.1 0.4 and 10.4 + 0.4%0 on average for 613C and 615N, respectively. The MC simulation applied to these isotope data has generated a possible food dependence pattern. Other possible combinations of foods were selected by adding further conditions, on energy/protein uptake ratios in addition to the isotope conditions. It has been shown that the estimated proportions of five food groups agree well with the observed food consumption rate based on other reported data. In conclusion, the present method is useful for analyzing a dietary mixing model, involving multisources of foods, using a small number of chemical or

158

M. Minagawa HOaNER H. (1986) Isotope effects of nitrogen in the soil and biosphere. In Handbook of Environmental Isotope Geochemistry (eds P. FRITZand J. CH. FONTES), pp. 361--426. Elsevier. LYON T. D. B. and BAXTER M. S. (1978) Stable carbon isotopes in human tissues. Nature 273, 750-751. MARIOTTI A. (1983) Atmospheric nitrogen is a reliable standard for natural 615N abundance measurements. Nature 303,685-687. MINAGAWAM., KARASAWAK. and KABAYAY. (1986) Carbon and nitrogen isotope abundances in human feeding ecosystem. Geochemistry 20, 79-88 (in Japanese with English abstract). MINAGAWA M. and WADA E. (1984) Stepwise enrichments of 15N along food chains; further evidence and the relation between 15N and animal age. Geochim, cosmochim. Acta 48, 1135-1140. MINAGAWA M., WINTER D. A. and KAPLAN I. R. (1984) Comparison of Kjeldhal and combustion method for measurement of nitrogen isotope composition in natural organic matters. Anal. Chem. 56, 1859-1861. MINISTRYOF HEALTHAND WELFARE, JAPAN (1987) Kokumin eiyou no genjyou (National Status of Nutrition), pp. 148. Daiiti-syuppan Tokyo (in Japanese). NAKAMURA K., SCHOELLER D. A., WINKLER F, J. and SCHMIDTH H.-L. (1982) Geographical variations in the carbon isotope composition of the diet and hair in contemporary man. Biomed. Mass Spec. 9, 390-394. PETERSON B. J. and FRY B. (1987) Stable isotopes in ecosystem studies. Ann. Rev. Ecol. Syst. 181,293-320.
RESOURCE COUNCIL SCIENCE AND TECHNOLOGY AGENCY REFERENCES

isotope tracers. This is applicable to p r o b l e m s involving simple mixing processes or stochastic p h e n o m ena. H u m a n feeding b e h a v i o r seems to b e described by a stochastic process, e v e n for m o d e r n h u m a n s . H e n c e , this analysis should b e very useful for estimating t h e feeding habits of prehistoric h u m a n s , w h o would d e p e n d o n n a t u r a l resources with less selectivity t h a n m o d e r n people, a n d using C a n d N isotope analysis of fossil b o n e to provide the data. T h e feeding ecology of wild animals a n d fish w h o s e diet is complicated by m a n y food sources should also b e elucidated by this dietary analysis. Acknowledgements--The author thanks all volunteers for providing hair samples, and Dr B. Chisholm of the Department of Anthropology in the University of British Columbia and Dr E. Wada of Mitsubishi Kasei Institute of Life Sciences for their critical readings to improve this article. Mrs Kyoko Karasawa-Tsuru and Miss Yuko Kabaya in Mitsubishi Kasei Institute of Life Sciences helped with chemical preparations of hair and food samples. The author is greatly indebted to reviewers, especially Professor H. R. Krouse, for their critical comments. This work was partly supported by the Grant-in-Aid for Scientific Research on Priority Areas, Ministry of Education, Science and Culture. Editorial handling: P. Fritz.

CHISHOLM B., NELSON D. E. and SCHWARCZH. P. (1983) Stable carbon isotope ratios as a measure of marine versus terrestrial protein in ancient diets. Science 216, 1131-1132. CRAIG H. (1957) Isotopic standards for carbon and oxygen and correction factors for mass-spectrometric analysis of carbon dioxide. Geochim. cosmochim. Acta 12, 133-149, DENIRO M. J. and EPSTEIN S. (1978) Carbon isotopic evidence for different feeding patterns in two hyrax species occupying the same habitat. Science 201,906-908. FARNSWORTHP., BRADYJ. E., DENIRO M. J. and MACNEISH R. S. (1985) A re-evaluation of the isotopic and archaeological reconstructions of diet in the Tehuacan valley. Am. Antiq. 50, 102-116. FOOD AND AGRICULTURE ORGANIZATION OF THE UNITED NATIONS (1984) Food Balance Sheets, 1979-81 average, Rome. GALIMOV E. M. (1985) The Biological Fractionation of Isotopes. Academic Press.

(1982) Standard Tables o f Food Composition in Japan, p. 702. Okurashou-Insatukyoku, Tokyo (in Japanese). ROUNICKJ. S. and WINTERBOURNM. J. (1986) Stable carbon isotopes and carbon flow in ecosystems. BioSci. 36, 171177.
SCHOELLERD. A., MINAGAWAM., SEATERR. and KAPLAN I.

R. (1986) Stable isotopes of carbon, nitrogen and hydrogen in the contemporary North American human food web. Ecol. Food Nutrition 18, 159-170. SCHOENINGER M. J., DENIRO M. J. and TAUBER H. (1983) 15N/14N ratios of bone collagen reflect marine and terrestrial components of prehistoric human diet. Science 220, 1381-1383. SMITH B. N. and EPSTEIN S. (1971) Two categories of 13Cl12C ratios for higher plants. Plant Physiol. 47, 380384. WAKIMOTO K. (1976) Algorithm for generating a random vector with restricted integer components and their extension to matrix. Essays in Probability and Statistics, Ogawa. WHELAN T., SAKETTW. and BENEDICTC. (1973) Enzymatic fractionation of carbon isotopes by phosphoenolpyruvate carboxylase. Plant Physiol. 51, 1051-1054.