点击显示 收起
1 From the Institute of Human Nutrition, University of Southampton, Southampton, United Kingdom (ME); Department of Child Health, Southampton University Hospitals National Health Service (NHS) Trust, Southampton, United Kingdom (PB and JM); and Public Health Nutrition Group, Obesity and Metabolic Division, Rowett Research Institute, Aberdeen, United Kingdom (DMJ)
2 Supported by the Wessex Medical Trust, United Kingdom (Hope; research grant G06; charity no. 274839) and by the Children's Research Fund, Panthenon Trust, and the Stainer Charitable Trust. 3 Reprints not available. Address correspondence to M Elia, Institute of Human Nutrition, Southampton General Hospital, Mailpoint 113, Tremona Road, Southampton, SO 16 6YD United Kingdom. E-mail: elia{at}soton.ac.uk.
ABSTRACT
Background: Intrauterine programming of body composition [percentage body fat (%BF)] has been sparsely examined with multiple independent reference techniques in children. The effects on and consequences of body build (dimensions, mass, and length of body segments) are unclear.
Objective: The study examined whether percentage fat and relation of percentage fat to body mass index (BMI; in kg/m2) in prepubertal children are programmed during intrauterine development and are dependent on body build. It also aimed to examine the extent to which height can be predicted by parental height and birth weight.
Design: Eighty-five white children (44 boys, 41 girls; aged 6.5–9.1 y) had body composition measured with a 4-component model (n = 58), dual-energy X-ray absorptiometry (n = 84), deuterium dilution (n = 81), densitometry (n = 62), and skinfold thicknesses (n = 85).
Results: An increase in birth weight of 1 SD was associated with a decrease of 1.95% fat as measured by the 4-component model (P = 0.012) and 0.82–2.75% by the other techniques. These associations were independent of age, sex, socioeconomic status, physical activity, BMI, and body build. Body build did not decrease the strength of the associations. Birth weight was a significantly better predictor of height than was self-reported midparental height, accounting for 19.4% of the variability at 5 y of age and 10.3% at 7.8 y of age (17.8% and 8.8% of which were independent of parental height at these ages, respectively).
Conclusions: Consistent trends across body-composition measurement techniques add strength to the suggestion that percentage fat in prepubertal children is programmed in utero (independently of body build and BMI). It also suggests birth weight is a better predictor of prepubertal height than is self-reported midparental height.
Key Words: Programming • body composition • birth weight • fat • protein • mineral • water • lean mass • 4-component model • height • parental height • shape • body build • leg • trunk • length
INTRODUCTION
A consistent link has been found between poor fetal growth, manifested by reduced birth weight and reduced height in both children (1-3) and adults (4-8). However, those studies have not taken into account the confounding effects of parental height (1-3), which is related to the height of the offspring through genetic and environmental factors (9-11). One way in which maternal height can influence birth weight is by its relation to pelvic shape and size (12), which is known to influence fetal growth and birth characteristics (13). Nutritional status and placental function also influence fetal growth and birth characteristics (14), but the overlap and relative importance of parental height on the one hand and other factors involved in programming of childhood or adult height on the other hand are unclear.
Body composition, including percentage fat and fat distribution, may also be programmed during early development and may contribute to the risk of cardiovascular disease, type 2 diabetes, and the metabolic syndrome in later life (15-18). Studies that investigated these influences in humans have largely relied on simple body-composition techniques, especially skinfold thicknesses, which have limitations (6), especially if programming alters the distribution of fat between subcutaneous and internal sites (19) or between body segments. Furthermore, although some studies have found a positive relation between birth weight and one or more skinfold-thickness measurements in preadolescent children (20), others have not (21). Three preadolescent studies have used dual-energy X-ray absorptiometry (DXA), which also has some limitations (22-24). One study found a significant relation between birth weight and fat-free mass but not fat mass (percentage fat was not reported) (25). Another study found a negative association between birth weight SD score (SDS) and percentage body fat (%BF), after adjusting for current weight SDS, but it did not adjust for other variables, such as age, sex, and physical activity (26). The third study found a positive relation between birth weight and both fat and fat-free mass [but again %BF was not reported or adjustments made for body mass or body mass index (BMI; in kg/m2), which might suggest subtle effects of programming on body composition] (27). Densitometry and deuterium dilution techniques assume a constancy of the density and hydration fraction of fat-free mass, respectively, which varies between persons and during growth and development (28), but they have not been used to examine the fetal programming hypothesis. A 4-component model of body composition (water, mineral, fat, and protein) (24, 29) is more robust than the individual reference methods (30), but it too has not been used to examine the fetal programming hypothesis. Therefore, although some of the classic body-composition techniques, which are based on independent principles, have not been used to examine the fetal programming hypothesis, techniques that were used have not always produced consistent results. Finally, although body build (or body shape, which can be defined in terms of length and mass per centimeter of various body segments) was implicated in altering relations of BMI to percentage fat (31, 32), its role in the fetal programming hypothesis has not been examined.
This study aimed to examine the following 2 interrelated hypotheses about programming: 1) birth weight, an endpoint of fetal growth, is a better predictor of height in preadolescent children than is midparental height, and 2) birth weight predicts percentage fat of prepubertal children independently of body build, BMI, and body-composition technique.
SUBJECTS AND METHODS
The study involved 85 white children (44 boys, 41 girls) 6.5–9.1 y of age, who were recruited from a larger sample of children (n = 786) participating in a school surveillance project on growth in the Southampton region of England. The surveillance project included all school classes of children with the desired age range. These children had routine records kept of their birth (including gestational age and birth weight) and of their weight and height at 1.5, 3, and 5 y of age, in accordance with the policy of the Local Health Authority. Only children who volunteered to participate in the study and who had a gestational age of 37 wk were included. The SDSs were calculated with the use of the 1990 UK growth charts as reference (33). The SDSs of the following variables did not differ significantly between the sample population and the surveillance population: birth weight (0.046 and 0.056, respectively; P > 0.50), current weight (0.61 and 0.35; P = 0.08), height (0.27 and 0.08; P = 0.11), and BMI (0.52 and 0.62; P > 0.50). The mean gestational age was the same (39.6 and 39.6 wk). The mean (±SD) measured weight of the 85 children at 7.8 y of age was 28.92 ± 6.83 kg, and their height was 1.277 ± 0.069 m.
The children, and their parents, were asked to attend the Clinical Research Facility at Southampton General Hospital. A questionnaire was used to obtain information about the socioeconomic status (SES) (34) and activity level [scored as 1) lower than their peers, 2) similar to their peers, and 3) more than their peers]. This 3-point scale is a simplified version of a 5-point scale, which was validated in children (35). Self-reported paternal and maternal heights were also recorded. SDSs were calculated (33) and midparental height (average of maternal and paternal SDSs) were established. The children had their weight measured electronically to the nearest 0.1 kg with the use of a Seca708 electronic weighing machine (Seca Medical Scales and Measurement Systems, Birmingham, United Kingdom), and height was measured to the nearest 0.1 cm with the use of a Leicester stadiometer (Invicta Plastics Ltd, Leicester, United Kingdom). Body composition was measured with 5 separate methods. DXA was undertaken with the use of Hologic Delphi and software version 12.2 (Vertec Scientific Ltd, Reading, United Kingdom). Air densitometry (use of whole-body air displacement plethysmography to establish body density compared with hydrodensitometry, which involves underwater weighing) was undertaken with the use of the BODPOD S/T system (Life Measurement Inc, Cranlea and Co, Birmingham, United Kingdom). Calculations assumed, first, that thoracic gas volumes (FRC + 1/2 tidal volume) were the mean values predicted by 3 child-specific equations (36-38) and, second, that the density of the fat-free body was 1.087 kg/L (30). Measurements of skinfold thickness (triceps + subscapular) were used to calculate %BF by using the equations of Slaughter et al (39). Deuterium dilution was used to measure total body water, assuming that deuterium oxide space is x1.04 water space (to account for exchangeable deuterium), and then converted to fat-free mass, assuming that it contains 76% water (30). Finally, body composition was assessed with the use of a 4-component model (29), which was applied to 58 children. The precision (1 SD) for measurement of body fat and fat-free mass was estimated to be 0.28 kg by deuterium dilution (1% body weight), 0.33 kg by air densitometry, and 0.39 kg by DXA. By using the 4-component model, the precision for measuring water was estimated to be 0.21 kg, 0.02 kg for mineral, and 0.22 kg for fat and protein.
Body dimensions were calculated with the use of DXA scans, which were photo-enlarged approximately 3-fold. Distances between specific points were measured to the nearest 0.5 mm with the use of a ruler and converted to absolute values by relating them to the dimensions of the scanned area. The following segment lengths were established: head and neck, from the top of the skull to the midpoint of a line joining the top of the acromial processes; trunk, from the lower end of the neck to the midpoint of a line joining the top of the midshaft of the femurs; and legs, as the difference between measured height and length of head, neck, and trunk. Leg length was also calculated as the sum of the upper leg (distance between upper end of the midshaft of the femur to the knee joint) and the lower leg (x1.15 distance from the knee joint to the ankle joint). This result was based on 14 measurements (1.15 ± 0.02; 10 boys, 4 girls; 7.8 ± 0.5 y of age) in children in which the position of the foot relative to the tibia allowed measurements between the bottom of the calcaneus and the ankle joint. The value agrees closely with that obtained in adults (6) (anthropometric measurements of knee joint to ankle joint and ankle joint to heal) and skeletons. The mean of the left and right sides were used in the calculations of lengths of body segments. The intraobserver CV for lengths of body segments, which were established with the measurements on separately photo-enlarged scans, were as follows: head and neck, 1.53%; trunk, 0.98%; femur, 1.93%; tibia, 1.53%; bihumeral breadth, 1.46%; and bifemoral breadth, 1.8%. The validity of the technique was assessed by scanning metal rods of known length (0.5–1.5 m). Body build, defined as the length and mass per centimeter of the legs and trunk (or head, neck, and trunk) and the bifemoral width, was accessed from the above-mentioned measurements of the various body segments.
Written informed consent was obtained from a parent of each subject. The study was approved and conducted according to the requirements of the South West Hampshire Local Ethics Committee.
Regression of body composition or body dimensions on birth weight was undertaken with and without adjustments for covariates, such as current weight, height, age, sex, SES, and physical activity. Graphical inspection of the relation between birth weight and percentage fat did not show a U-shaped curve, and, when each body-composition method was examined with linear and quadratic equations, the former were associated with better P values. Therefore, only linear regression results are reported here. The effect of body build was examined by adjusting body composition for trunk (or head, neck, and trunk) and leg characteristics (length and mass per centimeter of these segments) and for frame size (bifemoral breadth). To examine the relative importance of midparental height and birth weight on the height of the children (SDSs), correlations and semipartial correlations were calculated (40). The variables involved in calculating the semipartial correlation were height SDSs of the children (dependent variable) and birth weight SDSs and midparental height SDSs (independent variables). The difference between 2 non–independent regression coefficients was established as described by Steiger (41).
Analyses were performed with the use of the STATISTICAL PACKAGE FOR THE SOCIAL SCIENCES (version 12.0; SPSS Inc, Chicago, IL). A P value < 0.05 was considered to be significant. Values are reported as mean ± SD, except where otherwise stated.
RESULTS
The characteristics of the children studied and their weight, height, and BMI SDS at 1.5, 3, 5, and 7.8 y of age are shown in Table 1. Their current measured weight at 7.8 y of age (28.9 + 6.9 kg), height (1.28 ± 0.07 m), and BMI (17.6 ± 3.2) were all associated with positive SDSs. The mean height SDSs at 1.5, 3.5, and 6.5 y of age ranged from –0.136 to 0.271 m. The SDSs for birth weight, current weight, and current BMI did not differ significantly between sexes (44 boys, 41 girls). They also did not differ significantly between the group of 58 children (28 boys, 30 girls) whose body composition was measured with the 4-component model and the total group of 85 children. The %BF, measured by various techniques, and the density and hydration fraction of the fat-free mass are shown in Table 2.
View this table:
TABLE 1. Characteristics of children studied1
View this table:
TABLE 2. Percentage fat by various body composition techniques and density and hydration fraction of fat-free mass calculated with the 4-component model1
Birth weight SDS was a significant positive predictor of height at all these ages, explaining an increasing proportion of the variability up to 5 y of age (r2 = 0.194; r = 0.440; P < 0.001) and decreasing thereafter (r2 = 0.103 at 7.8 y of age; r = 0.320; P = 0.003). In bivariate analysis birth weight was found to be a stronger predictor of childhood height than was midparental height at all ages studied. The prediction based on parental height improved with age, from r2 of <0.01 (r < 0.1) at 1.5 y of age (NS) to r2 of 0.06 (r = 0.247; P = 0.037) at 5 y of age, when an apparent plateau was reached (P < 0.05) (Figure 1). The relation between birth weight SDS and height SDS was significantly better at 5 y of age than at 7.8 y (P < 0.05) and 1.56 y of age (P < 0.05). The relation between midparental height and height SDS at 5 y of age was not significantly stronger than at 1.5 and 3 y of age. No significant linear or quadratic trends over time were observed in the mean r2 values (between birth weight or parental height on the one hand and height on the other hand), except for the linear relation between birth weight and height of the offspring (r2 = 0.835; P = 0.030). Birth weight was not only found to be a better predictor of children's height than was midparental height SDS in bivariate regression analysis but also in multiple regression analysis involving both predictor variables; the semipartial r2 values at 7.8 y of age are shown in Figure 2. At 7.8 y of age, the relations with maternal SDSs (r2 = 0.045; r = 0.211; P = 0.050) were stronger than were paternal SDSs (r2 = 0.027; r = 0.165; P = 0.132).
FIGURE 1.. The extent to which midparental height SD scores (SDSs), birth weight SDSs, and the 2 in combination predict height SDSs of children at different ages (regression analysis; n = 71, 74, 72, 83, and 85 at consecutive time points). The r2 values > 0.0455 (n = 85) to 0.0545 (n = 71) are significant (P < 0.05; 2 tailed).
FIGURE 2.. Venn diagram shows the extent to which birth weight SD scores (SDSs) and midparental height SDSs predict height SDSs of children at 7.8 ± 0.6 y of age (n = 85). The diagram shows the values for r2 x 100 for bivariate analyses [10.3% (P = 0.003) and 6.1% (P = 0.023) of the variability, respectively], the associated semipartial r2 x 100 [8.8% (P = 0.005), 4.6% (P = 0.037) of the variability], and the multiple r2 x 100 [14.9% (P = 0.001) of the variability].
Measured height (127.7 ± 6.8 cm) at a mean age of 7.8 y correlated well (r = 0.95) with height (127.2 ± 6.7 cm) estimated by DXA. The length of all body segments tended to be greater in children with higher birth weight (Table 3). At 7.8 y of age, birth weight (SDS) was significantly related to leg length (height – length of head, neck, and trunk) and to bifemoral breadth (after adjustment for age and sex) but not to trunk length or to head, neck, and trunk length (Table 3). None of the relations between birth weight SDS and length of body segments remained significant after further adjustment for height or height SDS.
View this table:
TABLE 3. Regression analysis for change in length of body dimensions by increase in birth weight SD score, adjusted for age and sex (n = 85)
The %BF was inversely related to birth weight, both before and after adjustment for BMI, age, sex, physical activity, and SES. It remained inversely related to birth weight even when body build was included as an independent variable in multiple regression analysis (Table 4). Inverse relations between %BF and birth weight SDS were found with all body-composition techniques, and all these relations were significant, except for the technique involving deuterium dilution.
View this table:
TABLE 4. Regression analysis for change in percentage fat by increase in birth weight SD score1
The %BF (4-component model; n = 55) was not significantly related to changes in height SDS (from 1.5 to 3, 5, and 7.8 y of age) either before or after adjustment for age, sex, and current BMI. In addition, %BF was not significantly related to changes in weight SDS (from birth to 1.5, 3, 5, and 7.7.8 y of age) after adjustment for BMI SDS or for age, sex, and BMI SDS, except for the change in weight SDS between birth and 7.8 y of age [an increase of 1.447% (SE: 0.698%) body fat for each increase in weight SDS; P = 0.042; partial r = 0.274].
The effect of birth weight on fat mass, fat-free mass, and components of fat-free mass (total body water, protein, and mineral) were also examined (Table 5). With increasing birth weight an increase was observed in all the components of fat-free mass and a decrease in the fat mass. However, only some of these relations were statistically significant [ie, for fat mass and mineral mass in bivariate analysis and for fat-free mass in multiple regression analysis (see Table 5; protein mass was of borderline significance independently of age, sex, weight, height, physical activity, SES, and body build)].
View this table:
TABLE 5. Regression analysis for change in body composition (4-component model) by increase in birth weight SD score 1
DISCUSSION
This study suggests that birth weight is an important predictor of height in 6.5–9.1-y-old children and quantitatively more important than self-reported midparental height. The poorer prediction by midparental height may be due to the use of self-reported rather than measured height. Midparental height, however, may still be a better predictor of final adult height than that in early childhood. It is also possible that technical errors associated with measurement of height contribute to this possibility, because the same measurement error will result in a stronger relation in older children, who are taller, than younger children, who are shorter.
This study also shows that the relation between birth weight and %BF is already established as early as 6.5–9.1 y of age and is quantitatively similar to relations found in older men (6). Thus, in a study of older men (6), which used the same DXA and air densitometry machines as in this study, there was 4.4% more fat (as percentage of body weight) (mean of DXA and air densitometry results) in the low- than in the high-birth-weight groups (10th and 90th centiles of birth weight, respectively). This result was obtained after adjusting for differences in BMI between groups, which did not differ significantly in SES and physical activity. In this study involving prepubertal children, there was 5.0% more fat (as percentage of body weight) when birth weight was on the 10th than on the 90th centile (–1.65 and 1.65 SDSs), after adjusting for age, sex, BMI, physical activity, and SES. This is equivalent to 3.6% fat/kg difference in birth weight. These considerations have at least 2 implications. First, although BMI alone is frequently used as an indicator of adiposity or of %BF, it may provide misleading information about programming of body composition. For example, in a study of 7-y-old English children a positive relation was found between birth weight and obesity, which was defined according to BMI (42). Our study in children of a similar age provides evidence that children with a lower birth weight have more percentage fat than do children with a higher birth weight, after adjustment for current weight and height. Second, this information can help explain the increased risk of cardiovascular disease in persons with lower birth weight who, in comparison with persons with higher birth weight, are shown to have a greater %BF after controlling for BMI and other confounding variables (43, 44).
Previous studies that examined the association between birth weight and body composition have typically relied on simple body-composition techniques, which also have potentially important limitations. For example, it is possible for programming to alter the distribution of fat between subcutaneous and internal sites of the body (19) or the distribution between different subcutaneous sites, both of which could invalidate the use of skinfold thicknesses when used in a standard way to predict body composition (6). Similarly, it is theoretically possible for programming to alter the relative lengths and dimensions of body segments, which, in turn, could invalidate the use of bioimpedance or bioresistance measurements. Approximately 90% of whole-body bioimpedance arises from the limbs (45, 46), which means that changes in the composition or mass of the trunk may have little overall effect on the results. Even reference body-composition techniques, such as densitometry and deuterium dilution techniques, have their limitations (30). The 4-component model of body composition is more robust, because it does not depend on assumptions about the density or hydration fractions of the fat-free mass (47), which change during growth and which affect interpretation of classic densitometry and deuterium dilution techniques, respectively. The concerns about methods are emphasized by the significant differences in results found between several techniques. For example, %BF (see Results), which in some cases, was greater than the effect of birth weight SDSs (10th–90th centiles) on %BF, which was discussed earlier. This study is the first to apply the 4-component model of body composition to assess the fetal programming hypothesis, and it has used more independent body-composition techniques than any other previous study. However, despite the differences between methods, consistent trends on the effect of birth weight on body composition were found across a range of reference body-composition techniques that depend on different principles, including the 4-component model, which is difficult to obtain, and measurements of skinfold thicknesses, which are easy to obtain. This adds confidence to the findings of the study. The findings also add confidence to the use of skinfold-thickness measurements alone for measuring %BF in a population of this age rather than the need for other more complicated techniques. Quantitative differences between methods, in the relations between birth weight and percentage fat, reflect a combination of biological assumptions and technical errors associated with measuring body composition by various methods. The study also adds information on the relation of birth weight SDSs and components of the fat-free body, which do not necessarily change in the same proportions, reflecting biological differences between persons as well as some measurement imprecision.
Accelerated growth during early childhood was also proposed to increase risk of cardiovascular disease in later life (48, 49). However, this study found no significant association between %BF at 7.8 y of age and changes in height SDS (from 1.5 y of age onward) or weight SDS (from birth onward), after adjusting for age, sex, and BMI, apart from the weak relation with the increase in weight SDS between birth and 7.8 y of age. The lack of length measurements at birth is a limitation to this analysis, because accelerated growth in the first 1.5 y of life could be important.
Finally, because during growth and development important changes in segment lengths and body proportions occur, it can be hypothesized that relation of BMI to percentage fat in children (50) are at least partly explained by differences in body build, which may, in turn, be due to fetal programming (31, 32). Babies with a low birth weight not only end up being shorter and lighter adults than babies with high birth weight, but they also end up having a different body build, because of preferential preservation of the upper-body segment (and preferential loss in the legs which make up the lower-body segment). Malnutrition during postnatal life was also reported to preferentially shorten leg length rather than head and trunk length (51-55). A change in height as a result of shortening or lengthening of the trunk (without associated change in body composition) might be expected to alter BMI to a greater extent than the same change in height produced by lengthening or shortening the leg. This is because the change in height resulting from alterations in trunk length might be expected to be associated with a greater change in mass than are the same change in height because of alterations in the leg length. However, this approach is simplistic because it does not take into account important associations that may alter the BMI in subtle and unpredictable ways, such as associations between changes in length of segment lengths and mass per unit length. In this study, adjustment for such variables, as well as for frame size, did not reduce the strength of the relation between birth weight and % BF (even when adjustments for BMI were made). This suggests that the relations are independent of body build.
The study has a number of limitations. It involved a relatively small number of children, who volunteered to participate rather than being randomly selected from the general population; it did not examine sequential changes in body composition; it considered only white children; and it used self-reported parental height, which is not as accurate as measured height. However, the study has used a more robust method of body composition and a wider range of body-composition techniques than any previous study to examine the fetal programming hypothesis. The results support the concept that percentage fat is programmed, and it shows for the first time that body build is not responsible for this effect in children. It also suggests that programming of height of children 6.5–9.1 y of age occurs independently of midparental height.
ACKNOWLEDGMENTS
We thank the staff of the Wellcome Clinical Research Facility for their help with this study and Eric Milne for the deuterium analysis.
The author's responsibilities were as follows—ME, PB, and JM: contributed to the study design; DMJ: facilitated the deuterium analysis; JM and ME: contributed to data collection; and ME: undertook the statistical analysis. All authors provided intellectual input. None of the authors had a conflict of interest.
REFERENCES