Abstract
Objective
This study investigated 36‐year BMI trajectories in twins whose BMI in young adulthood was below, within, or above their genetically predicted BMI, with a focus on twin pairs with large intrapair BMI differences (within‐pair ΔBMI ≥ 3 kg/m2).
Methods
Together, 3227 like‐sexed twin pairs (34% monozygotic) were examined at age ~30 years in 1975 and followed up in 1981, 1990, and 2011. An individual's observed BMI in 1975 was considered within (±2.0), below (<−2.0), or above (>+2.0) genetically predicted BMI, measured by a polygenic risk score of 996,919 single nucleotide polymorphisms.
Results
In monozygotic and dizygotic twin pairs with large intrapair BMI differences, the co‐twin with a higher observed BMI in 1975 deviated above predicted BMI more frequently (~2/3) than the co‐twin with a lower BMI deviated below prediction (~1/3). Individuals below, within, and above prediction in 1975 reached, respectively, normal weight, overweight, and obesity by 2011, with a mean BMI increase of 4.5 (95% CI: 4.3–4.8).
Conclusions
Categorizing BMI as below, within, or above polygenic risk score‐predicted BMI helps identifying individuals who have been resistant or susceptible to weight gain. This may provide new insights into determinants and consequences of obesity.
Study Importance.
What is already known?
Monozygotic twin pairs with large intrapair differences in BMI form an ideal case‐control study design to find nongenetic contributions to BMI. However, findings are often interpreted from the perspective of the co‐twin with a higher BMI, even though the co‐twin with a lower BMI could also be the outlier in the pair, deviating below genetically predicted BMI.
What does this study add?
We extracted two types of twins with large intrapair BMI differences: each pair had a co‐twin whose BMI approximated the genetically predicted BMI (predicted by polygenic risk scores), but the other co‐twin deviated either (1) above or (2) below predicted BMI.
Twins aged ~30 years with a BMI below genetic prediction maintained a normal weight until age ~60 whereas twins within prediction changed from normal weight to overweight and twins above prediction increased from overweight to obesity.
How might these results change the direction of research?
Via categorizing an individual's BMI as below, within, or above the genetic predisposition to BMI, we can separately investigate individuals who have been either susceptible or resistant to weight gain, which could provide valuable new insights into the etiology and pathophysiology of obesity.
INTRODUCTION
Obesity prevalence has increased worldwide between 1975 and 2011 from 3.0% to 11.6% in men and from 6.6% to 15.7% in women [1]. Most people living with obesity have multifactorial obesity [2], meaning it results from a complex interaction between multiple genetic loci, epigenetic factors, and environmental influences.
Twin studies have demonstrated that genetic factors explain nearly 80% of individual variation in body mass index (BMI) in young adulthood [3]. Genome‐wide association studies have found a large number of common genetic variants associated with BMI [4]. The identified single nucleotide polymorphisms (SNPs) can be used to calculate a polygenic risk score (PRS) for BMI, which provides a personalized estimate of genetic susceptibility to obesity.
Monozygotic (MZ) co‐twins have a virtually identical genetic sequence. This unique setting allows the investigation of multiple pairs of two individuals with the same PRS for BMI. Additionally, dizygotic (DZ) twin pairs share, on average, 50% of their segregating genes (i.e., PRS correlates around 0.5). When reared together, both MZ and DZ pairs are exposed to similar environmental influences from their family, neighborhood, and school, for example. Comparing MZ and DZ twin pairs helps disentangle genetic and environmental contributions to BMI. Previously, we and others have studied exceptional adult MZ and DZ twin pairs with large within‐twin‐pair differences in BMI [5, 6, 7, 8, 9, 10]. However, these studies were cross‐sectional, had various definitions for large intrapair BMI differences, and did not consider the genetic predisposition to obesity.
Large within‐twin‐pair BMI differences can be a transitory phenomenon as the weight of one or both siblings may fluctuate over time. Although the co‐twin with a higher BMI is often considered to have gained weight relative to his or her sibling, this may not always be the case. In some cases, the twin with the higher BMI may have a body weight that aligns with the genetic predisposition, while the twin with a lower BMI may have a body weight below genetically predicted BMI. The latter group of co‐twins either have lost weight from their natural “genetic BMI level” or have never reached this level, suggesting that they have been experiencing a protection from weight gain. Thus, earlier studies on twin pairs with large intrapair BMI differences [5, 6, 7, 8, 9, 10] have not established whether the co‐twin with higher or lower BMI is the one who deviates more from the genetic predisposition.
First, we will categorize individuals by whether their observed BMI falls below, within, or above their genetically informed BMI, as predicted by a whole‐genome PRS. Second, we will give an example on how to separate twin pairs with large intrapair BMI differences into groups of twin pairs that include a co‐twin resistant to weight gain versus twin pairs that include a co‐twin susceptible to weight gain. Similarly, we will identify twin pairs with small intrapair BMI differences in which both co‐twins have been either prone to or protected against weight gain. Third, we will investigate the new groups' 36‐year BMI trajectories.
METHODS
Participants
The twin pairs in this 36‐year longitudinal study were selected from the Older Finnish Twin Cohort, established in 1974 [11] and consisting of twins born before 1958 and alive in 1974 in Finland (Figure 1). The 1975 and 1981 surveys targeted all twins in the cohort, whereas the 1990 survey was restricted to twins born between 1930 and 1957. Genotype data were collected mainly from the late 1990s onwards. Of the twins participating in 1975, 3227 complete same‐sex twin pairs had genotype data (34% MZ). The 2011 data collection targeted twins born between 1945 and 1957, with 943 of them (44% MZ) having genotype data. Data collection was approved by the ethics committee of the Hjelt Institute, University of Helsinki, and the ethics committee of the Helsinki and Uusimaa Hospital District, Finland, and the research was conducted in accordance with the principles of the Declaration of Helsinki. All participants gave informed consent.
FIGURE 1.
Participant flowchart. MZ, monozygotic; PRS, polygenic risk score; ssDZ, same‐sex dizygotic.
Twin characteristics
Zygosity was confirmed through genotyping information derived from blood samples. We removed 19 MZ pairs (taken away from the original MZ sample size, n = 1105) with low pi‐hat values for relatedness despite being labeled as MZ pair, mostly due to sample mix‐up. Personal characteristics were self‐reported through a questionnaire. BMI based on self‐reported weight and height is validated in this cohort and found to be highly reliable, with a correlation of 0.90 in women and 0.89 in men [12]. BMI was categorized as underweight (<18.5), normal weight (18.5 to <25), overweight (25 to <30), and obesity (30+) [13].
PRS calculation
The PRS for BMI was based on 996,919 common SNPs (minor allele frequency >5% in European individuals), calculated using HapMap3 SNPs, with 27,284 FINRISK individuals as a linkage disequilibrium reference panel [14, 15]. FINRISK is another Finnish cohort without sample overlap with our twin cohort. LDpred and PLINK were used to calculate the PRS. Technical details of genotyping, imputation, and PRS calculation have been described elsewhere [15]. We created standardized values (z scores) for PRS to simplify interpretation.
Statistical analyses
Stata/MP 17.0 (StataCorp LLC., College Station, Texas) was used for statistical analyses. We performed a linear regression model to estimate how well the PRS for BMI predicted BMI with age and sex as covariates. From the regression model, we predicted BMI values and calculated how much the observed BMI differed from the predicted BMI. Subsequently, the size and direction of the differences between observed and predicted BMI were used to develop novel groups of twin pairs discordant for PRS‐enriched BMI (two groups: one co‐twin's observed BMI 1. above or 2. below PRS‐predicted BMI, with their counterpart's observed BMI within PRS prediction) and twin pairs concordant for PRS‐enriched BMI (three groups: both co‐twins' observed BMI 1. within, 2. below, or 3. above predicted BMI), as elaborated upon in the results. We used independent t tests to compare differences between MZ and DZ pairs.
Repeated measures mixed‐effects linear regressions with unstructured covariance and maximum likelihood estimation were performed to compare the observed BMI trajectories between co‐twins. This method used all available datapoints regardless of missing values at one or more visits. We performed one model without specifications for the unadjusted values in observed BMI trajectories and one adjusted model that identified repeated measures of individuals and clustering within twin pairs. Only the adjusted values were reported in the main text. We used the “contrast” command to compare within‐pair observed BMI differences at each timepoint and to estimate whether a significant change in observed BMI occurred between baseline and final follow‐up in each group of co‐twins. We performed post hoc tests of partial interaction between intrapair observed BMI differences and adjacent year of visit (1975 vs. 1981, 1981 vs. 1990, and 1990 vs. 2011). Time points were coded as zero for 1975, six for 1981, 15 for 1990, and 36 for 2011.
RESULTS
Demographics
The 6454 twin individuals from the complete twin pairs in 1975 were in their 30s, on average, and in 2011 the remaining 1886 individuals were around age 60 (Table 1). The 36‐year follow‐up showed a lower average increase in age (60.5 − 33.4 = 27.1 years) due to a higher drop‐out rate among older individuals, which resulted from study design exclusions and mortality. The study had a balanced sex representation, with over 40% of participants being men at all surveys. In 1975, 1981, and 1990, normal weight was the norm, but by 2011 overweight surpassed normal weight. Over the 36‐year period, all individuals gained, on average, 3.2 kg/m2. The BMI categories were similar to those measured by NCD‐RisC (and age‐standardized in adults over age 20) in the same years in Finland (Table S1), except that the obesity prevalence was notably lower in most years in our study sample. In 1975, about 25% of the twin pairs had large intrapair BMI differences (≥3; of which 18% were MZ and 82% DZ). This proportion increased to nearly 50% in 2011 (of which 33% were MZ and 67% DZ). A total of 405 MZ pairs (37%) and 1230 DZ pairs (57%) had large within‐pair differences in BMI at some point during the study.
TABLE 1.
Characteristics of twin individuals from complete twin pairs per timepoint
| 1975, n = 6454 | 1981, n = 5852 | 1990, n = 4168 | 2011, n = 1886 | |
|---|---|---|---|---|
| Age in 1975 (y) | 33.4 (9.4) | 39.6 (9.5) | 47.4 (7.7) | 60.5 (3.7) |
| Women/men (women %) | 3584/2870 (56) | 3308/2544 (57) | 2396/1772 (57) | 1088/798 (58) |
| Zygosity, MZ frequency (%) | 2172 (34) | 2018 (34) | 1500 (36) | 824 (44) |
| Observed BMI (kg/m2) | 23.2 (3.2) | 23.9 (3.4) | 25.0 (3.9) | 26.4 (4.4) |
| Underweight, frequency (%) | 267 (4) | 145 (2) | 69 (2) | 18 (1) |
| Normal weight, frequency (%) | 4535 (70) | 3785 (65) | 2227 (53) | 746 (40) |
| Overweight, frequency (%) | 1441 (22) | 1639 (28) | 1490 (36) | 786 (42) |
| Obesity, frequency (%) | 211 (3) | 283 (5) | 382 (9) | 336 (18) |
| Intrapair BMI difference ≥3 kg/m2, frequency (%) | 1634 (25) | 1618 (28) | 1432 (34) | 836 (44) |
Note: Values are mean (SD) or frequency (percentage). BMI categories (kg/m2): underweight (<18.5), normal weight (18.5 to <25), overweight (25 to <30), obesity (30+).
Abbreviation: MZ, monozygotic.
Observed BMI trajectories by top and bottom PRS deciles
BMI differences between top and bottom PRS deciles were substantial at all timepoints (p < 0.001; Figure 2, Table S2a, b; mean BMI and participants per PRS decile are shown in Table S3a–c). The post hoc test for partial interaction provided evidence (p < 0.01) that persons in the top 10% of PRS showed a larger increase in BMI over 36 years (5.4: from 24.6 to 30.0) than did persons in the lowest 10% (3.6: 22.1 to 25.7).
FIGURE 2.
36‐year BMI trajectories by extreme PRS deciles. Unadjusted and adjusted values of observed BMI trajectories in top and bottom PRS deciles of all twin participants. PRS1 = mean PRS of individuals in the top PRS decile, PRS2 = mean PRS of individuals in the bottom PRS decile. Adjusted p values below 0.10 are displayed in the graph, derived from contrast tests and post hoc tests of partial interaction from the repeated measures mixed‐effects linear regressions. PRS, polygenic risk score.
PRS‐predicted BMI and observed BMI in 1975
In 1975, the PRS independently explained about 4.9% of the variation in BMI (Table S4a). However, when corrected for age and sex, PRS predicted 29.3% of the BMI variation (Table S4b). We used the PRS‐predicted BMI from this corrected model as an estimation of genetically informed BMI. On average, the observed BMI differed 2.0 from the PRS‐predicted BMI (Figure S1A, B). An individual's BMI in 1975 was categorized as below prediction (observed BMI more than 2.0 below the predicted BMI), within prediction (observed BMI within 2.0 from the predicted BMI in either direction), or above prediction (observed BMI more than 2.0 above the predicted BMI).
Developing twin pair groups discordant and concordant for PRS‐enriched BMI
We categorized twin pairs based on observed BMI and PRS‐predicted BMI. First, we grouped twin pairs as having either large (ΔBMI ≥ 3 ) or small (ΔBMI < 3) observed BMI differences within pairs. Additionally, we evaluated how both co‐twins' observed BMI related to their PRS‐predicted BMI in 1975 (example in Figure 3; scatterplot of all datapoints by zygosity in Figure S2). We refer to these new groups as twin pairs discordant and concordant for PRS‐enriched BMI. Two groups of twin pairs discordant for PRS‐enriched BMI were created: (a) one co‐twin with observed BMI within predicted BMI range, the other co‐twin with observed BMI above predicted BMI, (b) one co‐twin within BMI prediction, the other co‐twin below prediction. Three groups of twin pairs concordant for PRS‐enriched BMI were developed, with both co‐twins' observed BMI: (a) within, (b) below, or (c) above predicted BMI. All pairs who failed to fulfill these group requirements were excluded. In comparison with large or small intrapair differences based on observed BMI alone, this new selection captured 109/150 (73%) MZ and 468/667 (70%) DZ pairs with large intrapair BMI differences and 696/936 (74%) MZ and 977/1474 (66%) DZ pairs with small intrapair BMI differences, which left a total of 4500 individuals in 1975 for further investigation. All groups included individuals over a wide range of PRS scores (Figure S2).
FIGURE 3.
Grouping example of monozygotic twin pairs. Example of the development of twin pair groups discordant and concordant for PRS‐enriched BMI based on the deviation between observed BMI and PRS‐predicted BMI, plotted against PRS. PRS, polygenic risk score. [Color figure can be viewed at wileyonlinelibrary.com]
Characteristics of twin pair groups discordant and concordant for PRS‐enriched BMI
In 1975, the observed BMI of 2907 individuals was in line with PRS‐predicted BMI (65%), 829 deviated below (18%), and 764 (17%) deviated above BMI prediction. In the twin pairs discordant for PRS‐enriched BMI, it was more common that the co‐twin with higher observed BMI (1975) was the co‐twin above predicted BMI in both MZ (67%) and DZ (62%) twin pairs. In the remaining 33% and 38%, the co‐twins with lower observed BMI were the exceptional co‐twins, deviating below their BMI prediction. In most twin pairs concordant for PRS‐enriched BMI, both co‐twins of the pair followed their genetic predisposition to BMI in MZ (67%) and DZ (72%) pairs. The mean PRS and predicted BMI were similar between newly created twin groups, and the proportion of women ranged from 52% to 64% between groups (group details are in Tables S4–S9).
Repeated measures linear mixed models showed that observed BMI increased over time in all PRS‐enriched BMI groups (p < 0.001). The average (95% confidence interval [CI]) observed BMI increase in the 4500 individuals from all groups combined was 4.5 (4.3–4.8; p < 0.001). The mean observed BMI of all individuals above BMI prediction in 1975 (BMI = 28.0) passed the obesity threshold by 2011 (mean BMI = 32.5), whereas those within the range of BMI prediction in 1975 (BMI = 23.0) had, on average, overweight in 2011 (BMI = 27.3). The individuals whose BMI was below their genetic predisposition in 1975 (BMI = 19.6) had, on average, normal weight in 2011 (BMI = 24.1).
Observed BMI trajectories in twin pairs discordant for PRS‐enriched BMI
We performed repeated measures linear mixed models to compare the observed BMI trajectories between co‐twins of twin pairs discordant for PRS‐enriched BMI (number of twin pairs, sex proportion, and age by timepoint in Table S5a, b). The observed BMI increases over time in the co‐twin groups from twin pairs discordant for PRS‐enriched BMI ranged from 3.7 to 5.5 in MZ twins and 4.2 to 5.5 in DZ twins (p < 0.001 in all groups). In both MZ and DZ pairs, the co‐twins below prediction stayed in the normal weight BMI category over time (except DZ co‐twins reached overweight), the co‐twins within prediction went from normal weight to overweight, and the co‐twins above prediction went from overweight to obesity (Figure 4, Tables S10 and S11a, b).
FIGURE 4.
36‐year BMI trajectories in twin pairs discordant for PRS‐enriched BMI. Unadjusted and adjusted values of observed BMI over time in MZ and DZ twin pairs discordant for PRS‐enriched BMI (large within‐twin‐pair BMI differences [≥3 kg/m2] with an additional consideration of observed and PRS‐predicted BMI). Adjusted p values on top of each timepoint reflect contrast tests of within‐pair BMI differences in that year, and adjusted p values between two timepoints reflect post hoc tests of partial interaction effects between the BMI trajectories of the co‐twins during the adjacent years. n = number of pairs, PRS = mean polygenic risk score in both MZ co‐twins in a pair, PRS1 = mean polygenic risk score of the DZ co‐twins with higher BMI in 1975, PRS2 = mean polygenic risk score of DZ co‐twins with lower BMI in 1975. Adjusted p values below 0.10 are displayed in the graph, derived from contrast tests and post hoc tests of partial interaction from the repeated measures mixed‐effects linear regressions. DZ, dizygotic; MZ, monozygotic.
The MZ and DZ twin pairs discordant for PRS‐enriched BMI in 1975 maintained substantial intrapair observed BMI differences at all timepoints (MZ ΔBMI range = 2.1–4.5, p < 0.005, DZ ΔBMI range = 3.8–5.4, p < 0.001; Figure 4, Tables S10 and S11a, b). The intrapair observed BMI differences were larger in DZ than MZ twin pairs discordant for PRS‐enriched BMI at all timepoints (independent t test p < 0.005), except in the pairs with co‐twins below prediction in 1990 (p = 0.23) and 2011 (p = 0.18; BMI contrast values in Tables S10a, b vs. S11a, b). The post hoc test for partial interaction showed that the intrapair observed BMI differences diminished between 1975 and 1981 in all groups (Figure 4, Table S12a, b), which appeared more pronounced in MZ than DZ twin pairs. In twin pairs containing co‐twins with an observed BMI above prediction, the intrapair observed BMI difference decreased in MZ pairs by 2.0 (p < 0.001) and in DZ pairs by 0.5 (p = 0.004). In twin pairs with co‐twins having an observed BMI below prediction, the observed BMI difference diminished in MZ pairs by 1.1 (p = 0.014) and in DZ pairs by 0.5 (p = 0.004).
Observed BMI trajectories in twin pairs concordant for PRS‐enriched BMI
In twin pairs concordant for PRS‐enriched BMI (number of twins, sex proportion, and age by timepoint in Table S6a, b), the observed BMI increases over time in the co‐twin groups ranged from 4.0 to 4.4 in MZ twins and 4.1 to 4.9 in DZ twins (p < 0.001 in all groups). Overall, in most cases, the twin pairs below prediction stayed within the normal weight category, the twin pairs within prediction went from normal weight to overweight, and the twin pairs above prediction went from overweight to obesity (Figure 5, Tables S13 and S14a, b).
FIGURE 5.
36‐year BMI trajectories in twin pairs concordant for PRS‐enriched BMI. Unadjusted and adjusted values of observed BMI over time in MZ and DZ twin pairs concordant for PRS‐enriched BMI (small within‐twin‐pair BMI differences [<3 kg/m2] with an additional consideration of observed and PRS‐predicted BMI). Adjusted p values on top of each timepoint reflect contrast tests of within‐pair BMI differences in that year, and adjusted p values between two timepoints reflect post hoc tests of partial interaction effects between the BMI trajectories of the co‐twins during the adjacent years. n = number of pairs, PRS = mean polygenic risk score in both MZ co‐twins in a pair, PRS1 = mean polygenic risk score of the DZ co‐twins with higher BMI in 1975, PRS2 = mean polygenic risk score of DZ co‐twins with lower BMI in 1975. Adjusted p values below 0.10 are displayed in the graph, derived from contrast tests and post hoc tests of partial interaction from the repeated measures mixed‐effects linear regressions. DZ, dizygotic; MZ, monozygotic.
Contrast tests revealed intrapair observed BMI differences in 1975 in all MZ twin pair groups and in 1981 and 1990 in the MZ pairs within predicted BMI. Intrapair observed BMI differences in DZ pairs were present at all timepoints, except in 2011 for the groups below and above predicted BMI. Intrapair differences were similar between MZ and DZ twin pairs concordant for PRS‐enriched BMI (observed BMI contrasts in Tables S13 vs. S14a, b), except in pairs with both co‐twins below prediction in 1981 (independent t test p < 0.001) and 1990 (p = 0.0021), when the intrapair observed BMI differences were ~0.7 larger within DZ pairs. Among all types of MZ twin pairs concordant for PRS‐enriched BMI, the within‐pair observed BMI difference decreased between 1975 and 1981 (pairs below: 0.7, p < 0.001; within: 0.2, p = 0.037; above: 0.8, p = 0.016; Figure 5, Table S15a, b). Among the DZ pairs within predicted BMI, the intrapair observed BMI differences decreased by 0.3 (p = 0.001) between 1975 and 1981 (Figure 5, Table S15a, b), whereas no changes were seen for other DZ pairs.
DISCUSSION
To the best of our knowledge, this study is the first of its kind to determine in a pair of twins with large intrapair BMI differences whom of the co‐twins had acquired a BMI that deviated from their genetically informed BMI (i.e., PRS‐predicted BMI). Individuals' observed BMIs were labeled as being below (18%), within (65%), or above (17%) genetically predicted BMI in 1975, based on which, we defined groups of twin pairs discordant or concordant for PRS‐enriched BMI. We then traced their BMI over 36 years. In pairs discordant for PRS‐enriched BMI, one co‐twin's observed BMI was in line with the genetic predisposition, and the other co‐twin's observed BMI deviated either below or above genetically informed BMI. Although most discordant pairs contained one co‐twin with an observed BMI above genetic prediction, about one third of the pairs included a co‐twin with a BMI far below biological predisposition. In pairs concordant for PRS‐enriched BMI, most twins' BMI (~70%) was in line with their genetic predisposition. BMI increased similarly with time in all groups, thus the BMI during young adulthood at the first visit appeared to be the main determinant of BMI at the 36‐year follow‐up visit. The group of individuals with a BMI below genetic prediction maintained, on average, a normal weight over 36 years, whereas groups within or above prediction reached overweight and obesity, respectively.
The observed BMI of the divergent co‐twin in MZ and DZ twin pairs discordant for PRS‐enriched BMI deviated more commonly above genetically informed BMI (~2/3) than below (~1/3) in 1975. Thus, in MZ twins who share an identical genetic sequence, discordance for BMI in young adulthood more likely arose from environmental influences that stimulated weight gain in one of the co‐twins rather than protecting from weight gain or promoting weight loss. By grouping individuals' BMI by their biological predisposition to BMI, we can distinguish people who have been susceptible to weight gain or not, regardless of their genetic risk. A substantial proportion (18%) of all individuals had a BMI below prediction, suggesting they have been protected from weight gain in their environment. Investigating twin pairs discordant for PRS‐enriched BMI could provide new insights into etiological determinants of body weight and weight change, regardless of genetic influences.
BMI increased, on average, 4.5 between 1975 and 2011 in all 4500 individuals from the PRS‐enriched BMI groups. BMI is known to increase with aging from early (18–25 years) to middle (65 years) adulthood with about 3.0 to 3.5 in men and 4.0 to 5.7 in women, all born before 1958, as in our study [16, 17]. The increasing obesogenic environment in Finland may also contribute, which can be derived from the increasing age‐standardized adult obesity rates in Finland between 1975 and 2011 (Table S1) [1]. This was further exemplified by a Norwegian study showing that BMI throughout adulthood was higher in individuals born after 1970 compared with before, regardless of genetic predisposition to obesity [18].
Most twin individuals (65%) approximated their genetic predisposition to BMI during young adulthood, as expected from the high heritability of BMI [3]. The genetic drive toward BMI was supported by the MZ‐DZ comparisons in our study. First, the DZ pairs more frequently had intrapair BMI differences over 3 at some point in time than MZ pairs (57% vs. 37%). Second, the intrapair BMI differences in MZ twin pairs discordant for PRS‐enriched BMI were generally smaller than those of DZ pairs. Third, the intrapair differences in twin pairs discordant for PRS‐enriched BMI seemed to decrease more strongly in MZ than DZ pairs between 1975 and 1981, displaying a trend toward a more similar BMI in MZ than DZ pairs. Furthermore, in agreement with studies linking a higher genetic obesity susceptibility to a steeper BMI trajectory [18, 19], we found that individuals in the lowest genetic risk decile increased only 3.6 over 36 years versus 5.4 in the highest decile.
Whereas twin individuals in the highest PRS decile showed a steep upward BMI trajectory and attained a mean BMI of 30.0 in 2011, a similar number of individuals deviated above their genetically informed BMI and crossed the obesity threshold in 2011 as well (32.5). On the other end, twins in the bottom PRS decile had a modest increase in BMI and acquired overweight (25.7), but twins with an observed BMI below PRS prediction maintained a normal weight up until 2011 (24.1). Genetic risk was similar between twins with observed BMI below and above predicted BMI, confirming that genetic risk is probabilistic, not deterministic, and highlighting the important contribution of the environment to development and trajectories of weight gain and loss. Future studies should investigate the determinants and health consequences of low, middle, and high BMI trajectories consistent with genetic risk, in comparison with trajectories below or above biological predisposition.
This study has both strengths and limitations. The research design consisted of a large prospective cohort of twin pairs, who were examined on four occasions over the course of 36 years, with DNA samples collected from 3227 complete twin pairs, which were used to calculate a recent genome‐wide PRS. However, the PRS still needs improvement as an independent predictor, because it accounted for a relatively small proportion of the variation in BMI (~5%). Then again, after inclusion of age and sex in the model, the PRS prediction of BMI variation was substantial (~29%), similar to the SNP‐based heritability via whole‐genome sequencing of 0.30 [20]. Furthermore, the development of BMI will always depend on both genes and environment, and therefore, the PRS alone, as an independent predictor, cannot comprehensively explain BMI. The PRS‐predicted BMI in our study is not a fixed value based solely on DNA but depends on the environment and study sample at a certain time. We further acknowledge that BMI is an imperfect determinant of obesity in an individual, but we used it as a crude indication of adiposity differences at the group level.
CONCLUSION
Via the categorization of BMI as below, within, or above genetically predicted BMI, we identified twins who presumably have been protected from or prone to weight gain up until young adulthood. In about two thirds of twin pairs discordant for PRS‐enriched BMI, the co‐twin with higher BMI diverged above the genetically informed BMI, whereas in one third of the pairs, the co‐twin with lower BMI was below genetic prediction. Even though the environment seemed to stimulate weight gain in nearly 20% of all twin individuals, a similar proportion of twins experienced a protection from weight gain in their environment (and are thus leaner than their genetic prediction), offering an opportunity to uncover environmental elements benefiting BMI levels. Hence, grouping individuals by their observed BMI in relation to their genetically informed BMI forms a new avenue for research into weight gain susceptibility. After categorization, BMI steadily increased ~4.5 with aging in most groups between 1975 and 2011, regardless of genetic risk or baseline BMI. Due to the steady BMI increase, the BMI level in 1975 mainly determined the BMI level in 2011, implying the need to understand which factors are protective against weight gain before young adulthood. Twins below PRS‐predicted BMI maintained a normal weight over time, in contrast to twins within prediction, who went from normal weight to overweight, and twins above prediction who went from overweight to obesity. The determinants and health implications of regular BMI trajectories versus PRS‐enriched BMI trajectories require further investigation.
FUNDING INFORMATION
The study was supported by the Academy of Finland (Jaakko Kaprio: 263278, 264146, 265240, 308248, 312073, 336823, 352792; Kirsi H. Pietiläinen: 272376, 266286, 314383, 335443), the Wellcome Trust Sanger Institute (Jaakko Kaprio), the Broad Institute (Jaakko Kaprio), ENGAGE—European Network for Genetic and Genomic Epidemiology (Jaakko Kaprio: FP7‐HEALTH‐F4‐2007, 201413), Finnish Medical Foundation (Kirsi H. Pietiläinen), Finnish Diabetes Research Foundation (Kirsi H. Pietiläinen), Novo Nordisk Foundation (Kirsi H. Pietiläinen: NNF10OC1013354, NNF17OC0027232, and NNF20OC0060547), Gyllenberg Foundation (Kirsi H. Pietiläinen), Sigrid Juselius Foundation (Jaakko Kaprio and Kirsi H. Pietiläinen), Helsinki University Hospital Research Funds (Kirsi H. Pietiläinen), Jalmari and Rauha Ahokas Foundation (Kirsi H. Pietiläinen), Finnish Foundation for Cardiovascular Research (Kirsi H. Pietiläinen), Government Research Funds (Kirsi H. Pietiläinen), Helsinki University Hospital (Kirsi H. Pietiläinen), and the University of Helsinki (Kirsi H. Pietiläinen).
CONFLICT OF INTEREST STATEMENT
The authors declared no conflict of interest.
Supporting information
Data S1. Supporting Information.
ACKNOWLEDGMENTS
We would like to thank the study participants for their generous, long‐term, and invaluable contributions to our research.
Berntzen BJ, Palviainen T, Silventoinen K, Pietiläinen KH, Kaprio J. Polygenic risk of obesity and BMI trajectories over 36 years: A longitudinal study of adult Finnish twins. Obesity (Silver Spring). 2023;31(12):3086‐3094. doi: 10.1002/oby.23906
REFERENCES
- 1. NCD Risk Factor Collaboration . Worldwide trends in body‐mass index, underweight, overweight, and obesity from 1975 to 2016: a pooled analysis of 2416 population‐based measurement studies in 128.9 million children, adolescents, and adults. Lancet. 2017;390(10113):2627‐2642. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Rohde K, Keller M, la Cour PL, Bluher M, Kovacs P, Bottcher Y. Genetics and epigenetics in obesity. Metabolism. 2019;92:37‐50. [DOI] [PubMed] [Google Scholar]
- 3. Silventoinen K, Jelenkovic A, Sund R, et al. Differences in genetic and environmental variation in adult BMI by sex, age, time period, and region: an individual‐based pooled analysis of 40 twin cohorts. Am J Clin Nutr. 2017;106(2):457‐466. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Choi SW, Mak TS, O'Reilly PF. Tutorial: a guide to performing polygenic risk score analyses. Nat Protoc. 2020;15(9):2759‐2772. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Berntzen BJ, Jukarainen S, Bogl LH, Rissanen A, Kaprio J, Pietilainen KH. Eating behaviors in healthy young adult twin pairs discordant for body mass index. Twin Res Hum Genet. 2019;22(4):220‐228. [DOI] [PubMed] [Google Scholar]
- 6. Naukkarinen J, Rissanen A, Kaprio J, Pietilainen KH. Causes and consequences of obesity: the contribution of recent twin studies. Int J Obes (Lond). 2012;36(8):1017‐1024. [DOI] [PubMed] [Google Scholar]
- 7. Souren NYP, Tierling S, Fryns J‐P, Derom C, Walter J, Zeegers MP. DNA methylation variability at growth‐related imprints does not contribute to overweight in monozygotic twins discordant for BMI. Obesity (Silver Spring). 2011;19(7):1519‐1522. [DOI] [PubMed] [Google Scholar]
- 8. Makowski MR, Jansen CHP, Ebersberger U, et al. Influence of acquired obesity on coronary vessel wall late gadolinium enhancement in discordant monozygote twins. Eur Radiol. 2017;27(11):4612‐4618. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Mitchell KS, Mazzeo SE, Aggen SH, et al. Characteristics of monozygotic male and female twins discordant for overweight: a descriptive study. Eat Behav. 2008;9(3):366‐369. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. van Dongen J, Willemsen G, Heijmans BT, et al. Longitudinal weight differences, gene expression and blood biomarkers in BMI‐discordant identical twins. Int J Obes (Lond). 2015;39(6):899‐909. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Kaprio J, Bollepalli S, Buchwald J, et al. The older Finnish twin cohort‐45 years of follow‐up. Twin Res Hum Genet. 2019;22(4):240‐254. [DOI] [PubMed] [Google Scholar]
- 12. Korkeila M, Kaprio J, Rissanen A, Koshenvuo M, Sorensen TI. Predictors of major weight gain in adult Finns: stress, life satisfaction and personality traits. Int J Obes Relat Metab Disord. 1998;22(10):949‐957. [DOI] [PubMed] [Google Scholar]
- 13. WHO . Physical status: the use and interpretation of anthropometry. Report of a WHO expert committee. World Health Organ Tech Rep Ser. 1995;854:1‐452. [PubMed] [Google Scholar]
- 14. Yengo L, Sidorenko J, Kemper KE, et al. Meta‐analysis of genome‐wide association studies for height and body mass index in ∼700000 individuals of European ancestry. Hum Mol Genet. 2018;27(20):3641‐3649. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Masip G, Silventoinen K, Keski‐Rahkonen A, et al. The genetic architecture of the association between eating behaviors and obesity: combining genetic twin modeling and polygenic risk scores. Am J Clin Nutr. 2020;112(4):956‐966. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Dahl AK, Reynolds CA, Fall T, Magnusson PK, Pedersen NL. Multifactorial analysis of changes in body mass index across the adult life course: a study with 65 years of follow‐up. Int J Obes (Lond). 2014;38(8):1133‐1141. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Song M, Zheng Y, Qi L, Hu FB, Chan AT, Giovannucci EL. Longitudinal analysis of genetic susceptibility and BMI throughout adult life. Diabetes. 2018;67(2):248‐255. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Brandkvist M, Bjorngaard JH, Odegard RA, Asvold BO, Sund ER, Vie GA. Quantifying the impact of genes on body mass index during the obesity epidemic: longitudinal findings from the HUNT study. BMJ. 2019;366:l4067. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Buscot MJ, Wu F, Juonala M, et al. Longitudinal association of a body mass index (BMI) genetic risk score with growth and BMI changes across the life course: the cardiovascular risk in young Finns study. Int J Obes (Lond). 2020;44(8):1733‐1742. [DOI] [PubMed] [Google Scholar]
- 20. Wainschtein P, Jain D, Zheng Z, et al. assessing the contribution of rare variants to complex trait heritability from whole‐genome sequence data. Nat Genet. 2022;54(3):263‐273. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data S1. Supporting Information.