Adults prenatally exposed to the Dutch Famine exhibit a metabolic signature associated with a broad spectrum of common diseases

26

Exposure to famine in the prenatal period is associated with an increased risk of 27 metabolic disease, including obesity and type -2 diabetes. We employ ed nuclear magnetic 28 resonance (NMR) metabolomic profiling to provide a deeper insight into the metabolic changes 29 associated with survival of prenatal famine exposure during the Dutch Famine at the end of 30 World War II and explore their link to disease. 31

NMR metabolomics data were generated from serum in 480 individuals prenatally 32 exposed to famine (mean 58.8 years, 0.5 SD) and 464 controls (mean 57.9 years, 5.4 SD). We 33 tested associations of prenatal famine exposure with levels of 168 individual metabolic 34 biomarkers and compared the metabolic biomarker signature of famine exposure with those of 35 154 common diseases. 36

Prenatal famine exposure was associated with higher concentrations of branched -37 chain amino acids ((iso)-leucine), aromatic amino acid (tyrosine), and glucose in later life (0.2-38 0.3 SD, p < 3x10-3). The metabolic biomarker signature of prenatal famine exposure was 39 positively correlated to that of incident type -2 diabetes (r = 0.77, p = 3x10 -27), also when re-40 estimating the signature of prenatal famine exposure among individuals without diabetes ( r = 41 0.67, p = 1x10-18). Remarkably, this association extended to 115 common diseases for which 42 signatures were available (0.3 £ r £ 0.9, p < 3.2x10-4). Correlations among metabolic signatures 43 of famine exposure and disease outcomes were attenuated when the famine signature was 44 adjusted for body mass index. 45

Prenatal famine exposure is associated with a metabolic biomarker signature that 46 strongly resembles signatures of a diverse set of diseases , an observation that can in part be 47 attributed to a shared involvement of obesity. 48 49 50 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 3

51 Metabolomics is a powerful tool for illuminating molecular phenotypes underpinning disease 52 (1). With nuclear magnetic resonance (NMR) approaches, metabolomics is now possible within 53 large-scale epidemiologic studies and biobanks. Within these settings, NMR metabolomics is 54 revealing a range of common and unique features to a broad spectrum of diseases (2–5). Less 55 is known about how the metabol ome may reflect or mediate effects of prenatal exposure 56 histories on disease pathogenesis. Here, we investigate the long-term metabolomic sequelae of 57 gestational exposure to famine, an established risk factor for the development of metabolic 58 disease (6,7). 59 The Dutch Hunger Winter of 1944 -1945, a 6-month famine at the end of World War II, 60 provides a unique setting to study the long-term effects of an adverse prenatal environment 61 (6,8,9). Previous studies revealed that prenatal famine exposure is associated with an increased 62 risk in unfavourable metabolic phenotypes in adultho od including increased fasting glucose 63 and triglyceride levels, obesity, and type-2 diabetes (10–17). These associations have also been 64 observed for other historical famines (6). To date, a comprehensive view of metabolic changes 65 linked to prenatal famine exposure is lacking. In this study, we seek to provide a deeper insight 66 into the metabolomic profile associated with prenatal famine exposure. 67 We profiled samples for 944 participants from the Dutch Hunger Winter Families Study 68 using nuclear magnetic resonance (NMR) metabolomic s. We compared prenatal famine-69 exposed individuals to unexposed control participants on 168 different serum metaboli c 70 biomarkers and also compared the metabolome-wide signature of prenatal famine exposure to 71 an atlas of signatures marking risk of a range of common diseases in order to characterize the 72 phenotype of prenatal famine exposure . Our study reveals specific metaboli c biomarker 73 alterations and broader connections with a range of chronic diseases , providing new insights 74 into how prenatal famine exposure shapes health across the life course. 75 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 4 76

77 78 Study population 79 The Dutch Hunger Winter Families study (DHWFS) is described in detail elsewhere (18). In 80 short, historical birth records were retrieved from three institutions in famine-exposed cities of 81 all singleton births between 1 February 1945 and 31 March 1946 and a systematic sample of 82 births born in 1943 or 1947 . From these records we identified infants whose mothers were 83 exposed to the famine during or immediately preceding that pregnancy and unexposed time-84 controls born before or after the famine . These individuals were invited to participate in a 85 telephone interview and in a clinical examination, together with a same-sex sibling not exposed 86 to the famine (family-control). 87 The Dutch Hunger Winter Families study was approved by the Medical Ethic s Committee of 88 Leiden University Medical Center (P02.082) and the participants provided verbal consent at 89 the start of the telephone interview and written informed consent at the start of the clinical 90 examination. 91 We conducted 1,075 interviews and 971 clinical examinations between 2003 and 2005. 92 One non-biological sibling identified with genetic analyses was excluded from the cohort. 93 NMR metabolomics profiling was performed on serum samples of 962 individuals. Our sample 94 for this study included 944 individuals after excluding non-fasted samples (n = 17) and an 95 outlier in the metabolomics dataset as identified with principal component analysis (n = 1) 96 (Supplemental Fig. 1). 97 98 Famine exposure definitions 99 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 5 Food rations were distributed centrally and below 900 kcal/day between November 26, 1944 100 and May 15, 1945 (8). We defined famine exposure by the number of weeks during which the 101 mother was exposed to < 900 kcal/day after the last menstrual period (LMP) recorded on the 102 birth record (18). For analysis of timing of gestational exposure, we subdivide the gestational 103 period into units of 10 weeks. We considered the mother exposed in gestational weeks 1 -10, 104 11-20, 21-30, or 31 to delivery if these gestational time windows were entirely contained within 105 this period and had an average exposure of < 900kcal/day during an entire gestation period of 106 10 weeks. As the famine lasted 6-months some participants were exposed to famine during two 107 adjacent 10-week periods. In chronological order , pregnancies with LMP between 30 April 108 1944 and 24 August 1944 were considered exposed in weeks 31 to delivery; between 9 July 109 1944 and 15 October 1944 in pregnancy weeks 21-30; between 17 September 1944 and 24 110 December 1944 in pregnancy weeks 11-20, between 26 November 1944 and 4 March 1945 in 111 pregnancy weeks 1-10. Individuals with a LMP between 4 February and 12 May 1945 were 112 exposed to an average of < 900kcal/day for less than 10 weeks before conception and up to 8 113 weeks post-conception, are denoted as the weeks 9-0 weeks group. We defined individuals 114 exposed to one or at most two of these definitions exposed to ‘any’ gestational exposure. 115 116 Characteristics 117 Information on health history, including information on the use of cholesterol-lowering drugs, 118 was collected through telephone interviews. Measurement of height was carried out to the 119 nearest millimeter using a portable stadiometer (Seca), and body weight was measured to the 120 nearest 100 g by a portable scale (Seca). BMI was calculated from these measures (weight (kg) 121 / [height (m)]2). Cholesterol measures were reported previously (14) and were assessed using 122 standard enzymatic assays. LDL cholesterol was calculated for individuals with a triglyceride 123 concentration lower than 400 mg/dl using the Friedewald formula. A blood draw was 124 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 6 performed at the start of a 75 -g oral glucose test, and fasted glucose was quickly assessed in 125 serum by hexokinase reaction on a Modular P800 (Roche). The presence of type -2 diabetes 126 was either determined through previous health history or defined as fasting glucose ≥ 127 7.0mmol/l or 2hr glucose tolerance test ≥ 11.1mmol/l (19). 128 129 Metabolic biomarker quantification 130 Metabolic biomarkers were measured from serum samples using a high-throughput 1H-NMR 131 metabolomics platform developed by Nightingale Health Ltd. (Helsinki, Finland; 132 nightingalehealth.com; biomarker quantification version 20 21). Details of the procedure and 133 application of the NMR metabolomics platform have been described elsewhere (20,21). This 134

provides simultaneous quantification of 168 directly measured and 81 derived 135 metabolic biomarkers, including 37 clinically validated metabolic biomarkers certified for 136 diagnostics use . The metabolic biomarkers measured include amino acids, ketone bodies, 137 lipids, fatty acids, and lipoprotein subclass distribution, particle size and composition. A subset 138 of the biomarkers was selected for inclusion in the presented analysis, focusing on the 168 139 directly measured metabolic biomarkers. 140 Missing values were set to the minimum value for each metabolic measure. A value of one 141 was added to all metabolic biomarkers containing zeroes (i.e. x + 1), which indicated that they 142 were below the limit of quantification. All metabolic biomarkers were then natural logarithmic 143 transformed to obtain an approximately normal distribution. The metabolic biomarkers were 144 subsequently scaled to standard deviation (SD) units (mean 0, SD 1) for use in the analysis. 145 146 Genotype data generation and polygenic scores 147 From our metabolomics sample population, 931 individuals also had genotype data available. 148 Genotype data were measured using the Illumina Infinium TM Global Screening Array (GSA) 149 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 7 genotyping platform (version 24 v3.0. Illumina Inc., San Diego, USA) by the Human Genomics 150 Facility in the Genetic Laboratory Rotterdam (Rotterdam, the Netherlands) . Imputation was 151 performed using the 1000G P3v5 reference panel (22). Polygenic scores were calculated for 152 tyrosine, leucine and glucose levels with the PRSice-2 software using the independent hits of 153 publicly available genome-wide association study (GWAS) summary statistics (Supplemental 154 Table 3) (23). The base GWAS study utilized the same NMR platform and had participants 155 with the same ancestry (European) as those in our study (24). The polygenic scores were 156 residualized on the first ten genetic principal components and subsequently scaled to standard 157 deviation (SD) units (mean 0, SD 1) for analysis. 158 159 Statistical analysis 160 All analyses were performed in the R programming environment (R version 4.2.2). 161 For all linear regression analyses, we used linear regression within a generalized estimating 162 equations framework to account for the correlation between sibships (R geepack package, 163 version 1.3.9) (25) and adjusted for age, sex and cholesterol-lowering medication. 164 We first validated the NMR measurements by testing the consistency between glucose, 165 triglycerides, total cholesterol, LDL, and HDL cholesterol measured by routine clinical 166 chemistry and Nightingale Health NMR (Supplemental Fig. 2 ). Consistent with previous 167 studies, c orrelations were high (r ³ 0.9) for all metabolic biomarkers tested (26,27). We 168 subsequently tested whether previously observed associations between prenatal famine 169 exposure and these five metabolic biomarkers as measured by routine clinical chemistry were 170 consistently found when the same biomarkers were measured using Nightingale Health NMR. 171 Next, we performed a metabolome-wide association study of prenatal famine exposure by 172 assessing the relationship between famine exposure and 168 metabolic biomarkers. Due to the 173 correlated nature of the metabolic biomarkers, 95% of the variation in the 168 metabolic 174 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 8 biomarkers was explained by 14 principal components. Therefore, as previously described (28–175 30), we corrected for 14 independent tests using Bonferroni multiple testing correction (p value 176 = 0.05/14 = 3.57×10−3). Sensitivity analyses were performed to assess the robustness of the 177

of the metabolome-wide association study of prenatal famine. First, to assess the effect 178 of famine exposure independent of BMI or type -2 diabetes, the main model was additionally 179 adjusted for BMI and type -2 diabetes. Second, to determine the extent to which the observed 180 associations between prenatal famine exposure and metabolic biomarkers levels could be 181 attributed to genetics, the main model was additionally adjusted for the polygenic scores of the 182 metabolic biomarkers. Third, to check for potential differences between sexes, sex -stratified 183 analyses were performed adjusting for the same covariates as the main model and an interaction 184 term for sex and metabolic marker was included in the model to test whether potential 185 differences were statistically significant . Fourth, potential gestation timing specific effects of 186 famine exposure were examined by subdividing famine exposure into 5 gestational time 187 windows. In the regression analysis, the single indicator of famine exposure was replaced with 188 indicator variables identifying exposure within each of the gestational time windows. 189 To expand our analysis from focusing on individual metabolic biomarkers to broader 190 metabolic biomarker signatures associated with disease, we compared the metabolic biomarker 191 signature associated with prenatal famine exposure to the metabolic biomarker signature 192 predicting the future risk of type -2 diabetes. For this we utilized p ublished results from a 193 metabolome-wide study on incident type-2 diabetes using UK Biobank data (2). Specifically, 194 we correlated the effect sizes of famine exposure in DHWFS with the effect sizes in the UK 195 Biobank for incident type -2 diabetes across the 135 shared metabolic biomarkers in both 196 datasets. 197 We then extended our analysis and correlated the metabolic profile of prenatal famine to 198 the publicly availably metabolic signatures of a large set of diseases as estimated with UK 199 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 9 Biobank data (4). Out of 674 incident diseases available in the metabolomics atlas, we selected 200 those with at least 1000 cases (out of a total population ranging from ~103,300 to ~118,000) to 201 represent common diseases (n = 162), and those with at least one significant association with 202 a metabolic biomarker in the UK Biobank analysis (p < 5x10 -4) (n = 15 4), resulting in 154 203 diseases. The effect sizes of the overlapping 168 metabolic biomarkers were correlated between 204 prenatal famine and each disease. We corrected for multiple testing using Bonferroni correction 205 (p value = 0.05/15 4 = 3.2x10-4). Finally, we re-estimated the effect sizes of the association 206 between prenatal famine exposure and all 168 metabolic biomarkers, while additionally 207 adjusting for BMI. We then repeated the correlation analysis comparing this BMI -adjusted 208 prenatal famine metabolic biomarker signature with the 154 metabolic biomarker disease 209 signatures from the UK Biobank. 210 211

212 213 Population characteristics 214 Within the Dutch Hunger Winter Families Study, fasting NMR metabolomics data were 215 available for 944 study participants . Among these participants, 480 ( 51%) were prenatally 216 exposed to famine, and 464 (49%) were controls (including unexposed time controls born at 217 the same institution as the exposed individuals and unexposed same-sex sibling controls both 218 born either before or conceived after the famine ). As previously reported, f amine-exposed 219 participants had an increased BMI (11) and a higher prevalence of type -2 diabetes (16) and 220 controls were on average 0.9 years younger than famine -exposed. No differences were 221 observed in sex or the use of cholesterol lowering medication (Table 1). 222 223 Validation of NMR metabolomics measures 224 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 10 We first sought to validate the newly measured metabolomics measures in our study by 225 correlating them with previously measured clinical chemistry data available for five metabolic 226 biomarkers, namely fasted glucose, triglycerides, total cholesterol, LDL, and HDL cholesterol 227 (14,17). The correlations were all high (r ³ 0.9) in line with previous studies (Supplemental 228 Fig. 2) (4,31). Next, we examined the associations between prenatal famine exposure and these 229 five metabolic biomarkers, as measured by clinical chemistry or NMR , and found the effect 230 sizes to be consistent between the two measurements (Supplemental Table 1). 231 232 Metabolome-wide association study on prenatal famine exposure 233 Next, we examined the association of any prenatal famine exposure with all 168 metabolic 234 biomarkers individually. Prenatal famine exposure was associated with higher tyrosine (effect 235 size 0.28 SD), leucine (0.21 SD), glucose (0.23 SD), and isoleucine (0.18 SD) concentrations 236 (p<3x10-3; all analyses adjusted for age, sex, and use of cholesterol-lowering drugs) (Fig. 1A, 237 Supplemental Table 2). 238 We performed three sets of follow-up analyses for tyrosine, leucine, and glucose to gain 239 further insight into these associations. Isoleucine was excluded because it was highly correlated 240 with leucine (r = 0.9), both are branched-chain amino acids, while leucine showed the stronger 241 association with famine exposure ( Fig. 1B, Supplemental Fig. 3). First, the associations 242 between prenatal famine exposure and tyrosine, leucine, and glucose remained after including 243 BMI or type-2 diabetes as a covariate in the model (Fig. 1C). Second, we also considered 244 genetics as a potential explanatory factor for these associations . Since the polygenic scores of 245 each metabolic biomarker explained only approximately 1%-5% of the ir variance 246 (Supplemental Table 3), the associations were not affected by including the polygenic scores 247 as covariates (Fig. 1C). Third, we explored whether associations between famine exposure and 248 the three metabolic biomarkers were dependent on sex or the timing of exposure during 249 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 11 gestation (Supplemental Fig. 4). Effect sizes across different exposure timing subgroups were 250 similar to our estimates of the main analysis. In the sex stratified analysis, t he effect sizes for 251 glucose and tyrosine were lower in females than males, but we found no statistical evidence 252 for effect modification (interaction p values > 0.39). 253 254 Comparison to metabolic biomarker signatures of diseases 255 The individual metabolic biomarkers associated with prenatal famine exposure were previously 256 linked to type-2 diabetes (32). To further investigate whether these associations reflect an 257 increased risk of type-2 diabetes among individuals prenatally exposed to famine , we utilized 258 a previously reported metabolic biomarker signature of incident type -2 diabetes from UK 259 Biobank (2) and compared it to the complete set of biomarker associations in our study. 260 Specifically, we took the effect sizes of 135 metabolic biomarkers for the risk of type-2 diabetes 261 and compared them to the effect sizes we observed for prenatal famine. The metabolic 262 biomarker signature of p renatal famine exposure was highly correlated with that of incident 263 type-2 diabetes (r = 0.77, p = 3x10-27; Fig. 2A). This similarity persisted when we re-estimated 264 the effect sizes for prenatal famine exposure after excluding participants with type -2 diabetes 265 (r = 0.67, p = 1x10-18) (Fig. 2B). 266 To explore whether the metabolic biomarker profile of prenatal famine exposure may 267 reflect a risk of diseases beyond type-2 diabetes, we extended the analysis to recently published 268 atlas of signatures of a wide range of diseases in the UK Biobank (4). We focused on metabolic 269 biomarker signatures for the future onset of disease obtained from individuals not affected by 270 the disease of interest at baseline. For the analysis, we utilized a subset of 154 common incident 271 diseases that had at least one metabolic biomarker association in the UK Biobank (p < 5x10-4). 272 Remarkably, the metabolic biomarker signature of prenatal famine exposure was positively 273 correlated with the metabolic biomarker signature of 115 diseases (75%; 0.3 £ r £ 0.9, p < 274 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 12 3.2x10-4), and negatively correlated with 13 diseases (8%; -0.9 £ r £ -0.4, p < 3.2x10 -4) 275 (Supplemental Table 4). The metabolic biomarker signature of prenatal famine exposure 276 exhibited the strongest correlation with the signature of the future risk of myocardial infarction 277 (r = 0.9, p = 1.8x10-47). Other diseases with a strong correlation (r ³ 0.7) included those related 278 to the digestive system, diseases with an endocrine, nutritional and metabolic component, as 279 well as diseases of the nervous system (Fig. 3). The diseases displaying a negative correlation 280 were primarily associated with injury and other consequences of external causes , such as 281 fractures and open wounds (Supplemental Table 4). 282 We hypothesized that a potential common factor among the diseases with a similar 283 metabolic biomarker signature is obesity. To test this hypothesis, we re-estimated the effect 284 sizes for prenatal famine exposure while additionally adjusting for BMI and then re-calculated 285 the correlation between the resulting BMI -adjusted metabolic biomarker signature with the 286 signatures of the 154 common incident diseases (Supplemental Table 4). The strength of the 287 correlations was consistently attenuated across all diseases (mean = -58%; SD = 19%). Among 288 the 30 diseases whose metabolic biomarker signature was most similar to that of prenatal 289 famine, the attenuation ranged between 27-48% and the correlations remained moderate (0.4 £ 290 r £ 0.6). The degree of attenuation was not linked to whether the disease had an obvious 291 metabolic component (Fig. 4). 292 293

294 We further defined the metabolic phenotype associated with prenatal famine exposure using 295 nuclear magnetic resonance (NMR) metabolomic profiling. We show that prenatal exposure to 296 undernutrition is associated with specific metabolic differences later in life , including higher 297 levels of branched -chain amino acids (BCAA), an aromatic amino acid, and glucose. In 298 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 13 addition, we report that the metabolic biomarker signature of prenatal famine has marked 299 similarities to the signature of a wide range of common diseases. 300 Our study indicated specific differences in the metabolic profiles o f famine -exposed 301 individuals compared to controls. We observed that those exposed to famine prenatally have 302 higher levels of the aromatic amino acid tyrosine, the branched-chain amino acids leucine and 303 isoleucine, and glucose six decades after exposure . All four metabolic biomarkers have been 304 linked to type -2 diabetes and thus support the known association between prenatal famine 305 exposure and type -2 diabetes risk in adulthood (6,33). In addition, h igher levels of o ther 306 branched-chain and aromatic amino acids such as valine and phenylalanine that have also been 307 associated with type-2 diabetes showed a nominally significant association with prenatal 308 famine exposure, further supporting the link between famine and type-2 diabetes risk (33). 309 Interestingly, the metabolic biomarker associations with prenatal famine were independent of 310 BMI and type-2 diabetes status, indicating that they may not be fully driven by the higher BMI 311 and increased prevalence of type-2 diabetes among famine-exposed individuals. 312 The link with a higher type -2 diabetes risk among individuals exposed to famine in the 313 prenatal period was reinforced by investigating the complete range of metabolic biomarkers. 314 We observed a strong resemblance in the metabolic biomarker signature of prenatal famine 315 with that of the future onset of type -2 diabetes (33). Moreover, the strong correlation of the 316 famine signature with the incident type -2 diabetes signature persisted after excluding 317 participants who were already diagnosed with type -2 diabetes at the time of assessment . This 318

reinforces that type-2 diabetes is a main health outcome of prenatal famine exposure (6) 319 and indicates that even exposed individuals not diagnosed with type -2 diabetes have an 320 increased risk of developing this condition in the future. 321 Upon extending our analysis beyond type-2 diabetes, we observed a striking similarity 322 between the metabo lic biomarker signature of famine exposure and a wide range of other 323 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 14 incident disease signatures. This included conditions like disorders of lipoprotein metabolism, 324 obesity and type-2 diabetes, but also a priori less expected diseases like osteoarthritis, kidney 325 stones, and depressive disorders . Interestingly, these high correlations were substantially 326 attenuated for all incident diseases when we repeated the analysis using a metabolic biomarker 327 profile of prenatal famine that was adjusted for BMI. Our findings suggest that BMI is a shared 328 risk factor for or consequence of the diseases and that the metabolic biomarkers measured by 329 the NMR platform used may have a particularly strong association with BMI. Of note, after 330 accounting for BMI, moderate correlations between the metabolic biomarker profiles of 331 incident disease and prenatal famine remained. Our findings and previous studies highlight that 332 the NMR platform applied is especially useful for disease risk prediction, but of limited value 333 to gain new mechanistic insights. Further studies with more comprehensive metabolomics 334 platforms are needed to fully understand why the metabolic biomarker signature of prenatal 335 famine exposure links to a broad range of diseases, including effects independent of obesity. 336 337

338 Prenatal exposure to famine is associated with marked metabolic alterations later in life. 339 Differences in individual metabolic biomarkers include higher levels of branched-chain amino 340 acids, aromatic amino acids, and glucose. Moreover, the metabol ic biomarker signature 341 characteristic of prenatal famine strongly resembles that of a diverse set of diseases. Overall, 342 our findings underscore the broad impact of prenatal famine on adult health and highlight 343 obesity as a plausible contributing factor. 344 345 346 347 348 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 15 Abbreviations 349 BCAA: branched-chain amino acids 350 BMI: body mass index 351 DHWFS: Dutch Hunger Winter Families study 352 GWAS: genome-wide association study 353 LMP: last menstrual period 354 NMR: nuclear magnetic resonance 355 SD: standard deviation 356 357 Declarations 358 Ethics approval and consent to participate 359 The Dutch Hunger Winter Families study was approved by the Medical Ethics Committee of 360 Leiden University Medical Center (P02.082) and the participants provided verbal consent and 361 written informed consent. 362 Consent for publication 363 Not applicable. 364 Availability of data and materials 365 The datasets supporting the conclusions of this article are included within the article and its 366 additional files. The DHWFS data is available for replication purposes upon request to B. T. 367 Heijmans ([email protected]) and if replication is conducted within the secure Leiden 368 University Medical Center network environment. 369 Competing interests 370 The authors declare that they have no competing interests. 371 Funding 372 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 16 This research was supported by National Institute on Aging grants (R01AG066887). The NMR 373 metabolomics measurement of the DHWFS samples was funded by the BBMRI -NL 374 Metabolomics consortium (a research infrastructure financed by the Dutch government, NWO 375 184.021.007 and 184.033.111). DWB is a fellow of the CIFAR CBD Network. The funders 376 had no role in study design, data collection and analysis, decision to publish, or preparation of 377 the manuscript. 378 Authors’ contributions 379 M.J.T., D.W.B., L.H.L. and B.T.H. were involved in the conception, design, and conduct of 380 the study and interpretation of the results. M.J.T. performed the analyses and wrote the first 381 draft of the manuscript, and all authors edited, reviewed, and approved the final version of the 382 manuscript. B.T.H. is the guarantor of this work and, as such, had full access to all the data in 383 the study and takes responsibility for the integrity of the data and the accuracy of the data 384 analysis. 385

386 Not applicable. 387 388

389 1. Wishart DS. Metabolomics for Investigating Physiological and Pathophysiological 390 Processes. Physiol Rev. 2019 Oct 1;99(4):1819–75. 391 2. Bragg F, Trichia E, Aguilar-Ramirez D, Bešević J, Lewington S, Emberson J. 392 Predictive value of circulating NMR metabolic biomarkers for type 2 diabetes risk in 393 the UK Biobank study. BMC Med. 2022 Dec 1;20(1). 394 3. Bell JA, Richardson TG, Wang Q, Sanderson E, Palmer T, Walker V, et al. Effects of 395 general and central adiposity on circulating lipoprotein, lipid, and metabolite levels in 396 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 17 UK Biobank: A multivariable Mendelian randomization study. The Lancet Regional 397 Health - Europe [Internet]. 2022;21:100457. Available from: https://doi.org/10.1016/j. 398 4. Julkunen H, Cichońska A, Tiainen M, Koskela H, Nybo K, Mäkelä V, et al. Atlas of 399 plasma NMR biomarkers for health and disease in 118,461 individuals from the UK 400 Biobank. Nat Commun. 2023 Feb 3;14(1):604. 401 5. Guo Y, Chen S, Zhang Y, Wang H, Huang S, Chen S, et al. Circulating metabolites 402 associated with incident myocardial infarction and stroke: A prospective cohort study 403 of 90 438 participants. J Neurochem. 2022 Aug 15;162(4):371–84. 404 6. Lumey LH, Stein AD, Susser E. Prenatal famine and adult health. Annu Rev Public 405 Health. 2011;32:237–62. 406 7. Liu H, Chen X, Shi T, Qu G, Zhao T, Xuan K, et al. Association of famine exposure 407 with the risk of type 2 diabetes: A meta-analysis. Clinical Nutrition. 2020 408 Jun;39(6):1717–23. 409 8. Burger GCE, Drummond JC, Sandstead HR. Malnutrition and starvation in Western 410 Netherlands: September 1944 - July 1945. General State Print. Office., The Hague; 411 1948. 412 9. Stein Z, Susser M, Saenger G, Marolla F. Famine and Human Development: The 413 Dutch Hunger Winter of 1944-1945. New York: Oxford University Press; 1975. 414 10. Ravelli AC, van Der Meulen JH, Osmond C, Barker DJ, Bleker OP. Obesity at the age 415 of 50 y in men and women exposed to famine prenatally. Am J Clin Nutr. 1999 416 Nov;70(5):811–6. 417 11. Stein AD, Kahn HS, Rundle A, Zybert PA, van der Pal-de Bruin K, Lumey LH. 418 Anthropometric measures in middle age after exposure to famine during gestation: 419 evidence from the Dutch famine. Am J Clin Nutr. 2007 Mar;85(3):869–76. 420 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 18 12. Lumey LH, Ekamper P, Bijwaard G, Conti G, van Poppel F. Overweight and obesity 421 at age 19 after pre-natal famine exposure. Int J Obes (Lond). 2021 Aug;45(8):1668–76. 422 13. Ravelli GP, Stein ZA, Susser MW. Obesity in young men after famine exposure in 423 utero and early infancy. N Engl J Med. 1976 Aug 12;295(7):349–53. 424 14. Lumey LH, Stein AD, Kahn HS, Romijn JA. Lipid profiles in middle-aged men and 425 women after famine exposure during gestation: the Dutch Hunger Winter Families 426 Study. Am J Clin Nutr. 2009 Jun;89(6):1737–43. 427 15. Ravelli A, van der Meulen J, Michels R, Osmond C, Barker D, Hales C, et al. Glucose 428 tolerance in adults after prenatal exposure to famine. The Lancet. 1998 429 Jan;351(9097):173–7. 430 16. Lumey LH, Stein AD, Kahn H. Food restriction during gestation and impaired fasting 431 glucose or glucose tolerance and type 2 diabetes mellitus in adulthood: Evidence from 432 the DutchHunger Winter Families Study. J Dev Orig Health Dis. 2009 Apr;1:S164. 433 17. Tobi EW, Slieker RC, Luijk R, Dekkers KF, Stein AD, Xu KM, et al. DNA 434 methylation as a mediator of the association between prenatal adversity and risk 435 factors for metabolic disease in adulthood. Sci Adv. 2018 Jan;4(1):eaao4364. 436 18. Lumey LH, Stein AD, Kahn HS, van der Pal-de Bruin KM, Blauw GJ, Zybert PA, et 437 al. Cohort profile: the Dutch Hunger Winter families study. Int J Epidemiol. 2007 438 Dec;36(6):1196–204. 439 19. WHO, IDF. Definition and diagnosis of diabetes mellitus and intermediate 440 hyperglycaemia : report of a WHO/IDF consultation. World Health Organization; 441 2006. p. 442 20. Würtz P, Kangas AJ, Soininen P, Lawlor DA, Davey Smith G, Ala-Korpela M. 443 Quantitative Serum Nuclear Magnetic Resonance Metabolomics in Large-Scale 444 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 19 Epidemiology: A Primer on -Omic Technologies. Am J Epidemiol. 2017 Nov 445 1;186(9):1084–96. 446 21. Soininen P, Kangas AJ, Würtz P, Suna T, Ala-Korpela M. Quantitative serum nuclear 447 magnetic resonance metabolomics in cardiovascular epidemiology and genetics. Circ 448 Cardiovasc Genet. 2015 Feb;8(1):192–206. 449 22. 1000 Genomes Project Consortium, Auton A, Brooks LD, Durbin RM, Garrison EP, 450 Kang HM, et al. A global reference for human genetic variation. Nature. 2015 Oct 451 1;526(7571):68–74. 452 23. Choi SW, O’Reilly PF. PRSice-2: Polygenic Risk Score software for biobank-scale 453 data. Gigascience. 2019 Jul 1;8(7). 454 24. Smith CJ, Sinnott-Armstrong N, Cichońska A, Julkunen H, Fauman EB, Würtz P, et 455 al. Integrative analysis of metabolite GWAS illuminates the molecular basis of 456 pleiotropy and genetic correlation. Elife. 2022 Sep 1;11. 457 25. Homish GG, Edwards EP, Eiden RD, Leonard KE. Analyzing family data: A GEE 458 approach for substance use researchers. Addictive behaviors. 2010 Jun;35(6):558–63. 459 26. Julkunen H, Cichońska A, Tiainen M, Koskela H, Nybo K, Mäkelä V, et al. Atlas of 460 plasma NMR biomarkers for health and disease in 118,461 individuals from the UK 461 Biobank. Nat Commun. 2023 Feb 3;14(1):604. 462 27. Tikkanen E, Jägerroos V, Holmes M V, Sattar N, Ala-Korpela M, Jousilahti P, et al. 463 Metabolic Biomarker Discovery for Risk of Peripheral Artery Disease Compared With 464 Coronary Artery Disease: Lipoprotein and Metabolite Profiling of 31 657 Individuals 465 From 5 Prospective Cohorts. J Am Heart Assoc. 2021 Dec 7;10(23):e021995. 466 28. Al Rashid K, Taylor A, Lumsden MA, Goulding N, Lawlor DA, Nelson SM. 467 Association of the functional ovarian reserve with serum metabolomic profiling by 468 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 20 nuclear magnetic resonance spectroscopy: a cross-sectional study of ~ 400 women. 469 BMC Med. 2020 Aug 31;18(1):247. 470 29. Wang Q, Jokelainen J, Auvinen J, Puukka K, Keinänen-Kiukaanniemi S, Järvelin MR, 471 et al. Insulin resistance and systemic metabolic changes in oral glucose tolerance test 472 in 5340 individuals: an interventional study. BMC Med. 2019 Nov 29;17(1):217. 473 30. Santos Ferreira DL, Williams DM, Kangas AJ, Soininen P, Ala-Korpela M, Smith GD, 474 et al. Association of pre-pregnancy body mass index with offspring metabolic profile: 475 Analyses of 3 European prospective birth cohorts. PLoS Med. 2017 476 Aug;14(8):e1002376. 477 31. Tikkanen E, Jägerroos V, Holmes M V, Sattar N, Ala-Korpela M, Jousilahti P, et al. 478 Metabolic Biomarker Discovery for Risk of Peripheral Artery Disease Compared With 479 Coronary Artery Disease: Lipoprotein and Metabolite Profiling of 31 657 Individuals 480 From 5 Prospective Cohorts. J Am Heart Assoc. 2021 Dec 7;10(23):e021995. 481 32. Guasch-Ferré M, Hruby A, Toledo E, Clish CB, Martínez-González MA, Salas-482 Salvadó J, et al. Metabolomics in Prediabetes and Diabetes: A Systematic Review and 483 Meta-analysis. Diabetes Care. 2016 May;39(5):833–46. 484 33. Bragg F, Trichia E, Aguilar-Ramirez D, Bešević J, Lewington S, Emberson J. 485 Predictive value of circulating NMR metabolic biomarkers for type 2 diabetes risk in 486 the UK Biobank study. BMC Med. 2022 Dec 1;20(1). 487 488 489 490 491 492 493 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 21 Table 1. Population characteristics 494 Controls (n = 464) Famine-exposed (n = 480) p value Age, years (SD) 57.9 (5.4) 58.8 (0.5) 1.1x10-3 Sex, males, n (%) 200 (43.1) 225 (46.9) 0.24 Use of cholesterol-lowering medication, n (%) 55 (11.9) 61 (12.7) 0.69 Body mass index, kg/m2 (SD) 27.0 (4.2) 28.2 (4.8) 1.6x10-4 Type-2 diabetes, n (%) 38 (8.2) 61 (12.8) 0.02 Values are means (standard deviation) or numbers of subjects (valid %) shown for famine-exposed and 495 controls of the study population. Comparing the two categories by a two -sample t-test or chi -square 496 test, as appropriate. 497 498 499 500 501 502 503 504 505 506 507 508 509 510 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 22 511 Fig 1. Metabolome-wide association study on prenatal famine exposure . A. Association of prenatal famine 512 exposure with 168 metabolic biomarkers. Regression models were adjusted for age, sex and cholesterol-lowering 513 medication and correlation within sibships were controlled for (main model). Scattered points represent metabolic 514 biomarkers: the x-axis shows the effect size for the association of famine with the respective metabolic biomarker, 515 while the y-axis is negative log of the p value. The grey line represents the significance threshold for this analysis 516 (p value = 3.57x10−3). B. Heatmap showing the correlation of famine-associated metabolic biomarkers. Pearson’s 517 correlation was calculated for each metabolic biomarker pair. C. Sensitivity analyses on famine -associated 518 metabolic biomarkers. Main: main model; BMI-adjusted: main model additionally adjusting for BMI; Diabetes-519 adjusted: main model additionally adjusted for type -2 diabetes; polygenic score (PGS) -adjusted: main model 520 additionally adjusted for the polygenic score of the metabolic biomarkers. Effect estimates and 95% confidence 521 intervals are depicted for each model and are reported in standard -deviation (SD) units of the log -transformed 522 metabolic biomarkers. 523 524 525 526 527 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 23 528 Fig 2. Correlation analysis of the metabolic biomarker signature associated with famine and the metabolic 529 biomarker signature associated with incident type -2 diabetes. A. Overall metabolic biomarker signature 530 comparison of 135 metabolic biomarkers for prenatal famine exposure and incident type-2 diabetes (r = 0.77, p = 531 3x10-27) as established in the UK Biobank Study in a 12-year follow-up. B. Overall metabolic biomarker signature 532 comparison of 135 metabolic biomarkers for prenatal famine exposure (excluding 101 individuals with type -2 533 diabetes from the analysis) and incident type -2 diabetes as established in the UK Biobank Study in a 12 -year 534 follow-up (r = 0.67, p = 1x10 -18). The effect size estimates for each metabolic biomarker are shown as points. 535 Famine-associated metabolic biomarkers are indicated in blue. 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 24 551 Fig 3. Correlation analysis of the metabolic biomarker signature associated with famine and the metabolic 552 biomarker signature associated with various common diseases. Heatmap showing the effect size estimates of 553 the 30 most correlated diseases to prenatal famine exposure. The columns are clustered according to the metabolic 554 biomarker effect sizes and the rows are ordered according to the correlation of the metabolic biomarker signature 555 of the disease to prenatal famine exposure (Pearson r for IK21 Acute myocardial infarction = 0.85, Pearson r for 556 E11 Type-2 diabetes mellitus = 0.69). Only metabolic biomarkers that are nominally associated with prenatal 557 famine exposure are shown (p < 0.05). The diseases are shown with their ICD-10 (International Classification of 558 Diseases 10th Revision) classification. The full names of the metabolic biomarkers can be found in Supplemental 559 Table 2. 560 561 562 563 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint 25 564 Fig 4. Effect of additional adjustment of BMI in the correlation analysis of the metabolic biomarker 565 signature associated with famine and the metabolic biomarker signature associated with various common 566 diseases. The main model within the DHWFS cohort was adjusted for age, sex, and cholesterol -lowering 567 medication. The BMI-adjusted model within the DHWFS cohort was adjusted for age, sex, cholesterol-lowering 568 medication, and BMI. The effect sizes estimated for these two models of prenatal famine exposure were each 569 correlated to the effect sizes estimated for the risk of common diseases. The 30 diseases most correlated to the 570 metabolic biomarker signature of prenatal famine exposure are shown. 571 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 5, 2024. ; https://doi.org/10.1101/2024.04.04.24305284doi: medRxiv preprint