Childhood severe acute malnutrition is associated with metabolic changes in adulthood

This cohort study of 122 adult SAM survivors (SW=69, EM=53) and 90 age, sex and BMI-matched community participants (CPs) quantified serum metabolites using direct flow injection mass spectrometry combined with reverse-phase liquid chromatography. Univariate and sparse partial least square discriminant analyses (sPLS-DA) assessed differences in metabolic profiles and identified the most discriminative metabolites.


Introduction 61
Worldwide, an estimated 16.6 million children under the age of 5 years are severely wasted (1) 62 and severe acute malnutrition (SAM) remains a significant contributor to global mortality (1). As more 63 children survive episodes of SAM, there is a growing need to understand whether this early life exposure 64 is associated with long-term health risks including the development of non-communicable diseases 65 (NCDs). Unlike prenatal undernutrition, which has been associated with increased rates of type 2 diabetes 66 (T2D), hypertension, coronary heart disease and stroke in adulthood (2-4), the long-term consequences of 67 SAM in early childhood are poorly understood. 68 Limited data links early childhood growth failure to increased cardiovascular disease risk 69 (dyslipidemia, hypertension and glucose intolerance) in later life (5). Norwegian children with below-70 average weight and body mass index (BMI) in early life experienced more cardiovascular events by 64 71 years of age, but only after undergoing rapid weight rebound in later childhood (6). Childhood exposure 72 to the Chinese Famine (1959)(1960)(1961)(1962) was also shown to predict increased diabetes risk in adulthood, 73 specifically among those who experienced the most severe famine conditions (7). Similarly, adults 74 exposed to the Biafran Famine in early childhood had an increased prevalence of high blood pressure (8), 75 however, the study lacked birth weight data and was therefore unable to separate the effects of fetal and 76 infant famine exposure (8). 77 Acutely, SAM (defined by weight-for-height Z-score (WHZ) < -3, mid-upper arm circumference 78 (MUAC) < 115 mm and/or bilateral pitting edema) (9) is associated with several alterations in 79 intermediary metabolism, normally tightly regulated by hormones such as insulin and glucagon. Insulin 80 secretion is impaired during SAM (10,11) and was shown to remain impaired up to 6 weeks after the 81 body weight of children improved (12). While Malawian children diagnosed with SAM had profoundly 82 different metabolic profiles from controls after hospital discharge (13), we recently reported that 7 years 83

This is a secondary analysis of the JAMAKAS (Jamaica Marasmus and Kwashiorkor Adult Survivors) 108
Study, a cohort study that selected participants based on exposure to SAM in early childhood (detailed 109 flow chart presented in Figure 1). SAM diagnosis was based on the Wellcome criteria (21); the most 110 widely used classification for malnutrition up to 1999. Children presenting with a weight-for-age of 60-111 80% and edema were diagnosed with kwashiorkor, here referred to as edematous malnutrition, while those 112 with a weight-for-age < 60% without edema had marasmus, here referred to as severe wasting (21). 113

Selection of adult SAM survivors for metabolomics analysis:
The cohort was assembled by reviewing the 114 records of all 1,336 patients admitted to the Tropical Metabolism Research Unit of the University Hospital 115 of the West Indies between the years 1963 and 1993 with a diagnosis of SAM at age 6 months -5 years 116 (22). With an inpatient mortality rate of 3.5%, a total of 1289 surviving adults from the cohort were 117 theoretically available for tracing. Using the last recorded address and name of parents, community health 118 aides and nurses were able to identify a current address for 729 adult SAM survivors; the remaining 560 119 members of the cohort were not traced. We excluded all persons with an acute illness, using 120 glucocorticoids, a known history of a haemoglobinopathy, pregnant or lactating or unable to give written 121 informed consent. Of the 729 persons traced, a further 116 were unavailable to the study due to refusal (n 122 = 14), illness (n = 19), migration (n = 53), and pregnancy (n = 30), leaving 613 persons available for 123 recruitment. Of these, 316 adult SAM survivors enrolled in the JAMAKAS Study, and a subset of 122 124 SAM survivors submitted appropriate fasting blood samples for metabolomic analysis in the Jamaica 125 Metabolomics (JA-MET) Study (Figure 1). 126 127 128 Selection of community participants for metabolomics analysis: Community participants were 129 purposefully selected to be from the same socio-economic group as SAM survivors. Community Health 130 Aides recruited 159 participants from within the same communities where each adult SAM survivor lived 131 as follows: starting on the street where the SAM survivor lived, visits were conducted house to house 132 alternately on either side of the road. If unsuccessful, adjacent streets were similarly visited. Height and 133 weight were measured in the field using a stadiometer and a digital scale that was calibrated daily. 134 Community participants were matched based on age ± 5 years, sex, and BMI ± 2 kg/m 2 that fell within 135 the same BMI class as the SAM survivor being matched (i.e., underweight, less than 18.5 kg/m 2 ; normal, 136 18.5 to 24.9 kg/m 2 ; overweight, 25 to 29.9 kg/m 2 ; and obese, greater than 30 kg/m 2 ). Like SAM survivors, 137 community participants were asked about their general health status using a standardized questionnaire 138 (23). Additional exclusion criteria for community participants were a reported history of SAM and being 139 related to a SAM participant within the study. A total of 90 community participants submitted fasting 140 blood samples for metabolomic analysis in the JA-MET Study. 141 142

Measurements 143
All assessments and measurements were done contemporaneously in both SAM survivors and community 144 participants as follows: 145 Anthropometry: Body weight was measured to the nearest 0.1 kg using a portable balance (Seca 770,146 Hamburg, Germany). Using a minimum of two readings, height was measured to the nearest 0.1 cm using 147 a stadiometer (Invicta, London, UK) with the participant's head held in the Frankfurt plane. Waist 148 circumference was measured to the nearest 0.1 cm using a standardized protocol i.e. at the midpoint 149 between the iliac crest and the lowest rib with the participant standing erect (24). 150 sphingomyelins (p = 10), and phosphatidylcholines (p = 9)), acylcarnitines (p = 40), organic acids (p = 196 17), monosaccharides (p = 1), histidines (p = 2) and others (p = 4). The method combined the derivatization 197 and extraction of analytes, and the selective mass-spectrometric detection using multiple reaction 198 monitoring (MRM) pairs. Isotope-labeled internal standards and other internal standards were used for 199 metabolite quantification. Data analysis was done using Analyst 1. 6.2. 200 Each identified metabolite was included in the data analysis if it passed the following quality control 201 cutoffs: 1) a mean coefficient of variability < 25% across experimental batches, 2) more than 90% detected 202 values, and 3) above or equal to the lower limit of detection (LOD) in at least 50% of either study group 203 (i.e., adult SAM survivors or community participants). If a specific metabolite measure was below the 204 detection range, the value was replaced by half the LOD of that metabolite. 205 206

Statistical Analysis 207
The sample size was convenience based, restricted by the number of stored serum samples from the cohort. 208 The primary analysis aimed to compare the metabolomic profiles of adult SAM survivors and community 209 participants. Secondary analyses were conducted to: 1) compare the metabolite profiles of adult survivors 210 that had experienced edematous malnutrition versus severe wasting, and 2) relate the metabolic profiles 211 to markers of T2D, hypertension and fatty liver disease. 212 Principal component analysis (PCA) was used on standardized values to detect sample outliers and 213 examine inherent clustering and correlations. To assess differences in metabolite profiles, we conducted 214 both univariate and partial least-squares multivariate analyses. Generalized linear regression models were 215 performed on Box-Cox transformed variables while adjusting for age, sex, and BMI. Additional models 216 were also ran with further adjustment for income. Model fit was assessed through inspection of residuals 217 and p-values were corrected for multiple testing using Benjamini and Hochberg false-discovery-rate 218 draft of the manuscript, RB and CB provided main edits of content and RB had responsibility for final 241 content. All authors read and approved the final manuscript. 242

Participant characteristics 245
The clinical characteristics of the adult SAM survivors (n = 69, survivors of severe wasting; n = 53, 246 survivors of edematous malnutrition) and 90 community participants selected for metabolomic analysis 247 are described in Table 1. This subset of 122 adult SAM survivors were similar (with respect to age, sex 248 and BMI) to the the other 194 participants with whom they were enrolled and the 297 subjects that could 249 have been available for enrollment. SAM survivors and community participants had similar socio-250 economic status, and, in line with findings from the full cohort, this subset of SAM survivors had shorter 251 stature (p = 0.03) and higher L/S ratio, i.e. less liver fat (p < 0.01) compared to community participants. 252 Furthermore, survivors of severe wasting weighed less (p < 0.001) and had lower BMI (p < 0.001), waist 253 circumference (p < 0.01), fat mass (p < 0.01), android fat (p = 0.013) and lean mass (p = 0.04) than 254 survivors of edematous malnutrition. Also, in this subset, survivors of severe wasting still tended towards 255 worse pancreatic beta-cell function (IGI) compared to survivors of edematous malnutrition (p = 0.052). 256 Clinical differences between survivors of SAM and community participants where most apparent in males. 257 A sex-stratified analysis showed that male survivors of SAM were: shorter (170.6 ± 7.8 cm vs 176.0 ± 258 7.1 cm; p < 0.001), weighed less (63.6 ± 12.5 kg vs 68.1 ± 9.4 kg; p = 0.03), had lower lean mass (53.5 ± 259 7.8 kg vs 57.5 ± 6.6 kg; p = 0.003), but higher android-to-gynoid fat (AG) ratio (0.92 ± 0.19 vs 0.85 ± 260 0.19; p = 0.047) and L/S ratio, i.e. less liver fat (1.23 ± 0.1 vs 1.17 ± 0.1; p = 0.034) than male community 261 participants (Supplementary Table 1). Furthermore, this "small" phenotype among male SAM survivors 262 was mainly driven by those who had experienced severe wasting as they were: lighter (60.3 ± 11 kg vs 263 69.1 ± 14 kg; p = 0.005), with a lower ] kg/m 2 vs 22.9 [20-24.8] kg/m 2 ; p = 0.02) 264 and lean mass (51.5 ± 7.6 kg vs 56.8 ± 7 kg; p = 0.007) than survivors of edematous malnutrition (data 265 not shown). Similar to males, the "small" phenotype was also observed in women who survived severe 266 wasting as they had: lower weight (64.1 ± 18 kg vs 75.2 ± 19 kg; p = 0.027), lower  total, essential, aromatic (AAA), branched-chain (BCAA), glucogenic and ketogenic amino acids) and 280 thirteen ratio variables such as the kynurenine: tryptophan (KT) and the Fischer ratio. Only two values 281 were below the detection range and were replaced by half the LOD of that metabolite (hippuric acid). No 282 sample outlier or inherent clustering was detected by PCA (data not shown). 283 284 A subset of SAM survivors are metabolically distinguishable from community participants 285 Several metabolite variables differed between SAM survivors and community participants, and overall, 286 metabolite profiles could be used to distinguish most SAM cases from those of community participants 287 (Figure 2). While correcting for age, sex and BMI, (and age and sex only), seventy-seven metabolite 288 variables met the feature selection criteria of being both FDR-significant in univariate linear models 289 ( Figure 2A) and were selected by the sparse PLS-DA models using cross validation designed to 290 distinguish between SAM survivors and community participants (Figures 2B and 3). Based on 291 permutation testing, group classification by the PLS-DA model was better than random (p < 0.001) 292 (Supplementary Figure 1). The mean balanced error rate based on centroid distance was 18% ± 0.7, 293 with an R 2 = 0.38, and Q 2 = 0.39 (indicating acceptable consistency between the predicted and original 294 data) (33)  Specifically, the 77 differential variables were: 16/40 acylcarnitines, 20/22 amino acids, 4/10 299 sphingomyelins, 11/14 lysophosphatidylcholines, 7/14 phosphatidylcholines, 7/16 organic acids and 300 12/20 summary and ratio variables. Results from univariate analyses are detailed in Supplementary Table  301 2 and the top 15 metabolites that best distinguished SAM survivors and community participants based on 302 variable importance in the projection score are listed in Table 2 and presented in Figure 4. As seen in the 303 correlation plot (Figure 3), the mean concentrations of most amino acids, namely leucine, aspartic acid, 304 glutamic acid, valine, threonine and related summary values (e.g., total essential amino acids, urea cycle 305 amino acids, BCAA, BCAA / AAA), were higher in SAM survivors. However, tryptophan was a notable 306 exception as it was lower in adult SAM survivors compared to community participants, and this was linked 307 to SAM survivors having higher kynurenine-tryptophan (KT) ratios. Similarly, choline and a subset of 308 phosphatidylcholines, sphingomyelins and lysophosphatidylcholines (PC ae C36:0, PC aa C36:0, PC aa 309 C36:6, SM(OH) C24.1, lysoPC a 16:0) were also higher in SAM survivors. In contrast, the mean 310 concentration of most acylcarnitines (including C5:1-DC, C3:1 and C14) and certain sphingomyelins and 311 lysophosphatidylcholines (SM(OH) C22.1, SM(OH) C22.2 and lysoPC a 20:3) were lower in SAM 312 survivors than in community participants. Additionally, the ratio of acylcarnitine to free carnitine (C2/C0), 313 a marker of fatty acid beta oxidation, was lower in SAM survivors as was beta-hydroxybutyric acid. While 314 the overall sPLS-DA model suggests that the metabolic profile of many SAM survivors can be 315 distinguished from those of community participants, the observed fold change between groups of 316 individual metabolite variables were very small. Some SAM survivors seemed more readily 317 distinguishable from control participants; however, this potential subgroup was not related to SAM-type 318 (i.e., severe wasting vs. edematous malnutrition) (Supplementary Figure 2). Additionally, while roughly 319 43% of female SAM survivors and 56% of male SAM survivors were below average height for the 320 population (i.e. females < 160.8 cm, males < 171.8 cm), SAM survivors of either sex who were below 321 average height and those who were at or above average height had similar metabolic profiles (data not 322 shown). 323

Survivors of severe wasting and edematous malnutrition have similar metabolic profiles. 325
Our PLS-DA models adjusted for age, sex and BMI could not distinguish the metabolic profiles of SAM 326 survivors who experienced edematous SAM from those who experienced severe wasting ( Figure 5). 327 Permutation testing showed that classification of the two SAM phenotypes by the PLS-DA model was not 328 better than random (p = 0.71). The mean balanced error rate based on centroid distance was 56.3% ± 0.03, 329 R 2 = 0.063, Q 2 = -0.35 and AUC for PLS-component1 = 0.76. 330 Similarly, age, sex and BMI-adjusted (and age and sex adjusted) univariate linear models did not identify 331 any differential metabolite variable between the two SAM phenotypes. Results from generalized linear 332 models additionally adjusted for birth weight or income were also non-significant for all metabolites tested 333 (data not shown). 334 335 Specific metabolite variables are associated with cardiometabolic risk factors. 336 Metabolite variables that differ with SAM exposure and are related to cardiometabolic risk were tested 337 for association with fat mass, mean diastolic pressure, mean systolic pressure, HOMA-IR, WBISI, oDI 338 and estimates of liver fat while adjusting for age, sex and BMI. Boxplots of these selected metabolite 339 variables (i.e., BCAA: AAA ratio, KT ratio, urea cycle metabolites, choline, betaine, glutamic acid, 340 C2/C0, β-hydroxybutyric acid) are presented in Figure 4. After accounting for multiple testing, we found 341 that both β-hydroxybutyric acid and C2/C0 were associated with measures of liver fat ( Figure 6). These 342 variables were highly correlated (ρ = 0.88, p < 0.0001) and inversely associated with both mean liver 343 attenuation (ρ = -0.34, p < 0.0001) and L/S ratio (ρ = -0.33, p = 0.0002). The interactive effect between 344 SAM exposure and mean liver attenuation (i.e., difference in slope) was tested but only tended towards 345 significance (interaction term for C2/C0 model, p = 0.077 ( Figure 6D); for β-hydroxybutyric acid model, 346 p = 0.052 (Figure 6E)). Overall, models explained a relatively modest proportion of variance. The models 347 for β-hydroxybutyric acid and mean liver attenuation or L/S ratio both had an adjusted R 2 of 0.21 348 (Supplementary Table 3), but the partial R 2 for SAM exposure was only 0.060 and 0.049, respectively. 349 The C2/C0 models were slightly stronger (mean liver attenuation, adj. R 2 0.33; L/S ratio, adj. R 2 0.40). 350 However, SAM exposure still explained less than 7% of variance in C2/C0 (partial R 2 0.058 and 0.068, 351 respectively) (Supplementary Table 3 To further explore participant subclustering in an agnostic manner, we ran a similarity network fusion 357 (SNF) analysis (Figure 7). This method first builds networks of similar participants based on each dataset 358 separately and then fuses these single-data networks through iterations of spectral clustering. This 359 unsupervised method was used to integrate the 3 different datatypes available (i.e., clinical features, body 360 composition and metabolite variables) (Supplementary Table 4) and reveal participant subclusters based 361 on the similarity between subjects. 362 As visualised by the split in dark (male) vs. pale (female) colors in the top horizontal border of heatmaps, 363 the clusters obtained from clinical features ( Figure 7A) and body composition ( Figure 7B) mostly aligned 364 with sex (NMI, 0.40 and 0.61, respectively). Body composition clusters were mainly associated with the 365 following measures of adiposity (truncal fat mass, android fat mass and total fat mass; p < 0.001) and lean 366 mass (total lean mass and lean mass in trunk, leg, arm compartments; p < 0.001). However, while males 367 mostly grouped together in Body Composition Cluster 3 (BMI 21.9 ± 2.5; fat mass 6.3 kg ± 4.2; lean mass 368 56 kg ± 6.0), females were further split into 2 subclusters that differed mainly by measures of adiposity 369 where Cluster 1 (n=30) was composed of lean individuals (BMI 19.5 ± 2.2; fat mass 10.6 kg ± 4.8; lean 370 mass 34.5 kg ± 4.8) while Cluster 2 (n = 51) mostly grouped females that tended towards overweight and 371 obesity (BMI 28.4 ± 5.1; fat mass 30.4 kg ± 10.7; lean mass 41.0 kg ± 6.4, p<0.001). Similarity clustering 372 based on metabolite variables generally split SAM survivors vs. community participants (blue vs. red, 373 NMI, 0.33); the small subcluster of 8 participants identified ( Figure 7C) is possibly related to deviations 374 from the study protocol such as incomplete fasting. The integration of all datatypes ( Figure 7D) revealed 375 that the two most prominent participants clusters (K = 2) were strongly related to sex (NMI 0.90,K=2 376 Cluster 1: n = 72, female 99% (light grey nodes); K = 2 Cluster 2: n = 83, 99% males (dark grey nodes)). 377 The long separating edges (red and yellow) between these clusters were mostly related to body 378 composition. When participants were grouped into 4 clusters (K=4), the NMI was 0.53 for the groups 379 cross-split by both sex and SAM exposure. However, as seen in the alluvial plot which traces how 380 individuals flow between clusters split, the female dominated groups contained a mix of survivors and 381 community participants with K = 4 Cluster 2 (n = 43) containing 49% female cases and 51% community 382 participants (light grey nodes) and K = 4 Cluster 3 (n = 32) with 53% female cases and 34% community 383 participants (blue nodes). However, the male dominated group (K = 2 Cluster 2) split along SAM exposure 384 as K = 4 Cluster 4 (n = 39) contained 87% male survivors (white nodes) while K = 4 Cluster 1 (n = 41) 385 contained 78% male community participants (dark grey nodes). Also, the similarity grouping subclusters 386 tended to be driven by metabolic features as illustrated by for example the tight net of short blue edges 387 between nodes of K = 4 Cluster 1 and K = 4 Cluster 4. Thus, subclusters associated with having 388 experienced SAM in childhood were more evident in males. 389 390

Discussion 391
This study is one of the first to investigate the metabolic profiles of adults who were hospitalized with 392 severe malnutrition in early childhood using targeted metabolomic analyses. As we hypothesized, the 393 metabolic profiles of adult SAM survivors differed from community participants, and several of the 394 distinguishing metabolite variables had recognized associations with cardiometabolic risk factors. 395 We report that Jamaican adult SAM survivors (age 28.4 ± 8.8 years, > 20 years post hospital discharge) 396 showed differences from community participants of overall similar age, BMI and body composition in 77 397 metabolite variables measured in fasting serum. The profile differences were related to increases in most 398 amino acids but not tryptophan, increases in choline and certain phosphatidylcholines, sphingomyelins 399 and lysophosphatidylcholines together with decreases in many acylcarnitines in SAM survivors compared 400 to community participants. It is to be noted, however, that the observed fold change of individual 401 metabolite variables was very small in the fasted state. These differences might be amplified with age or 402 after a metabolic challenge. Additionally, some SAM survivors were more readily distinguishable from 403 community participants than others and might thus be more vulnerable to cardiometabolic risk. 404 In contrast, we have previously shown that younger Malawian SAM survivors (aged 9·6 ± 1·6 years, 7 405 years post hospital discharge) did not show differences in their metabolic profiles compared to community 406 and sibling participants (14). Thus, the metabolic signatures linked to NCDs that we describe may start to 407 manifest as SAM survivors age. Our cohort is itself relatively young in terms of developing NCDs, yet 408 differences that could set adult SAM survivors on a potentially unfavorable health trajectory were already 409 detected. Differences between the two settings could also be linked to factors other than age such as 410 specific environmental exposures (pollution, water quality, obesity and adult dietary patterns) or 411 differences in diagnostic criteria and treatment strategies. Additionally, in our previous study, >15% of the 412 children were known HIV-positive and >31% had unknown HIV status, whereas HIV-positive individuals 413 were excluded from this current study. 414

415
Our secondary hypothesis was that survivors of severe wasting would have a distinct metabolic profile 416 from survivors of edematous malnutrition, especially given that severe wasting is associated with lower 417 birth weight (22). However, while survivors of severe wasting had lower BMI, waist circumference, lean 418 mass, fat mass and android fat than survivors of edematous malnutrition, metabolic differences were not 419 found between survivors of these two phenotypes. It is notable that although these young adult SAM 420 survivors were not overweight generally (mean BMI < 25kg/m 2 ), survivors of severe wasting had lower 421 lean muscle mass. Thus, the observed changes in body composition between SAM phenotypes might not 422 be sufficient, at this stage, to differentially influence their metabolic profiles in a way that can be 423 detected with a static measure of fasting serum. However, survivors of severe wasting could still be at 424 greater long-term risk, especially considering the link between reduced lean muscle mass and the 425 development of NCDs later in life (34,35). Additionally, we acknowledge that the effects of aging 426 and/or a metabolic challenge in this group will be important to evaluate in future studies as many of 427 these young and mostly lean participants might still be either be suffering from low nutrition quality (at 428 worst) or not exposed to a sufficiently obesogenic diet. 429 430

The association between SAM exposure and adult body size and composition is sex-specific 431
We report a significant interaction between sex, SAM exposure and adult anthropometry and body 432 composition. Male SAM survivors showed a "small" phenotype being of shorter stature, weighing less, 433 and having less lean mass, while having a greater android/gynoid (AG) fat ratio than males from the 434 community. This finding might have intrauterine origins as boys grow faster than girls from an early stage 435 of gestation, and this makes them more vulnerable if their nutrition is compromised (36). Greater AG fat 436 ratio and reduced lean mass in male SAM survivors might have important clinical and metabolic 437 implications, as these factors are associated with an increased risk for metabolic syndrome in healthy 438 adults (37). Additionally, they might also be at risk for later sarcopenic obesity. In keeping with the idea 439 of greater long-term risk in male SAM survivors, adult Mexican males who experienced malnutrition in 440 their first year of life were shown to be more glucose intolerant and hyperinsulinemic compared to 441 controls, using OGGT (17). While female SAM survivors did not differ in body composition to females 442 from the community, clustering analysis revealed that females may show more diverse body type 443 subgroups which could mask the effects of SAM exposure. Also, this divergent sex effect could be due to 444 BMI representing slightly different body composition in males (muscle mass per unit height) vs females 445 (fat mass per unit height). These differences, particularly in height, lean mass and body fat distribution 446 could impose an additional cardio-metabolic risk particularly in male SAM survivors especially if they 447 become obese in later life. 448

Metabolic profiles in relation to risk of type 2 diabetes 449
We questioned specifically whether metabolic perturbations linked to having experienced SAM in early 450 childhood could persist and/or be associated with the cardiometabolic risk profiles of adult survivors. 451 As previously demonstrated (19), some members of this cohort of adult SAM survivors had similar insulin 452 sensitivity and beta-cell function to community participants. However, this subset of SAM survivors had 453 higher concentrations of BCAAs and AAAs, five of which (isoleucine, leucine, valine, tyrosine and 454 phenylalanine) have reported associations with diabetes risk in normoglycemic individuals (38). 455 Additionally, adult SAM survivors had lower tryptophan and an associated higher KT ratio which has 456 been identified as a predictor of incident T2D and coronary events, with the dysregulation of the KT 457 metabolic pathway described as one of the mechanisms of insulin resistance (39). Further, SAM survivors 458 had higher median concentrations of urea cycle amino acids (arginine, citrulline, ornithine, aspartic acid 459 and urea) which have been associated with T2D (40, 41). The higher glutamic acid seen in SAM survivors 460 has also been associated with both increased 2-hour plasma glucose and higher tertiles of HOMA-IR (42). 461 Furthermore, two phosphatidylcholine subclasses (diacyl and acyl-alkyl phosphatidylcholines) were 462 higher in SAM survivors than community participants with PC ae C36:0 showing the greatest difference. 463 These structural lipids (i.e. constituents of cell membranes) are also involved in cell signaling and 464 metabolic control (43), and, together with other choline-containing phospholipids such as 465 lysophosphatidylcholines and sphingomyelins, have been linked to increased risk of T2D (44). Some 466 studies report lower acyl-alkyl-phosphatidylcholines in subjects with insulin resistance (45), but these 467 results were not replicated in our cohort. 468 Taken together, these metabolic findings could suggest greater risk of glucose dysmetabolism and eventual 469 T2D in SAM survivors, albeit in the current absence of overtly impaired insulin sensitivity or clinical 470 disease which may develop with obesity and age to oral glucose tolerance testing (46). 471

Metabolic profiles in relation to risk of fatty liver disease 472
The hydrolysis of lipid stores and the oxidation of fatty acids are a key acute adaptive response to SAM, 473 evidenced by high circulating levels of free fatty acids (FFA), ketones and even-numbered acylcarnitines 474 (47). Ultimately, in both SAM phenotypes, hepatic mitochondrial function is reduced and associated with 475 hepatic steatosis (48) (49). While hepatic fat does accumulate during a SAM episode, it resolves 476 completely (albeit slowly) with recovery (50). However, it is unclear whether any lingering metabolic 477 perturbations could impact later hepatic fat metabolism. 478 Non-alcoholic fatty liver disease (NAFLD) results from either excess FFA delivery from diet or peripheral 479 stores (secondary to peripheral insulin resistance), or decreased intrahepatic FFA oxidation and increased 480 de novo lipogenesis (51). While SAM survivors had less liver fat than community participants, the 481 difference was small, and both groups failed to meet the criteria for moderate-to-severe fatty liver (L/S < 482 1). However, compared to community participants, SAM survivors had lower β-hydroxybutyric acid 483 concentrations and lower C2/C0 (related to lower acylcarnitine (C2) concentrations). Beta-hydroxybutyric 484 acid and C2 were highly correlated in our data and are both known to regulate β-oxidation of FFA. β-485 hydroxybutyric acid is a marker of hepatic ketogenesis post FFA oxidation, and its production rate is the 486 major determinant of its concentration in serum (52, 53). Carnitine is essential for the uptake of fatty acids 487 into mitochondria prior to β-oxidation and is known to be reduced in persons with NAFLD (54). 488 Interestingly, the associations between liver fat and both β-hydroxybutyric acid and C2/C0 are weaker in 489 SAM survivors compared to community participants, suggesting that the diagnosis of SAM might 490 somehow diminish the association between these metabolites and liver fat. 491 These metabolic findings suggest that while early life SAM may limit β-oxidation, at this stage, the impact 492 on the development of NAFLD is inconclusive. This is consistent with a recent study in rodents where 493 protein restriction after weaning and subsequent feeding of a high fat and carbohydrate diet did not induce 494 hepatic steatosis (55). However, decreased fatty acid oxidation in SAM survivors could represent a 495 harbinger for later hepatic steatosis in these currently still young, lean SAM survivors. Additionally, the 496 timing and severity of the early nutritional insult (prenatal vs postnatal) might variably influence the 497 development of NAFLD, and, as intrauterine growth restriction has been associated with subsequent 498 NAFLD (56), the combined insults may be additive. Changes in metabolites other than those related to β-499 oxidation could also potentially affect the risk of developing NAFLD. For example, the higher KT ratio 500 seen in SAM survivors might be evidence of defective indoleamine 2,3-dioxygenase activity and this has 501 been associated with liver inflammation and fibrosis (57). 502 503

Metabolic evidence of risk of hypertension 504
SAM survivors and community participants had similar systolic and diastolic blood pressure. Although 505 20.8% of our participants had elevated blood pressure readings (21.3% with measured diastolic blood 506 pressure ≥ 80 mm Hg, 3.9% with measured systolic blood pressure ≥ 140 mm Hg), we did not demonstrate 507 a significant association between any targeted metabolite and elevated systolic blood pressure, elevated 508 diastolic blood pressure or a combination of the two. However, choline and choline-containing molecules 509 have a reported association with the development of hypertension (58) and both choline and 510 phosphatidylcholine concentrations were higher in SAM survivors compared to community participants. 511 Several other metabolites with reported associations with elevated blood pressure ([alpha]-1 acid 512 glycoproteins (58) and serum free fatty acids i.e. heptanoic, oleic, non-anoic, eicosanoic and hexanoic 513 acids (59)) were not targeted for analysis in this study, thus our findings may be inconclusive. 514 515 516 517

Limitations and Strengths 518
This study has limitations. First, although representative of the full cohort based on age, sex and BMI, the 519 number of participants analyzed here included less than half of those enrolled in the main cohort; thus 520 certain sub-analyses might be underpowered. Additionally, CT scans are less reliable at evaluating mild 521 liver fat accumulation. Also, metabolites measured at fasting must be interpreted with some caution as 522 their concentrations do not reflect pathway flux (increased production versus decreased utilization). 523 Missing data in certain variables (e.g. birth weight for community participants, specific dietary 524 information, income and L/S ratio) made it difficult to account for additional suspected confounders and/or 525 include all participants for certain analyses such as SNF. Despite these limitations, the study was 526 strengthened by several factors: 1) the cohort is very well characterized with rich longitudinal clinical 527 data, 2) we ensured high analytical sensitivity by using targeted mass spectrometry, and measuring serum 528 as opposed to plasma, and 3) data was analyzed using conventional statistical approaches together with 529 both supervised and unsupervised machine learning methods for feature selection, controlling false 530 discovery rates, mitigating overfitting and describing general patterns. Importantly, we demonstrated for 531 the first time that early-life SAM could lead to metabolic derangements more than 20 years later and these 532 may be tracked over time to identify patients at particular risk of developing NCDs. 533 534

Conclusions 535
This study is the first to investigate the metabolic profiles of adults who were hospitalized for severe acute 536 malnutrition in early childhood. Our data provides evidence that metabolic profiling can distinguish adult 537 survivors of SAM from unexposed participants living in the same communities and of similar age, sex 538 and BMI. Some metabolite variables that are greater in adult SAM survivors are associated with the risk 539 of T2D and may be signatures of reduced hepatic fatty acid oxidation and possibly NAFLD. These 540 findings should be validated further using stable isotopes that can capture pathway flux, by exposing SAM 541 survivors to specific metabolic challenges and by repeating studies in older cohorts. Our findings support 542 the hypothesis that persons exposed to SAM in early life have long term metabolic consequences that 543 could impact risk for NCDs and thus, they may benefit from targeted clinical management. 544 545 Acknowledgements 546 We gratefully acknowledge the men and women who took part in the study. We also recognize Professor    Assay mass spectrometry-based analysis between SAM survivors (n = 122) and age, sex and BMI-764 matched community participants (n = 90). Boxplots summarize medians (midline) and interquartile ranges 765 (IQRs); circles represent outlying data points; false discovery rate corrected p-values are presented. Grey 766 shaded box includes the top 15 metabolite variables that were most differential between groups; black box 767 highlights differential metabolites previously associated with NCDs. AAs, amino acids; AAA, aromatic 768 amino acids; BCAA, branched-chain amino acids; LYSOPC, lysophosphatidylcholines; SM, 769 sphingomyelins, PCaa, phosphatidylcholine di-acyl; PCae, phosphatidylcholines acyl-alkyl; SAM, severe 770 acute malnutrition.