Placental insulin-mTOR and stress–inflammatory signaling patterns are associated with childhood adiposity Authors: Ellen C. Francis, Kristen E. Boyle, Lauren E. Gyllenhammer, Dana Dabelea, Thomas Jansson, Wei Perng Published: 2026 ABSTRACT BackgroundPlacental signaling pathways regulate nutrient transport and fetal growth, with potential long-term consequences for offspring metabolic health. Most prior human studies have focused on individual placental markers, limiting insight into the role of coordinated activity across multiple pathways in relation to offspring outcomes. Our objective was to identify patterns across placental nutrient signaling pathways and assess whether the latent placental signaling patterns were associated with early childhood adiposity, and secondarily, explore associations of adiposity-associated patterns with metabolic biomarkers.MethodsAmong 108 mother-child pairs from the Healthy Start cohort, we quantified 33 placental signaling proteins and their phosphorylated-to-total protein ratios involved in nutrient sensing, insulin/growth factor signaling, stress/inflammation, and mitochondrial biogenesis using Simple Western assays of term placental villus tissue. We applied unsupervised methods to identify latent patterns and LASSO regression was used to select patterns associated with %fat mass at age 4. Multivariable linear regression was used to estimate associations adjusting for offspring age, sex, and maternal pre-pregnancy BMI. These same models were used in exploratory analysis of fasting levels of adiponectin, leptin, insulin, glucose, and lipids at age 4.ResultsWe identified two placental signaling patterns associated with %fat mass. The insulin-mTOR-energy sensing pattern was associated with lower childhood %fat mass (β = −2.46, 95% CI − 4.84, − 0.09) and adiponectin, and the stress-inflammatory MAPK pattern was associated with higher %fat mass (β = 1.28, 95% CI 0.05, 2.51), leptin, and triglycerides; however, the FDR p-values ranged from 0.06 to 0.13.ConclusionTwo placental signaling patterns were associated with childhood %fat mass and metabolic markers. These findings indicate that capturing placental activity across several signaling pathways may yield insights into early origins of adiposity and metabolic health.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12916-026-04896-8. Background The activation of placental signaling pathways that regulate nutrient transport is central to fetal development and may have lasting implications for growth and metabolic health across the life course [1]. Animal studies have demonstrated that disruptions in placental signaling pathways can impair fetal growth. For example, increased placenta nutrient transport via mTOR activation has been associated with fetal overgrowth, while greater placental stress activation (e.g., JNK) has been observed in intrauterine growth restriction—a condition linked to increased fat mass and a less favorable cardiometabolic profile later in life [2–5]. Human epidemiologic studies have demonstrated that cardiometabolic risk factors present in childhood track across the life course and are predictors of future obesity [6–10], type 2 diabetes, and metabolic syndrome [11–15], highlighting the importance of early identification of biological risk before overt complications arise [16]. In human pregnancies, variation in the above placental signaling pathways have also been associated with neonatal and childhood metabolic outcomes, including adiposity, lipid profiles, and insulin sensitivity [17–24]. However, much of this work has focused on discrete proteins or single pathways, often in the context of maternal complications such as obesity or gestational diabetes (GDM) [25]. As a result, there is limited understanding of how the interrelationship among these pathways in healthy, low-risk pregnancies—where subtle signaling variation may still be biologically meaningful. Even subtle shifts in placental signaling may nonetheless influence long-term health trajectories, given that gestation is a sensitive period for developmental programming. As biological pathways operate in concert, understanding their coordination may be particularly important in normal low-risk pregnancies where severe dysregulation of a single pathway is likely to be absent. As such there is a need to identify latent patterns across interacting placental nutrient signaling pathways, to provide a broader-level view of molecular signaling across several pathways which would complement traditional single-protein models [26, 27]. The objectives of this study were to identify latent patterns across placental nutrient signaling pathways associated with early childhood fat mass and secondarily, as an exploratory objective, evaluate whether the placental signaling patterns associated with fat mass are also associated with biomarkers of metabolic risk and BMI percentile. Methods Participants and sample collection We used data from the Healthy Start BabyBUMP Project, a subsample of the Healthy Start Study (ClinicalTrials.gov; NCT02273297). The larger Healthy Start pre-birth cohort study enrolled 1,410 women aged ≥ 16 years of age at < 24 weeks of gestation from prenatal clinics at the University of Colorado Hospital between 2009 and 2014. Women with prior diabetes, a history of prior preterm birth < 25 gestational weeks or fetal death, asthma with active steroid management, serious psychiatric illness or multiple gestation were excluded [28]. The ancillary Healthy Start BabyBUMP Project comprised a convenience subsample of placentas collected immediately after delivery. Trophoblast villous tissue was dissected, snap-frozen in liquid nitrogen, and stored at − 80 °C for biomarker analysis. Placental samples were obtained from 111 mother–offspring pairs. After excluding two preterm deliveries and one neonate missing all outcome data, 108 pairs were eligible for analysis. Complete placental protein measurements were available for 93 placentas, which were used to derive latent signaling patterns. Offspring were invited to a follow-up research visit in early childhood (mean age 4.8 ± 0.6 years), during which anthropometry and body composition were assessed. Sixty-two children completed the visit. We evaluated potential differences in maternal, pregnancy, and child characteristics between participants with and without observed fat mass measurements. Because these characteristics were similar across groups, analyses of offspring outcomes were restricted to complete cases. The final analytic sample for associations with fat mass included 57 mother–offspring pairs with both complete placental protein and follow-up outcome data. Placental samples and signaling The placental proteins were selected a priori based on their roles in insulin/growth factor signaling, inflammation, nutrient transport, and energy sensing [19, 29]. We homogenized ~ 20 mg of frozen trophoblast placental villus tissue in 75µL ice-cold buffer D (250mM sucrose, 10mM HEPES, pH 7.4) that had a 1:100 dilution of protease and phosphatase inhibitors. Next, we used Simple Western Assays with WES (ProteinSimple, Santa Clara, CA) to measure the phosphorylation and total abundance of the following proteins: ribosomal protein S6 (RPS6), p70 ribosomal protein S6 kinase 1 (S6K1), eukaryotic translation initiation factor 4E-binding protein 1 (4E-BP1), protein kinase B (Akt), protein kinase C-α (PKCα), insulin receptor β (IRβ), insulin-like growth factor 1 receptor (IGF-1r), glycogen synthase kinase-3 beta (GSK3β), extracellular signal-regulated kinase (ERK), 1/2 pro-caspase 1, signal transducer and activator of transcription 3 (STAT3), c-Jun N-terminal kinase 1 and 2 (JNK1 and JNK2), interleukin-1 beta (IL-1β), p38 mitogen-activated protein kinase (p38 MAPK), AMP-activated protein kinase (AMPK), eukaryotic initiation factor 2 alpha (eIF2α), O-linked N-acetylglucosamine transferase (OGT), phosphoinositide 3-kinase (PI3K), 11β-hydroxysteroid dehydrogenase type 1 (11β-HSD1), and peroxisome proliferator-activated receptor gamma coactivator 1-alpha (PGC-1α). We used a 0.1 mg/mL total protein concentration (Additional File: Fig. 2) and ran the WES plates according to the manufacturer’s instructions, with slight modification (200 V, 55 m separation time). We included an equalizer sample on each plate with clean median values for each protein to control for batch variation. We also multiplexed a loading control (vinculin or β-actin) in each capillary and normalized the protein levels to it. Placental protein expression was normalized to total protein levels, and phosphorylated-to-total ratios were computed to capture signaling activation. Total protein expression values were also retained to assess co-expression patterns. Maternal and child data Maternal lifestyle and clinical data Maternal race and ethnicity, educational attainment, parity, and smoking status during pregnancy were self-reported via questionnaire. GDM status was abstracted from medical records and all data collection are reported in detail elsewhere [30, 31]. We calculated maternal pre-pregnancy BMI based on pre-pregnancy weight obtained from medical records (89%) or self-report (11%), and measured height at the first research visit. Gestational weight gain (GWG) was estimated by subtracting pre-pregnancy weight from the last clinically measured weight during pregnancy and categorized based on the Institute of Medicine 2009 Guidelines [32]. Child outcomes The primary outcome of interest in this analysis was % fat mass. This variable is based on direct measures of fat tissue (i.e., adipose tissue) and calculated as the proportion of fat tissue relative to fat free mass (i.e., water, bone and non-bone mineral, and protein). We focused on adiposity because for the majority of youth, development of excess adiposity is the initial event in a sequence of events leading unfavorable changes in metabolic profile [33]. Early childhood fat mass was measured using the BOD POD Pediatric Option, which is has been validated in children against hydrodensitometry and the four-compartment model [34]. Measurements for each participant were taken in triplicate, and the average of the two closest measures was used for analyses. Secondary outcome included: fasting serum was used to measure triglycerides (TGs), total cholesterol, high-density lipoprotein (HDL), low-density lipoprotein (LDL), and glucose using manufacturer prepackaged enzymatic kits and the AU400e Chemistry Analyzer (Olympus America). Insulin was measured using a radioimmune assay, and leptin and adiponectin were measured using a Multiplex assay kit, all by Millipore Corporation. We included child body mass index (BMI) percentile based on height and weight were measured by trained nurses, and. calculated according to the World Health Organization (WHO) growth reference [35, 36], as a secondary outcome to provide insight into whether associations with fat mass percentage were also reflective of differences in overall size. Data and statistical analysis Placental protein concentrations were log-transformed and standardized prior to analysis. The analytical workflow is shown in Fig. 1. Latent placental signaling patterns were derived using unsupervised approaches to characterize underlying structure in the protein data without conditioning on child outcomes. An unsupervised approach can identify placental signaling patterns across pathways that may not be apparent using outcome-driven feature selection. Three complementary methods were applied: principal component analysis (PCA), which identifies orthogonal axes of maximal variance; weighted gene co-expression network analysis (WGCNA), which detects correlation-based modules; and consensus clustering, which identifies stable discrete groupings through internal resampling [37, 38]. These three approaches were used to assess the reproducibility of signaling patterns across the different methods. Specifications for each approach are provided in Additional File: Fig. 1, and analytic R packages are listed in Additional File: Table 1. Fig. 1Analytic workflow for evaluating placental signaling patterns and their association with early childhood metabolic health. The analysis proceeded in five sequential steps. [1] Placental protein data were preprocessed by normalizing expression levels to total placental protein and calculating phospho-to-total protein ratios to capture pathway activation (n = 93). [2] To derive latent variables summarizing placental signaling, we applied three unsupervised data reduction methods—principal component analysis (PCA), weighted gene co-expression network analysis (WGCNA), consensus clustering —each generating summarized scores from protein expression profiles. [3] We used LASSO regression with 10-fold cross validation to select the signaling patterns predictive of 4 year fat mass percentage (n = 57), adjusting for age and sex as unpenalized terms. Separate LASSO models were specified for each of the three approaches. The patterns that were not penalized to zero were retained for use in multivariable regression modeling. [4] The proteins that grouped onto each retained patterns were compared to assess whether similar signaling patterns were identified across methods. [5] Summary scores from the selected patterns were used in adjusted models to examine associations with early childhood fat mass, metabolic outcomes, and BMI. Analytic workflow for evaluating placental signaling patterns and their association with early childhood metabolic health. The analysis proceeded in five sequential steps. [1] Placental protein data were preprocessed by normalizing expression levels to total placental protein and calculating phospho-to-total protein ratios to capture pathway activation (n = 93). [2] To derive latent variables summarizing placental signaling, we applied three unsupervised data reduction methods—principal component analysis (PCA), weighted gene co-expression network analysis (WGCNA), consensus clustering —each generating summarized scores from protein expression profiles. [3] We used LASSO regression with 10-fold cross validation to select the signaling patterns predictive of 4 year fat mass percentage (n = 57), adjusting for age and sex as unpenalized terms. Separate LASSO models were specified for each of the three approaches. The patterns that were not penalized to zero were retained for use in multivariable regression modeling. [4] The proteins that grouped onto each retained patterns were compared to assess whether similar signaling patterns were identified across methods. [5] Summary scores from the selected patterns were used in adjusted models to examine associations with early childhood fat mass, metabolic outcomes, and BMI. To identify which latent patterns were most strongly and consistently associated with childhood fat mass percent, we applied regularized regression (LASSO, α = 1) to each set of derived patterns (principal components, WGCNA modules, and consensus clusters). Three separate LASSO models were specified for each of the three approaches. For instance, the WGCNA LASSO model included only the WGCNA modules and covariates. The LASSO approach imposes a penalty to shrink coefficients, reducing overfitting and instability arising from correlated predictors and modest sample size [39, 40]. The regularization parameter λ was determined via 10-fold cross-validation, and we retained variables with non-zero coefficients at λmin (the value that minimized the mean cross-validated error). Because LASSO was used for pattern selection rather than clinical prediction, we utilized λmin to prioritize pattern selection of biologically relevant signaling patterns with moderate effect sizes, which could be inadvertently excluded by more conservative penalties (e.g., λse) [41]. Selected placental patterns were interpreted and labeled based on the signaling pathways that were most predominately represented by the proteins in each pattern. To qualitatively assess reproducibility across the three unsupervised approaches, we visualized protein membership across methods and evaluated concordance in the proteins assigned to each pattern. These selected patterns were then evaluated in multivariable linear regression models to estimate associations with childhood fat mass. Model 1 adjusted for child sex and age at follow-up. Model 2 additionally adjusted for pre-pregnancy BMI. Associations are reported as β coefficients with 95% confidence intervals. Bayesian Information Criterion (BIC) was used to compare relative model fit across patterns. Finally, as an exploratory analysis, we investigated associations of the % fat mass associated placental patterns with concurrently assessed metabolic markers and BMI using linear regression (model 1 and model 2 adjustments as detailed above) and visualized results with heatmaps. Given the limited sample size for these outcomes (n = 23–31), these analyses were considered secondary. To account for multiple comparisons, Benjamini–Hochberg false discovery rate–adjusted q-values are reported alongside nominal P-values. We provide interpretation of results where nominal P-values < 0.05 and indicate whether these passed FDR correction. All analyses were conducted in R (version 4.0.0). Results Maternal and child characteristics are presented in Table 1. Most of the pregnancies were from non-Hispanic white females who had completed at least some college. The mean pre-pregnancy BMI was 24.9 ± 5.7 kg/m2 and 5% of women developed GDM. The mean gestational age of offspring was at term (277 ± 8.4 days), birth weight-for-gestational age percentile of 30.4 (24.5), and 54% were males. At the follow-up research visit, the mean age of children was 4.8 ± 0.6 years, and BMI percentile was 46.2 ± 23.2. Table 1Maternal and offspring characteristicsCharacteristics; mean (SD), n (%)n = 94 Maternal Age at index pregnancy (years)29.2 (6.1)Race-ethnicityNon-Hispanic White60 (64.5%)Non-Hispanic Black9 (9.7%)Hispanic20 (21.5%)Non-Hispanic Other4 (4.3%)EducationCollege Degree22 (23.7%)Grad Degree24 (25.8%)Some college25 (26.9%)HS or less22 (23.7%)Pre-pregnancy BMI (kg/m2)24.9 (5.7)Gestational diabetes mellitusYes5 (5.4%)Total gestational weight gain (lbs)13.5 (7.8)Smoked during pregnancy5 (5.4%) Neonatal Male sex50 (53.8%)Gestational age at delivery (days)277 (8.4)Weight-for-gestational age percentile b30.4 (24.5) Childhood Age (years)4.8 (0.6)BMI, (kg/m2) c15.2 (1.3)BMI, (percentile)46.2 (23.2)a Other includes, Asian, American Indian/Alaska natives, Hawaiian/Pacific Islanders.b Birthweight z-score specific to gestational age derived from U.S. natality reference [42].c BMI specific to age derived from WHO growth standards [43].HEI; Healthy Eating Index. Table 1Maternal and offspring characteristicsCharacteristics; mean (SD), n (%)n = 94 Maternal Age at index pregnancy (years)29.2 (6.1)Race-ethnicityNon-Hispanic White60 (64.5%)Non-Hispanic Black9 (9.7%)Hispanic20 (21.5%)Non-Hispanic Other4 (4.3%)EducationCollege Degree22 (23.7%)Grad Degree24 (25.8%)Some college25 (26.9%)HS or less22 (23.7%)Pre-pregnancy BMI (kg/m2)24.9 (5.7)Gestational diabetes mellitusYes5 (5.4%)Total gestational weight gain (lbs)13.5 (7.8)Smoked during pregnancy5 (5.4%) Neonatal Male sex50 (53.8%)Gestational age at delivery (days)277 (8.4)Weight-for-gestational age percentile b30.4 (24.5) Childhood Age (years)4.8 (0.6)BMI, (kg/m2) c15.2 (1.3)BMI, (percentile)46.2 (23.2)a Other includes, Asian, American Indian/Alaska natives, Hawaiian/Pacific Islanders.b Birthweight z-score specific to gestational age derived from U.S. natality reference [42].c BMI specific to age derived from WHO growth standards [43].HEI; Healthy Eating Index. Maternal and offspring characteristics a Other includes, Asian, American Indian/Alaska natives, Hawaiian/Pacific Islanders. b Birthweight z-score specific to gestational age derived from U.S. natality reference [42]. c BMI specific to age derived from WHO growth standards [43]. HEI; Healthy Eating Index. The placental signaling proteins showed a modest correlation structure, with a median pairwise Spearman correlation of 0.12 (IQR: 0.06–0.21). Across the unsupervised approaches (PCA, WGCNA, consensus cluster), several proteins repeatedly grouped together and the visualization of this is provided in Additional File: Fig. 1. Proteins involved in insulin and metabolic growth signaling pathways, including IGF1r, AKT, and PGC1α often grouped together. Proteins associated with cellular stress or inflammatory signaling, such as total levels of STAT3, eIF2a, p38MAPK, and JNK also tended to group together. Phophorylation of mTORC1 downstream targets, 4E-BP1 and RPS6, were frequently grouped. A single cluster identified from consensus clustering (Cluster 3) and three components from PCA (PC6, PC7, and PC8) had beta estimates that were not penalized to zero LASSO (Additional File: Table 2). The WGCNA modules had beta estimates penalized to zero and therefore, were not included in further analyses. From these latent variables (Cluster 3, PC6, PC7, PC8), two signaling patterns were identified. The first represented an insulin-mTOR-energy sensing pattern, characterized by acitviation of AMPK and mTORCI translational targets. This pattern was represented in protein groupings identified by both consensus clustering (Cluster 3) and PCA (primarily Principal Component 8, with partial overlap with Principal Component 6). The second represtend a stress-inflammatory MAPK signaling pattern, characterized by JNK, ERK, and related stress-response proteins, and was identified by PCA component 7 only. The proteins assigned to Cluster 3, and the most representative proteins of Component 8, 6, and 7 (loading score ≥ 75th percentile) are shown in Fig. 2. For PCA, protein loading scores were conditionally shaded based on the magnitude and the direction of loading. Fig. 2Placental Signaling Patterns and Characteristic Proteins. Numbers are the protein loading scores. The top loading proteins (absolutely value ≥ 75th percentile) were selected as characteristic of each component. AMPK, 4EBP1, and RPS6 were assigned to Cluster 3 (C3), these protiens were top loading proteins on component 8. AMPK and RPS6 loaded on component 6. The proteins captured signaling in energy sensing and mTORC1 downstream targets. Component 7 proteins were distinct from C3 and captured signaling in stress, inflammation, and MAPK. Darker purple represents strong positive loadings, darker blue represents stronger negative loadings. *indicates a protein identified in several patterns. Placental Signaling Patterns and Characteristic Proteins. Numbers are the protein loading scores. The top loading proteins (absolutely value ≥ 75th percentile) were selected as characteristic of each component. AMPK, 4EBP1, and RPS6 were assigned to Cluster 3 (C3), these protiens were top loading proteins on component 8. AMPK and RPS6 loaded on component 6. The proteins captured signaling in energy sensing and mTORC1 downstream targets. Component 7 proteins were distinct from C3 and captured signaling in stress, inflammation, and MAPK. Darker purple represents strong positive loadings, darker blue represents stronger negative loadings. *indicates a protein identified in several patterns. In multivariable regression analyses, higher scores for the insulin-mTOR-energy sensing pattern (both methods) were associated with lower childhood fat mass percentage (consensus Cluster 3: β = −2.46, 95% CI − 4.84, − 0.09; PC8 showed similar effect sizes and model fit) (Table 2). In contrast, the stress-inflammatory MAPK pattern (PCA method) was associated with higher fat mass percentage (β = 1.28, 95% CI 0.05, 2.51). Although PCA Component 6 was retained as predictive of fat mass percentage, the estimates were smaller and less consistent (higher BIC values and confidence intervals including the null). Table 2Association of Placental Signaling Clusters with Childhood Fat Mass PercentagePattern, latent variableModel 1Model 2 N BETA (95% CI)BIC N BETA (95% CI)BIC Insulin-mTOR -energy sensing pattern Consensus Cluster 3 57-2.44 (-4.79, -0.09)384.157-2.46 (-4.84, -0.09)388.1 Principal Component 8 57-1.56 (-3.04, -0.07)384.157-1.56 (-3.06, -0.06)388.1 Principal Component 6 570.91 (-0.33, 2.15)386.4570.91 (-0.35, 2.16)390.4 Stress-inflammatory MAPK pattern Principal Component 7 571.24 (0.04, 2.45)384.2571.28 (0.05, 2.51)388.1Abbreviations: mTOR, Mechanistic Target of Rapamycin, MAPK, Mitogen-Activated Protein KinaseModel 1: Age and sexModel 2: Model 1 + maternal pre-pregnancy body mass index (kg/m2) Table 2Association of Placental Signaling Clusters with Childhood Fat Mass PercentagePattern, latent variableModel 1Model 2 N BETA (95% CI)BIC N BETA (95% CI)BIC Insulin-mTOR -energy sensing pattern Consensus Cluster 3 57-2.44 (-4.79, -0.09)384.157-2.46 (-4.84, -0.09)388.1 Principal Component 8 57-1.56 (-3.04, -0.07)384.157-1.56 (-3.06, -0.06)388.1 Principal Component 6 570.91 (-0.33, 2.15)386.4570.91 (-0.35, 2.16)390.4 Stress-inflammatory MAPK pattern Principal Component 7 571.24 (0.04, 2.45)384.2571.28 (0.05, 2.51)388.1Abbreviations: mTOR, Mechanistic Target of Rapamycin, MAPK, Mitogen-Activated Protein KinaseModel 1: Age and sexModel 2: Model 1 + maternal pre-pregnancy body mass index (kg/m2) Association of Placental Signaling Clusters with Childhood Fat Mass Percentage Abbreviations: mTOR, Mechanistic Target of Rapamycin, MAPK, Mitogen-Activated Protein Kinase Model 1: Age and sex Model 2: Model 1 + maternal pre-pregnancy body mass index (kg/m2) In analyses of associations between the fat mass associated placenta signaling patterns and child metabolic markers and BMI, the insulin-mTOR-energy sensing pattern (associated with lower fat mass percentage) was associated with lower adiponectin. In contrast, the stress-inflammatory MAPK signaling pattern (associated with greater fat mass percentage) was associated with higher leptin and triglycerides; however, while some of the effect estimates were large (e.g., triglyceride β = 4.60, 95% CI 0.44, 8.76; P-value = 0.03), the FDR corrected Q-values were > 0.05. To facilitate interpretation of these models, the beta estimates, P-and q-values, were displayed in Fig. 3 with conditional shading of the estimates based on the magnitude and the direction. Fig. 3Linear Models of Insulin-mTOR-energy sensing and Stress-inflammatory MAPK patterns and Childhood Metabolic Health Indicators Insulin-mTOR-energy sensing (Cluster 3 and PC8, PC6), and Stress-inflammatory MAPK (PC7) Panel A: Beta coefficients, P-value, and FDR Q-value of from a model of latent placenta variables and offspring metabolic health indicators adjusted for offspring age and sex. Darker red colors indicate positive larger beta coefficients, darker blue indicate larger negative beta coefficents. Panel B: additional adjustment for maternal pre-pregnancy BMI Linear Models of Insulin-mTOR-energy sensing and Stress-inflammatory MAPK patterns and Childhood Metabolic Health Indicators Insulin-mTOR-energy sensing (Cluster 3 and PC8, PC6), and Stress-inflammatory MAPK (PC7) Panel A: Beta coefficients, P-value, and FDR Q-value of from a model of latent placenta variables and offspring metabolic health indicators adjusted for offspring age and sex. Darker red colors indicate positive larger beta coefficients, darker blue indicate larger negative beta coefficents. Panel B: additional adjustment for maternal pre-pregnancy BMI Discussion In this pregnancy cohort with longitudinal follow-up from birth through early childhood (median offspring age 4.79 ± 0.6 years), we utilized three unsupervised analytic approaches: PCA, WGCNA, and consensus clustering to identify latent placental signaling patterns. Two placental signaling patterns that captured multiple pathways relevant to early-life programming of future adiposity and metabolic risk were identified: an insulin-mTOR-energy sensing pattern, associated with lower childhood fat mass percentage; and a stress-inflammatory AMPK signaling pattern associated with higher childhood fat mass percentage. In addition to being associated with offspring adiposity, these signaling patterns also exhibited biologically aligned associations with several childhood metabolic markers (adiponectin, leptin, triglycerides) The insulin-mTOR-energy sensing pattern, characterized by activation of RPS6, 4E-BP, and AMPK, was inversely associated with early childhood fat mass percentage and adiponectin. These findings align with prior research reporting inverse associations of AMPK activation and birthweight [44, 45], as well as broader evidence linking AMPK activity with improved metabolic outcomes [46]. On the other hand, the association of this placental pattern with lower adiponectin is less clear. Lower levels of adiponectin have been observed in the context of obesity [47], and among adults, lower adiponectin is often associated with worse glycemic control [48]. However, prior studies in neonates and children have found similarly paradoxical correlations of higher adiponectin and greater weight, weight gain, and sum of skinfolds in childhood [49–51]. The exact role of adipokines in early childhood are unknown, given that adiponectin influences adipose deposition (e.g., subcutaneous, visceral) [52], during periods of rapid development, such as early childhood, lower adiponectin may not signal greater or even pathologic insulin resistance as it does in adult populations [52]. In the context of our study, the clustering of AMPK, 4E-BP1, and RPS6—along with their inverse association with fat mass and adiponectin, may reflect coordinated regulation of placental growth and nutrient signaling. Given that most prior studies on placental mTOR and AMPK signaling have been conducted in high-risk pregnancies (e.g., maternal obesity, hyperglycemia, or hypoxia), where placental function is often pathologically disrupted [25, 45], 53– [55], an important contribution of this work is that the Healthy Start cohort is primarily comprised of low-risk pregnancies, allowing us to observe signaling patterns more likely to reflect physiological variation rather than overt pathology. A particularly notable aspect of this pattern was the co-activation of AMPK with mTORC1 downstream targets RPS6 and 4E-BP1. This may appear counterintuitive, as AMPK is classically thought to inhibit mTOR signaling, particularly in energy-stressed states [56]. However, placental signaling does not always follow traditional nutrient-sensing paradigms observed in liver, adipose, or skeletal muscle. Indeed, prior animal studies have shown that placental mTOR can be inhibited even in the absence of AMPK activation, suggesting distinct regulatory mechanisms [22, 57]. Co-activation of AMPK and mTOR effectors may represent placenta-specific regulation that balances energy sensing with protein synthesis [58]. For example, we have previously speculated that 4E-BP1 activation may preferentially support placental growth and biosynthesis, whereas RPS6 signaling may be more directly involved in nutrient transport, oxidative metabolism, and fetal growth [29]. Given the complex regulation of placental signaling by several inputs (e.g., glucose, amino acids, blood flow, oxygenation), these data underscore the need for future studies to disentangle how integrated placental signaling networks shape fetal nutrient exposure and influence early metabolic trajectories [59]. We found that Stress-inflammatory AMPK pattern—marked by higher JNK1 activation, IL-1β, IGF-1r and PGC1α signaling was positively associated with childhood fat mass, and to some extent higher triglyceride levels and BMI percentile. Higher BMI, greater fat mass percentage, and elevated triglycerides typically characterizes a state of metabolic dysfunction or adiposity-related risk; [33] however, the FDR corrected P-values were > 0.05. The clustering of IGF-1r and PGC1α is biologically plausible, as IGF-1r is a known upstream regulator of PGC1α expression and mitochondrial biogenesis [60]. Beyond the role of PGC1α in mitochondrial remodeling, it also influences lipid metabolism including long-chain fatty acid oxidation and hepatic triglyceride storage and secretion [61, 62]. The grouping of these proteins with placental stress and inflammatory signaling pathways, may indeed capture responses to inflammation or oxidative stress in utero that may be associated with alterations in lipid metabolic pathways that could relate to later elevated triglyceride levels in childhood and potential risk for a worse adiposity trajectory Strengths and limitations A major strength of this study is the use of absolute quantification for a panel of placental proteins, including phosphorylation at specific residues indicative of pathway activation, enabling direct comparison of signaling activity across participants and enhancing biological interpretability. An additional strength is the focus on a generally healthy pregnancy cohort with low prevalence of GDM (n = 5) and maternal obesity (n = 15), which allowed characterization of physiological signaling coordination rather than overt pathologic dysregulation. This perspective is particularly valuable given the limited foundational knowledge of normal placental signaling regulation [63], and that many canonical signaling relationships observed in other metabolic tissues—such as liver, adipose, or skeletal muscle—do not always translate directly to the placenta. By analyzing protein clusters in a healthy cohort, we were able to examine signaling patterns that may reflect normal physiological coordination of placental function, an area often underexplored in favor of pathophysiologic models However, these strengths must be considered alongside limitations. The clustering patterns and their associations with offspring outcomes observed in this study may not generalize to populations with greater metabolic or obstetric risk. Future work is needed to examine whether these placental signaling profiles are preserved, amplified, or disrupted in pregnancies complicated by conditions such as obesity, diabetes, or hypertension. We acknowledge the derivation of latent variables involve trade-offs and may oversimplify complex biological relationship or obscure single protein-outcome associations. Thus, the use of latent variables versus traditional single protein models should be informed by the study objectives. In this study, LASSO was used as a pattern-selection step to identify placental signaling patterns most strongly related to childhood adiposity, after which those selected patterns were examined in conventional regression models. Although this approach allowed us to estimate associations without interpreting shrunken penalized coefficients directly, the reported confidence intervals and p-values may not fully account for uncertainty introduced by selecting and evaluating patterns in the same dataset. Accordingly, these findings should be viewed as hypothesis-generating and in need of replication in independent cohorts. In addition, the relatively small sample size may have reduced the precision of effect estimates, as evidenced by some strong effect estimates that were not significant in our exploratory analysis of metabolic outcomes. For transparency, we present all models tested and have highlighted that findings should be considered hypothesis-generating, and none should be regarded as conclusive evidence [64]. The sample size for offspring outcomes, particularly metabolic biomarkers (n = 23–31), was modest, which limited precision of effect estimates. Given the number of candidate placental patterns and outcomes, evaluating all latent variables across all biomarkers or conducting outcome-specific penalized selection would substantially increase multiple comparisons and risk overfitting, potentially yielding imprecise and non-reproducible associations. Therefore, analyses of metabolic biomarkers were restricted to placental patterns selected for the primary adiposity outcome and were interpreted as exploratory and hypothesis-generating. Future studies with larger samples are warranted to evaluate outcome-specific placental signaling patterns across a broader range of metabolic endpoints Conclusions We identified two placental signaling patterns, an insulin-mTOR-energy sensing pattern and a stress-inflammatory MAPK placental signaling pattern. These patterns captured signaling across several pathways and were associated with early childhood adiposity and select metabolic markers at median age 4.8 years These findings highlight the value of characterizing patterns of placental signaling as determinants of early metabolic health trajectories. While these associations should be interpreted cautiously in light of the modest sample size and observational design, our findings lay a foundation for future mechanistic research that could inform etiological insight and early-life predictors of long-term cardiometabolic risk Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1: Additional File Fig. 1. Visualization of placenta protein assignments to clusters, Additional File Fig. 2. Summary of representative images of proteins from WES runs, Additional File Table 1. R (version 4.5.1) Packages Used in the Current Study, Additional File Table 2. LASSO Selection of Latent Placental Profiles Supplementary Material 1: Additional File Fig. 1. Visualization of placenta protein assignments to clusters, Additional File Fig. 2. Summary of representative images of proteins from WES runs, Additional File Table 1. R (version 4.5.1) Packages Used in the Current Study, Additional File Table 2. LASSO Selection of Latent Placental Profiles