Fetal Growth Disorders Detection During First Trimester Gestation Through Comprehensive Maternal Circulating DNA Profiling

Early diagnosis, close follow-up and timely delivery constitute the main 30 elements for appropriate detection and management of Fetal Growth Disorders (FGD). We 31 hypothesized that fetoplacental FGD-associated alterations can be detected in circulating DNA 32 (cirDNA) samples isolated from maternal blood, as early as the first gestational trimester. To 33 study whether markers in maternal cirDNA may facilitate FGD early detection, we profiled 34 plasma cirDNA from maternal samples prospectively collected during first gestational trimester. 35

Plasma cirDNA was isolated from samples prospectively collected during first 36 trimester gestation (n=56). Small, Large and Appropriate for Gestational Age (SGA n=11, LGA 37 n=18, and AGA n=29, respectively) status was determined at birth according to weight and 38 gestational age. cirDNA amount, fragmentation, mitochondrial/nuclear ratio and cirDNA 39 methylation profiles were quantified using qPCR-based assays. Machine learning approaches 40 were applied to build a molecular signature for prediction of LGA and SGA. Prediction accuracy 41 was assessed by Receiving-Operating Curve (ROC) analysis, and Positive and Negative 42 Predictive values (PPV and NPV , respectively) were calculated. 43

Total concentration of plasma cirDNA, cirDNA fragmentation and ratio of 44 mitochondrial/nuclear cirDNA were increased in SGA and LGA compared to AGA pregnancies. 45 DNA methylation profiles also shown distinctive patterns. Out of the 10 selected loci, we 46 detected 5 genes ( HSD2, RASSF1, CYP19A1, IL10, and LEP) showing significant differential 47 methylation differences (p<0.05) across the SGA, AGA and LGA samples at first trimester. We 48 combined these molecular and epigenetic cirDNA markers in a signature that reliably 49 discriminates between FGD and AGA pregnancies with high accuracy (AUC>0.95), achieving 50 88.8% PPV and 85.7% NPV . 51 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 4

Our findings show that maternal blood cirDNA profiles accurately detects 52 early gestation FGD. The proposed novel marker panel hold great potential for implementation 53 of low invasive approaches for reliable prediction of FGDs, enabling a disruptive path toward 54 precision medicine in FGD. 55 56 TRIAL REGISTRATION: 57 Not applicable 58 59 FUNDING 60 Internal startup funds from the University of Missouri to RC. The funding source was not 61 involved in the study design; in the collection, analysis and interpretation of data; in the writing 62 of the report; and in the decision to submit the article for publication. 63 64 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 5 MAIN TEXT 65

66 According to the United Nations, nearly 400,000 babies are born daily worldwide. Of those, it is 67 estimated that 8.6% are born small for gestational age (SGA, 90th percentile)(1). Many potential adverse health outcomes that could affect the mother 70 and fetus are associated with fetal growth disorders (FGD) during pregnancy (2, 3). 71 72 Notwithstanding, we are unaware of any intervention that is currently available and can improve 73 outcomes in SGA and LGA pregnancies targeting the affected molecular and physiological 74 pathways. Hence, early diagnosis, close monitoring and timely delivery constitute the main 75 elements for the appropriate detection and management of pregnancies with FGD(4). Nowadays, 76 screening of fetal growth abnormalities is based on fetal biometric measurements using 77 ultrasound in which serial sonographic assessments of fetal size over time are used to estimate 78 fetal growth and identify any deviation from normative trajectories. However, besides the 79 unprecedented technological development in the field, errors and approximations still hinder 80 FGD detection and assessment (5). Hence, there is an unmet need for reliable early pre-natal 81 biomarkers of fetal growth that can be assessed using low-invasive approaches. 82 83 FGDs are associated with fetal (6), placental (7), and maternal anomalies (8). However, whether 84 those anomalies have different weight in FGD etiology throughout the extent of pregnancy 85 remains to be determined. Hence, it is reasonable that an analyte that will reflect such 86 complexity, such as circulating DNA (cirDNA) in maternal blood may hold the potential for 87 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 6 sensitive and specific assessment of FGDs at different gestational stages, paving the way for the 88 development of molecular diagnostic tools and providing insights on the temporal differences in 89 fetoplacental and maternal pathophysiology in FGD. 90 91 Since the initial report by Dennis Lo and cols (9) , numerous studies have been conducted to 92 determine the origin and dynamics of the fetal cirDNA in maternal plasma (reviewed in (10)). 93 cirDNA in plasma derives from the fetus, the placenta and the mother (11), and the relative 94 contribution of each source to the total amount of DNA recovered from maternal plasma or 95 serum is highly dynamic and followed by rapid clearance after birth (12). Thus, the assessment 96 of cirDNA in maternal blood can provide valuable evidence for fetoplacental health and disease 97 throughout pregnancy (10) . 98 99 In this study, we hypothesized that fetoplacental FGD-associated alterations can be detected in 100 cirDNA samples isolated from maternal blood. As a corollary, we propose that cirDNA markers 101 can predict the occurrence of SGA and LGA pregnancies, as early as in first trimester of 102 gestation, when assessed either individually, or in combination in a marker panel generated via 103 machine learning approaches. 104 105

106 Molecular characterization of plasma cirDNA in first trimester SGA, LGA and AGA pregnancies 107 108 The total concentration of plasma cirDNA in SGA (mean cirDNA=9.40 ± 5.49 ng/mL) and LGA 109 (mean cirDNA=12.36 ± 4.24 ng/mL) was not significantly higher than in AGA pregnancies 110 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 7 (mean cirDNA=8.48 ± 3.16 ng/mL) (p=0.281; Kruskal-Wallis test) (Figure 2A). The fraction of 111 fragmented DNA was significantly increased in LGA (mean DFI=2.78 ± 1.46) and in SGA 112 (mean DFI=2.37 ± 0.45) compared to AGA pregnancies (mean DFI=1.56 ± 0.71) (p=0.001; 113 Kruskal-Wallis test; p LGA-AGA=0.004, and p SGA-AGA=0.019, Dunn’s test) (Figure 2B). 114 Noteworthy, we also detected significant differences (p=0.005, Kruskal-Wallis test) in the ratio 115 of mitochondrial/nuclear DNA among SGA, LGA and AGA pregnancies (Figure 2C). Compared 116 with AGA pregnancies, p ost hoc analysis revealed a significant increase in LGA (mean 117 MNR=1.20 ± 1.18 and 0.23 ± 0.22, for LGA and AGA pregnancies, respectively, p=0.004, 118 Dunn’s test), and a non-significant increase in SGA (mean MNR=0.78 ± 1.02 and 0.23 ± 0.22, 119 for SGA and AGA pregnancies, respectively, p=0.274, Dunn’s test). 120 121 Next, we investigated cirDNA methylation of the 10 selected candidate genes (13–22) (Figure 122 2D). Significant differences in 5 out of the 10 genes were present (p HSD2=0.003, pRASSF1=0.018, 123 pCYP19A1=0.008, p IL10=0.017 and, p LEP=0.002; Kruskal-Wallis test). Post hoc analysis revealed 124 significant differences between SGA and AGA in 4 genes (p HSD2=0.034, p RASSF1=0.0182, 125 pCYP19A1=0.013, and pLEP=0.004, Dunn’s test), between LGA and AGA in 2 genes (p HSD2=0.005, 126 and pLEP=0.028, Dunn’s test), and between SGA and LGA in the IL10 gene (p=0.013, Dunn’s 127 test). 128 129 Multivariate analysis of plasma cirDNA profiles in first trimester SGA, LGA and AGA 130 pregnancies 131 132 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 8 Principal Component Analysis (PCA; Figure 3) showed that samples corresponding to SGA, 133 LGA and AGA pregnancies clustered separately based upon the variables assessed in cirDNA 134 isolated from first trimester maternal blood (Figure 3A). As shown in Figure 3B, decrease in 135 cirDNA methylation in IL10, SLC36A1, PTPRN2 and LEP underlie sample distribution shifts 136 towards the SGA samples. Conversely, decreased cirDNA methylation in the HSD2 gene and 137 increased DNA fragmentation dragged sample distribution towards LGA samples. Correlation 138 with clinical variables also demonstrated a positive correlation between the levels of 139 mitochondrial DNA in plasma and the presence of maternal hypertension. Remarkably, cirDNA 140 methylation in SLC16A10, IRS1, PTPRN2 and SLC36A1 was positively correlated with maternal 141 age, but not with other clinical variables (Figure 3C). 142 143 Building and testing a molecular signature for non-invasive early diagnosis of FGD. 144 145 Whether the variables assessed in cirDNA samples, can be combined as a marker panel for early 146 detection of FGD pregnancies (Figure 4) was explored using generalized logistic models (GLM) 147 (Supplementary Table S2). Such models aimed at distinguishing between AGA and LGA/SGA 148 pregnancies. The performance of each model was evaluated using three statistical methods (i.e., 149 empirical, binomial, and non-parametric) and a cross-validation strategy. The first model 150 (Figure 4A) has a high power (86.9% accuracy, 80.0% sensitivity and 92.3% specificity) to 151 discriminate FGDs in general, including both SGA and LGA, from AGA pregnancies 152 (AUCs=0.96, 0.98, and 0.96, for empirical, binomial, and non-parametric ROCs, respectively). 153 The performance of the model was further evaluated according to positive predictive and 154 negative predictive values for FGDs (PPV and NPV, respectively), achieving 88.8% PPV and 155 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 9 85.7% NPV. A second model (Figure 4B) has high power (87.5% accuracy, 75.0% sensitivity 156 and 91.7 % specificity) to differentiate specifically SGA pregnancies (AUCs=0.95, 0.96, and 157 0.93 for empirical, binomial, and non-parametric ROCs, respectively; PPV=75.0% and 91.7 % 158 NPV). The third model (Figure 4C) has high power (93.3% accuracy, 66.7% sensitivity, and 159 100.0 % specificity) to differentiate specifically LGA pregnancies (AUCs=0.97, 0.98, and 0.96 160 for empirical, binomial, and non-parametric ROCs, respectively; PPV=100.0% and 92.3 % 161 NPV). Lastly, we generated a fourth model (Figure 4D) with high power (81.2% accuracy, 75.0 162 % sensitivity, and 83.3 % specificity) to differentiate between SGA and LGA pregnancies 163 (AUCs=0.88, 0.92, and 0.87 for empirical, binomial, and non-parametric ROCs, respectively; 164 PPV=60.0% and 90.9 % NPV). 165 166

167 Using a comprehensive cirDNA profiling of maternal blood samples, we report here a novel 168 marker panel enabling accurate prediction of FGDs that can be implemented as early as the first 169 gestational trimester. Our marker panel combines the total amount of cirDNA, with physical 170 characterization of the cirDNA (i.e., fragmentation analysis and ratio of mitochondrial DNA) 171 which are representative of the increase shedding of cirDNA into circulation upon cellular 172 turnover. Furthermore, the proposed panel includes DNA methylation assessment of 10 genes 173 that were selected a priori due to their direct roles in fetal growth and placental homeostasis. 174 Analyses of circulating nucleic acids (i.e., DNA and RNA) in maternal blood are widely used as 175 a non-invasive prenatal diagnostic test (NIPT). Numerous tests and applications have been 176 developed during the last two decades demonstrating the utility and precision of such tests for 177 determining fetal health (e.g., chromosomal alterations), as well as for predicting obstetric 178 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 10 complications, such as pre-eclampsia and premature birth. Nevertheless, some controversy 179 remains regarding the clinical utility of cirDNA testing during pregnancy based on the limited 180 fraction of fetal DNA in the total circulating DNA of maternal blood, especially in early gestation 181 (23). Hence, despite the tremendous advances in the field, the use of liquid biopsies for 182 monitoring fetal growth and FGD detection is currently unavailable. A recent study including 183 over 16,000 individuals(24), demonstrated that the amount of cirDNA in maternal blood shows a 184 steady increase during pregnancy with inflection points at the 10th, 19th and 30th week of 185 gestation. Importantly, this study demonstrated that the concentration of cirDNA in maternal 186 blood reached the standard NIPT requirements already at the 9 th week of gestation(24), 187 encouraging the interrogation of early fetal markers using high sensitivity approaches. 188 Large-scale DNA methylation profiles have been proven as a successful source for candidate 189 biomarkers, but their application as molecular diagnostics is hampered by costs and other 190 intrinsic limitations (e.g., the generation of information that is beyond the clinical application). 191 Hence, clinical assay design requires the reduction of the number of measured loci to a 192 manageable level. In this study, we reported a 15-marker signature that can differentiate with 193 high precision (> 90.0% accuracy) between FGD and normal pregnancies in first trimester 194 maternal blood samples. Noteworthy, we did not detect significant differences in the maternal 195 weight, height and BMI at baseline among the SGA, AGA and LGA groups, nor the markers 196 quantification correlated with maternal phenotype. These findings suggest that the molecular 197 signature is not biased towards the identification of constitutionally large or small babies, but 198 rather identify clinically relevant FGDs. As the number of patients per group in this study does 199 not provide enough statistical power to further stratify the SGA and LGA groups according to 200 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 11 maternal baseline assessment, further investigation of the performance of the molecular signature 201 in constitutionally large or small babies is warranted. 202 The performance of the markers discovered in the present study and their use in a clinical setting 203 will be formally assessed in future studies (i.e., assay development, verification, and validation) 204 for single or multi-marker panels using larger datasets. The clinical utility of such marker panels 205 will be further enhanced by combining them with clinical and demographic information (e.g., 206 BMI, maternal and gestational age, diabetes, hypertension, etc.) using multivariate models 207 towards the operationalization of personalized diagnostic and monitoring algorithms in FGD. 208 The application of such algorithms in a clinical setting holds the potential to enable a disruptive 209 path toward precision medicine in FGD. 210 The prese nce of feta l DNA in matern al plasma has be en r eported a lmost 3 0 years ago by Denni s 211 Lo and cols (9). Since then, numerous studie s have been conducted to det ermine the origin and 212 dynamics of the fetal cirDNA in m aternal plasma, r evi ewed in (10). cirDNA is shed into 213 maternal blood via apoptosis , necros is, and oxidative str ess( 11). The se ph ysical characteristics 214 are faithfully represe nted in the ma rker panel by inclusion of the c irDN A fragmentation and 215 mitochondrial/nuclear cirDNA ratio. Experimental e videnc e indicates th at cirDNA in plasma 216 derive s from the fetu s, the p lacenta and the mother, and there i s a dy namic on the relative 217 contribution of eac h source to the t otal amount of DNA recovered from maternal plasma or 218 serum, with a r apid clearance after bi rth (12) . T hus, the assessmen t of c irD NA in maternal blood 219 can therefore provide valuabl e e vi dence for fetoplac ental health and diseas e throughout 220 pregnancy (10). 221 Since epigenetic alterations represent a common trait in most complex diseases, they hold great 222 potential as robust and precise biomarkers in tissues and bodily fluids(25). In particular, the 223 analysis of cirDNA in maternal blood is of outstanding clinical relevance and it is currently a 224 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 12 standard practice in NIPT. Several features of DNA methylation support their potential as early 225 diagnostic biomarkers, including the large number of DNA methylation changes in affected cells, 226 their early occurrence in disease development, and their stability enabling detection in small 227 amounts of DNA. DNA methylation is de facto being already used in clinical practice(25). DNA 228 methylation biomarkers are instrumental in oncology and several commercial tests based on 229 DNA methylation markers are already implemented in the clinics. 230 231 This study provided empirical evidence demonstrating that cirDNA isolated from maternal blood 232 samples at the first gestational trimester has a diagnostic value for the occurrence of FGD. By 233 combining the cirDNA quantification with physical and epigenetic markers, the proposed 234 molecular signature provides a comprehensive profiling of cirDNA in maternal blood. The 235 relatively low number of markers in the panel (i.e., 15 markers) would enable the development of 236 molecular diagnostic assays for early FGD diagnostics that can be implemented in the clinical 237 practice. 238 Notwithstanding the high potential for the findings, the molecular signature should be further 239 validated using larger independent larger datasets. We anticip ate that this signature wil l be the 240 foundation for the development FGD molecular tests fol lowing the Desi gn Control Guide line s 241 for assay dev elopment , analytical v erification, and clinical validation(26) in future studies . 242 Moreover, the development of diagnostic algorithms combining molecular and clinical data will 243 require a larger number of subjects per group than those in the present study. 244 In conclusion, our findings show that maternal blood cirDNA profiles accurately detects early 245 gestation FGD. The proposed novel marker panel hold great potential for implementation of low 246 invasive approaches for reliable prediction of FGDs as early as the first trimester of pregnancy. 247 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 13 248

249 Subject recruitment. Sex as a biological variable was not considered in this study. Pregnant 250 participants were recruited for this study. Figure 1 summarizes the study design. Pregnant 251 subjects were prospectively recruited at their initial Obstetrics visit, blood samples were 252 collected by a certified phlebotomist during their first 9-13 week of gestation and plasma 253 fractions were separated as detailed below. Maternal demographic and clinical information 254 (Table 1), as well as ultrasound data from pr enatal controls was prospectively retrieved from 255 existing electronic medical records. After birth and based on the outcome of the pregnancy, 256 subjects were separated into three groups SGA (n=11), LGA (n=18) and AGA (n=29). SGA and 257 LGA pregnancies were defined taking in account gestational age and weight at delivery, 258 following the revised reference chart for the US(27). Pregnancies with babies above 90% and 259 below 10% percentiles were regarded as LGA and SGA, respectively. No significant differences 260 (p>0.05) were detected in maternal age, ethnicity, height, weight and BMI at NOB among the 261 subjects on the SGA, AGA and LGA groups (Table 1). 262 Plasma cirDNA isolation from whole blood maternal samples . Plasma and cellular fraction 263 from whole blood samples was separated by centrifugation at 1,800 rpm for 20 minutes. cirDNA 264 from the plasma fraction was isolated using the Circulating Nucleic Acids kit (Qiagen, Valencia, 265 CA), according to manufacturer’s instructions. Isolated cirDNA was stored at -80°C until use. 266 Plasma cirDNA characterization. cirDNA amounts and fragmentation for each sample was 267 quantitively assessed using SYBR-green based qPCR assays, as previously reported (28). In 268 brief, total cirDNA amount was assessed by single-locus (i.e., KRAS) amplification and 269 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 14 extrapolation to calibration curves. cirDNA fragmentation was assessed using a combination of 270 three qPCR assays targeted to the same locus (i.e., KRAS). These assays share the same reverse 271 primer, but the length of the assay differs due to the forward primer expanding to i) 357 bp 272 representing long DNA fragments, ii) 145 bp representing nucleosome-sized DNA fragments, 273 and iii) 60 bp representing highly fragmented DNA. The relative enrichment of these three 274 fractions was calculated as the ratio between the amplification (i.e., DDCt) and the DNA 275 Fragmentation Index (DFI) was calculated for each sample. Mitochondrial DNA release into 276 maternal blood was quantified by the Mitochondrial-Nuclear DNA ratio (MNR) according to the 277

of Otandault, et. al(29). cirDNA methylation was measured using Methylation-Sensitive 278 Restriction Enzyme qPCR (MSRE-qPCR), as previously reported (30), and encompassed 10 279 candidate genes selected according to previous reports on their involvement in placental 280 homeostasis and fetal development (HSD2, RASSF1, CYP19A1, IGF2R, SLC16A10, IRS1, IL10, 281 PTPRN2, SLC36A1, LEP) (13–22). Primers for all qPCR assays are provided in Supplementary 282 Table S1. 283 Building of molecular signature. All data analysis was conducted in R (version 4.2.2). Principal 284 component analysis and variable plots were conducted using the factoextra package version 285 1.0.7. The molecular signature was extracted based on cirDNA markers and clinical attributes, 286 and combined in generalized logistic models (GLM), according to the following comparisons: i) 287 AGA vs. LGA/SGA pregnancies, ii) AGA vs. SGA pregnancies, iii) AGA vs. LGA pregnancies, 288 and iv) SGA vs. LGA pregnancies (Supplementary Table S2). GLMs were build using the 289 riskRegression (version 1.3.7) package. Empirical, binormal, and non-parametric statistical 290 methods, each using randomForest (version 4.7.1), were the parameters leveraged in the 291 construction of three ROC curves using R-package ROCit (version 2.1.1). 292 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 15 Statistics: Univariate data analysis and plotting were conducted using GraphPad Prism 10 293 (version 10.1.2; GraphPad Software, Boston, MA). Differences in continuous clinical variables 294 and cirDNA quantity, fragmentation, mitochondrial DNA content and methylation between 295 AGA, SGA and LGA were assessed using Kruskal-Wallis test, followed by multiple test 296 comparisons using Dunn’s test implemented in the R software (version 4.4.2). Differences in 297 categorical clinical variables were assessed using the χ2 test implemented in the R software 298 (version 4.4.2). Principal component analysis was conducted using the R-package factoextra 299 (version 1.0.7). Correlations were calculated using Pearson or Biserial methods implemented in 300 the R software (version 4.4.2). Molecular signatures were built and tested as described above. 301 Study approval: This project has been evaluated and approved by the Institutional Review Board 302 (IRB) of the University of Missouri (protocol number: 2016992_MU) and it was conducted 303 following the ethical standards for human experimentation established in the Declaration of 304 Helsinki. A written informed consent was received from each participant prior to participation. 305 Data sharing: To ensure independent interpretation of clinical study results and enable authors 306 to fulfill their role and obligations under the International Committee of Medical Journal Editors 307 (ICMJE) criteria, the authors grant all external authors access to data pertinent to the 308 development of the publication. Scientific and medical researchers can access to molecular and 309 anonymized clinical data and supporting analytic code utilized in this publication by contacting 310 the corresponding author. 311 312 AUTHOR’S CONTRIBUTION 313 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 16 R.C. participated in the conceptual framework of the project, performed experiments, analyzed 314 data, and wrote the manuscript. K.C., G.S., M.O. and M.R. performed experiments. J.H. 315 conducted data analysis. H.W., D.G. and J.R.G provided medical expert guidance, participated in 316 data analysis and interpretation of results, and served as blinded observers. All authors have 317 reviewed and approved the final version of the manuscript. 318 319 ACKNOWLEDGMENTS 320 Authors would like to acknowledge the Research Success Center (RSC) of the Department of 321 Obstetrics, Gynecology & Women’s Health at the University of Missouri for patient recruitment 322 and sample collection. 323 Parts of this study were presented at the 90 th Annual Meeting of the Central Association of 324 Obstetricians and Gynecologists. Nashville TN. USA. October 26-28, 2023. 325 326 327 328 329 330 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 17

331 1. Hediger ML, et al. Growth of infants and young children born small or large for gestational age: 332 findings from the Third National Health and Nutrition Examination Survey. Arch Pediatr Adolesc Med. 333 1998;152(12). https://doi.org/10.1001/ARCHPEDI.152.12.1225. 334 2. Salam RA, Das JK, Bhutta ZA. Impact of intrauterine growth restriction on long-term health. Curr 335 Opin Clin Nutr Metab Care. 2014;17(3):249–254. 336 3. Scifres CM. Short- and Long-Term Outcomes Associated with Large for Gestational Age Birth Weight. 337 Obstet Gynecol Clin North Am. 2021;48(2):325–337. 338 4. Lees CC, et al. ISUOG Practice Guidelines: diagnosis and management of small-for-gestational-age 339 fetus and fetal growth restriction. Ultrasound Obstet Gynecol. 2020;56(2):298–312. 340 5. Salomon LJ, et al. ISUOG Practice Guidelines: ultrasound assessment of fetal biometry and growth. 341 Ultrasound Obstet Gynecol. 2019;53(6):715–723. 342 6. Nowakowska BA, et al. Genetic Background of Fetal Growth Restriction. Int J Mol Sci. 2021;23(1). 343 https://doi.org/10.3390/IJMS23010036. 344 7. Zur RL, et al. The Placental Basis of Fetal Growth Restriction. Obstet Gynecol Clin North Am. 345 2020;47(1):81–98. 346 8. Monaco-Brown M, Lawrence DA. Obesity and Maternal-Placental-Fetal Immunology and Health. 347 Front Pediatr. 2022;10. https://doi.org/10.3389/FPED.2022.859885. 348 9. Dennis Lo YM, et al. Presence of fetal DNA in maternal plasma and serum. Lancet. 349 1997;350(9076):485–487. 350 10. Taglauer ES, Wilkins-Haug L, Bianchi DW. Review: Cell-free fetal DNA in the maternal circulation 351 as an indication of placental health and disease. Placenta. 2014;35(SUPPL):S64-8. 352 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 18 11. Tjoa ML, et al. Trophoblastic Oxidative Stress and the Release of Cell-Free Feto-Placental DNA. Am 353 J Pathol. 2006;169(2):400–404. 354 12. Chiu RWK, Lo YMD. Cell-free fetal DNA coming in all sizes and shapes. Prenat Diagn. 355 2021;41(10):1193–1201. 356 13. Cao J, Chen Y , Wang H. 11β -hydroxysteroid dehydrogenases and biomarkers in fetal development. 357 Toxicology. 2022;479. https://doi.org/10.1016/J.TOX.2022.153316. 358 14. Zhang L, et al. Dynamic Changes of Fetal-Derived Hypermethylated RASSF1A and Septin 9 359 Sequences in Maternal Plasma. Reprod Sci. 2021;28(4):1194–1199. 360 15. Stejskalova L, Pavek P. The function of cytochrome P450 1A1 enzyme (CYP1A1) and aryl 361 hydrocarbon receptor (AhR) in the placenta. Curr Pharm Biotechnol. 2011;12(5):715–730. 362 16. Cassidy FC, Charalambous M. Genomic imprinting, growth and maternal-fetal interactions. J Exp 363 Biol. 2018;221(Pt Suppl 1). https://doi.org/10.1242/JEB.164517. 364 17. Uusküla L, et al. Mid-gestational gene expression profile in placenta and link to pregnancy 365 complications. PLoS One. 2012;7(11). https://doi.org/10.1371/JOURNAL.PONE.0049248. 366 18. Ustianowski P, et al. COBLL1 and IRS1 Gene Polymorphisms and Placental Expression in Women 367 with Gestational Diabetes. Biomedicines. 2022;10(8). https://doi.org/10.3390/BIOMEDICINES10081933. 368 19. Ding J, et al. Trophoblast-derived IL-6 serves as an important factor for normal pregnancy by 369 activating Stat3-mediated M2 macrophages polarization. Int Immunopharmacol. 2021;90. 370 https://doi.org/10.1016/J.INTIMP.2020.106788. 371 20. Awamleh Z, et al. Exposure to Gestational Diabetes Mellitus (GDM) alters DNA methylation in 372 placenta and fetal cord blood. Diabetes Res Clin Pract. 2021;174. 373 https://doi.org/10.1016/J.DIABRES.2021.108690. 374 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 19 21. Chen PY , et al. Prenatal Growth Patterns and Birthweight Are Associated With Differential DNA 375 Methylation and Gene Expression of Cardiometabolic Risk Genes in Human Placentas: A Discovery-376 Based Approach. Reprod Sci. 2018;25(4):523–539. 377 22. Mansell T, et al. Genetic variation, intrauterine growth, and adverse pregnancy conditions predict 378 leptin gene DNA methylation in blood at birth and 12 months of age. Int J Obes (Lond). 2020;44(1):45–379 56. 380 23. Carbone IF, et al. Circulating Nucleic Acids in Maternal Plasma and Serum in Pregnancy 381 Complications: Are They Really Useful in Clinical Practice? A Systematic Review. Mol Diagn Ther. 382 2020;24(4):409–431. 383 24. Zhou Y , et al. Analysis of cell-free fetal DNA in 16,843 pregnant women from a single center in China 384 using targeted sequencing approach. Placenta. 2022;122:18–22. 385 25. Raj S, Kumar D. Early epigenetic markers for precision medicine. Prog Mol Biol Transl Sci. 386 2023;198:153–164. 387 26. Design Control Guidance For Medical Device Manufacturers | FDA [Internet]. 388 https://www.fda.gov/regulatory-information/search-fda-guidance-documents/design-control-guidance-389 medical-device-manufacturers. Accessed May 22, 2021. 390 27. Duryea EL, et al. A revised birth weight reference for the United States. Obstetrics and gynecology. 391 2014;124(1):16–22. 392 28. Cortese R, et al. Tumor circulating DNA profiling in xenografted mice exposed to intermittent 393 hypoxia. Oncotarget. 2015;6(1):556–569. 394 29. Otandault A, et al. Hypoxia differently modulates the release of mitochondrial and nuclear DNA. Br J 395 Cancer. 2020;122(5):715–725. 396 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 20 30. Cortese R, et al. Epigenetic markers of prostate cancer in plasma circulating DNA. Hum Mol Genet. 397 2012;21(16):3619–3631. 398 399 400 401 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 21 FIGURE LEGENDS 402 Figure 1: Study design. Participants were prospectively recruited at their initial Obstetrics visit 403 and first gestational trimester (gestational week 9-13) maternal blood samples were collected. 404 Plasma cirDNA was isolated and characterized by assessing quantity, fragmentation, 405 mitochondrial DNA content and DNA methylation status in 10 candidate genes using qPCR-406 based methods. Participants were grouped based on the outcome of the pregnancy taking in 407 account gestational age and weight at delivery, following the revised reference chart for the 408 US(27), into three groups SGA (90th percentile, n=18) and AGA 409 (10th-90th percentile, n=29). Plasma cirDNA markers were evaluated individually and combined 410 in molecular signatures differentiating among AGA, SGA, and LGA pregnancies. 411 Figure 2: cirDNA profiling in SGA, LGA and AGA pregnancies. A) Total plasma cirDNA 412 concentration was increased in SGA and LGA compared with AGA pregnancies (mean 413 cirDNA=9.40 ± 5.49 ng/mL, 12.36 ± 4.24 ng/mL, and 8.48 ± 3.16 ng/mL, for SGA, LGA and 414 AGA respectively), although the differences were not statistically significant (p=0.281). B) 415 Fraction of fragmented (apoptotic) cirDNA was significantly increased (p=0.001) in SGA and 416 LGA compared with AGA pregnancies (mean DFI= DFI=2.37 ± 0.45, 2.78 ± 1.46, and 1.56 ± 417 0.71, for SGA, LGA and AGA respectively). C) Ratio of mitochondrial/nuclear DNA was 418 significantly different (p=0.005) in SGA and LGA compared with AGA pregnancies (mean 419 MNR= 0.78 ± 1.02, 1.20 ± 1.18 and 0.23 ± 0.22, for SGA, LGA and AGA pregnancies, 420 respectively). D) Differential cirDNA methylation in 10 candidate genes We observed significant 421 differences among SGA, LGA and AGA pregnancies in 5 out of the 10 genes (p HSD2=0.003, 422 pRASSF1=0.018, p CYP19A1=0.008, p IL10=0.017 and, p LEP=0.002; Kruskal-Wallis test). SGA, AGA 423 and LGA groups are represented by red, blue and green bars, respectively. p-values correspond to 424 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 22 Kruskal-Wallis test. Significance of ad hoc comparisons was calculated using Dunn’s test and are 425 expressed as **: p<0.05, *: p<0.01, n.s.: non-significant. 426 Figure 3: Multivariate analysis in SGA, LGA and AGA pregnancies. A) Samples 427 corresponding to SGA (red), LGA (green) and AGA (blue) pregnancies clustered separately in 428 PCA analysis. SGA, AGA and LGA groups are represented by red, blue and green dots and 429 elipses, respectively. B) Graphic of variables depicting the contribution of each variable in 430 sample distribution. Positively and negatively correlated variables point to the same or opposite 431 side of the plot, respectively. The contributions to the sample discrimination are color coded in a 432 gradient from red (higher contribution) over yellow to light blue (lower contribution). C) 433 Correlation matrix of cirDNA profiles with clinical variables. Correlation coefficients (r) are 434 represented in a color scale from 0.5 (red) to -0.5 (blue). Correlations were calculated using 435 Pearson (1) or Biserial (2) tests. *: p<0.05, #: p<0.1. 436 Figure 4: Receiving-Operating Curve (ROC) analysis assessing the performance of the 437 molecular signature for the discrimination between SGA, LGA and AGA pregnancies. The 438 performance of the signature was assessed by calculating the Area Under the Curve (AUC) when 439 applying three different models: Empirical, Binormal and Non-Parametric, represented as black, 440 red, and blue lines, respectively. A) ROC analysis assessing the performance of the molecular 441 signature for the discrimination between AGA and SGA/LGA pregnancies. AUC=0.96, 0.98, and 442 0.95 for the empirical, binormal, and non-parametric modelling, respectively. B) ROC analysis 443 assessing the performance of the molecular signature for the discrimination between AGA and 444 SGA pregnancies. AUC=0.95, 0.96, and 0.92 for the empirical, binormal, and non-parametric 445 modelling, respectively. C) ROC analysis assessing the performance of the molecular signature 446 for the discrimination between AGA and LGA pregnancies. AUC=0.97, 0.98, and 0.96 for the 447 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 23 empirical, binormal, and non-parametric modelling, respectively. D) ROC analysis assessing the 448 performance of the molecular signature for the discrimination between SGA and LGA 449 pregnancies. AUC=0.88, 0.92, and 0.87 for the empirical, binormal, and non-parametric 450 modelling, respectively. 451 452 SUPPLEMENTARY MATERIAL 453 Supplementary Table S1: qPCR primers for cirDNA characterization. 454 Supplementary Table S2: Generalized logistic models for AGA, SGA and LGA molecular 455 signatures. 456 457 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint 24 TABLES 458 Table 1: Subjects demographics 459 Group SGA AGA LGA p-value n 11 29 18 Maternal age [years ± SD] 28.12 ± 7.71 29.6 ± 4.51 28.46 ± 4.39 0.709 a Maternal ethnicity [White (%)] 9 (81.8%) 28 (96.5%) 18 (100%) 0.084 b Maternal height [meters ± SD] 1.59 ± 6.23 1.64 ± 8.44 1.63 ± 5.41 0.108 a Maternal weight [kilograms ± SD] 75.78 ± 12.22 78.42 ± 19.91 85.35 ± 26.24 0.716 a Maternal BMI at NOB [kg/m2 ± SD] 30.0 ± 3.86 29.0 ± 7.14 31.2 ± 9.83 0.547a Gestational Age at Birth [weeks ± SD] 37.9 ± 1.64 38.8 ± 1.12 39.2 ± 1.11 0.078 a Birth weight [g ± SD] 2433.32 ± 506.63 3219.91 ± 340.46 4030.54 ± 221.23 <0.001a a: Kruskal-Wallis test b: χ2 test, white vs. non-white 460 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for thisthis version posted December 24, 2024. ; https://doi.org/10.1101/2024.12.21.24319487doi: medRxiv preprint