Metabolomic signatures of SSRI exposure during neural differentiation and correlation of lysophosphatidylcholines with early symptoms of neurodevelopmental disorders Authors: Abishek Arora, Kristina Vacy, Cátia Marques, Mihai-Ovidiu Degeratu, Francesca Mastropasqua, Jenny Humphrey, Xuan Ye, Marika Oksanen, Peter Vuillermin, Anne-Louise Ponsonby, Ingela Lanekoff, Kristiina Tammimies, the Barwon Infant Study Investigator Group Published: 2026 ABSTRACT SummaryBackgroundSelective serotonin reuptake inhibitors (SSRIs) are often prescribed during pregnancy. Epidemiological studies examining in-utero SSRI exposure and neurodevelopmental outcomes such as autism and ADHD have produced mixed results, in part due to challenges in accounting for underlying maternal mental health conditions and other confounding factors. The molecular pathways through which SSRIs may affect early neurodevelopment remain poorly understood.MethodsWe exposed neuroepithelial stem cells derived from four human induced pluripotent stem cell (iPSC) lines to fluoxetine, citalopram, sertraline, and paroxetine. We then assessed cellular viability, reactive oxygen species (ROS) levels, mitochondrial function using adenosine triphosphate (ATP) assays, and performed high-throughput metabolomics at two timepoints: proliferation and neural differentiation stages. The key metabolic findings were validated in the in-vitro model and in a complementary population-based cohort, the Barwon Infant Study, consisting of 1074 mother-child pairs with analysed cord-blood metabolomes.FindingsSertraline and paroxetine significantly decreased ROS and ATP levels in-vitro, indicating mitochondrial alteration. Metabolomic profiling revealed consistent elevation of three lysophosphatidylcholines (LPCs 16:0, 18:0, 18:1) in cells exposed to all SSRIs except citalopram. We further observed elevated LPC levels in the cord blood of infants prenatally exposed to SSRIs compared to those who were not (LPC 16:0 sn1: β = 0.173, 95% CI = 0.011–0.334, p = 0.037; LPC 16:0 sn2: β = 0.175, 95% CI = 0.007–0.343, p = 0.042 and LPC 18:0 sn2 β = 0.174, 95% CI = 0.018–0.330, p = 0.029), with a dose-dependent correlation to autism (LPC 16:0 sn2: overall p = 0.01) and ADHD-related symptoms (LPC 16:0 sn1: overall p = 0.007; LPC 16:0 sn2: overall p = 0.002) at age two.InterpretationThese findings provide insights into SSRI-induced molecular changes in human iPSC derived neural cell cultures and highlight candidate metabolites that were validated in a clinical cohort. This may warrant their further investigation as indicators of SSRI exposure and emphasise the need for exploring prenatal SSRI exposure effects and neurodevelopmental outcomes in a wider context of other more well-established liability factors, including genetic background.FundingVetenskapsrådet, Swedish Foundation for Strategic Research, Hjärnfonden, H.K.H. Kronprinsessan Lovisas förening för barnasjukvård, StratNeuro, Swedish Foundation for International Cooperation in Research and Higher Education, Karolinska Institutet and the European Research Council.Keywords: Selective serotonin reuptake inhibitors, Induced pluripotent stem cells, Clinical cohort, Neurodevelopmental disorders, Metabolomics, Cellular assays Research in context Evidence before this study While anti-depressants, including selective serotonin reuptake inhibitors (SSRIs), are considered safe for use during pregnancy, there are some indications from epidemiological research that SSRIs may have a potential impact on the developing brain. Risk stratifying the possible effect of various SSRIs on early brain development is crucial to informing an evidence-based approach for the management of maternal depression. Empirical studies have aimed to evaluate such effects but their use of animal models and lack of comparison between different SSRIs in the same experimental setup makes it difficult to draw substantial conclusions regarding safety profiles and extrapolate their findings to humans. Nevertheless, current research indicates the lack of a clear causal relationship between in-utero SSRI exposure and a negative impact on human brain development. To contextualise this study, we performed a literature search in PubMed and MEDLINE up to October 2025 using terms such as “SSRI”, “prenatal exposure”, “pregnancy”, and “neurodevelopmental disorders”. This search confirmed that epidemiological findings remain inconsistent, and that many reported associations weaken after accounting for maternal mental health, familial factors, and genetic liability. Overall, robust causal evidence linking prenatal SSRI exposure to neurodevelopmental disorders is lacking, and mechanistic studies in human neural models are limited. Added value of this study Our study identified both unique and shared molecular signatures in response to exposure to four commonly prescribed SSRIs in human neural stem cell cultures, which were hitherto unknown. The key results from the pre-clinical experiments were further validated using population-based clinical data to identify patterns related to metabolic changes mediated by the SSRIs. We found a significant positive association between adverse neurodevelopmental outcomes at age two and the lysophospholipid LPC 16:0 (autism, LPC 16:0 sn2: overall p = 0.01; and ADHD, LPC 16:0 sn1 and sn2: overall p = 0.007 and 0.002, respectively) in the clinical cohort, which we first identified in the SSRI-exposed human neural stem cells. By linking SSRI-related metabolomic findings from differentiating human neural cells to cord-blood metabolites in a population-based cohort, this study identifies lysophosphatidylcholines as candidate metabolites that may be sensitive to prenatal SSRI exposure, without implying causality. Implications of all the available evidence Our findings provide additional insights into SSRI-mediated effects during early brain development, with a particular impact on phospholipid metabolism. When viewed alongside existing epidemiological evidence, which does not demonstrate a robust causal link between prenatal SSRI exposure and neurodevelopmental disorders, these results should be interpreted as metabolic patterns that merit further investigation, rather than an indication of adverse developmental effects. Larger mechanistic and genetically informed studies will be important for determining how these lipid changes relate to broader liability factors, including maternal mental health, genetics, and perinatal environment. The benefits of SSRI treatment during pregnancy remain well established, and our findings do not recommend the alteration of current clinical practice. Introduction Selective serotonin reuptake inhibitors (SSRIs) are widely prescribed antidepressants that block the reuptake of serotonin by binding to the serotonin transporter, resulting in elevated levels of serotonin in the synaptic cleft.1 Depression affects ∼28% of pregnant women2 and SSRI use during pregnancy has increased over recent decades.3 While the benefits of treating clinical depression with SSRIs clearly outweigh any potential risks, it is critical to understand their impact on foetal development. Although SSRIs are considered relatively safe for use during pregnancy, several epidemiological studies have reported associations between in-utero SSRI exposure and neurodevelopmental disorders (NDDs), such as autism and attention-deficit/hyperactivity disorder (ADHD). Some meta-analyses have reported modest increases in relative risk; however, these findings are difficult to interpret because many associations are substantially attenuated after adjustment for maternal mental health, shared familial factors, and genetic liability.4, 5, 6 Other systematic reviews have found no consistent evidence for adverse cognitive or developmental outcomes once these confounders are considered.7,8 Beyond NDDs, a recent review has identified associations between different SSRIs and several adverse foetal health outcomes,9 including an increased burden of cardiovascular complications in neonates.10,11 Sertraline, citalopram, paroxetine, and fluoxetine are the most commonly prescribed SSRIs to women of reproductive age.12,13 These SSRIs vary in pharmacological properties such as half-life (approximately 21 h to a few days), the relationship between dose/blood levels and the dose range of prescription.14 SSRIs can cross the placenta15 and subsequently alter foetal serum serotonin levels,16 with a plausible but uncertain impact on the developing brain.17,18 Neural development at the early stages of gestation, prior to the formation of the intact blood–brain barrier19 and widespread synaptic connectivity, relies on regulation by neurotransmitters.20 Serotonergic activity is one of the building blocks of neuronal organisation, with serotonin being one of the first neurotransmitters to emerge.21 Although many uncertainties remain around SSRIs and their mechanism of action, few studies have demonstrated wide-ranging cellular effects, including an alteration in reactive oxygen species (ROS) levels22 and an impact on mitochondrial oxidative phosphorylation.23 During neural development, ROS functions as a signalling molecule contributing to neuronal growth and synaptogenesis.24 However, excessive ROS-induced oxidative stress can lead to genetic mutations, lipid peroxidation, and mitochondrial dysfunction.25 Furthermore, in a previous study, we demonstrated that fluoxetine exposure alters lipid and energy metabolism, as observed in the transcriptome and metabolome of differentiating neuroepithelial stem (NES) cells.26 These findings, together with evidence that SSRIs accumulate in cellular membranes and can perturb phospholipid composition,27 support plausible indications that there are cellular metabolic effects of SSRIs. Despite these efforts, there are no systematic studies investigating the potential molecular mechanisms of SSRI exposure during early neural development and how these could potentially link to NDDs. In this study, we hypothesised that SSRI exposure could alter markers of oxidative stress and energy metabolism in developing neural cells, lead to consistent lipid signatures across multiple SSRIs, and that these signatures might associate with SSRI exposure and early neurodevelopmental traits in a clinical birth cohort. To test these hypotheses, we used human induced pluripotent stem cell (iPSC)- derived NES cells undergoing undirected differentiation into neurons and exposed them to four commonly prescribed SSRIs. We, first, evaluated the effects of SSRI exposure on cellular viability, ROS, and oxidative phosphorylation of adenosine triphospate (ATP). After characterising these in-vitro effects, we performed untargeted mass spectrometry analysis to develop metabolomic profiles of the general and specific effects of each SSRI. Finally, we analysed the top three upregulated lipids from the cellular models in cord blood from a population-based cohort, the Barwon Infant Study (BIS), consisting of 1074 mother-child dyads, and investigated the association between SSRI exposure, the cord blood lipid levels and correlation with autism and ADHD traits. This integrated approach aims to advance understanding of SSRI-mediated effects during pregnancy, including their potential impact on neural cell development, differentiation, and associated metabolic pathways, and to inform future epidemiological and clinical evaluations. Methods Cell culture For the purpose of this study, previously generated and described human iPSC lines from tissue samples of a male: CTRL-9-II – CTRLMale28 and female: AF22 – CTRLFemale29 neurotypical donors and two male donors with known genetic variants for NDDs including autism: HNRNPUdel/+ – ASDHNRNPU30 and ASDCASK_SS – ASDCASK,31 were used (Supplementary Table S1). Information pertaining to the generation and quality control of the iPSC lines, as well as their relevant clinical background are provided in the cited publications. These lines were selected to capture conserved SSRI exposure effects across sex and neurodevelopmental genetic backgrounds rather than to model population-level heterogeneity or genotype-specific responses. The iPS cells were differentiated into a neural progenitor state and maintained as NES cells, as previously described.29,32 While NES cells are neural progenitors with both neurogenic and gliogenic potential, upon differentiation they are primarily neurogenic unless they are specifically differentiated to glia.29,32 The iPSC lines and this NES-to-neuron differentiation system have been previously validated and characterised for pluripotency, neural progenitor identity and neuronal differentiation in multiple independent studies and were therefore not re-characterised in the present work.28, 29, 30, 31 The NES cells were seeded (40,000 cells/cm2) on tissue culture treated plates (Sarstedt), coated with 20 μg/mL poly-L-ornithine (P3655, Sigma–Aldrich) and 1 μg/mL Laminin2020 (L2020, Sigma–Aldrich). These were grown in DMEM/F-12 Glutamax basal medium (31331-093, Thermo Scientific) supplemented with 0.05× B27 (17504044, Thermo Scientific), 1× N-2 (17502001, Thermo Scientific), 10 U/ml Penicillin-Streptomycin (15140122, Thermo Scientific), 10 ng/mL recombinant human bFGF (13256-029, Thermo Scientific) and 10 ng/mL recombinant human EGF (AF-100-15, Peprotech). The undirected differentiation of NES cells to neural cell population was achieved by the withdrawal of growth factors from the culture medium.29,32 Here, the NES cells were again seeded (40,000 cells/cm2) on tissue culture treated plates (Sarstedt), coated with 20 μg/mL poly-L-ornithine (P3655, Sigma–Aldrich) and 1 μg/mL Laminin2020 (L2020, Sigma–Aldrich). However, these were grown in DMEM/F-12 Glutamax basal medium (31331-093, Thermo Scientific) supplemented with 0.5× B27 (17504044, Thermo Scientific), 1× N-2 (17502001, Thermo Scientific) and 10 U/ml Penicillin-Streptomycin (15140122, Thermo Scientific). The cells were differentiated for 5 days or 28 days post induction of neural differentiation, followed by sampling based on the downstream application. Day 5 represents an early neuronal progenitor differentiation stage, whereas day 28 reflects a more mature neuronal lineage enriched for post-mitotic neurons in this NES-based differentiation system.28, 29, 30, 31 Representative images of the morphological changes that occur during key cell differentiation stages are available in Supplementary Fig. S1A. Selective serotonin reuptake inhibitors (SSRIs) We evaluated the effects of fluoxetine hydrochloride (FH; F132, Sigma–Aldrich), citalopram hydrobromide (CH; C7861, Sigma–Aldrich), sertraline hydrochloride (SH; S6319, Sigma–Aldrich) and paroxetine hydrochloride (PH; P9623, Sigma Aldrich) in our cell lines. SSRI stock solutions of 0.01 M concentration were prepared in absolute ethanol (SH) or MilliQ H2O (FH, CH and PH). Working solutions were prepared in MilliQ H2O at 0.001 M. For the in-vitro exposures at the required concentration(s), the SSRI working solutions were subsequently mixed with DMEM/F-12 Glutamax basal medium (31331-093, Thermo Scientific) supplemented with 0.5× B27 (17504044, Thermo Scientific), 1× N-2 (17502001, Thermo Scientific) and 10 U/ml Penicillin-Streptomycin (15140122, Thermo Scientific). Vehicle controls were also prepared, as required. Freshly prepared exposure medium was replaced on every second day during the period of neural differentiation. Cytotoxicity assay To select the appropriate treatment concentration of the SSRIs for our cell lines, we performed cytotoxicity testing. The treatment concentration ranges (0.00–30.00 μM) for the SSRIs were determined based on those optimised in our previous study using FH,26 followed by testing in the CTRLMale cell line using the MTS assay (G3580, Promega). The NES cells were seeded (12,000 cells/well) in 96-well plates (Sarstedt) and exposed to a concentration range of the SSRI being evaluated. Growth factors were withdrawn from the cell culture, to initiate neural differentiation. The exposure media was changed every second day. At 24- and 120-h (5 days) post exposure, the cells were treated with MTS (3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium) reagent, based on instructions from the manufacturer. After 3 h of incubation at 37 °C and 5% CO2, absorbance was recorded at 490 nm using a spectrophotometric microplate reader (FLUOstar Omega, BMG Labtech). For this assay, one independent experiment was performed at each timepoint, with a suitable number of technical replicates as visualised in the plots. Quantification of percentage viability at each concentration and timepoint was done in R (v4.1.2).33 The response was represented as a regression line generated using the Loess method for local regression fitting, and the shaded area denoted the 95% confidence interval. Two-way ANOVA, followed by Tukey’s post hoc correction for multiple comparisons, was used to test for significance (adj. p < 0.05) of the reported findings. Cellular ROS assay The cellular ROS assay (ab113851, Abcam) was used to determine total ROS levels at day 5 and 28 of SSRI exposures (FH, CH and PH: 3.0 μM; SH: 1.5 μM). In short, NES cells were seeded (12,000 cells/well) in dark walled, clear bottom 96-well plates (Sarstedt) and allowed to adhere overnight at 37 °C, 5% CO2 incubation. Thereafter, neural differentiation was induced by withdrawing growth factors from the culture medium accompanied by exposure to the SSRIs, with the culture medium being changed on every alternative day. The culture medium was aspirated, and the cells were exposed to the fluorogenic dye, DCFDA (2',7'-dichlorodihydrofluorescein diacetate), at a concentration of 10 μM, in 100 μL of 1× Buffer (from the kit) and incubated at 37 °C, 5% CO2 for 45 min. Fluorescence at excitation wavelength 485 nm and emission wavelength 520 nm was recorded using a spectrophotometric microplate reader (FLUOstar Omega, BMG Labtech). For these assays, two independent experiments were performed for each differentiation timepoint, with a suitable number of technical replicates as visualised in the plots. All readings were reported as percentage of control. Normality testing was performed using the Shapiro–Wilk Test. Subsequently, Kruskal–Wallis rank sum test, followed by Nemenyi’s post hoc correction for multiple comparisons, was used to test for the significance (adj. p < 0.05) of the reported findings in R (v4.1.2).33 Mitochondrial ToxGlo assay To test and correct for the Crabtree effect,34 the mitochondrial ToxGlo assay (G8000, Promega) was used to check for mitochondrial vulnerability, with effects on oxidative phosphorylation and subsequent ATP generation, to the potent mitotoxin sodium azide.35 In short, NES cells were seeded (12,000 cells/well) in white walled, clear bottom 96-well plates (Greiner) and allowed to adhere overnight at 37 °C, 5% CO2 incubation. The assay was used to determine cell membrane integrity and ATP levels after 24 h, in accordance with the optimised manufacturer’s instructions (Promega). On the next day, the culture medium was replaced with glucose free DMEM (11966025, Thermo Scientific) supplemented with 10 mM galactose (G5388, Sigma–Aldrich), instead of DMEM-F12 Glutamax basal medium (31331-093, Thermo Scientific), 90 min prior to the assay. The galactose induced the cells to primarily rely on oxidative phosphorylation to produce ATP, instead of glycolysis. Cell membrane integrity was assessed by measuring a distinct protease activity associated with necrosis using a fluorogenic peptide substrate, bis-AAF-R110. After incubation at 37 °C, 5% CO2 for 30 min, fluorescence at excitation wavelength 485 nm and emission wavelength 520 nm was recorded using a spectrophotometric microplate reader (FLUOstar Omega, BMG Labtech). Next, the ATP detection reagent was added, causing cell lysis, and producing a luminescent signal proportional to the amount of total ATP present. Immediately after, luminescence was recorded using a spectrophotometric microplate reader (FLUOstar Omega, BMG Labtech). For this assay, one independent experiment was performed, with a suitable number of technical replicates as visualised in the plots. All readings were reported as percentage of control. Two-way ANOVA followed by Tukey’s post hoc correction for multiple comparisons, was used to test for significance (adj. p < 0.05) of the reported findings in R (v4.1.2).33 Cellular ATP assay The CellTiter Glow 2.0 assay (G9242, Promega) was used to quantify the total ATP levels generated by oxidative phosphorylation in response to the SSRI exposures. This was done at day 5 (FH, CH and PH: 3.0 μM; SH: 1.5 μM) and day 28 of SSRI exposures (FH, CH, SH and PH: 1.5 μM). In short, NES cells were seeded (12,000 cells/well) in white walled, clear bottom 96-well plates (Greiner) and allowed to adhere overnight at 37 °C, 5% CO2 incubation. Thereafter, neural differentiation was induced by withdrawing growth factors from the culture medium and exposure to SSRIs, with the culture medium being changed at every alternative day. On day 5 and day 28 post induction of neural differentiation, the culture medium (100 μL/well) was replaced with glucose free DMEM (11966025, Thermo Scientific) supplemented with 10 mM galactose (G5388, Sigma–Aldrich), instead of DMEM-F12 Glutamax basal medium (31331-093, Thermo Scientific), 90 min prior to the assay. Thereafter, the cells were incubated at room temperature for 30 min followed by the addition of the CellTiter Glo 2.0 reagent (100μL/well) and mixing on an orbital shaker for 2 min to induce cell lysis. The cells were then incubated at room temperature for another 10 min followed by recording luminescence using a spectrophotometric microplate reader (FLUOstar Omega, BMG Labtech). For these assays, two independent experiments were performed for each differentiation timepoint, with a suitable number of technical replicates as visualised in the plots. All readings were reported as percentage of control. For the day 5 results, normality testing was performed using the Shapiro–Wilk Test. Subsequently, Two-way ANOVA followed by Tukey’s post hoc correction for multiple comparisons, was used to test for significance (adj. p < 0.05) of the reported findings. For the day 28 results, normality testing was also performed using the Shapiro–Wilk Test. Subsequently, the Dunn (1964) Kruskal–Wallis multiple comparison with p-values adjusted with the Holm method was used to test for significance (adj. p < 0.05) of the reported findings in R (v4.1.2).33 Direct infusion electrospray ionisation mass spectrometry (ESI-MS) Untargeted metabolomic profiling was performed using the direct infusion probe (DIP) for ESI-MS36 on snap-frozen cells at day 5 (N = 120) and day 28 (N = 120) of SSRI exposures during neural differentiation of NES cells, with two biological replicates having three technical replicates each for the treated and untreated groups in all four cell lines: CTRLMale, CTRLFemale, ASDHNRNPU and ASDCASK. For the day 5 samples, the cells were exposed to 3.0 μM of FH, CH and PH, and 1.5 μM of SH, as per the previously determined concentrations based on the cytotoxicity assays (Supplementary Fig. S1B). For the day 28 samples, the cells were exposed to a standardised dose of 1.5 μM to enable direct comparison of long-term exposure related metabolomic effects induced by each SSRI. The cell culture plates were stored at −80 °C prior to lysing with the electrospray solvent containing 9:1 methanol:water, with 0.1% formic acid and 28 internal standards. To enable comparisons, all cell samples were normalised to total cell counts per sample. The samples were analysed on a QExactive™ Orbitrap instrument (Thermo Scientific) at mass resolving power of 140,000 (m/Δm) using untargeted full scanning between 70 and 1000 Da in positive mode. The resulting data was extracted using 30 s data per sample and sorted in MZmine 2.37 One-point quantification of metabolite concentrations in the sample was calculated using the intensity ratio of endogenous metabolite to the internal standard and multiplied with the concentration of the internal standard.36 Differential metabolomics and feature selection Principal component analysis (PCA) was performed in R (v4.1.2)33 followed by visualisation with the factoextra package (v1.0.7) (https://cran.r-project.org/package=factoextra). Prior to the PCA, any missing values (0.03% of the dataset) were imputed using the missMDA package (v1.18)38 in R (v4.1.2).33 Here, the number of dimensions of the PCA were estimated by cross-validation using the k-fold method followed by imputation of missing values using the PCA model. To test for significance of the resulting metabolite concentrations (μM) following the mass spectrometry analysis workflow, a mixed linear model using the nlme package (v3.1-162)39 in R (v4.1.2)33 was applied to the dataset for each SSRI (FH, CH, SH and PH) at each timepoint (day 5, day 28). Any missing values were treated as NA values. Here, the mixed linear model used was lme(concentration ∼ exposure, random = ∼ 1 | cell line). The data was filtered prior to the application of the mixed linear model, based on the SSRI exposure. The obtained p values were adjusted for multiple comparisons using the FDR method in R (v4.1.2).33 Significantly modulated metabolites (Treatment vs. Control) were selected based on a significance threshold of adjusted p < 0.05 and were visualised using the ggplot2 package (v3.3.5)40 in R (v4.1.2).33 Feature selection of shared SSRI effects on the metabolites was performed using the statistical meta-analysis module of MetaboAnalyst 5.0.41 Metabolite concentration lists were provided for each SSRI at day 5 and day 28, respectively. Followed by performing an automated data integrity check, normalisation was performed using log2 transformation and data-scaling. Data scaling adjusts each variable/feature by a scaling factor computed based on the dispersion of the variable, in this case auto-scaling was used (mean-centred and divided by the standard deviation of each variable). For the selection of significant features that were shared by all exposure conditions, p values from each analysis were combined using the Fisher’s method, −2∗∑Log(p). A combined T statistic and combined p value was provided for each metabolite (feature), which were used for selecting significant shared features (combined p < 0.05). In addition to overall models, we conducted exploratory descriptive analyses stratified by clinical diagnosis background (ASD lines: ASDHNRNPU, ASDCASK; non-ASD lines: CTRLMale, CTRLFemale) to identify potential trends in exposure responses. Given the small number of lines per stratified group, these analyses were not powered for formal interaction testing. Pathway enrichment analysis Detected metabolites were tested for enrichment in the Small Molecule Pathway Database (SMPDB)42 using the quantitative enrichment analysis (QEA) approach to metabolite set enrichment analysis (MSEA) from the enrichment analysis module of MetaboAnalyst 5.0.41 Any missing values were imputed using the in-built K-Nearest Neighbour (KNN) algorithm for similar samples. The metabolite labels of 172 out of 189 metabolites were successfully mapped to PubChem (https://pubchem.ncbi.nlm.nih.gov) and annotated with PubChem CIDs. The remainder (17) were excluded from the analysis. The concentration data was normalised using data-scaling (auto-scaling, as done in the feature selection). The cut-off threshold for entries from metabolite sets to be included from the metabolite set library was set to 2. QEA uses a generalised linear model to estimate a Q-statistic for each metabolite set, which describes the correlation between compound concentration profiles, X, and experimental outcomes, Y. The Q-statistic for a metabolite set is the average of the Q-statistics for each metabolite in the set. The cut-off threshold for entries from metabolite sets to be included was set to 2. Based on the statistical testing, enrichment dot-plots and pathway networks were generated. In the pathway networks, 2 or more nodes were only connected by an edge when the number of shared metabolites was >25% of the combined metabolites contributing to each node. Lipid exposure and cPLA2 inhibition Solutions of the selected LPCs (LPC 16:0, 855675P, Sigma–Aldrich; LPC 18:0, 845875P, Sigma–Aldrich; and LPC 18:1, 855775P, Sigma–Aldrich) were prepared in methanol (stock: 0.01 M; 322415, Sigma–Aldrich) and used immediately by warming up at 37 °C or stored at −20 °C for future use. Working solutions were prepared based on the determined concentration of the LPCs at day 5 of SH exposure (LPC 16:0–2.0 μM, LPC 18:0–1.8 μM and LPC 18:1–0.8 μM), from the mass spectrometry based-metabolomics. Vehicle controls were also prepared, as required. The cells were exposed to a 2× mixture of the LPCs at the initiation of undirected differentiation and every media change (alternative feeding days). LPC synthesis was inhibited by tandemly exposing the differentiating cells with 10.0 μM AACOCF3 (arachidonyl trifluoromethyl ketone; ab120350, Abcam). AACOCF3 is a phospholipase-modulating compound with reported PLA2-inhibitory activity, and was used to perturb LPC-related lipid pathways.43, 44, 45 Cellular ROS and cellular ATP assays were performed at day 5 and 28 of exposure, as described in the previous sections. For these assays, two independent experiments were performed for each differentiation timepoint, with a suitable number of technical replicates as visualised in the plots. All readings were reported as percentage of control. Two-way ANOVA followed by Tukey’s post hoc correction for multiple comparisons, was used to test for significance (adj. p < 0.05) of the reported findings in R (v4.1.2).33 Population-based mother-child cohort The Barwon Infant Study (BIS) consists of 1074 mother-child pairs (including 10 sets of twins) recruited from the Barwon region (Victoria, Australia) recruited antenatally between July 2010 and July 2013. Cohort details, including population characteristics, eligibility and exclusion criteria, and data collection methods, have been described elsewhere.46 The BIS cohort is broadly representative of the Australian population, with a slight over-representation of families from English-speaking backgrounds.46 Information on antidepressant use, including the type used, was collected in a medication questionnaire during the 28-week antenatal review (1st and 2nd trimester). In addition, the Edinburgh Postnatal Depression Scale (EPDS)47 was administered during the 28-week antenatal review. This is a widely used and validated measure of depression that assesses mood over a one-week period. Further details are described earlier.48 Lipidomic measures were obtained from cord blood serum sample, obtained from the umbilical cord blood at birth, then placed into serum clotting tubes and analysed with ultra-high-performance-tandem mass spectrometry (UHPLC-MS/MS). The time from blood collection to processing and storage, as well as the total duration of storage at −80 °C, was recorded and considered in the subsequent analysis. Maternal blood contamination in cord blood samples was assessed using DNA methylation profiling based on previously established methods.49,50 The details of the UHPLC-MS/MS analysis are described elsewhere.51 Briefly, 776 lipid features were quantified, including LPC 16:0, LPC 18:0, and LPC 18:1, with the fatty acyl moiety at either the sn1 or sn2 position. These were chosen a priori based on the results of the in-vitro experiments. Lipids were extracted using a 1:1 butanol solution with 10 mM ammonium formate, which included a mixture of internal standards. The analysis was conducted using UHPLC-MS/MS on an Agilent 6490 QQQ mass spectrometer coupled with an Agilent 1290 series UHPLC system. Separation was achieved using two ZORBAX Eclipse Plus C18 columns (21 × 100 mm, 18 mm, Agilent) at a temperature of 45 °C. Mass spectrometry was performed in both positive and negative ion modes with dynamic scheduled multiple reaction monitoring. Lipid species concentrations were determined by comparing them to the corresponding internal standards. Further details, data processing and quantification have been described previously.49,51 ADHD and autism-related symptoms were assessed using the Child Behaviour Checklist for ages 1.5–5 (CBCL) at 2 years and the Strengths and Difficulties Questionnaire P4-10 (SDQ) at 4 years, methods described elsewhere.49 Specifically, the DSM-5-oriented subscales for attention-deficit/hyperactivity problems (CBCL-ADHP) and autism spectrum problems (CBCL-ASP) from the CBCL were utilised, along with the SDQ subscales for hyperactivity (SDQ-hyper), peer problems (SDQ-peer), and prosocial behaviour (SDQ-prosocial). The questionnaires were completed by the child’s parent or guardian. Multivariate regression models were applied to assess the association between prenatal SSRI use, antidepressant use, and EPDS score and LPC levels in cord blood. Multivariate regression models, with LPCs split into categories based on quartiles, to allow for nonlinear associations with dose–response and neurodevelopmental outcomes, were also assessed. Covariates included child sex, gestational age, time sample spent on the bench and in the freezer before analysis, contamination of the child’s blood with maternal blood, and child’s birth weight. LPC-neurodevelopmental outcomes were additionally adjusted at the child’s age the CBCL or SDQ was completed. Nominal statistical significance was considered at p < 0.05 and trends at p < 0.10. Statistics All statistical analyses were performed in R (v4.1.2).33 The statistical models and tests used for the analyses, along with any inclusion/exclusion criteria are described in the methodology relevant to the experimental technique, in the sections above. All plots, unless otherwise stated, were created using the ggplot2 package (v3.3.5)40 or the pheatmap package (v1.0.12, https://CRAN.R-project.org/package=pheatmap) in R (v4.1.2).33 Ethics The study was conducted in accordance with the Declaration of Helsinki. The Swedish Ethical Review Authority (Study cohort: 2016/1452-31 and iPSC reprogramming: 2012/208-31/3) gave ethical approval for the research conducted at Karolinska Institutet. Written informed consent, including for iPSC research, was obtained from all participants or their caregivers, based on their age. The Barwon Infant Study was approved by the Barwon Health Human Research and Ethics Committee (HREC 10/24). Written informed consent was obtained from all participants or their caregivers, based on their age. Role of funders The funders did not play any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Results Increased comparative cytotoxicity is observed with sertraline We studied the in-vitro effects of four SSRIs, namely fluoxetine hydrochloride (FH), citalopram hydrobromide (CH), sertraline hydrochloride (SH) and paroxetine hydrochloride (PH) using four human iPSC derived NES cell lines, of which two were from individuals with NDDs and known pathogenic genetic variants.28, 29, 30, 31 The NES cells underwent undirected differentiation for up to 28 days, with and without SSRI exposure. First, we tested the optimal concentrations for SH, CH, and PH using the CTRLMale cell line, similar to our prior work with FH.26 Using an exposure range between 0.0 and 30.0 μM, cell viability was assessed using the MTS assay at 24- and 120-h post-differentiation. The highest concentration with no significant reduction in viability (Tukey post hoc p > 0.05) or observable changes in cellular morphology under phase-contrast microscopy at day 5 was selected. Based on these criteria, the optimal concentrations were 3.0 μM for FH, CH, and PH and 1.5 μM for SH (Supplementary Fig. S1B). SH demonstrated increased cytotoxicity at 120 h post-differentiation compared to the other SSRIs, for which its optimal concentration was adjusted. Sertraline and paroxetine decrease cellular ROS and ATP levels Total ROS levels were detected using the DCFDA (2',7'-dichlorodihydrofluorescein diacetate) assay at day 5 and 28 of differentiation in the four cell lines. At day 5 and day 28, a significant (Nemenyi post hoc p = 0.005) reduction in ROS levels was detected for SH exposure (Fig. 1A). For PH, significantly reduced total ROS levels were detected at day 28 (Nemenyi post hoc p = 0.007) (Fig. 1A). In-vitro assays and differential metabolomics following selective serotonin reuptake inhibitor (SSRI) exposure in differentiating neuroepithelial stem cells. A.In-vitro assay determined total ROS levels at day 5 (n = 24 per condition) and day 28 (n = 24 per condition) of fluoxetine (FH), citalopram (CH), sertraline (SH) and paroxetine (PH) exposure (Kruskal–Wallis Rank Sum, Nemenyi post hoc p: ∗<0.05, ∗∗<0.01, ∗∗∗<0.001). B.In-vitro assay determined total ATP levels at day 5 (n = 24 per condition; two-way ANOVA, Tukey’s post hoc adjusted p: ∗<0.05, ∗∗<0.01, ∗∗∗<0.001) and day 28 (n = 24 per condition; Dunn (1964) Kruskal–Wallis, Holm post hoc p: ∗<0.05, ∗∗<0.01, ∗∗∗<0.001) of FH, CH, SH and PH exposure. C. Distribution of significant metabolites across metabolite classes at day 5 and day 28 (mixed linear model, FDR adjusted p). D. Heatmap of estimate values from the applied mixed linear model for metabolites and SSRIs, with a significant overlap across at least two SSRIs (FDR adjusted p: a < 0.05, b < 0.01, c < 0.001). To correct for the Crabtree Effect, where cells in culture prefer glycolysis over oxidative phosphorylation,34 the cells were assayed in a cell medium in which glucose was replaced with galactose to promote ATP generation by oxidative phosphorylation (Supplementary Fig. S1C). CH significantly increased (Tukey post hoc p = 0.0004) total ATP levels, while PH caused a decrease (Tukey post hoc p < 0.0001) at day 5 (Fig. 1B). At day 28, no difference was observed with CH compared to the control; however, both SH (Holm adj. p < 0.0001) and PH (Holm adj. p = 0.005) decreased total ATP levels (Fig. 1B). Early metabolic changes are induced by sertraline Mass spectrometry-based untargeted metabolomics revealed over 2500 metabolite signals, including adducts, in our cell lysates. Of these, 189 unique metabolites met the signal threshold and were quantified at both day 5 and day 28 post neural differentiation and SSRI exposure. At day 5, cells were exposed to FH, CH, and PH at 3.0 μM and SH at 1.5 μM, and the exposure was validated in the metabolomics data (Supplementary Fig. S1D). PCA revealed overlapping metabolic profiles across the SSRI treatments, with PC1 and PC2 accounting for 43.5% of the total variability (32.6% and 10.9%, respectively) (Supplementary Fig. S2A). Herein, the SSRI mediated effects are not driven by large-scale metabolomic shifts but specific metabolite changes which can be visualised using volcano plots (Supplementary Fig. S3A). Using a mixed linear model accounting for the genetic background and the exposure followed with FDR correction, we report that SH modulated the largest number of metabolites (52 metabolites), followed by PH (19 metabolites), FH (9 metabolites), and CH (1 metabolite) at day 5 (Supplementary Table S2A). For SH, the metabolites spanned seven classes: acylcarnitines, amino acids, lysophospholipids, phosphatidylcholines, phosphatidylethanolamines, plasmanylcholines and sugars (Fig. 1C). Amino acids such as glutamic acid (SH, adj. p = 0.001; PH, adj. p = 0.019), methionine (SH, adj. p = 0.004; PH, adj. p = 0.027), threonine (SH, adj. p = 0.001; PH, adj. p = 0.014), and lysine (SH, adj. p = 0.004; PH, adj. p = 0.049) were reduced by both SH and PH (Supplementary Table S2A and Fig. 1D). Feature selection analysis of day 5 data highlighted ten significant metabolites across SSRI treatments (Supplementary Table S2D). The lysophosphatidylcholines – LPC 16:0, LPC 14:0, and LPC 18:0 emerged as the top three features, each with a combined p-value of <0.0001. Paroxetine and sertraline share metabolic profiles following prolonged exposure For the day 28 metabolomics analysis, all four cell lines were treated with SSRIs at a standardised concentration of 1.5 μM, based on the SH treatment level. These exposures were confirmed in the samples (Supplementary Fig. S1D). Again, PCA did not show clear clustering by treatments (Supplementary Fig. S2B). PC1 and PC2 explained 38.1% of the variance (22.8% and 15.3%, respectively). Furthermore, specific metabolite changes can be visualised using volcano plots (Supplementary Fig. S3B). The mixed linear model analysis detected the highest number of metabolite changes after PH treatment (64 metabolites), followed by SH (61 metabolites), FH (28 metabolites), and CH (1 metabolite) (Supplementary Table S3A). Most metabolites affected by SH, FH, and CH were also modulated by PH (Fig. 1C). For instance, PC 30:0 was significantly reduced by all SSRIs (FH, adj. p = 0.004; CH, adj. p = 0.040; SH, adj. p = 0.018; PH, adj. p = 0.0001). Amino acids such as glutamine (FH, adj. p = 0.007; SH, adj. p = 0.003; PH, adj. p = 0.005), proline (FH, adj. p = 0.004; SH, adj. p = 0.003; PH, adj. p = 0.004), and threonine (FH, adj. p = 0.002; SH, adj. p = 0.001; PH, adj. p = 0.001) were decreased across these treatments (Fig. 1D). GABA was specifically reduced by SH (adj. p = 0.023) and PH (adj. p = 0.035) (Supplementary Table S3A and Fig. 1D). LPCs (18:0, 18:1, 16:0) were consistently elevated: LPC 18:0 (FH, adj. p = 0.002; SH, adj. p < 0.0001; PH, adj. p < 0.0001), LPC 18:1 (FH, adj. p = 0.001; SH, adj. p < 0.0001; PH, adj. p = 0.001), and LPC 16:0 (SH, adj. p < 0.0001; PH, adj. p = 0.008) (Fig. 1D). Feature selection using metabolite data from day 28 identified 24 significant metabolites (Supplementary Table S3D). The top three features were LPC 18:0 (combined p < 0.0001), LPC 18:1 (combined p = 0.001), and LPC 16:0 (combined p = 0.002). Exploratory analysis of diagnosis-related SSRI exposure effects To complement the primary analyses of SSRI exposure effects across all four cell lines, we examined responses stratified by the clinical background of the cell donors to identify exploratory diagnosis-related patterns in cellular readouts and metabolomics. This analysis revealed only modest diverging patterns. In terms of total ROS levels, a significant decrease was observed at day 28 in the non-ASD lines following SH (Nemenyi post hoc p = 0.022) and PH (Nemenyi post hoc p = 0.002) exposures (Supplementary Fig. S4A). At day 5, both non-ASD and ASD lines showed increased total ATP with CH (non-ASD, Tukey post hoc p = 0.004; ASD, Tukey post hoc p = 0.040), while PH decreased total ATP in both groups, more prominently in the ASD lines (non-ASD, Tukey post hoc p = 0.017; ASD, Tukey post hoc p = 0.0002) (Supplementary Fig. S4B). At day 28, sertraline caused a significant total ATP decline in both groups, with a larger reduction in the non-ASD lines (non-ASD, Holm adj. p = 0.007; ASD, Holm adj. p = 0.019) (Supplementary Fig. S4B). In the metabolomic signature analysis, across timepoints, the ASD lines exhibited fewer changes at Day 5 (SH: 18, CH: 4, FH: 2, PH: 1; Supplementary Table S2B and Fig. S4D), when compared with the non-ASD lines (SH: 24, PH: 10, FH: 6, CH: 0; Supplementary Table S2C and Fig. S4C). However, at Day 28 this pattern reversed, with the ASD lines showing more extensive modulation (SH: 55, PH: 29, FH: 18, CH: 0; Supplementary Table S3B and Fig. S4F) relative to the non-ASD lines (SH: 17, PH: 14, FH: 5, CH: 0; Supplementary Table S3C and Fig. S4E). Some diverging patterns also emerged, for instance, fatty acids FA 10:0 and FA 20:1 were increased, while PC 26:0 was decreased only in the ASD lines, with CH and SH at Day 5 (Supplementary Table S2B and Fig. S4D). At Day 28, GABA decreased only with SH in both the ASD and non-ASD lines (Supplementary Table S3B and C), and the short-chain acyl-carnitine, C3 carnitine decreased with SH and PH (Supplementary Table S3B and C, Supplementary Fig. S4E and F). Overall, these exploratory comparisons largely support shared lipid effects, with limited diagnosis-specific features. Phospholipid biosynthesis and amino acid metabolism are key pathways modulated by SSRI exposure Using metabolite set enrichment analysis (MSEA) and quantitative enrichment analysis (QEA), significant biochemical pathways affected by SSRI exposures were identified. At day 5, exposure to FH (Fig. 2A, Supplementary Table S4A) resulted in the enrichment of ten pathways, including phospholipid biosynthesis (adj. p = 0.0001), galactose metabolism (adj. p = 0.0001), and sphingolipid metabolism (adj. p = 0.001). SH exposure significantly enriched 48 pathways (Supplementary Table S4C), notably phospholipid biosynthesis (adj. p < 0.0001), purine metabolism (adj. p = 0.001), and tryptophan metabolism (adj. p = 0.001) (Fig. 2A and Supplementary Fig. S2C). Similarly, PH exposure enriched 50 pathways (Supplementary Table S4D), including phospholipid biosynthesis (adj. p < 0.0001), purine metabolism (adj. p = 0.014), and tryptophan metabolism (adj. p = 0.027) (Fig. 2A and Supplementary Fig. S2D). In contrast, CH exposure did not result in significant pathway enrichment (Supplementary Table S4B). Enrichment of detected metabolites in biochemical pathways. A. Enrichment dot-plot of top 10 significantly enriched metabolic pathways (adjusted p ≤ 0.05) following fluoxetine (FH), sertraline (SH) and paroxetine (PH) exposure at day 5 and day 28. B. Interaction network of top 50 enriched metabolic pathways following SH exposure at day 28. C. Interaction network of top 50 enriched metabolic pathways following PH exposure at day 28. At day 28, similar results were observed. FH exposure enriched ten pathways (Supplementary Table S5A), such as galactose metabolism (adj. p = 0.0003), sphingolipid metabolism (adj. p = 0.003), and phospholipid biosynthesis (adj. p = 0.007) (Fig. 2A). SH exposure led to the enrichment of 54 pathways (Supplementary Table S5C), including phospholipid biosynthesis (adj. p < 0.0001), purine metabolism (adj. p = 0.0002), and phenylalanine and tyrosine metabolism (adj. p = 0.003) (Fig. 2A and B). PH exposure enriched 46 pathways (Supplementary Table S5D), including phospholipid biosynthesis (adj. p = 0.0001), purine metabolism (adj. p = 0.018), and phenylalanine and tyrosine metabolism (adj. p = 0.041) (Fig. 2A and C). CH exposure continued to show no significant pathway enrichment (Supplementary Table S5B). These findings emphasise the consistent enrichment of key biochemical pathways, particularly phospholipid biosynthesis and amino acid metabolism, in response to FH, SH, and PH exposures. This highlights the central role of these pathways in SSRI-induced metabolic effects. Selected LPCs associate with in-vitro and in-vivo SSRI exposures As LPC 18:0, LPC 18:1, and LPC 16:0 were the most consistent findings within the cellular exposure metabolic screening at both time points (Figs. 1D and 3A), we continued to investigate their effects both in-vitro and in-vivo. First, we hypothesised that LPCs could contribute to changes in ROS and ATP levels that were observed after the SH and PH exposures (Fig. 1A and B). To test this, we inhibited LPC production using the phospholipase A2 inhibitor arachidonyl trifluoromethyl ketone (AACOCF3, 10 μM) for 24 h, as well as treated the cells with a combination of the LPCs (18:0, 18:1, 16:0) and SH. The 24-h inhibition of the LPC production attenuated the reduction of ROS both at day 5 and day 28 in the SH-treated cells; however, there was no significant difference between the two conditions (Fig. 3B). For ATP production, the LPC production inhibition only slightly attenuated the effect of SH on day 5 and seemed to enhance the reduction on day 28 (Fig. 3C). Interestingly, the combination of LPCs increased the ROS and ATP levels at both time points (Supplementary Fig. S2E and F). Our findings indicate that the three LPCs and inhibition of their synthesis were associated with changes in ROS and ATP levels, though inconsistently, suggesting more complex regulatory patterns. Association of selected lysophosphatidylcholines (LPCs) with selective serotonin reuptake inhibitor (SSRI) exposures and neurodevelopmental disorder outcome measures. A. Detected concentration of selected LPCs following fluoxetine (FH), citalopram (CH), sertraline (SH) and paroxetine (PH) exposure at day 5 (n = 24 per condition) and 28 (n = 24 per condition) (mixed linear model, FDR adjusted p: ∗<0.05, ∗∗<0.01, ∗∗∗<0.001). B.In-vitro assay determined total ROS levels at day 5 (Control, n = 6; SH, n = 8; SH + AACOCF3, n = 8) and day 28 (Control, n = 7; SH, n = 8; SH + AACOCF3, n = 4) of LPC inhibition (two-way ANOVA, Tukey’s post hoc adjusted p: ∗<0.05, ∗∗<0.01, ∗∗∗<0.001; NS: non-significant). C.In-vitro assay determined total ATP levels at day 5 (n = 8 per condition) and day 28 (n = 8 per condition, except SH + AACOCF3, n = 4) of LPC inhibition (two-way ANOVA, Tukey’s post hoc adjusted p ∗<0.05, ∗∗<0.01, ∗∗∗<0.001; NS: non-significant). D. Adjusted estimated mean change in doubling of LPC concentration (with 95% confidence interval, CI), per exposure. E. Adjusted estimated mean change in CBCL (Child Behaviour Checklist) Autism Spectrum Problems and CBCL Attention Deficit Hyperactivity Problems (with 95% CI) by quartiles of LPC concentrations (Q1, reference group). Next, we used metabolomiic and neurodevelopmental outcome data from the BIS cohort,49 comprising 1074 mother-child pairs, to analyse the LPC effects in a population-based cohort. Detailed cohort characteristics relevant to the present analyses are provided in Supplementary Table S6A. First, we investigated the associations between maternal SSRI use (n = 40) and any antidepressant use (n = 48) within the first and second trimesters on the levels of the selected LPCs (Fig. 3D, Supplementary Table S6B) in the infant cord blood. Our findings indicate that prenatal SSRI use was associated with elevated levels of cord blood LPC 16:0 sn1 (β = 0.173, 95% CI = 0.011–0.334; p = 0.037), and LPC 16:0 sn2 (β = 0.175, 95% CI = 0.007–0.343; p = 0.042), and LPC 18:0 sn2 (β = 0.174, 95% CI = 0.018–0.330; p = 0.029). Additionally, we observed trends (p < 0.1) suggesting higher levels of LPC 18:0 sn1 and LPC 18:1 sn2 with prenatal SSRI use. When analysing the effects of any antidepressant use, which added eight additional mothers on other antidepressants, these associations were attenuated (Fig. 3D, Supplementary Table S6B). There were no significant correlations between maternal prenatal EPDS scores and any of the LPCs studied (Fig. 3D, Supplementary Table S6B). Furthermore, no significant associations were found between prenatal SSRI use and maternal LPC levels at 28 weeks of gestation or with the infant's LPC levels at 6 months, 12 months, and 4 years. We previously published significant positive associations between LPCs and CBCL autism and ADHD symptom scales, although not all remained significant after FDR correction.49 Here, we further examined the relationship between the three consistently upregulated LPCs and the outcomes by categorising each LPC measure into quartiles. This approach allowed us to explore potential non-linear associations with CBCL-autism and CBCL-ADHD scores and assess dose–response relationships. Significant associations were found for LPC 16:0 sn2 with CBCL-autism scores (overall p = 0.01) and for both LPC 16:0 sn1 and LPC 16:0 sn2 with CBCL-ADHD scores (overall p = 0.007 and 0.002, respectively) (Fig. 3E, Supplementary Table S6C). When investigating outcomes older than two years of age, we did not observe significant associations between any of the LPCs and autism/ADHD-related measures from the strength and difficulties questionnaire (SDQ) at age 4 for hyperactivity difficulties, peer problem difficulties, prosocial behaviour strengths, or parent-reported autistic traits (Supplementary Table S6D). Discussion We have investigated the effects of four commonly prescribed SSRIs on human iPSC-derived neuroepithelial stem cells undergoing undirected differentiation for 5 and 28 days towards neurons using four genetically distinct human iPS cell lines. Exposure to SH and PH, of which the latter is not clinically recommended during pregnancy,52 notably decreased total ROS and ATP levels, compatible with altered oxidative phosphorylation, though direct mitochondrial function was not assessed. The metabolomic analysis further identified multiple significantly modulated metabolites predominantly driven by SH and PH, enriched in energy metabolism and amino acid pathways. Importantly, we identified elevated lysophosphatidylcholines (LPC 16:0, LPC 18:0, and LPC 18:1) as robust molecular changes following SSRI exposures, consistent across cellular models and clinical samples, underscoring their potential clinical significance. The data from the population-based BIS cohort confirmed that LPC 16:0 and LPC 18:0 could be potential candidate biomarkers of prenatal SSRI exposure. Our findings suggest LPCs may be associated with SSRI exposure and early neurodevelopmental traits, but further validation in larger cohorts is required. There is a lack of clarity regarding the ability of SSRIs to alter ROS levels and their mechanism of action. The association between SSRIs and ROS has been reported predominantly in epidemiological and in-vivo studies of depression.53, 54, 55, 56 In clinical depression, the antioxidant systems are known to be inherently suppressed22,57 suggesting that the antioxidant, ROS-reducing properties of SSRIs apply only under oxidative stress.22,58 We hypothesise that reduced ROS availability during key neural developmental stages could influence ROS-mediated signalling pathways that are important for neurite growth and synaptogenesis.24 We also highlighted differences in in-vitro ATP levels, which we use as a surrogate marker for normal mitochondrial function. Both SH and PH have been previously shown to induce mitochondrial damage with depleted ATP levels, but in astrocytes.59 Another in-vitro study on human placental cells also suggests that SH alters ROS and ATP levels.60 While our results align with some observations reported in the literature, the precise, underlying mechanisms by which SSRIs affect ROS and ATP levels remain largely unclear. In addition to ROS and ATP levels, we detected several pathways mediated by FH, SH, and PH exposure related to energy metabolism. The SSRIs, specifically escitalopram (enantiomer of CH) and FH, have been shown to localise in cellular phospholipid membranes, cytoplasm, and endoplasmic reticulum, influencing various aspects of cellular physiology.27 Interestingly, in human immortalised cell lines of brain origin, FH was demonstrated to activate the AMPK-ACC-CPT1 pathway,61 which is known to play an important role in energy-related metabolic homeostasis through the regulation of mitochondrial ROS.62 FH has also been shown to enhance energy metabolism in the hippocampus of an in-vivo model of major depressive disorder.63 A comprehensive multi-omics study has also supported the link between FH and increased energy metabolism across multiple rat brain regions.64 FH exposure in mice has been shown to induce the gradual accumulation of sphingomyelin,65 a pathway we highlighted in our earlier study26 and this study. While there are fewer studies with CH, SH, and PH in this domain, CH has been known to cause in-vitro phospholipidosis,66 and PH has been reported to cause energy metabolism changes in animal models.67, 68, 69 We also reported enrichments in pathways related to the metabolism of different amino acids crucial for normal brain function. In an animal model, PH demonstrated an impact on metabolites involved in amino acid metabolism.68 This is noteworthy as we have previously shown differences in urinary metabolites derived from amino acids in individuals diagnosed with autism.70 Others have also reported an association of tryptophan, purine, phenylalanine, and tyrosine metabolism with autism.71, 72, 73, 74, 75 Intriguingly, CH did not seem to have any major effects on cellular metabolism in our study despite a recent report that prenatal exposure is associated with higher rates of ADHD diagnosis.76 More work is needed, potentially testing a range of CH doses, to further investigate its effects in-vitro. In addition to the enrichment of metabolic pathways, our main finding was the consistent increase of the three LPCs (LPC 16:0, LPC 18:0, and LPC 18:1) following SSRI exposure in neural cells and in cord blood. While these observations suggest altered phospholipid metabolism, the underlying mechanisms remain unclear. LPC 18:0 has been demonstrated to be an agonist for the peroxisome proliferator-activated receptor-gamma (PPAR-γ) pathway in a mouse model.77 In the brain, the PPAR-γ pathway has been connected with motivation and associative learning.78 Treatment of rodent models of autism with an agonist of the PPAR-γ pathway was shown to improve several phenotypic features.79,80 We also reported a decrease in the levels of several phosphotidylcholines, which has been previously associated with autism81 and can be attributed to an increased activity of cytosolic phospholipase A2 (cPLA2).81,82 The enzyme cPLA2 hydrolyses phospholipids to LPCs and other bioactive lipids,83 potentially contributing to the observed increase in LPC levels that we have reported. As cPLA2 activity and PPAR-γ signalling were not measured in this study, further research is required to connect these downstream effects with SSRI exposure that could potentially lead to premature neural differentiation84 during early development. We acknowledge that elevated LPC levels in the cord blood are not the only metabolites associated with autism and ADHD traits as there were several other metabolites associated with these outcomes in the BIS cohort.49 However, pending further investigation, this could be more specific for in-utero SSRI exposure and thus related to a broad number of outcomes also reported in the offspring. For instance, higher levels of LPC 18:0 in maternal serum have been indicated as a biomarker for congenital heart defects (CHDs) in infants.85 Furthermore, LPC 16:0 and LPC 18:1 levels in cord blood were positively associated with birth weight,86, 87, 88 a known liability factor for different cardiometabolic conditions.89 This, in turn, could be connected with PH, as its early in-utero exposure has been correlated with an increased liability of cardiac malformations in infants.52,90 There is a link between CHDs and adverse neurodevelopmental outcomes,91,92 which warrants further investigation of the in-utero SSRI exposure effects through the LPCs. Additionally, in-utero SSRI exposure was found to be associated with increased anxiety and depression-related symptoms in adolescents, along with changes in the activation of fear-related brain circuitry.93 Nevertheless, it is also essential to highlight that despite potentially adverse indicators of SSRI-mediated effects in-utero, a recent large-scale study demonstrated a reduction in depression-related mental health outcomes following postnatal SSRI treatment, with subsequent benefits observed on child externalising behaviours during early childhood as well.94 Although our study has many strengths, we must also recognise its limitations. Despite employing four different cell lines, the small cell line panel size limits our ability to distinguish biological heterogeneity from stochastic variability, and we lacked sufficient power to make strong comparisons regarding sex or genetic background in relation to exposure effects, indicating a need for further research. Nevertheless, we have accounted for probable genetic differences in our findings by also including cell lines derived from individuals with NDDs and reported the exploratory analyses of putatively different exposure outcomes based on clinical diagnoses status of the iPSC-donors. Additionally, our in-vitro system does not model maternal pharmacokinetics, placental transfer, foetal metabolism, or maternal mental health, and the SSRI concentrations, therefore, representing cellular exposure paradigms rather than exact foetal brain or serum levels. In the context of our findings, it is important to reiterate the fact that there is contradictory epidemiological evidence regarding the association of in-utero SSRI exposure with neurodevelopmental outcomes. Based on clinical best practice, SSRI use during pregnancy is safe and an effective therapeutic intervention in depression and allied neuropsychiatric conditions. While the BIS cohort provides a robust clinical validation for our in-vitro model, its observational design is susceptible to confounding factors, including genetic influences. Additionally, the relatively small number of infants exposed to SSRIs (n = 40) limits the generalisability of our findings. Furthermore, BIS is broadly representative of the Australian population, though families from English-speaking backgrounds are slightly over-represented,46 which should be considered when extrapolating the findings. Furthermore, the antidepressant exposure in BIS was ascertained by maternal self-report via a medication questionnaire completed at the 28-week antenatal review. This approach may introduce self-reporting and recall inaccuracies. Although prospective reporting reduces differential recall bias, misclassification is possible and is likely non-differential, which would bias exposure–outcome associations toward the null. Dosage, adherence, and medication use outside the reporting window were not captured and could contribute to exposure heterogeneity. We also need to acknowledge that our study did not cover the issue of SSRI withdrawal, which has been associated with neuroinflammation in a murine model and a potential elevation of ROS levels.95 The metabolomic and neurological impact of withdrawal in early life, either at or prior to weaning, also requires evaluation. Overall, larger studies are required to determine whether LPC 16:0 and LPC 18:0 levels in cord blood statistically mediate an association between maternal SSRI use and autism-related outcomes in infants. In terms of policy, the advantages and disadvantages of SSRI use in pregnancy requires continual revaluation as new findings, such as those described here, occur. In conclusion, this translational investigation provides novel insights into the metabolomic impact of prenatal SSRIs on human neural differentiation and neuronal development. We identify LPC 16:0 and LPC 18:0 as candidate metabolites of altered lipid metabolism following SSRI exposures. Future studies should clarify the molecular mechanisms underlying these metabolomic alterations, the association with neurodevelopmental symptoms and other outcomes in infants, as well as assess their clinical utility in liability prediction and targeted interventional strategies. Contributors Conceptualisation: A.A., P.V., A-L.P. and K.T.; Methodology: A.A., P.V., A-L.P. and K.T.; Software development: A.A., K.V. and K.T.; Validation: A.A., K.V., C.M., M.-O.D. F.M., J.H., X.Y. and M.O.; Formal Analysis: A.A., K.V., C.M., M.-O.D., F.M. and J.H.; Investigation: A.A., K.V., C.M., M.-O.D., F.M., J.H., X.Y. and M.O.; Resources: P.V., A-L.P., I.L. and K.T.; Data Curation: A.A., K.V., C.M. and F.M.; Writing – Original Draft: A.A., M.-O.D, J.H. and K.T.; Writing – Review & Editing: A.A. and K.T.; Visualisation: A.A., K.V. and K.T.; Supervision: A-L.P., I.L., and K.T.; Project Administration: A.A. and K.T.; Funding Acquisition: A.A. and K.T. All authors have read and approved the final version of the manuscript. A.A. and K.T. have accessed and verified the underlying data of the manuscript. The members of the BIS investigator group steering committee are David Burgner, Erica Grundell, Katherine Drummond, Len Harrison, Luba Sominsky, Martin O’Hely, Mimi Tang, Peter Sly, Richard Saffery, Samantha Dawson and Toby Mansell. The BIS investigator group is responsible for the recruitment of their study participants and the subsequent data/sample collection related to them. Data sharing statement The data and code required for differential metabolomics analysis are available on GitHub (https://github.com/Tammimies-Lab/SSRI-Metabolomics). The raw data from the mass spectrometry-based metabolomics is available on MetaboLights (MTBLS12645) or available upon request from the corresponding author (kristiina.tammimies@ki.se). The data from the Barwon Infant Study are available under restricted access to protect participant confidentiality. Researchers may request access by contacting the BIS Steering Committee via Anne-Louise Ponsonby at the Florey Institute of Neuroscience and Mental Health (annelouise.ponsonby@florey.edu.au). Requests are considered based on scientific and ethical considerations, and approved applications require a collaborative research agreement. Deidentified data can be shared in either Stata or CSV file formats. Further details about the cohort, data access process, and project overview are available at: www.barwoninfantstudy.org.au. Declaration of interests A.A., K.V, C.M., M.-O.D., F.M., J.H., X.Y., M.O., P.V., A-L.P., I.L. and K.T. declare no conflicts of interest. The foundational work and infrastructure for the BIS was sponsored by the Murdoch Children's Research Institute, Deakin University, and Barwon Health. Subsequent funding was secured from the Minderoo Foundation, the European Union's Horizon 2020 research and innovation programme, National Health and Medical Research Council of Australia (NHMRC) and Agency for Science, Technology and Research Singapore, The William and Vera Ellen Houston Memorial Trust Fund (via HOMER Hack), The Shepherd Foundation, The Jack Brockhoff Foundation, the Scobie & Claire McKinnon Trust, the Shane O'Brien Memorial Asthma Foundation, the Our Women Our Children's Fund Raising Committee Barwon Health, the Rotary Club of Geelong, the Ilhan Food Allergy Foundation, Geelong Medical and Hospital Benefits Association, Vanguard Investments Australia Ltd, the Percy Baxter Charitable Trust, and Perpetual Trustees. The Cotton On Foundation and Creative Force offered their support in-kind. Acknowledgements The authors would like to thank the donors and their parents for providing valuable material for this study as a part of the RATSS (the Roots for Autism and ADHD Twin Study in Sweden) cohort. The authors acknowledge the RATSS team, especially Prof Sven Bölte, Dr Karl Lundin Remnelius, Dr Johan Isaksson and Dr Janina Neufeld for their support. The authors also acknowledge the BIS cohort participants and assistance from Lada Holland for data cleaning and verification. We thank Prof Peter J. Meikle, Dr Satvika Burugupalli and the Metabolomics laboratory at the Baker Institute for the conduct and advice on lipidomic lab measures for BIS. The project was supported by the Swedish Research Council – Vetenskapsrådet (I.L., and K.T.), Swedish Foundation for Strategic Research (K.T.), the Swedish Brain Foundation – Hjärnfonden (K.T.), H.K.H. Kronprinsessan Lovisas förening för barnasjukvård och Stiftelsen Axel Tielmans minnesfond (A.A.), Strategic Research Area Neuroscience – StratNeuro (K.T.), The Swedish Foundation for International Cooperation in Research and Higher Education STINT (K.T.), Board of Research at Karolinska Institutet (K.T.), and the European Union – European Research Council (I.L.). Views and opinions expressed are, however, those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them. Open access funding is provided by Karolinska Institutet. The funders did not play any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Footnotes Supplementary data related to this article can be found at https://doi.org/10.1016/j.ebiom.2026.106291. Contributor Information Kristiina Tammimies, Email: kristiina.tammimies@ki.se. the Barwon Infant Study Investigator Group: David Burgner, Erica Grundell, Katherine Drummond, Len Harrison, Luba Sominsky, Martin O’Hely, Mimi Tang, Peter Sly, Richard Saffery, Samantha Dawson, and Toby Mansell Supplementary data Supplementary Figs. S1–S4 and Table S1 Supplementary Table S2 Supplementary Table S3 Supplementary Table S4 Supplementary Table S5 Supplementary Table S6 Study_Group_Members References Associated Data Supplementary Materials Supplementary Figs. S1–S4 and Table S1 Supplementary Table S2 Supplementary Table S3 Supplementary Table S4 Supplementary Table S5 Supplementary Table S6 Study_Group_Members