In 2005, the <exposome=, an idea that expands and brings together the already established <-ome= research fields, was introduced by Wild (2005) . The concept can be summarized as investigating the totality of chemicals that an individual is exposed to during their lifetime, and the consequent health-related effects (Miller and Jones 2014) . The main focus of current exposome research lies on xenobiotics and their biotransformation products, and can be further complemented with other -omic techniques, e.g. proteomics or metagenomics, to characterize the biological impact of these exogenous chemicals (Vitale et al. 2021, Kalia et al. 2022) .

Exposure to various xenobiotics during one´s lifetime, especially during the first thousand days , significantly influences human health and the development of chronic diseases (Heindel et al. 2015) . For example, exposure to endocrine disrupting chemicals, such as bisphenol A, have been associated with an increased risk of diseases like obesity and neurodevelopmental disorders (Braun 2017) . Early-life exposure to mycotoxins, such as aflatoxins and fumonisins, can adversely influence child growth (IARC 2015, Rasheed et al. 2021) . Conversely, exposure to other xenobiotic classes, for instance polyphenols, can have potentially beneficial health effects such as reduced risk for breast cancer due to soy isoflavones consumption in early life (Wu et al. 2008) . The diet of infants is a major source of exposure to both potentially toxic or beneficial xenobiotics. For example, consumption of complementary foods rather than breast milk can expose infants to higher levels of mycotoxins ( Krausová et al. 2023 ).

Exposome research has mainly focused on adverse exposures and often does not include molecules that might be able to mitigate these actions. One vital class of natural xenobiotics is polyphenols, which are prevalent in plant- and fungi-based foods. They can impact the microbiome by, for example, promoting the growth of beneficial bacteria (Gowd et al. 2019) . Additionally, polyphenols are bioactive compounds in humans, exhibiting properties that have shown to be either positive, like antimicrobial or anti-inflammatory properties (Pandey and Rizvi 2009) , or potentially negative, such as estrogenic effects of isoflavones ( KYíová 2019 ). Polyphenols can also reduce the toxicity of other xenobiotics, as seen with mycotoxins (Rasouli et al. 2022) . Therefore, assessing human exposure to polyphenols is essential. However, this is a challenging task as polyphenols are an immense class of molecules that, once ingested, yield a variety of biotransformation products, from both human and microbial metabolism (Oesterle et al. 2021) .

breast milk and matched infant stool samples. Details of study location and sample collection were previously described by Ayeni et al. (under review). The same set of samples was used before to investigate specific exposure classes: polyphenols in the breast milk samples (Berger et al. submitted), and mycotoxins in the stool samples (Krausová et al. 2022) and breast milk samples (Ayeni et al. under review). Briefly, samples were provided by eleven Nigerian infantmother pairs from Ilishan-Remo, Ogun state. Breast milk and infant stool samples were collected at the same time point by the mothers, and temporarily stored at 4°C prior to further cold storage. Samples were then stored at -20°C until analysis. Food frequency questionnaires were administered to the mothers to ascertain dietary patterns and health status of the infants (unpublished data). Ethical approval was obtained from the Ethical Committee of Babcock University (BUHREC 421/21R, BUHREC 466/23). All mothers were properly informed before providing their written consent to be included in the study.

consisting of an Agilent 1290 Infinity II UHPLC and a Sciex ZenoTOF 7600 MS was used to analyze the samples. The LC parameters used were previously optimized (Oesterle et al. 2022) . In brief, a Waters Acquity HSS T3 (2.1 x 100 mm, 1.8 ým) column with a Waters Vanguard precolumn at a temperature of 30°C, an autosampler temperature of 4°C, and an injection volume of 5 µL was used. The eluents were composed of H2O with 0.1% v/v FA as eluent A and ACN with 0.1% v/v FA as eluent B, at a flow rate of 0.6 mL/min, and the gradient given in Table A.3.

The MS was operated in negative polarity, and the source parameters consisted of a CAD gas of 9 arb. unit, a curtain gas of 35 psi, an ion sou rce gas 1 and 2 of 50 psi, a source temperature of 550°C, and a spray voltage of -4500 V. The TOF MS was operated with a scan window of m/z 100 to 1000, an accumulation time of 0.25 s, a declustering potential of -70 V, and a collision energy of -10 V. The SWATH parameters used involved scanning for fragments from m/z 100 to 1000 with 10 windows, a declustering potential of -70 V, an accumulation time of 0.05 s, and different collision energies for each window. The SWATH windows were optimized for each matrix separately using the total ion chromatogram from the injection of a pooled quality control and are listed in Table A.4 along with the collision energies applied to each window.

converted to ABF file format with Reifycs Analysis Base File Converter (v2011-2020) , before they were further processed in MS-Dial (v4.9.221218) (Tsugawa et al. 2015) . MS-Dial was used for feature pre-processing, e.g., building extracted ion chromatograms, and for feature annotation with a spectral library created by MSDial that combines various databases such as GNPS and MassBank (Tsugawa et al. 2015) . The parameters used in MS-Dial for the two biological matrices are listed in Table A.4. Features were further annotated with in silico fragmentation using MS-Finder (v3.52) (Tsugawa et al. 2016) . Prior to feature annotation, the feature lists were filtered by MS1 matching with the Exposome-Explorer (Neveu et al. 2020) and Phytohub (Giacomoni et al. 2017) databases. R (v4.3.1) (R Core Team 2023) was used for MS1 matching and feature clean-up with the process blank and pooled quality control (QC) samples.

annotation were given based on the levels previously defined by Schymanski et al. (2014) . In brief, features identified with authentic reference standards were labeled as Level 1 and features annotated with spectral libraries as Level 2a. Level 3 was then split in a similar manner as described in Oesterle et al. (2023) , and features annotated by in silico fragmentation were labeled as Level 3a, and features putatively annotated by their MS1 with the two databases as Level 3b. The chemical classes of the features were also determined using the ChemRICH MeSH prediction tool (Barupal and Fiehn 2017) , the classes listed in MS-Finder and MS-Dial, and the classes listed in the entries of the two online databases (PhytoHub and Exposome-Explorer).

biological matrix. The stool QC was prepared by pooling aliquots of each infants9 wet stool in respective quantities (Table A.2). The breast milk QC was prepared by combining 10 µL of each sample. The QC samples were processed following the same procedure as the experimental samples. For each biological matrix, the respective QC was used to condition the LC column prior to the acquisition of the experimental samples. Moreover, after column conditioning, three technical replicates of the QCs were measured. The QCs were routinely analyzed after every five experimental samples to continuously check the reliability of the instrument as well as to correct for any signal drifts that may occur within the acquisition batch. In addition, for each QC, a five-point serial dilution series with a constant dilution factor of four was prepared with the respective dilution solvent of the biological matrix9s sample preparation procedure. Moreover, matrixmatched calibration curves for the 242 reference standards were created with the QCs (Table A.5). For each biological matrix, along with the samples, a process blank was prepared by leaving the micro-reaction tube empty rather than taking an aliquot of the biological matrix. For the stool protocol, the process blank was assumed to contain 20 mg of dry weight, therefore 40 µL of H2O and 160µL of ACN:MeOH (1:1) + 1% v/v FA were added. Both process blanks were measured in triplicates.

Features with an average chromatographic peak area in the process blank measurements greater than one-third of the average in the samples were removed (Kirwan et al. 2014) . In addition, features were removed unless they were detected in at least two of the triplicate measurements of the QCs and their peak area in the QC triplicate measurements had a relative standard deviation <30% (Dudzik et al. 2018) . Features with an average signal-to-noise value of less than three, as calculated by MS-Dial, were also removed. Finally, the QC dilution series was developed to assess the reliability of extracted features, meaning a feature's signal in the QCs should decrease as the QC is diluted. Therefore, Spearman rank correlation was applied to the QC dilution series for assessing the association between a feature´s signal and the overall dilution factor, whereby features with a correlation value less than 0 were removed.

methods in negative mode yielded a total of 12192 features acquired in the breast milk samples and 16461 features in the stool samples.

After quality control, 4347 and 6905 features remained in breast milk and stool, respectively. A suspect screening workflow optimized for polyphenols but also capable of covering many other xenobiotics and endogenous metabolites (Oesterle et al. 2023) was then used to extract features of interest in the biological matrices.

This workflow involved a suspect list from two databases: PhytoHub, an online database containing 2268 phytochemicals, mainly polyphenols (Giacomoni et al. 2017) , and Exposome-Explorer, a database containing 1262 chemicals known to be biomarkers for exposure to environmental and lifestyle factors such as diet, pollutants, or contaminants (Neveu et al. 2020) . The application of this workflow resulted in a total of 542 matched features in the breast milk and 864 in the stool samples. From the 542 breast milk features, 34 were identified as Level 1, 51 annotated as Level 2a, 416 as Level 3a, and 41 as Level 3b. While from the 864 features in the stool, 68 were identified as Level 1, 78 annotated as Level 2a, 652 as Level 3a, and 66 putatively annotated as Level 3b. The annotated features are listed in Table A.6 for breast milk and Table A.7 for stool.

In the breast milk, many of the features detected were fatty acids, peptides, saccharides, or amino acids (Figure 1a). This was expected as breast milk is rich in nutrients, especially lipids , proteins, and carbohydrates, required for infant growth and development (Boudry et al. 2021) .

The breast milk also contained polyphenols such as flavonols, flavones, and isoflavones. Several studies have shown that polyphenols are abundant in human breast milk and they could be beneficial to the infants (Lu et al. 2021, Song et al. 2013, Carregosa et al. 2023) . In contrast, toxi c xenobiotics, such as mycotoxins, were detected at a low prevalence in breast milk. This class of xenobiotics are typically found at low concentrations in breast milk (Adejumo et al. 2013, Braun et al. 2020, Braun et al. 2022) .

Moreover, the heatmap of the features detected in the breast milk (Figure 1b) depicts that the features cluster in two main groups, though there is no clear separation of the types of features in each of the two groups.

More diverse chemical classes, predominantly of plant-based origin, were present in the stool (Figure 2a). One of the main chemical classes detected were polyphenols, which was expected as the LC-MS method was originally optimized for these analytes (Oesterle et al. 2022, Oesterle et al. 2023) . A wide range of different polyphenol classes were detected, including 55 flavonols, 32 flavones, 29 isoflavones, four chalcones, and 27 hydroxycinnamic acids. Besides polyphenols, several potentially toxic xenobiotics were also detected. For instance, one feature (m/z 254.952, retention time: 5.8 min, Level 3b) was putatively annotated as a polychlorinated biphenyl-16. It was previously reported that prenatal exposure to PCBs were associated with adverse effects on birth weight (Govarts et al. 2020) . In addition, naphthalene epoxide (Level 3a), a metabolite of naphthalene, was detected. Naphthalene exposure has been associated with various negative health implications, especially for respiratory health (Cakmak et al. 2014) . Several fatty acids were also detected in the stool samples, including arachidonic acid (Level 2a) milk samples obtained from Nigerian mothers at thre e different time points. B) Heatmap with Ward clustering displaying the features detected in all the breast milk samples. C) Box plots showing quantitative differences (Table A.8) over time for selected analytes identified at Level 1 that represent both potentially beneficial and toxic xenobiotics in the breast milk samples. The boxplots of all the identified features are shown in Figure B.2. samples from Nigerian infants at three different ti me points. B) Heatmap with Ward clustering displaying the feature s detected with the suspect screening workflow in the stool samples. C) Box plots of the semi-quantification (Table A.8) at the three distinct infant ages of several i dentified features (Level 1) that represent both po tentially beneficial and toxic xenobiotics in the stool samples. These plots show that com plementary foods increase exposure to various xenobiotics, and breast feeding keeps exposure leve ls low, despite potential lactational transfer. The box plots of all the identified features are shown in Figure B.3.

and eicosapentanaenoic acid (Level 3a). The levels of these two fatty acids were previously reported to be higher in stool of infants that were breastfed compared to infants fed that were formula-fed (Sillner et al. 2021) .

Similar to the breast milk heatmap, the features in the stool (Figure 2b) also form two clusters, with one cluster from the samples during the time of breastfeeding (month 1) and the other for those acquired after the introduction of complementary foods (months 6 and 12). In addition, many of the features detected in the stool samples were conjugated with sugar moieties. This could be attributed to phase II metabolism, as it was previously reported that metabolites of genistein included hexose and pentose conjugates in cells (Flasch et al. 2022a) ; or to their low intestinal absorption, as, for example, seen with the bioavailability of large polyphenols like proanthocyanidins (Scalbert et al. 2002) . Moreover, the bioavailability in infants may be further reduced as metabolic pathways, e.g. phase I and II metabolism (Lu and Rosenbaum 2014) , or their microbiome (Wopereis et al. 2014) are not yet fully developed.

were then compared with each other with a retention time deviation of 0.1 min and a mass error of 20 ppm. This resulted in a total of 149 features that were detected in both matrices, of which 32, 24, 91, and four were annotated as Levels 1, 2a, 3a, and 3b, respectively. One of the most prevalent chemical classes in both biological matrices was phenolic acids. Phenolic acids are a polyphenol class comprising of many human and microbial metabolites, including metabolites from other larger polyphenols such as anthocyanins (de Ferrars et al. 2014) . Besides phenolic acids, several other biotransformation products were detected in both matrices including caffeic acid-3-ß-D-glucuronide, daidzein-7-ß-Dglucuronide (Figure 3b), and genistein-7-ß-Dglucuronide. This observation suggests that the chemicals were either transferred directly from the breast milk to the infant and the corresponding stool samples or that the infant9s metabolism conjugated the parent compounds in the colon or liver by UDPglucuronosyltransferases (Rowland et al. 2013) . The overlap in features between the two matrices showed that xenobiotics, for example polyphenols, are transferred from the mother to the infant. Though scant data exist, the lactationa l transfer of various xenobiotics have been previously described in humans. This includes persistent organic pollutants (Haddad et al. 2015) , ellagitannins and their metabolites (Henning et al. 2022) , and pharmaceutical drugs and environmental pollutants (Dubbelboer et al. 2023) . In addition, the heatmap (Figure B.1) of all the features detected in both matrices showed that the two matrices are highly different, as only 149 from the 1259 total features are present in both matrices. This further underscores the differences previously seen in the types of chemical classes found in each matrix (Figure 1a and 2a).

antagonistic effects, correlations of the features in the breast milk and stool samples were explored.

Hence, Spearman rank correlation (p. adj < 0.05) was applied among the 34 identified features in the breast milk (Table A.9, Figure 5a), and the 68 in the stool (Table A.10, Figure 5b). In the breast milk samples, many of the polyphenols showed moderate to strong correlations ( ò = 0.37 - 0.88) (Table A.9). For example, the two microbial metabolites of matairesinol, enterolactone and enterodiol, strongly correlated with each other as expected (ò = 0.67). Additionally, polyphenols also correlated with other xenobiotics, e.g. ellagi c acid and dihydrocaffeic acid showed moderate correlation with alternariol monomethyl ether ( ò = 0.38 and 0.38, respectively). Estradiol-17glucuronide also showed moderate to strong correlations with numerous polyphenols, including enterolactone (ò = 0.41), hydroxybenzoic acid (ò = 0.58), and salicylic acid (ò = 0.70). It was previously observed in mice that enterolactone activates estrogensensitive reporter gene expression (Penttinen et al. 2007) . Hippuric acid showed negative correlation with other phenolic acids, such as benzoic acid (ò = -0.73), which is most likely due to hippuric acid being a metabolite of benzoic acid (Lees et al. 2013) .

4breast milk samples. B) Spearman rank correlation (p. adj < 0.05) heatmap between the 67 identified features (Level 1) in the stool samples.

All the identified features in the stool, except for one combination, positively correlated with one another, especially features within the same chemical class. For example, two lignan metabolites, enterodiol and enterolactone, moderately correlated with each other ( ò = 0.52); and two isoflavone metabolites, daidzein 7-ß-Dglucuronide and genistein 7-ß-D-glucuronide, showed strong correlation (ò = 0.82).

Additionally, phloretin strongly correlated with diosmetin (ò = 0.87), kaempferol ( ò = 0.81), and quercetin (ò = 0.81). Besides polyphenols, the potentially toxic xenobiotics: nivalenol, amoxicillin, mono-2-ethylhexyl-phthalate, alternariol monomethyl ether, trimethoprim, pnitrophenol, and florfenicol, showed moderate to strong correlations with one another ( ò = 0.40 0.90) (Table A.10). Of the xenobiotics, the highest correlation was recorded between mono2-ethylhexyl phthalate and p-nitrophenol ( ò = 0.90). Apart from individual effects, mixture toxicity of these xenobiotics can lead to more severe health consequences especially during early-life (Hamid et al. 2021, Krausová et al. 2023) . The only negative correlation was between alternariol monomethyl ether, an Alternaria mycotoxin, and genistein-7-sulfate, a metabolite of genistein (ò = -0.41). Previously, it was observed in vitro that genistein had antagonistic effects on the genotoxicity of alternariol, another Alternaria mycotoxin (Aichinger et al. 2017) .

Spearman rank correlation was further applied to explore possible correlations between the identified features in breast milk and infant stool. Only samples obtained at month 1 were selected because the infants predominantly consumed breast milk at this time point. No significant correlations were found (p. adj > 0.05) (Table A.11 and Figure B.4) which may be attributed to the low sample size (n = 10).

However, with the raw p-value (p < 0.05), many of the polyphenols in the breast milk positively correlated with polyphenols in the stool, especially between phenolic acids. For example, benzoic acid in breast milk strongly correlated with 4-hydroxybenzoic acid ò( = 0.79) in the infant stool. Alternariol monomethyl ether in the breast milk did not correlate with alternariol monomethyl ether in the stool, but showed strong negative correlation with nivalenol ( ò = -0.72).

Interestingly, alternariol monomethyl ether in the breast milk had strong negative correlations with several polyphenols in the stool. For example, alternariol monomethyl ether strongly negatively correlated with (-)-epicatechin ( ò = -0.98) and urolithin A (ò = -0.69). Urolithin C, also a metabolite of ellagitannins and structurally similar to urolithin A, was shown to influence the metabolism of alternariol in vitro (Crudo et al. 2021) . Overall, further investigations are necessary to access the toxicological potentials of many of the observed correlations in in vitro models.

correlation with Escherichia-Shigella (ò = -0.70).

Extracts containing vulgaxanthin I exhibited antibacterial activities against Gram-negative bacteria including Escherichia (Vuli et al. 2013) . The alkaloid, 1,3-dimethylurate, had strong negative correlations with Clostridioides amplicon sequencing (with their species and ASV ide ntifier given) and the features that showed significance by ANOVA (with their m/z, retention time, and identification level given). B) A network of the Spearman rank correlation results with microbes (with their genus-level classification and ASV identifier given) colored in green and features (with their m/z, retention time, and identification level given) i n orange.

(ò = -0.73) and Eggerthella (ò = -0.69).

Similarly, the terpenoids (+)-khusitene negatively correlated with Romboutsia (ò = 0.68) and pseudolaric acid H with Streptococcus (ò = -0.72). The negative correlations observed for alkaloids and terpenoids may be attributed to their antibacterial properties (Yang et al. 2020, Cushnie et al. 2014) . While the causal link between the features derived from xenobiotics and members of the infant gut microbiome remains to be elucidated, the results highlight the need to explore xenobiotic-microbiome interactions in less complex in-vitro models and its consequent health implications.

consider have become apparent in the applied analytical workflow and the overall study design. Many of the analytes investigated, especially polyphenols, have a wide range of isomers, making annotation of the features a complex task. This is reflected in the suspect screening results where many features (Table A.6 and A.7) have the same annotation though they were distinct analytes. Additionally, DIA MS 2 spectra typically have more noise than DDA MS 2 spectra (Guo and Huan 2020) , further complicating annotation. The MS 2 spectra with increased noise have a strong influence on in silico fragmentation, hence a high number of features annotated at level 3a is reported. The uncertainty in feature annotation also impacts the biological relevance of the statistical and correlation analyses. In addition, the number of participants in the study was small (n = 11) and as such, the correlation results must be interpreted with caution. Moreover, due to the low sample size, all time points were included in the correlations between the infant stool and gut microbiome. Therefore, the correlations may reflect indirect drivers such as the dynamic development of the microbiome in early life or changes in diet with age. A larger sample size and comparison among cohorts from other geographical settings would be needed for a more comprehensive analysis. Data is sparse on longitudinal metabolomic/exposomic profiles of healthy Nigerian mother-infant pairs anchored on highresolution mass spectrometric analysis of breast milk and matching stool samples. Thus, the data herein gives an important snapshot of the chemical exposome in biofluids of a selected population that has been considered to be at a relatively high exposure levels to many beneficial and adverse xenobiotics.

chemicals in early life is known to have a considerable impact on the development of humans. This study provides a comprehensive overview of potentially beneficial as well as potentially toxic xenobiotics in breast milk and stool from Nigerian mother-infant pairs. Several xenobiotics detected in the breast milk were also present in the corresponding stool samples, although the stool samples contained, as expected, a higher number of different xenobiotics. Correlations were observed between xenobiotics and certain members of the gut microbiome of the infants. Exposure to xenobiotics and their impact on the health of the infants significantly increased with the introduction of complementary foods. However, the toxicological relevance of these correlations needs to be further explored in larger cohorts and validated in in vitro models. Despite the limited sample size, the longitudinally study design and the advanced exposomic/metabolomic workflow applied allowed for the detailed assessment of the chemical exposome in breast milk and stool. The next steps should be the application of such workflows in larger cohorts and in different populations, especially in long-term studies, to better characterize the influence that exposure to various chemicals and their impact on health and microbiome development have.

and their infants for providing the samples. They would also like to thank and acknowledge all the members of their working groups, the Warth and Rompel labs, for their help, support, and feedback. The authors would like to express their gratitude to the Mass Spectrometry Center of the Faculty of Chemistry at the University of Vienna for technical support during the measurements, and to the Joint Microbiome Facility of the University of Vienna and the Medical University of Vienna. This work was supported by the University of Vienna through the Exposome Austria Research Infrastructure (B.W.), the Austrian Federal Ministry of Education, Science and Research (project DigiOmics4AT, B.W.), the Austrian Federal Ministry for Climate Protection, Environment, Energy, Mobility, Innovation and technology (BMK, B.W.), and the Austrian Science Fund (FWF) grant P32326 (A.R.).

the MetaboLights data repository (MTBLS8792). The 16S rRNA gene amplicon data is available on the BioProject accession number PRJNA1013123.