Methods of reducing occurrence of neurodevelopmental disorders

Abstract

The disclosure provides a method of reducing occurrence of a neurodevelopmental disorder (ND) in a subject at risk for developing an ND, the method comprising administering a composition comprising an equol-producing bacterium or gut-health-promoting bacterium to the subject.

Description

METHODS OF REDUCING OCCURRENCE OF NEURODEVELOPMENTAL DISORDERSBACKGROUND

[0001] Neurodevelopmental disorders (ND) exert profound and lasting impacts on central nervous system maturation, frequently emerging during childhood. They encompass conditions such as autism spectrum disorder (ASD), attention-deficit / hyperactivity disorder (ADHD), intellectual disabilities, and communication disorders. ASD, characterized by social difficulties and restricted and repetitive behaviors and interests, affects 1 -2% of the population, while ADHD’s prevalence reaches 7% in children 1 and 3.4% in adults2, leading to significant physical and mental health burdens3. Communication disorders affect 5-7% of people45and can cause problems with speaking, formulating words, and understanding abstract ideas.

[0002] Various prenatal and early life factors67may contribute to ND etiology, making early diagnosis challenging due to the lack of specific biomarkers. Genetic and environmental influences may disrupt neural adaptability, potentially causing abnormal neuronal homeostasis8, while immune dysregulation, inflammation, and autoantibodies may also contribute9 10. A substantial subset of individuals with ASD especially experience cooccurring gastrointestinal (Gl) symptoms, implicating a gut-brain connection11-13.

[0003] Gut bacteria play a significant role in metabolism, Gl health, neurological health14-16, and immune function17-21, suggesting their potential involvement in NDs. Gut-brain communication occurs various ways, including along the vagus nerve and through transport of short-chain fatty acids (SCFAs), cytokines, amino acids, and neurotransmitter precursors22. Microbes can locally synthesize neurotransmitters2324, and enteroendocrine cells release hormones and dendritic cells that modulate immune and microglia function, ultimately influencing neuroactive metabolites that impact behavior2526. Bacterial strains can improve social and anxiety-like behaviors62728, via impacts on gut permeability, the metabolome, and synaptic plasticity. Improvements in Gl and ASD-related symptoms were observed in a small pilot study of ASD-diagnosed children after microbiome transfer therapy31, with some long-lasting benefit observed up to two years later32. However, the exact contribution of microbiome disruptions to ND etiology and pathophysiology remains unclear, with divergent findings across cross-sectional studies on ASD31and ADHD32.

[0004] The connection between early gut microbiota and cognition remains poorly understood, with only three existing investigations and still no prospective microbiome study of medically documented NDs. These studies explored microbiome diversity and its correlation with ASQ-3 scores33and cognitive function in typically developing infants34, aswell as social behavior in three-year-olds35. However, none of these studies made a formal ND diagnosis the primary outcome.

[0005] While significant interest surrounds the microbiome, longitudinal prospective studies on early-life microbiomes and future ND diagnoses are lacking36.SUMMARY

[0006] The disclosure provides a method of reducing occurrence of a neurodevelopmental disorder (ND) in a subject at risk for developing an ND, the method comprising administering a composition comprising an equol-producing bacteria to the subject.

[0007] The disclosure also provides a method of reducing occurrence of a neurodevelopmental disorder (ND) in a subject at risk for developing an ND an undergoing treatment for otitis media, the method comprising administering a composition comprising Coprococcus to the subject.BRIEF DESCRIPTION OF THE FIGURES

[0008] Figure 1 . Early life environmental and medical risk factors implicated in neurodevelopmental disorders (ND). NDs across the ABIS cohort, with mean diagnosis age and 95% confidence interval.

[0009] Figure 2A-2L. Cord serum metabolomic and stool fatty acid differences in future ASD (ABISASD). (A-K) Polar metabolites (A-K) significantly enriched / depleted in the cord serum of ABISASD (n=27) versus controls (n=24). Controls were selected for ABISASD by propensity score matching (1 :1 nearest neighbor) on prenatal factors from the birth questionnaire (encompassing infectious disease or other infection, severe life event, smoking, caffeine intake, vitamins / minerals and iron supplements, consumption of milk, dairy, and eggs during pregnancy, education level of the mother and father) and gestational age, in weeks. Significance was determined by Kruskal Wallis in R. See also Table 1 . (L) Comparison of palmitic acid (16:0) molar percentage in stool samples at one year of age between ABISASD (n=23) and matched ABISControls (n=23). Targeted metabolomics employed mass spectrometry with selected reaction monitoring (SRM) for 22 fatty acids. Controls were selected for stool metabolomics via propensity score matching on biological sex and municipality, with no difference in age at stool collection (controls: 10.89 ±3.51 months, future ASD: 11.19 ±2.37, p=0.726).

[0010] Figures 3A-3F. Association of Citrobacter and Coprococcus prevalence, and Phascolarctobacterium abundance, with early otitis infection and future neurodevelopmental (ND) outcomes. Figure 3A: Odds ratios for the prevalence of Citrobacter and Coprococcus by early otitis and future NDs. Figure 3B: Prevalence of Citrobacter in the full dataset, byhistory of otitis and future NDs. Figure 3C and Figure 3E: Prevalence of Coprococcus (Figure 3C) and Citrobacter (Figure 3E) in the matched dataset, considering early otitis and future NDs. Figure 3D and Figure 3F: Difference in relative abundance of Phascolarctobacterium (Figure 3D) and Coprococcus (Figure 3F) in the matched dataset, based on early-life otitis infection.

[0011] Figures 4A and 4B. Preliminary results of normalized protein expression at birth, controls versus (Figure 4A) future autism (n=167), (Figure 4B) male subset, diagnosed prior to 13y (n=20).DETAILED DESCRIPTION

[0012] The present disclosure is based, at least in part, on the discovery that equol- producing bacteria present in the gut of a subject demonstrated a protective effect in subjects at risk for developing a neurodevelopmental disorder.

[0013] The data provided in the Exampls show an inverse correlation of certain commensal bacteria with ND progression, early-life Gl symptoms, mood issues, and HLA alleles linked to autoimmune disease, proposing a strong link between neurodevelopment, gut barrier function, and the immune system. Collectively, the data provided herein suggest that an inflammatory stage, mediated by gut bacteria, may contribute to ND risk very early in life.

[0014] Microbial and metabolomic imbalances that were observed at birth and one year have several important implications. First, that cord serum metabolome differences persisted after controlling for immunostimulatory events and psychosocial stressors underscores inherent biological variations at birth rather than just external influences. Notably, a decrease in crucial lipids, like linolenic acid (LA) and a-linolenic acid (ALA), bile acids, and triglycerides in neonates with future ASD, together suggest pro-inflammatory events present at birth. LA and ALA are precursors to long-chain polyunsaturated fatty acids (PUFAs) like docosahexaenoic acid (DHA, 22:6n-3), which have anti-inflammatory effects on the brain4950and regulate autophagy, neurotransmission, and neurogenesis51. Their modulation of the endocannabinoid system, through inhibition of neurotransmitter release (e.g., GABA), influences synaptic function and plasticity52. We observed cord serum triglyceride depletion in infants with future ASD, and these levels were found to be correlated with Bifidobacterium abundance at one year. During the last trimester, the fetal brain rapidly accumulates PUFAs, particularly DHA53, implicated in many brain disorders5455. Regarding bile acids, UDCA, also lower in future ASD, has shown therapeutic promise in conditions spanning metabolic disease, autoimmune disease, chronic inflammatory disease, and neuropathology56-59. Its ability to cross the blood-brain barrier is notable60, with anti-inflammatory and anti-apoptotic mechanisms possibly linked to dopamine and mitochondrial regulation57. Moreover, the environmental pollutant, PFDA, was significantly higher in future ASD. Per- and polyfluoroalkyl substances including PFDA exert significant influence on immune responses61. These substances are known to be associated with chronic inflammation, oxidative stress62, immune suppression, and possible involvement in autoimmune diseases63.

[0015] Second, the onset of gut microbial dysbiosis occurs very early in life (in infancy), and is significantly correlated with the disruption of essential metabolites (e.g., vitamins, fatty acids, and neurotransmitter precursors). Although acute, chronic Gl inflammation and increased intestinal permeability have been observed in ASD6465and first-degree relatives64, microbial differences were evident here in ABIS years before diagnosis, characterized by a depletion of anti-inflammatory microbes and fortifiers of the gut barrier. Recent research has demonstrated the remarkable potential of Akkermansia muciniphila in promoting intestinal health66, athough it has not been studied in neurodevelopment. Akkermansia muciniphila was absent in ABIS infants later diagnosed with ASD or comorbid ASD / ADHD and inversely correlated with Gl and mood symptoms in early childhood. Intriguingly, Akkermansia was not associated with future ADHD, suggesting that disruptions in mucin health have a more robust connection to ASD. Akkermansia muciniphila facilitates mucin, produces folate67, propionate, and acetate6869, is known for enhancing enterocyte monolayer integrity and fortifying compromised gut barriers70, and possesses immunomodulatory properties71. Microbiome-derived 3,4-dihydroxy-phenyl-propionic acid, an epigenetic modifier that downregulates IL-6 cytokine production72, showed a correlation with Akkermansia.Bifidobacterium was also depleted in future NDs, across conditions. Bifidobacteria promote healthy immune responses and enhance dopamine production by elevating phenylalanine32. In stool, Bifidobacterium abundance correlated with 4-hydroxyphenyllactate, which is a metabolite of tyrosine - a catecholamine precursor associated with cognitive function7374. Species of Coprococcus75, Akkermansia, Roseburia, and Turicibacterwere inversely associated with mood symptoms at age five. Coprococcus have potent anti-inflammatory properties and, in ABIS, were inversely associated with NDs and linked to protective factors, including lower vulnerability scores, fewer antibiotics (none in the first year), and infant diets with fewer snacks. Roseburia also possesses potent anti-inflammatory properties, playing roles in colonic motility and the immune system76. Turicibacter sanguinis plays roles in serotonin utilization77and pathways of steroid and lipid metabolism.

[0016] Although several key butyrate producers68were inversely associated with future NDs ( Faecal i bacterium prausnitzii75'75, Roseburia5579, Anaerostipes79and Acidaminococcales), no marked difference in butyrate levels were observed in stool(p=0.38). However, it isn’t known whether butyrate turnover rates differed between controls and future cases. Citrate levels were higher in future ASD, consistent with cross-sectional study of children with ASD, finding increased succinate and citrate in urine80, while uridine levels in stool positively correlated with Faecalibacterium. Treatment with uridine can promote tissue regeneration and repair by metabolic adaptation, improving mitochondrial activity81, and may reduce inflammation and oxidative stress82. Acidaminococcales suppresses inflammation and its low abundance aligns with a cross-sectional study of ASD83.

[0017] Several known equol producers were consistently higher in controls, including Slackia and Adlercreutzia equolifaciens. Coriobacteriaceae, a family involved in lipid metabolism84and equol production, were also depleted in infants with future NDs. Eggerthella and Slackia positively influence host lipid and xenobiotic metabolism85. A significant decrease was observed in a potential equol signal in future ASD cases. Equol has been studied for anti-inflammatory effects and its estrogenic activity. In preclinical models, equol exhibits blood-brain barrier permeability capability86, anti-neuroinflammatory properties, and neuroprotective effects, protecting microglia against LPS-induced inflammation86and neurotoxins87.

[0018] Third, the dysbiosis that was observed persisted even after adjusting for environmental factors, like infant diet, psychosocial vulnerability (i.e., maternal smoking during pregnancy), and antibiotic use. While diets can vary during the transition from breast milk to solid foods, it’s improbable that diet alone can account for the substantial differences we observed as the microbial DNA was derived from one-year-old infants and findings persisted in light of diet differences. Thus, collectively, the findings here address a critical gap in literature studying children already diagnosed with NDs88. The findings indicate that dysbiosis is not solely a result of post-diagnosis dietary changes but rather existed prior to diagnosis, providing valuable insights into the early development of these conditions. In many cases, the infant microbiome differences pointed to early mood and Gl symptoms, as well, at 2.5 to 5 years. Both adjusted and unadjusted approaches showed enrichment of some bacteria in controls (e.g., Coprococcus eutactus, Akkermansia muciniphila, Blautia obeum) and cases (e.g., Corynebacterium variabile). Corynebacterium spp. are generally considered pathobionts, some susceptible only to vancomycin or aminoglycosides89, and have been linked to increased respiratory infection in newborns delivered by C-section90. Certain Bifidobacterium breve ASVs lost significance after adjustment, as well as Ruminococcaceae CAG-352, Adlercreutzia equolifaciens, and Roseburia species, suggesting their role in NDs is heavily mediated by environmental factors.

[0019] Fourth, the extent of the microbial dysbiosis seen in ABIS infants may be linked to the effects of HLA, stress, and infection / repeated antibiotics in childhood. Immunostimulatory events and psychosocial stressors have lasting effects on fetal health, development, and immunity, the former disrupting cell differentiation, migration, and synaptic maturation37. In ABIS, prenatal and early-life stress and psychosocial adversity heightened ND risk, aligning with other studies91 92. In preclinical models, stress-induced prenatal events activates GABAergic delay, mediated by proinflammatory cytokine interluekin-6 (IL-6) and impacting adult microglia93. Repeated infections and antibiotic exposures in early life were also a significant risk factor across the first several years of life. C-section was a risk for speech disorder and intellectual disability. Although ABIS lacks intrapartum antibiotic data, we suspect all caesarean births (1 1 .9% of births in ABIS) entailed antibiotics, as well94. HLA, which significantly impacts the immune system and, especially, microbial interactions9596, is often overlooked in neurodevelopmental research. Its contribution to synaptic function, central nervous system development, and neurological disorders is increasingly demonstrated97. Nearly thirty years ago, DR / 31 *0401 was first implicated in ASD98. Here, the following was observed: DR4-DQ8 homozygosity (linked to autoimmune diseases including CD and T 1 D) increased ND and ASD risk by 1 .8- and 2.8-fold. Children heterozygous for DR3-DQ2 / DR4-DQ8 or homozygous for DR4-DQ8 shared deficits in Roseburia, Phascolarctobacterium faecium, and Coprococcus species, as did infants with future NDs.

[0020] Young children later diagnosed with ASD or exhibiting significant autism traits tend to experience more ear and upper respiratory symptoms99. In ABIS, infants who had otitis in their first year were found to be more prone to aquiring NDs if they lacked detectable levels of Coprococcus or harbored Citrobacter. The absence of Coprococcus, despite comparable levels in controls irrespective of otitis, raises questions about microbial community recovery. This potential failure of the microbiome to recover following such events may serve as a mechanism connecting otitis media to ND risk. Moreover, antibiotic-resistant Citrobacter'00was more prevalent in these infants. The presence of strains related to Salmonella and Citrobacter, labeled in this investigation as SREB, was signfiicantly higher in infants who later developed comorbid ASD / ADHD (21%), compared to controls (3%). This disruption may have consequence on neurodevelopment during a critical period. Salmonella and Citrobacter have shown the ability to upregulate the Wingless (Wnt) signaling. The Wnt pathway is vital for immune dysregulation and brain development, and its disruption has been implicated in ASD pathogenesis101-104.

[0021] Lastly, early dysbiosis points to disruption of several metabolites in stool, including amino acids, fatty acids, vitamins, and neurotransmitter precursors. In ABIS, significant depletions of semi-essential amino acid L-arginine and essential amino acid lysine wereobserved in infants with future ASD, aligning with amino acid disruptions in children already diagnosed105 106. In ABIS, positive correlations between L-arginine and Roseburia, Coprococcus, and Akkermansia abundances were observed. L-arginine’s association with improved innate immune response and barrier function has been documented107. Both lysine and arginine are critical for growth metabolism, immune function, histone modifications, and the production of nitric oxide. Lysine is involved in fatty acid metabolism, calcium absorption, immune function, and protein deposition, with high rates of metabolism in splanchnic tissues108. Its degradation pathway is closely linked to that of an important neurotransmitter precursor, tryptophan109. Arginine, a semi-essential amino acid, is vital for infant health, contributing to cardiovascular, neurological, and intestinal functions110, including neurogenesis regulation111. Depletion of arginine, especially during bacterial challenges, substantially hampers neonates’ capacity to generate an adequate immune response, thereby elevating susceptibility to infections, particularly those originating in the gastrointestinal tract112. Promising effects of amino acid supplementation have been demonstrated in premature neonates113 114.

[0022] Two fatty acid differences were notable in the stool of future ASD versus controls: omega-7 monounsaturated palmitoleic acid, (9Z)-hexadec-9-enoic acid (below the level of detection in 87.0% of future ASD but present in 43.5% of controls) and palmitic acid (elevated in future ASD). Palmitoleic acid has been associated with a decreased risk of islet and primary insulin autoimmunity115. Conversely, palmitic acid, a saturated fatty acid, has been linked to neuronal homeostasis interference116. Its effects are partially protected by oleic acid116, which although approaching significance, was lower in the cord serum of future ASD.

[0023] Few metabolites were higher in stool of infants with future ASD, but there are a few notable examples: a-d-glucose, pyruvate, and 3-isopropylmalate. Coprococcus inversely correlated with 3-isopropylmalate, suggesting gut-brain connections117and a possible imbalance in BCAA pathways given the role of 3-isopropylmalate dehydrogenase in leucine and isoleucine biosynthesis118. An increase in dehydroascorbate suggests potential disruptions in vitamin C metabolism, crucial for neurotransmitter synthesis and antioxidant defense, and elevated pyruvate, early disturbance of neurotransmitter synthesis or energy production. Pimelic acid elevation, found in disorders of fatty acid oxidation119, suggests disruption of mitochondrial pathways for fatty acid oxidation.

[0024] Akkermansia and Coprococcus, absent or reduced in infants with future NDs, positively correlated with signals in stool representing neurotransmitter precursors and essential vitamins in stool. Specifically, Akkermansia correlated with tyrosine and tryptophan (i.e., catecholamine and serotonin precursors, respectively) and Coprococcus with riboflavin.Disruption of BCAA metabolism in ASD has been documented120 121, involving coding variants in large amino acid transporters (LAT) and reduced utilization of trypotphan and large aromatic amino acids120, along with increased glutamate and decreases in tyrosine, isoleucine, phenylalanine, and tryptophan in children with ASD121. Oxidative stress, a diminished capacity for efficient energy transport121, and deficiencies in vitamins (like vitamin B2) essential for neurotransmitter synthesis and nerve cell maintenance have been implicated122. Riboflavin, as an antioxidant, reduces oxidative stress and inflammation123, demonstrating neuroprotective benefits in neurological disorders124, possibly through maintenance of vitamin B-6125, necessary for glutamate conversion to glutamine and 5- hydroxytryptophan to serotonin.

[0025] Together, these findings support a hypothesis of early-life origins of NDs, mediated by gut microbiota.

[0026] In one aspect, described herein is method of reducing occurrence of a neurodevelopmental disorder (ND) in a subject at risk for developing an ND, the method comprising administering a composition comprising an equol-producing bacterium or guthealth-promoting bacterium to the subject.

[0027] Exemplary equol-producing bacteria include, but are not limited to Slackia, Coriobacteriaceae, Eggerthella, and Adlercreutzia equolifaciens. Exemplary gut-health promoting bacteria include, but are not limited to, Akkermansia muciniphila, Roseburia horn inis, Faecalibacterium prausnitzii, Erysipelotrichiaceae UCG-003 spp., Alistipes putredinis, Phascolarctobacterium, Turicibater, Anaerostipes caccae, Coprococcus, Acidaminococcales, and Bifidobacterium sp.

[0028] In some embodiments, the neurodevelopmental disorder is autism, attention- deficit / hyperactivity disorder, communication disorders, or intellectual disability. In some embodiments, the neurodevelopmental disorder is an autism spectrum disorder.

[0029] In another aspect, described herein is a method of reducing occurrence of a neurodevelopmental disorder (ND) in a subject at risk for developing an ND an undergoing treatment for otitis media, the method comprising administering a composition comprising Coprococcus to the subject. In some embodiments, the Coprococcus is an amoxicillinresistant strain of Coprococcus.

[0030] In some embodiments, the subject is human. In some embodiments, the subject is an infant (e.g., less than 12 months in age). In some embodiments, the subject is a child (i.e., > 12 months and < 18 years in age). In some embodiments, the subject is an adult (> 18 years in age).

[0031] In some embodiments, methods comprise identifying a subject (e.g., a child) as a subject at risk for developing an ND (e.g., autism). As demonstrated in the Examples, elevated levels of HEXIM1 and NLIDC in a cord blood were present in children that eventually developed autism compared to healthy controls. As such, a child identified as having elevated levels of HEXIM1 and / or NLIDC in cord blood compared to healthy controls would be identified as a child being at risk for developing autism.

[0032] The determining step of the methods described herein optionally comprises comparing the measurement of HEXIM 1 and / or NUDC in the cord blood of a subject to a reference measurement of HEXIM 1 and / or NUDC and scoring the measurement from the sample as elevated based on statistical analysis or a ratio relative to the reference measurement. In some embodiments, the reference measurement comprises at least one of the following (a) HEXIM 1 and / or NUDC protein level(s) in a cord blood sample from a subject that did not develop autism, (b) HEXIM1 and / or NUDC protein level(s) in a collection of comparable cord blood samples from subjects that did develop autism; or (c) HEXIM1 and / or NUDC protein level(s) in an arbitrary standard optionally further including statistical distribution information for the multiple measurements, such as standard deviation. In some embodiments, the methods described herein comprise comparing the level of a protein biomarker described herein (i.e., HEXIM1 and / or NUDC) in a blood sample from the subject to the level of the protein biomarker in a cord blood sample from a healthy control, or from a reference control set of comparable subjects who did not develop autism, wherein an elevated level compared to the reference standard identifies the subject as a subject that would benefit from treatment with the composition comprising an equol-producing bacterium as described herein.

[0033] Detection of HLA alleles

[0034] Detection of HLA alleles can be used to determine whether a subject, such as, for example, a human subject, will respond to administration of an equol-producing or guthealth-promoting bacterial strain. This is especially relevant to those possessing DR4-DQ8, which was found to be more prevalent in children with a future neurodevelopmental disorder diagnosis. It was found that children heterozygous for DR3-DQ2 / DR4-DQ8 or homozygous for DR4-DQ8 shared deficits in Roseburia, Phascolarctobacterium faecium, and Coprococcus species, which may also be deficient in those with a future neurodevelopmental disorder. Assays such as RT-PCR, PCR, qPCR, DNA and RNA sequencing, microarray analysis and any other genome-based analyses known in the art, along with any suitable immunoassay, may be used to detect an HLA allele described herein in a sample from the subject, for example, stool, plasma, serum, cerebrospinal fluid, sputum, saliva, breast milk, tears, bile, semen, vaginal secretion, amniotic fluid, urine, stool,leukocytes, bone marrow cells, buccal cells, fibroblasts and lung or other tissue biopsies. In some embodiments, the sample is a stool sample. In addition, such analyses may be qualitative or quantitative.

[0035] Subjects may be screened for the presence of an HLA allele through the use of quantitative reverse transcriptase-polymerase chain reaction (qRT-PCR), which utilizes competitive techniques employing an internal homologous control that differs in size from the target, for example, by a small insertion or deletion. Non-competitive and kinetic quantitative PCR or RT-PCR may also be used. Experiments may combine real-time, kinetic PCR or RT- PCR detection together with an internal homologous control that can be simultaneously detected alongside the target sequences. In one non-limiting embodiment, real time quantitative PCR may provide the capability of measuring the level of an HLA allele gene product amplified through PCR. Quantitative PCR may require only a nominal amount of a sample to perform such experiments.

[0036] Quantitative amplification is based on the monitoring of a signal (e.g., fluorescence of a probe) representing copies of a template in cycles of an amplification (e.g., PCR) reaction. In the initial cycles of the PCR, a very low signal is observed because the quantity of the amplification product formed does not support a measurable signal output from the assay. After the initial cycles, as the amount of formed amplification product increases, the signal intensity increases to a measurable level and reaches a plateau in later cycles when the PCR enters into a non-logarithmic phase. Through a plot of the signal intensity versus the cycle number, the specific cycle at which a measurable signal is obtained from the PCR reaction can be deduced and used to back-calculate the quantity of the target before the start of the PCR. The number of the specific cycles that is determined by this method is typically referred to as the cycle threshold (Ct). Exemplary methods are described in, e.g., U.S. Pat. Nos. 6,180,349; 6,033,854; and 5,972,602.

[0037] In some embodiments, nucleic acid microarrays, or gene chip technology, may be used to screen and identify subjects having a protective (or non-protective) HLA haplotype allele. A "microarray" is an array of distinct polynucleotides, oligonucleotides, polypeptides, peptides, or antibodies affixed to a substrate, such as paper, nylon, or other type of membrane; filter; chip; glass slide; or any other type of suitable support.

[0038] In some embodiments, the microarray technology involves the positioning of highly condensed and ordered arrays of nucleic acid probes, for example, DNA oligonucleotides, on a substrate, for example, a glass slide or nylon membrane. Each oligonucleotide may comprise a nucleotide sequence that is complementary to a portion of an HLA allele, wherein the oligonucleotide can be placed on a single glass slide or nylon membrane. Forexample, and not by way of limitation, up to 50,000 DNA fragments, may be placed on a single glass slide and up to 5,000 placed on a nylon membrane. The resulting microarrays can then be used to screen for the presence an HLA allele transcription product expressed in a sample to be screened.

[0039] In some embodiments, a nucleic acid microarray may be utilized by preparing labeled nucleic acid from a sample to be screened and hybridized such labeled nucleic acid with the array. In addition, labeled nucleic acid of a designated control sequences may be prepared (or in the event that the array is sold as part of a kit, could be supplied to the user). Radioactive, colorimetric, chemiluminescent or fluorescent tags may be used for labeling of nucleic acid sequences from the sample and for the control. Numerous techniques for scanning arrays, detecting fluorescent, chemiluminescent, or colorimetric output, are known in the art and may be used for detecting hybridization of a nucleic acid from a test sample to the microarray. For example, a low-cost, high-throughput fluorescent microarray scanning system (ScanArray®, PerkinElmer Life And Analytical Sciences, Inc., Waltham, MA, USA), or a colorimetric microarray scanner (Arraylt® SpotWare™, TeleChem International, Inc., Sunnyvale, CA, USA) may be used. Numerous protocols for the preparation of labeled nucleic acid sequences are publicly available and may be used.

[0040] In some embodiments, the identifying step comprises screening for an HLA allele through sequencing (i.e. determining the nucleotide sequence of a given DNA or RNA fragment) of a genomic DNA or HLA allele expression product present in a sample taken from the subject. Any sequencing methods known in the art may be used to determine the nucleotide order of the HLA allele DNA or RNA. For example, but not by way of limitation, an HLA allele may be identified by first performing PCR and then sequencing the product of PCR to determine the specific allele (see Listgarten et al., 2008, PloS omput. Biol. 4(2):e 1000016).

[0041] The sequencing of a nucleic acid sample (i.e., determining the nucleotide order of a given DNA or RNA fragment) is not limited to any one technique. The present invention contemplates the use of any sequencing technique known in the art.

[0042] In some embodiments, the identifying step comprises the use of antibodies in an immune assay to detect protein isoforms resulting from expression of a protective (and optionally non-protective) HLA haplotype allele in a sample from the subject. Exemplary immunoassays include, but are not limited to, enzyme linked immunosorbent assays (ELISAs), Western blots and radioimmunoassays (RIA).

[0043] Compositions and Routes of Administration

[0044] The bacteria described herein (e.g., equol-producing bacteria or Coprococcus, and gut-health-promoting bacteria such as Akkermansia muciniphila) can be for any suitable route of administration. For example, the bacteria described herein (e.g., equol-producing bacteria or Coprococcus, and gut-health-promoting bacteria such as Akkermansia muciniphila) is formulated into a composition that can be administered to the subject via oral administration, rectal administration, transdermal administration, intranasal administration or inhalation. In some embodiments, the equol-producing bacterium is administered to the subject orally.

[0045] Suitable oral dosage forms for the compositions described herein include tablets, capsules, solutions, suspensions, syrups, lozenges, and dry powders. Tablets can be made using compression or molding techniques well known in the art. Gelatin or non-gelatin capsules can prepared as hard or soft capsule shells, which can encapsulate liquid, solid, and semi-solid fill materials, using techniques well known in the art.

[0046] In some embodiments, the composition further comprises an acceptable carrier. The term "carrier" includes, but is not limited to, lipids, phospholipids, salts, emulsifiers, excipients, diluents, preservatives, binders, lubricants, disintegrators, swelling agents, fillers, stabilizers, and combinations thereof. Polymers used in the dosage form include hydrophobic or hydrophilic polymers and pH dependent or independent polymers.Exemplary hydrophobic and hydrophilic polymers include, but are not limited to, hydroxypropyl methylcellulose, hydroxypropyl cellulose, hydroxyethyl cellulose, carboxy methylcellulose, polyethylene glycol, ethylcellulose, microcrystalline cellulose, polyvinyl pyrrolidone, polyvinyl alcohol, polyvinyl acetate, and ion exchange resins. “Carrier” also includes all components of the coating composition, which may include plasticizers, pigments, colorants, stabilizing agents, and glidants.

[0047] The compositions described herein can be prepared using one or more excipients, including diluents, preservatives, binders, lubricants, disintegrators, swelling agents, fillers, stabilizers, and combinations thereof.

[0048] In some embodiments, the composition comprises at least about 104colony forming units (cfu) of a bacteria described herein (e.g., equol-producing bacteria or Coprococcus) . In some embodiments, the composition comprises at least about 104, 105, 106, 107, 108, 109, 1010, 1011, 1012, or 1013cfu, including ranges between any of the listed values, for example 104-108cfu, 104-109cfu,104-1010cfu,- 104-1011cfu, 104-1012cfu, 104-1013cfu, 105-108cfu, 105-109cfu, 105-101° cfu, 105-1011cfu, 105-1012cfu, 105-1013cfu, 106-108cfu, 106-109cfu, 106-101° cfu, 106-1011cfu, 106-1012cfu, 106-1013cfu, 107-108cfu, 107-109cfu, 107-101° cfu, 107-1011cfu, 107-1012cfu, 107-1013cfu, 108-109cfu, 108-101° cfu, 108-1011cfu, 108-1012cfu, or 108-1013cfu of abacteria described herein (e.g., equol-producing bacteria or Coprococcus).

[0049] In any of the methods of preparing a probiotic composition described herein, the method may further comprise administering the composition to a subject or ingestion by the subject.

[0050] In some embodiments, the compositions described herein is incorporated into a food product. The term “food product” as used herein refers to any substance containing nutrients that can be ingested by an organism to produce energy, promote health and wellness, stimulate growth, and maintain life. Exemplary food products include, but are not limited to, baked goods (cakes, cookies, crackers, breads, scones and muffins), dairy-type products (including but not limited to cheese, yogurt, custards, rice pudding, mousses, ice cream, frozen yogurt, frozen custard), desserts (including, but not limited to, sherbet, sorbet, water-ices, granitas and frozen fruit purees), spreads / margarines, pasta products and other cereal products, meal replacement products, nutrition bars, trail mix, granola, beverages (including, but not limited to, smoothies, water or dairy beverages and soy-based beverages), and breakfast type cereal products such as oatmeal.

[0051] In some embodiments, the food product is a meal replacement product. The term “meal replacement product” as used herein refers to a food product that is intended to be eaten in place of a normal meal. Nutrition bars and beverages that are intended to constitute a meal replacement are types of meal replacement products. The term also includes products which are eaten as part of a meal replacement weight loss or weight control plan, for example snack products which are not intended to replace a whole meal by themselves, but which may be used with other such products to replace a meal or which are otherwise intended to be used in the plan. These latter products typically have a calorie content in the range of from 50-200 kilocalories per serving.

[0052] All of the U.S. patents, U.S. patent application publications, U.S. patent applications, foreign patents, foreign patent applications and non-patent publications referred to in this specification and / or listed in the Application Data Sheet, are incorporated herein by reference, in their entireties.

[0053] From the foregoing it will be appreciated that, although specific embodiments of the invention have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the invention.EXAMPLES

[0054] Materials and Methods:

[0055] Human Participants: ABIS is a prospective, general population-based cohort study to which parents of all children born in southeast Sweden during the period 1 October 1997- 1 October 1999 were invited. Of the 21 ,700 families asked, 17,055 agreed to participate (78.6%), giving informed consent after receiving oral and written information. The original motivation behind ABIS was to prospectively study the etiology of immune-mediated diseases and the effects of environmental and genetic factors. The children have been followed from birth. Participating parents completed questionnaires at birth and at 1 , 2-3, 5, 8-10, 14-16, and 17-19 years, with diaries kept during the first year of life126. ABIS children also completed questionnaires at 8, 10-12, and 17-19 years. Data include, but are not limited to, family medical history, antibiotic use, medications, diet, lifestyle, mood and disposition, home environment, environmental exposures, and psychosocial vulnerability.

[0056] For this investigation, we analyzed the earliest questionnaires derived from parents of ABIS children at one, three, and five years of age. The latest diagnoses considered in this investigation were obtained when the children were 21 to 23 years of age. The ABIS cohort showed a relatively balanced distribution of biological sex (48.2% female, 51 .8% male).Although data on race, ancestry, and ethnicity were not collected, we anticipate homogeneity given that 89.2% of ABIS children had both parents born in Sweden, 7.7% had one parent born outside of Sweden, and only 3.1% had both parents born outside of Sweden. In the late 1990s, the majority of the population consisted of Ethnic Swedes, followed by Finns.

[0057] Regarding socioeconomics, 6.6% of ABIS children exhibited the greatest psychosocial vulnerabilities (vulnerability index scores >2). Disposable household income for ABIS families in 2000, 2006, and 2012 was reported as 305,360.7 SEK (95% Cl: 297,279- 313,462.5), 41 1 ,253.2 SEK (408,306.9-414,199.5), and 554,889.7 SEK (549,889.1 - 559,890.3), respectively.

[0058] Given the influence of biological sex on diagnosis, especially in boys, we explored microbiome associations stratified by sex and age of diagnosis (ASD only). Caution is warranted due to the limited sample size after this stratification. The compounded influence of biological sex on cohort-wide risk factors was not explicitly considered and may be a limitation to generalizability. It is important to note that, by design, the ABIS cohort consists of individuals from Sweden, and therefore, the findings may not be directly applicable to more diverse or non-European populations. Further research in populations with varied demographics is warranted to enhance the external validity of these findings.

[0059] Diagnoses: International Classification of Diseases diagnoses (ICD-10) were derived from the National Patient Register (NPR)127, capturing diagnoses through December 2020. Psychiatric diagnoses were set according to the Diagnostic and Statistical Manual ofMental Disorders, fourth edition (DSM-IV128) or fifth edition (DSM-V130), depending on diagnosis date (Figure 1 B). Qualifying diagnosis codes included ASD (F84.0, F84.1 , F84.2, F84.3, F84.4, F84.5, F84.8, F84.9), ADHD (F90.0, F90.0A, F90.0B, F90.0C, F90.0X), speech disorders (F80.0, F80.0A, F80.0B, F80.0C, F80.1 , F80.1A, F80.1 B, F80.1 C, F80.2, F80.2A, F80.2B, F80.2C, F80.3, F80.8, F80.8A, F80.8B, F80.8C, F80.8D, F80.8W, F80.9), and intellectual disability (F70.0, F70.1 , F70.8, F70.9, F71.0, F71.1 , F71.8, F71.9, F72.0, F72.1 , F72.8, F72.9, F73.0, F73.1 , F73.8, F73.9, F78.0, F78.1 , F78.8, F78.9, F79.0, F79.1 ,F79.8,F79.9, F83.9).

[0060] Acquisition of cord serum metabolites: Cord serum samples were obtained from maternal donations, with a total of 120 samples from mothers whose infant also donated stool at one year. The analysis encompassed two methods: lipidomics and hydrophilic (water-soluble) metabolite profiling. The latter included examination of free fatty acids, bile acids and amino acids. Bile acids measured included 12-oxo-litocholic acid, 7-oxo- deoxycholic acid, 7-oxo-hyocholic acid, beta-muricholic acid, chenodeoxycholic acid, cholic acid, deoxycholic acid, dihydroxycholestanoic acid, glycochenodeoxycholic acid, glycocholic acid, glycodehydrocholic acid, glycodeoxycholic acid, glycohyocholic acid, glycohyodeoxycholic acid, glycolitocholic acid, glycoursodeoxycholic acid, hyocholic acid, hyodeoxycholic acid, litocholic acid, omega / alpha-muricholic acid, perfluorooctanoic acid, tauro-alpha-muricholic acid, tauro-beta-muricholic acid, taurochenodeoxycholic acid, taurocholic acid, taurodehydrocholic acid, taurodeoxycholic acid, taurohyodeoxycholic acid, taurolitocholic acid, tauro-omega-muricholic acid, tauroursodeoxycholic acid, trihydroxycholestanoic acid, and ursodeoxycholic acid. Free fatty acids included decanoic acid, myristic acid, linolenic acid, palmitoleic acid, linoleic acid, eicosapentaenoic acid, palmitic acid, oleic acid, stearic acid, and arachidic acid. Polar metabolites included fumaric acid, glutamic acid, aspartic acid, malic acid, phenylalanine, ferulic acid, citric acid, tryptophan, 3-indoleacetic acid, 3-hydroxybutyric acid, isovaleric acid, indole-3-propionic acid, salicylic acid, isocaproic acid, and succinic acid, serine, threonine, glutamine, proline, valine, lysine, methionine, syringic acid, and isoleucine / leucine.

[0061] Lipidomic analysis of cord serum: A total of 360 cord serum samples were randomized and analyzed as described below. 10 pl of serum was mixed with 10 pl 0.9% NaCI and extracted with 120 pl of CHCh: MeOH (2:1 , v / v) solvent mixture containing internal standard mixture (c = 2.5 pg / ml; 1 ,2-diheptadecanoyl-sn-glycero-3-phosphoethanolamine (PE(17:0 / 17:0)), N-heptadecanoyl-D-erythro-sphingosylphosphorylcholine (SM(d18:1 / 17:0)), N-heptadecanoyl-D-erythro-sphingosine (Cer(d18:1 / 17:0)), 1 ,2-diheptadecanoyl-sn-glycero- 3-phosphocholine (PC(17:0 / 17:0)), 1 -heptadecanoyl-2-hydroxy-sn-glycero-3-phosphocholine (LPC(17:0)) and 1-palmitoyl-d31 -2-oleoyl-sn-glycero-3-phosphocholine (PC(16:0 / d31 / 18:1 ))and, triheptadecanoylglycerol (TG(17:0 / 17:0 / 17:0)). The samples were vortexed and let stand on the ice for 30 min before centrifugation (9400 ref, 3 min). 60 pl of the lower layer of was collected and diluted with 60 pl of CHCI3: MeOH. The samples were kept at -80 °C until analysis.

[0062] The samples were analyzed using an ultra-high-performance liquid chromatography quadrupole time-of-flight mass spectrometry (UHPLC-QTOFMS from Agilent Technologies; Santa Clara, CA, USA). The analysis was carried out on an ACQUITY UPLC BEH C18 column (2.1 mm x 100 mm, particle size 1.7 pm) by Waters (Milford, USA). Quality control was performed throughout the dataset by including blanks, pure standard samples, extracted standard samples and control plasma samples. The eluent system consisted of (A) 10 mM NH4AC in H2O and 0.1% formic acid and (B) 10 mM NH4AC in ACN: IPA (1 :1 ) and 0.1% formic acid. The gradient was as follows: 0-2 min, 35% solvent B; 2-7 min, 80% solvent B; 7-14 min 100% solvent B. The flow rate was 0.4 ml / min.

[0063] Quantification of lipids was performed using a 7-point internal calibration curve (0.1-5 pg / mL) using the following lipid-class specific authentic standards: using 1 -hexadecyl- 2-(9Z-octadecenoyl)-sn-glycero-3-phosphocholine (PC(16:0e / 18:1 (9Z))), 1 -(1Z-octadecenyl)- 2-(9Z-octadecenoyl)-sn-glycero-3-phosphocholine (PC(18:0p / 18:1 (9Z))), 1 -stearoyl-2- hydroxy-sn-glycero-3-phosphocholine (LPC(18:0)), 1 -oleoyl-2-hydroxy-sn-glycero-3- phosphocholine (LPC(18:1 )), 1 -palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (PE(16:0 / 18:1 )), 1 -(1 Z-octadecenyl)-2-docosahexaenoyl-sn-glycero-3-phosphocholine (PC(18:0p / 22:6)) and 1-stearoyl-2-linoleoyl-sn-glycerol (DG(18:0 / 18:2)), 1-(9Z- octadecenoyl)-sn-glycero-3-phosphoethanolamine (LPE(18:1 )), N-(9Z-octadecenoyl)- sphinganine (Cer(d18:0 / 18:1 (9Z))), 1 -hexadecyl-2-(9Z-octadecenoyl)-sn-glycero-3- phosphoethanolamine (PE(16:0 / 18:1 )) from Avanti Polar Lipids, 1-Palmitoyl-2-Hydroxy-sn- Glycero-3-Phosphatidylcholine (LPC(16:0)), 1 ,2,3 trihexadecanoalglycerol(TG(16:0 / 16:0 / 16:0)), 1 ,2,3-trioctadecanoylglycerol (TG(18:0 / 18:0 / 18:)) and 3p-hydroxy-5- cholestene-3-stearate (ChoE(18:0)), 3p-Hydroxy-5-cholestene-3-linoleate (ChoE(18:2)) from Larodan, were prepared to the following concentration levels: 100, 500, 1000, 1500, 2000 and 2500 ng / mL (in CHCI3:MeOH, 2:1 , v / v) including 1250 ng / mL of each internal standard.

[0064] The International Lipid Classification and Nomenclature Committee (ILCNC) introduced the LIPID MAPS129, a comprehensive classification framework for lipids. This chemically based system organizes lipids into eight classes: fatty acyls, glycerolipids (GL), glycerophospholipids (GP), sphingolipids (SP), saccharolipids (SL), polyketides (PK), prenol lipids (PR) and sterol lipids (ST). Classification is based on lipid class, fatty acid composition, carbon count, and double bonds content. Specific lipid species, such as PCs, Pes, Pis, SMs and ceramides, possess two fatty acyl groups attached to their head group, while lysoPCsand lysoPEs are characterized by a single fatty acyl group. Classes including CEs, DGs, and TGs feature varying numbers of fatty acyl groups (one, two and three, respectively). Furthermore, lipids are also classified into classes like phophatidylcholines (PC), lysophophatidylcholines (lysoPC), phosphatidylethanolamines (PE), di- and triacylglycerols (DG, TG), sphingomyelins (SM), ceramides (Cer), phosphatidylinositols (PI), phosphatidylglycerols (PG), monohexosylceramides (HexCer), lactosylceramides (LacCer). Additional categorization of TGs is based on the fatty acid composition, distinguishing between saturated, monounsaturated, and polyunsaturated species. Notably, PCs and PEs exhibit subcategories, including alkylether PCs or PEs (plasmalogens), each possessing slightly different structure and biological function. In cases where detailed structure has not been determined, naming is based on carbon and double bond sums.

[0065] Data were processed using MZmine 2.53

[0030] . The identification was done with a custom data base, with identification levels 1 and 2, i.e. based on authentic standard compounds (level 1) and based on MS / MS identification (level 2) based on Metabolomics Standards Initiative. Quality control was performed by analysing pooled quality control samples (with an aliquot pooled from each individual samples) together with the samples. In addition, a reference standard (NIST 1950 reference plasma), extracted blank samples and standards were analysed as part of the quality control procedure.

[0066] Analysis of polar and semipolar metabolites: 40 pl of serum sample was mixed with 90 pl of cold MeOH / H2O (1 :1 , v / v) containing the internal standard mixture (Valine-d8, Glutamic acid-d5, Succinic acid-d4, Heptadecanoic acid, Lactic acid-d3, Citric acid-d4. 3- Hydroxybutyric acid-d4, Arginine-d7, Tryptophan-d5, Glutamine-d5, 1-D4-CA,1-D4-CDCA,1- D4-CDCA,1 -D4-GCA,1 -D4-GCDCA,1 -D4-GLCA,1 -D4-GUDCA,1-D4-LCA,1-D4-TCA, 1-D4- UDCA) for protein precipitation. The tube was vortexed and ultrasonicated for 3 min, followed by centrifugation (10000 rpm, 5 min). After centrifuging, 90 pl of the upper layer of the solution was transferred to the LC vial and evaporated under the nitrogen gas to the dryness. After drying, the sample was reconstituted into 60 pl of MeOH: H2O (70:30).

[0067] Analyses were performed on an Acquity UPLC system coupled to a triple quadrupole mass spectrometer (Waters Corporation, Milford, USA) with an atmospheric electrospray interface operating in negative ion mode. Aliquots of 10 pL of samples were injected into the Acquity UPLC BEH C18 2.1 mm x 100 mm, 1 ,7-pm column (Waters Corporation). The mobile phases consisted of (A) 2 mM NH4AC in H2O: MeOH (7:3) and (B) 2 mM NH4AC in MeOH. The gradient was programmed as follows: 0-1 min, 1% solvent B; 1- 13 min, 100% solvent B; 13-16 min, 100% solvent B; 16-17 min, 1% solvent B, flow rate 0.3 mL / min. The total run, including the reconditioning of the analytical column, was 20 min.

[0068] Quantification of BAs and PFAS were performed using a 7-point internal calibration cur59 metabolites specified in Table 1 .

[0069] The identification was done with a custom data base, with identification levels 1 and 2, based on Metabolomics Standards Initiative. Quality control was performed by analysing pooled quality control samples (with an aliquot pooled from each individual samples) together with the samples. In addition, a reference standard (NIST 1950 referenceplasma), extracted blank samples and standards were analysed as part of the quality control procedure.

[0070] Human leukocyte antigen genotype and analysis across ND subtypes: Human leukocyte antigen (HLA) class II genotype was determined using sequence-specific hybridization with lanthanide-labelled oligonucleotide probes on blood spots on 3,783 children. Due to the genetic overlap and comorbidities observed in the literature with autoimmune disease, prevalence of risk alleles commonly reported in autoimmunity, specifically DR4-DQ8 and DR3-DQ2, was compared across NDs and controls using odds ratios in Python 3.1 1 .4.

[0071] Stool sample collection and preservation: Stool samples were collected from 1 ,748 participating infants at one year. With a sterile spatula and tube provided by the WellBaby Clinic, samples were obtained from the diaper. Immediate freezing followed collection, either at the infant’s home or the clinic. For samples collected at home, the use of freeze clamps facilitated frozen transport to the WellBaby Clinic, where subsequent dry storage at -80°C was maintained. Stool samples were collected at an average of age 11 .93 ± 2.94 months, with no significant age differences between the control and ND groups at the time of collection.

[0072] Sequencing and quantification of microbial abundances: Extraction and Sequencing. DNA extraction from stool samples and subsequent 16S rRNA-PCR amplification targeting the V3-V4 region were carried out. A total of 1 ,748 samples were sequenced in ten pools using Illumina MiSeq 2x300 bp at the Interdisciplinary Center for Biotechnology Research (ICBR) at the University of Florida, Gainesville, Florida, USA, following established protocols126. Amplicons for targeted V3-V4 16S rRNA sequencing were produced using Standard Illumina Read 1 sequencing / indexing primers 341 F (NNNNCCTACGGGAGGCAGCAG) and 806R (GGGGACTACVSGGGTATCTAAT).

[0073] Forward primer: 5'- P5 - Adapter - Linker - SBS3 - 16S -3'

[0074] 5' - AATGATACGGCGACCACCGAGCIWTHTAYGGIAARGGIGGGIATHGGIAA - 3' (SEQ ID NO: 1 )

[0075] Reverse primer: 5' - P7 Adapter - Linker - Barcode - SBS12 - 16S - 3'

[0076] 5'-CAAGCAGAAGACGGCATACGAGAT-(BARCODE)- GTGACTGGAGTTCAGACGTGTGCTCTTCCGSTCTGGGGACTACVSGGGTATCTAAT - 3' (SEQ ID NO: 2)

[0077] For pooling, barcodes 1 1 nucleotides in length were used. Each PCR sample was spin-colum purified and quantified by Qubit prior to pooling.

[0078] Paired end Joining and Demultiplexing. Amplicons were first processed in Qiimel131, involving paired-ends read joining (join_paired_ends.py), de-multiplexing (split_libraries_fastq.py), and generation of separate fastq files (split_sequence_file_on_sample_ids.py). In the initial QIIME1 processing, we were mindful of the potential impact of filtering on the meaning of downstream quality scores. Therefore, we intentionally set our demultiplexing options in the split_libraries_fastq.py step to be as lenient as possible so that error correction could be carried out using one, consistent method. Specifically, we chose the following parameters: -q (maximum unacceptable Phred quality score): Set to “0,” allowing the most lenient passage of reads for downstream processing; -r (maximum number of consecutive low-quality base calls allowed before truncating a read): Set to “0,” meaning no consecutive low-quality base calls were tolerated, however, at a Phred quality score of “0”, thus passing reads of all quality scores forward; -n (maximum number of ambiguous or undetermined bases, or “N” characters, allowed in a sequence): Set to “100.” These lenient parameter choices were made with the intention of conducting most of the filtering and quality control in DADA2 R package.

[0079] Filtering and Sample Inference. Using DADA2, sequences with poor quality were removed and high-resolution amplicon sequence variants (ASVs) were derived. DADA2 incorporates an error model to estimate error rates at each position in the sequence and thereby distinguish biological variations from sequencing errors. Using this package, initial inspection of the raw, demultiplexed reads (plotQuality Profile), filtering and trimming (filterAndTrim), subsequent inspection of the filtered reads (plotQualityProfile), learning of error rates (learnErrors, plotErrors), and sample inference (dada) were carried out. In the filterAndTrim step, we employed the quality control process with the following parameters: truncation length (truncLen=c(421)) set to discard reads shorter than 421 bases, trimming (triml_eft=21 ) set to remove the first 21 bases were removed to eliminate barcodes, resulting in reads of 400 bases in length, maxN set to “0” (the default) to remove all ambiguous or undetermined bases (represented as “N” characters), maximum expected error (maxEE) set to “2” to discard reads with expected errors exceeding this threshold (EE = sum(10A(-Q / 20)), removal of reads matching against the phiX genome, and multithreading enabled to filter the input files in parallel. Subsequently, five samples with very low reads were excluded. Sample inference was conducted using the dada function with multithreading, and sequence tables from the ten pools were merged before applying the consensus-method chimera removal. Taxonomic assignment was determined using the Silva 138133database and verified using the National Center for Biotechnology Information (NCBI) Basic Local Alignment Search Tool (BLAST)134as needed.

[0080] In total, 102,972,740 reads (average number of reads per sample: 61 ,994.4; median: 57,860; maximum: 776,158; minimum: 10,059) were observed, encompassing 12,844 unique ASVs. Among these, 4,938 ASVs were shared across at least two children, and 2,444 were present in five or more children. For identification of core taxa, our rarefied dataset was employed, where counts were standardized to a sequencing depth of 21 ,800 reads, resulting in the removal of 27 control samples and 636 corresponding ASVs.

[0081] For microbiome analyses investigating ND status, samples with low read counts and samples from infants who later received an autoimmune diagnosis in the absence of a comorbid ND were removed, given the microbial associations that we have seen at this age in infants with future autoimmune disorders, resulting in a dataset of 1 ,661 samples. Of these, 116 were later diagnosed with an ND (ABISND), with 87 children acquiring two or more NDs while 1 ,545 were deemed controls (ABIScontrois) in the absence of a future diagnosis. Of the 116 ABISND, 14 children were later diagnosed with a speech disorder, 7 with intellectual disability, 85 with ADHD, and 39 with ASD. For the analysis of environmental factors, all samples from ABIS were considered, after removing those with low counts (n=1 ,743).

[0082] Global and targeted metabolomics on stool at one year: Untargeted LC-MS metabolomic analysis was performed on stool samples from a subset of 46 individuals (n=23 ABISASD and n=23 ABISc), selected by propensity score matching on biological sex at birth and municipality, with no difference in age at stool collection (ABIScontrois: 10.89 ±3.51 months, ABISASD: 11.19 ±2.37, p=0.726). Stool samples, averaging 1 1.4 mg in weight, underwent cellular extraction and pre-normalization to sample protein content.

[0083] Global metabolomics profiling was conducted at the Southeast Center for Integrated Metabolomics (SECIM) at the University of Florida, Gainesville, FL. Full metabolomic methods (chromatography and MS) were performed as described previously135. Briefly, a Thermo Q-Exactive Oribtrap mass spectrometer with Dionex UHPLC and autosampler employed positive and negative heated electrospray ionization, each with a mass resolution of 35,000 at m / z 200, as separate injections. Separation was achieved using an ACE 18-pfp 100 x 2.1 mm, 2 pm column with a consisting of 0.1% formic acid in water (mobile phase A) and acetonitrile (mobile phase B). This polar embedded stationary phase provides comprehensive coverage. It should be noted that there are some limitations of this technique in the coverage of very polar species. The flow rate was 350 pL / min, with a column temperature of 25°C, and injection volumes were 4 pL for negative ions and 2 pL for positive ions.

[0084] A total of 4,105 features were detected, with 1 ,250 in positive mode and 2,855 in negative mode. Feature identification, deisotoping, alignment, and gap filling for features thatmay have been missed in the first alignment algorithm were performed using MZmine136freeware. Data were processed to remove adducts and complexes. Metabolomic features were referenced against SECIM’s internal retention time metabolite library of 1 ,414 compounds for identification based on existing metabolomic databases (Metlin, ChemSpider, pubChem).

[0085] Targeted metabolomics was also performed at the SEIC, University of Florida, Gainesville, FL, on fatty acids (22 metabolites) and trytophan (six metabolites), employing selected reaction monitoring (SRM). Fatty acids were quantified and identified using LipidMatch137in silico libraries.

[0086] QUANTIFICATION AND STATISTICAL ANALYSIS

[0087] Environmental risk factors from pregnancy to early childhood: The ABIS cohort was stratified into four distinct neurodevelopmental conditions: ASD, ADHD, Speech Disorders, and Intellectual Disabilities. Children with multiple diagnoses were included in each corresponding diagnostic category, accounting for potential overlap. Within each of these conditions, the prevalence of environmental factors (including family medical history, infections, exposures, and living conditions) using IBM SPSS version 29. These environmental factors were derived from ABIS parent questionnaires administered at birth, one year, three years, and five years.

[0088] To avoid the potential for inflated false discovery rates, we adopted a targeted approach. First, we identified factors with a minimum of 15% difference in prevalence across the conditions. Subsequently, odds ratio calculations were performed for each selected factor and diagnostic group combination where appropriate using Python 3.11 .4. The number of comparisons within each comparative group was deliberately limited to fewer than 50. This deliberate restraint was exercised to maintain the robustness of the findings and obviate the need for false discovery corrections given the focused nature of the comparisons. Likewise, only factors involving family medical history, infections, exposures, and living conditions were considered.

[0089] Mood and Gl symptom clusters at three and five years: Parents participating in the ABIS study completed an extensive questionnaire at the child’s three and five year visits, including a set of 11 -12 binary-response questions concerning growth, mood, and gastrointestinal issues. Symptoms were structured around the question, "Do you think that the child suffers from or is affected by..." to assess poor growth, poor weight gain, poor appetite, stomachache, bloated / gassy stomach, diarrhea three times or more per day, vomiting three times or more per day, constipation, fatigue, general irritation, cranky mood / screaming, and poor sleep quality.

[0090] Principal component analysis (PCA) was conducted using IBM SPSS version 29 on binary responses to the twelve symptoms reported at the child's 3-year visit and eleven symptoms at the 5-year visit. Components were subjected to Varimax rotation with Kaiser normalization, and component scores were computed through regression. This approach allowed interpretatation of the components, with KMO measures indicating satisfactory sampling adequacy (KMO = 0.854 at 3 years and 0.839 at 5 years), and significant Bartlett’s tests of sphericity (both p’s < 0.001 ), confirming data suitability for PCA.

[0091] To evaluate individual symptoms and cumulative occurrence across the three- and five-year components, odds ratios were calculated using Python 3.11 .4.

[0092] Otitis and comparative prevalence of microbes: Differences in the microbiome based on otitis infection in the first twelve months of life were sought within a propensity match group (n=576). Children were selected in a 1 :1 fashion based on mode of delivery, antibiotics and smoking during pregnancy, and total months of breastfeeding by otitis infection status. The unmatched approach included all ABIS samples with available otitis data (n=1307) . Genera were selected for prevalence testing based on DESeq2 results following FDR correction. Based on these results, odds ratios for the prevalence of Citrobacter and Coprococcus were calculated in Python 3.11 .4 to assess association with early otitis and future ND outcomes.

[0093] Confounds of the gut microbiome: To examine the influence of confounding factors on gut microbiota composition ( -diversity), a permutational multivariate analysis of variance (PERMANOVA; adonis function, vegan package138; R Foundation) was conducted using the Bray-Curtis distance on compositional-transformed counts, employing 1000 permutations. The analysis aimed to evaluate influence of several factors, including biological sex a, mode of delivery, maternal smoking during pregnancy, geographic region (county of Sweden), infant antibiotic usage, early-life respiratory or gastrointestinal infections, and the child’s vulnerability index score.

[0094] Differential abundance between ABIScontrois and ABISND: Differential expression analyses of gut microbiome variations between controls and future NDs were performed using the DESeq2139package. Children with future autoimmune conditions, which could also reflect dysbiosis, were excluded from the analysis. Estimations of size factors and dispersion were computed, and negative binomial general linear models (GLM) were fitted using Wald statistics with local type fitting of dispersions to mean intensity.

[0095] The GLM analysis yielded Iog2 fold change (FC) and standard error (IfcSE) values. To account for multiple comparisons, false discovery rate (FDR)-adjusted p-values were calculated within each analysis by the Benjamini-Hochberg method140. Significance wasassessed for each analysis across different comparative groups and taxonomic ranks. Comparisons were made between controls and NDs collectively (diagnosis of any qualifying ND) and then stratified by diagnosis types (i.e., ASD, ADHD, speech disorder).

[0096] Differential abundance across symptom clusters and risk factors, as well as HLA: The full microbiome cohort (n=1743) was employed to investigate associations involving symptoms, risk factors, and HLA, using the DESeq2139package with FDR correction within each comparison. The prospective assessment of microbial taxa at one year of age in relation to symptoms at three and five years involved binary categorizations of symptoms such as fatigue, stomachache, general irritation, diarrhea, and sleep quality, as well as symptom clusters that exhibited the strongest links with development of one or more future NDs. Risk and protective factors from the year one and birth surveys were dichotomized for analysis. Risk factors encompassed gastroenteritis, infection requiring antibiotics, otitis in the first year; increased psychosocial vulnerability; maternal smoking during pregnancy; fewer months of total / exclusive breastfeeding; more frequent chocolate, fries, and chips in the first year; and birth by caesarean section, while protective factors represented the inverse (i.e., no infection; reduced or no psychosocial vulnerability; no smoking; longer periods of total / exclusive breastfeeding; no or fewer servings of chocolate, fries, and chips in the first year; birth by vaginal delivery). Infant diet features were dichotomized based on frequency (daily or 3-5 times weekly, versus seldom or 1 -2 times weekly). Months of breastfeeding (total / exclusive) were dichotomized as one to four months compared to five or more. Psychosocial vulnerability was dichotomized for high / low risk based on the total index.

[0097] For the HLA investigation, two separate analyses within DESeq2 were conducted: first, comparing DR4-DQ8 homozygotes and those without DR4-DQ8, and second, comparing individuals carrying DR3-DQ2 / DR4-DQ8 to those lacking either of these risk alleles.

[0098] Differential abundance by age of diagnosis within ASD group, stratified by sex at birth: To address the potential impact of early ASD diagnosis on phenotype severity, we also explored differences based on diagnosis age within the ABISASD subgroup. These analyses were conducted separately for males and females, considering the average age of diagnosis was much lower in males (13.4 ±4.0 years, males; 17.6 ±2.6 years, females).

[0099] Differential abundance after controlling for microbiota confounders and ND risk factors (ABISND-Match): Propensity score matching was applied using the matchit R package to control for confounds affecting neurodevelopmental risk or gut microbiome composition. The confounds considered were biological sex, mode of delivery, geography (region / county of Sweden), toxic exposure (e.g., smoking during pregnancy), total psychosocial vulnerabilityindex, and infant diet (including total months of breastfeeding and frequency of consumption of beef, chocolate, other candy, chips / cheese doodles in the first 12 months of life). Cases with missing data in these confounds were excluded, leaving 82 ND cases (ABISND-Match). The controls (n=163) were selected using nearest neighbor propensity score matching to balance the distribution of these confounds. It was found that other additional confounds including HLA genotype (p's > 0.162) and antibiotic use (p’s > 0.203) were inherently balanced between ABISND-Match and these selected controls. The dataset contained 15,120,951 reads, with an average of 61 ,718.2 reads per sample (median 57,208, maximum 328,839 and minimum 11 ,440) spanning 3,847 unique ASVs. 1 ,573 ASVs were seen in at least two children, and 878 in four children. Within the ABISND-Matc group, 26 infants were diagnosed with future ASD, 59 with future ADHD, and 13 with a future speech disorder. Most infants received only one ND diagnosis later in life, with 43 with ADHD, 11 with a speech disorder, eight with ASD, and one with intellectual disability, while 19 received multiple ND diagnoses.

[0100] The DESeq2 analysis was repeated on this matched group of infants as previously described. Furthermore, overall microbiome community composition differences were assessed using the PIME R package, which filters taxa based on prevalence intervals to achieve optimal classification. This was run at the genus level for differences between matched NDs and controls and at the ASV level for separating ND subtypes (i.e., multiple diagnoses, ASD, ADHD, and speech disorder).

[0101] Selection of core microbiota across ND subgroups: Considering the complexity of the microbiome, our analysis also extended to assess the prevalent and widely shared bacterial species based on ND status. Core taxa, determined for controls and future NDs (ASD, ADHD, speech disorder, and intellectual disability), were assessed using the microbiome R package on compositionally transformed ASV counts.

[0102] The approach encompassed two datasets: the complete and the rarefied dataset (the latter restricted to 21 ,800 counts for an even sampling depth. For this approach, taxa that never achieved a prevalence exceeding a minimum of 10% (20% for speech and intellectual disability) were excluded. To pinpoint the key microbial taxa consistently present across ND subgroups, we employed the core_members function from the microbiome R package. The core microbes were defined as those taxa displaying relative abundances surpassing 0.01 % within over 50% of the samples belonging to a specific ND group. The resulting core taxa were further visualized using the plot_core R function.

[0103] To bolster the reliability of our findings, the analysis encompassed both the original, unrarefied dataset and the rarefied dataset. Notably, the process of rarefaction didnot exert any discernible influence on the outcomes of the core taxa identification. For consistency, the rarefied dataset was adjusted to a uniform sequencing depth of 21 ,800 reads.

[0104] Stool metabolomics analysis and correlations with microbial abundances at one year: Metabolite concentrations were assessed for association with future ASD in stool samples (n=23 ABISASD and n=23 ABIScontrois) using MetaboAnalyst1425.0, focusing on their relationship with ND outcomes, particularly ASD and comorbid ASD / ADHD. Subsequently, selected metabolites were compared against bacterial relative and absolute abundances (the latter, calculated using qPCR). The data underwent normalization to the sum of metabolites per sample, followed by log-10 transformation of metabolite concentrations. Stool metabolites were analyzed separately in positive and negative ion mode, including integrated peak height intensities of 571 known metabolites (300 in negative and 271 in positive ionization mode). Metabolites exhibiting enrichment or depletion in cord serum or stool samples from infants diagnosed later with ASD were identified using Kruskal Wallis / Mann-Whitney U, partial least squares-discriminant analysis (PLS-DA), and foldchange analyses. Hierarchical clustering and point biserial correlation were also conducted.

[0105] Fatty acid differences in the stool samples were assessed for association with future ASD status using Mann-Whitney II tests in R. The groups did not differ significantly in terms of the sum of peaks (p=0.722). Molar percentages were calculated and compared. Notably, an observation was made that children with future ASD often exhibited a deficiency in palmitoleic acid. To investigate this further, we calculated the difference in the prevalence of palmitoleic acid using a chi square analysis.

[0106] Machine learning was utilized to extract the top twenty stool metabolomic features in the negative-ion mode that best predicted the relative abundance of Bifidobacterium, Roseburia, Faecalibacterium, Akkermansia, and Coprococcus, separately. Random forest regressors were run on each of the genera separately to identify the top twenty metabolites predicting abundance. Subsequently, the metabolites were linked to bacterial abundance using Spearman correlations. This analysis was carried out in Python 3.11 .4.

[0107] Additional associations with stool metabolite concentrations were evaluated for Akkermansia and Coprococcus using Spearman correlations within the MetaboAnalyst 5.0143framework. This analysis incorporated the total counts of 16S copies for Akkermansia and ASV-88 Akkermansia muciniphila, the relative abundance of Coprococcus, and normalized metabolite concentrations.

[0108] Cord serum metabolomics and correlations with microbial abundances at one year: Because of the prenatal programming influence of maternal infections37, smoking38,stress / severe life events39, diet4041, and coffee intake42during pregnancy on the cord serum and fetus, we controlled for these factors with propensity score matching to select controls (n=27) for future cases of ASD (n=27). For this we employed, a 1 :1 nearest neighbor method on the following variables from the birth survey: infectious disease or other infection, severe life event, smoking, caffeine intake, vitamins / minerals and iron supplements, and consumption of milk, dairy, and eggs during pregnancy, as well as education level of the mother and father and birth week of the child. We examined the relationship between concentrations of known, polar metabolites detected in the 1 14 cord serum samples and the relative abundances of several of the taxa most consistently associated with NDs in this investigation. Pearson correlations were performed in R. Differences between future ASD and controls were determined by Kruskal Wallis / Mann Whitney testing in R.

[0109] Results:

[0110] An array of biomarkers and questionnaires was analyzed to determine factors associated with future ND diagnosis. While some factors were associated with increased risk of ND, other factors were associated with decreased risk. Questionnaire results are presented, followed by Biomarker results and Integrated results, where we consider their interactions.

[0111] Questionnaires revealed many ND risk factors early in life, spanning infection and antibiotic events, chemical exposures, family history of disease, other medical issues, and serious life events.

[0112] The ABIS questionnaire data were obtained from the parents of participating children at several time periods, starting with pregnancy and continuing throughout childhood. Here, we analyzed the data up until five years of age in the broader ABIS material (see Figure 1 ). The questionnaire data were detailed in scope.

[0113] Infection and antibiotic events. Infections during early childhood (birth to five years) significantly correlated with heightened risks of ADHD or ASD - most notably, otitis and repeated eczema in the first year. Those experiencing three or more penicillin-requiring infections during this period were prone to future NDs, e.g., speech disorder (OR=3.89 [2.14- 7.05, 95% Cl]), ADHD (OR=3.27 [2.29-4.67], or intellectual disability (OR=2.44 [1.18-5.06]). Penicillin use between 1 -2.5 years resulted in a 1.6-fold [1 .2-2.1 , 95% Cl] higher likelihood of future ASD (p=0.0030). Future ASD were also more likely to have used non-penicillin antibiotics during this period (OR=1.5 [1 .0-2.1 ], p=0.0292), while 23.8% of future intellectual disability had other antibiotics (OR=2.2 [1.1 -4.4, 95% Cl], p=0.0337) besides penicillin (p=0.2917). Children experiencing frequent otitis episodes (three or more times from 1 -2.5 years) were 2.13 [1.1 -4.13, 95% Cl], 1.74 [1 .21 -2.51 , 95% Cl], and 1.75 [1.33-2.30, 95% Cl]times more likely to later be diagnosed with intellectual disability, ASD, or ADHD, respectively. From 2.5-5 years, increased paracetamol antipyretics (six or more times) raised ASD risk (OR=1.82 [1.16-2.88, 95% Cl]). Penicillin use during this period increased risks of ADHD by 1 .54-fold and ASD by 1 .76-fold. Children with at least three, or six, infections requiring antibiotics were 1 .58-2.39 times more likely to develop NDs, especially ADHD (OR=1 .62 to 2.9). Future speech disorder were 1 .85-2.27 times more likely to have had three or more instances of otitis. This was also seen in future intellectual disability. Frequent gastroenteritis (three or more times) from 2.5-5 years was reported in 33.8% of future ND and 35.8% of future ADHD, versus 26.9% of controls.

[0114] Chemical Exposures. Maternal smoking during pregnancy posed risks for NDs cumulatively (OR=3.0 [2.33-3.87, 95% Cl]), as well as ASD (OR=3.72 [1 .92-7.21 , 95% Cl]), and ADHD (OR=3.31 [2.52-4.34, 95% Cl]) separately, especially with ten or more cigarettes daily. Likewise, maternal use of analgesics during pregnancy increased the risk of ADHD (OR=1 .41 [1 .23-1 .62, 95% Cl]) and ASD (OR=1 .46 [1.19-1 .78, 95% Cl]). Exposure to parental smoking from 1 to 2.5 years was dose-dependently associated with increased risk across all NDs except speech disorders. Most striking was ADHD in children whose mother smoked >15 cigarettes / day (OR=4.88 [3.23-7.36]) and ASD in those whose father smoked >15 cigarettes / day (OR=3.47 [2.01-6.01]). Toddlers exposed to parental smoking or smoking in general in the home exhibited a dose-dependent, elevated ND risk across all diagnosis groups except speech disorders. For instance, while only 8.8% and 10.2% of controls had a father or mother who smoked, respectively, throughout this period, 20.5% of toddlers with future intellectual disability had a father who smoked and 30% had a mother who smoked. The highest risk was observed in future ADHD, with a 6.05 [3.68-9.94, 95% Cl]-increased likelihood of ADHD if the mother smoked more than 15 cigarettes / day.

[0115] Family history of disease. Children born to parents with a family history of asthma, Celiac disease (CD), or type-1 diabetes (T 1 D) showed an increased risk of NDs (OR=1 .28-1 .46). Paternal asthma notably displayed the strongest link to risk of ASD (OR=1 .71 , [1 .2-2.42, 95% Cl]), as well as ADHD (OR=1 .56, [1 .20-2.02, 95% Cl]). However, these factors did not show significant association for speech disorders or intellectual disabilities.

[0116] Other medical issues. Early gastrointestinal problems, and to a lesser extent, mood issues, were evident in children with future NDs. Symptoms at 2.5 and five years were reduced, by PCA, into 'Mood and unrest,' 'Child growth,' and 'G I problems' components, explaining a cumulative variance of 57.7% at 2.5 years and 58.3% at five years. At 2.5 years, 'Mood and unrest' symptoms correlated with future ND (ORs, 2.05-3.46, 95% Cl [1 .35-1 .74 to 2.90-6.91]), particularly pronounced in future intellectual disability (OR=3.46,95% Cl [1.74-6.91], 17.9% cases vs. 5.9% controls). Within the 'Gl problems' cluster, symptoms like stomachache, bloated or gassy stomach, or constipation were seen in 21 .3% of future intellectual disability (OR=2.4, 95% Cl [1 .2-4.9], p=0.013). Although not significant for future ASD, 5.6% had two or more symptoms compared to 3.5% of controls (p=0.077). At five years of age, 'Stomach pain' and 'Stomachache' were more prevalent in future NDs (13.5-24.6%) compared to controls (8.0-8.9%), especially prominent in future comorbid ASD / ADHD (OR=3.39-3.45, p<0.0001). Irritability or cranky moods were more common in future ASD or ADHD (5.0-7.7% in future cases vs. 2.4-2.5% in controls), though less frequent than Gl symptoms.

[0117] Serious life events. Serious life events from birth to five years (e.g., separation / divorce, death in the family, serious illness / accident, unemployment) increased the likelihood of future ND by 1 .98-times (1 .6-2.44, 95% Cl), affecting 32.8% of children, compared to 19.8% of controls. Severe life events during pregnancy were also associated with future ADHD (OR=1 .62 [1 .33-1 .98, 95% Cl]). Remarkably, within the group of children whose mothers smoked during pregnancy, 85% of future ASD experienced such an event by age five, compared to only 40% of controls. Infants born prematurely exhibited a 1 .4-fold [1.12-1 .87, 95% Cl] increased likelihood of a future ND diagnosis (except ASD). Also, the intellectual disability group showed a substantial association with preterm birth, occurring in 15.2% versus 4.2% of controls (OR=4.13 [2.51 -6.77, 95% Cl]). Children with speech disorders or intellectual disabilities were 1 .93-2.02 times more likely to be delivered via caesarean section (C-section).

[0118] Biomarker data revealed significant differences in the cord serum metabolome, HLA genotype, infant gut microbiome, and stool metabolome.

[0119] Biomarker data was collected from cord serum at birth and from stool samples, on average, at one year of age. Differences in HLA and the infant gut microbiome were found both by aggregated ND diagnosis as well as by subtypes. The cord serum metabolome, cord serum lipidome, and stool metabolome were explored in ASD only.

[0120] Cord serum metabolome and lipidome. To assess cord serum lipidome and metabolome differences, prenatal risk factors including maternal infections , smoking38, stress / severe life events39, diet4041, vitamins / supplements, and coffee intake42, as well as education of the parents and gestational age, were controlled for using propensity score matching to select matched controls (n=27) for future ASD cases (n=27).

[0121] Linolenic acid was the strongest inverse association with future ASD after matching on the prenatal factors (p=0.00288; Figure 2A). Conversey, perfluorodecanoic acid (PFDA) was the most significant positive association with future ASD (p=0.0087; Figure 2B).Zealarone was also higher in future ASD (p=0.01 ; Figure 3C). Higher in controls were 3- carboxy-4-methyl-5-propyl-2-furanpropanoate, C16:1 , decanoic acid, and palmitic acid (Figure 2D-G). Although approaching significance, octanoic acid (p=0.054), oleic acid (p=0.10), and ursodeoxycholic acid (UDCA) were higher in matched controls (Figure 2H,J- K), while TbMCA was higher in future cases (Figure 2I). Lipidomics revealed eight triglycerides enriched in the cord serum of controls (p’s<0.05): TG(14:0 / 18:2 / 18:2), mass-to- charge ratio (m / z) 822.75, TG(46:2) / (16:1 / 14:0 / 16:1 ), TG(48:3), TG(49:1), TG(49:2), TG(51 :1), TG(51 :2), m / z 862.79, TG(51 :3), TG(53:2), TG(53:3), m / z 888.80, TG(53:4), and TG(58:1).

[0122] HLA genotype. Familial autoimmune disease is more prevalent in individuals with NDs (ASD especially43-45), and is, in large part, driven by immune-mediating HLA genetics. HLA also plays important roles in host-microbe interactions. Thus, we performed HLA Class II haplotyping on 3,809 children in ABIS, including 327 with future NDs. DR4-DQ8 homozygosity (seen in 2.9% of controls) was more pronounced in future NDs (5.1%; OR=1.8 (1.03-3.13), p=0.039) and future ASD, especially (7.6%; OR=2.8 (1.44-5.25), p=0.0021 ). The DR4-DQ8 allele is classically associated with incidence and severity of autoimmune disease46, especially T 1 D and CD.

[0123] Infant gut microbiome. Stool was collected from 1 ,748 infants at 11 .9±2.9 months, with an average 61 ,994 reads / sample. Neither age at stool collection nor copies of 16S rRNA / gram of stool differed between cases and controls, cumulatively or by ND subtype (p’s >0.62). We sought to identify differentially abundant bacteria based on future ND status. For generalizability, first, we compared all available controls with future NDs and then matched for risk factors and microbiome diversity confounds, i.e., mode of delivery (p=.003), geography (p=.001), first-year antibiotic exposure (p=.005), first-year otitis (p=.006). Psychosocial vulnerability index (p=.O8) and biological sex (p=.10) approached significance, while other infections and smoking during pregnancy showed no association (data not shown).

[0124] Future ND status was differentiable by both individual taxa and the common bacterial core47. Bacteria enriched in controls included Akkermansia muciniphila, Roseburia hominis, Erysipelotrichiaceae UCG-003 spp., Adlercreutzia equolifaciens, Alistipes putredinis, Phascolarctobacterium, Coprococcus, and Bifidobacterium sp. (Figure 4A). Veillonella parvula, Megamonas funiformis, ASV-77 Bacteroides sp., and ASV-103 Klebsiella sp. were higher in future cases. Table S2 reports the greatest differences. Some core species were stable, while others were unique, including Bacteroides uniformis and Bacteroides vulgatus (all four NDs), Veillonella parvula (ASD and ADHD), Escherichia- Shigella (ASD, ADHD, and intellectual disability), and Enterobacter (ASD and intellectualdisability). Differences were not due to sequencing depth and persisted in both rarefied and unrarefied datasets.

[0125] Microbial differences after confound adjustment demonstrated distinct core microbial communities in ABISND-Match (n=82) and matched controls (n=163). At 60% microbial prevalence, separation between matched groups was observed. Lactobacilliaceae, Akkermansiaceae, Christensenellaceae, Erysipelotrichaceae UCG-003, AV-81 Bifidobacterium breve, Phascolarctobacterium faecium, among others, were enriched in matched controls, while Bacteroidota, Enterobacteriaceae, Prevotella 9, Parabacteroides merdae, and Barnesiellaceae were enriched in ABISND-Match.

[0126] Infant gut microbiome (by ND subtype). Given the spectrum of severity and diverse characteristics of NDs, we explored differences by subtype, i.e., ASD, ADHD, speech disorder, and multiple NDs. The core microbiome diverged significantly by subtype, though some differences were shared. Principal components separating subtypes (n=114) demonstrated that while the core microbiota of infants with multiple future NDs (ABISmuitipie) overlapped with future ASD (ABISASD), both ADHD (ABISADHD) and speech disorder (ABISspeec ) differed substantially at 65% prevalence. Taxa significantly more abundant across NDs included Megamonas funiformis (all NDs, ABISADHD), ASV-86 Enterobacter sp. (ABISASD), and in ABISspeech, Sutterella, ASV-252 Clostridium sensu stricto 1 neonatale, ASV-184 Blautia sp., and ASV-155 Bacteroides sp.

[0127] To address core variability, controls were separately compared to future ASD, ADHD, speech disorders, or multiple NDs. Across all subtypes, 178 bacteria, mainly Clostridia, were more abundant in controls. Notably, Akkermansia muciniphila, Phascolarctobacterium faecium, Roseburia hominis, Coprococcus eutactus, Coprococcus comes, Bacteroides ovatus, Bifidobacterium breve, and Alistipes putredinis, were consistently less abundant in future NDs, irrespective of subtype. ASV-184 Blautia sp. and ASV-155 Bacteroides sp. were higher in ABISspeech. Carnobacteriaceae and ASV-86 Enterobacter sp. Were higher in ABISASD, and Megamonas funiformis and ASV-77 Bacteroides sp. in ABISADHD. Neither Adlercreutzia nor Christensenella were observed in ABISspeech. Coprococcus eutactus and Bacteroides stercoris, while lower in prevalence in all ABIS, were never observed ABISASD, nor comorbid ASD / ADHD. Interestingly, Bifidobacterium prevalence differed in future speech disorder, seen in 85.7% of ABISspeech compared to 98.1% of controls (OR=0.11 , p=0.0059), but not in other NDs.

[0128] ABISASD were further subdivided, considering potential differences in severity and a higher prevalence of ASD among males48. Males with late diagnoses (n=12) had higher Veillonellales-Selenomonadales, Coprococcus, Akkermansia muciniphila, andRuminococcus gauvreauii compared to males with an early diagnosis (n=11 ). Males with early diagnoses had higher Parabacteroides distasonis, Sutterella wadsworthensis, Prevotella 9 copri, ASVs-18 and 82 Bacteroides sp., ASV-86 Enterobacter sp., and ASV-103 Klebsiella sp. In females, early diagnosis of ASD (n=8) was associated with increased abundance of ASVs belonging to Blautia, Bacteroides sp., Citrobacter, Enterococcus sp., Veillonella sp., and Subdoligranulum sp., while late diagnosis (n=8) was associated with higher Bacteroides ASVs 77, 82, and 155.

[0129] Infant stool metabolome. Differences in the stool metabolome were observed in a subset of future ASD (n=23) versus controls (n=23), selected by propensity score matching on biological sex and municipality. The positive mode revealed 19 significantly different metabolites, and the negative mode identified 124 (p’s <0.05) by fold-change analysis.Future ASD exhibited higher levels of 3-isopropylmalate (FC=6.3, p=0.0024) and quinate (FC=4.0, p=0.035), while controls showed higher levels of 2-hydroxyphenylacetic acid (FC=0.41 , p=0.032), L-lysine (FC=0.69, p=0.016), and glutarate (FC=0.4, p=0.032). Other significant differences were observed (p’s <0.05), most notable being various ion formations of glutarate and L-arginine, L-serine, L-cysteine-S-sulfate, nicotinate, and picolinic acid (higher in controls), and 3-dehydroshikimate, pyridoxamine, N-acetyl-DL-serine, glutaric acid, a-aminoadipate-N-methyl-L-glutamate, (R,R)-tartaric acid, proline, N-BOC-L-aspartic acid, and 5-hydroxymethyl-2-furaldehyde (higher in future ASD).

[0130] Investigating the abundance of key compounds, intriguing patterns were observed that highlight distinct trends in equol and butyrate production. Given that several producers of of equol were consistently higher in controls, we sought to identify potential equol signals. A potential m / z for equol was identified (m / z 241 .0866). Albeit a low signal, the ppm mass error (1 .66) was line with our instrument accuracy. The decrease in this compound from future cases compared to controls was significant (4168.98 ±743.38 and 4814.55 ±984.03, p=0.008).

[0131] Shifting focus to short-chain fatty acids and associated metabolites, it was found that, despite an increase in butyrate producers, butyrate signals were not substantially different between groups (p=0.38). This contrasted with the observed equol trend, suggesting that the increase in butyrate producers did not translate into a corresponding rise in butyrate levels.

[0132] Partial Least-Squares Discriminant Analysis (PLS-DA) analysis identified 24 and 75 significant features in the positive and negative ion modes, respectively, with VIP scores >2. Noteworthy metabolites higher in controls in the positive-mode included m / z 171.1490, m / z 144.1018 (likely N-methylpipecolic acid), m / z 185.1283 (likely N-(3-acetamidopropyl)pyrolidin-2-one), m / z 167.0563 (likely 1 -methylxanthine), m / z 148.0967 (likely N-(2- Hydroxyethyl)-morpholine N-oxide), m / z 144.1018 (likely N-methylpipecolic acid), and m / z 159.0440 (likely 1 ,2-Napthoquinone). Conversely, m / z 149.0446 (likely 3-hydroxyglutaric acid), Ser-Pro (m / z 203.1026), and m / z 253.093 (likely galactosylglycerol) were higher in future ASD. Among notable negative-mode metabolites elevated in controls are m / z 241.1193 (likely pyroglutamylleucine), His-Pro (m / z 251 .1147), m / z 132.0666 (likely N- methyl-threonine or 3-hydroxyvaline), and m / z 182.0126 (likely homocysteic acid or 8- hydroxythioguanine). Glutarate, L-glutamine, and riboflavin were higher in controls, while N- methyl-L-histidine, 3-aminoisobutanoate, 3-isopropylmalate, and N-acetyl-DL-glutamic acid were higher in future ASD.

[0133] Of the 22 fatty acids detected, two notable differences were observed. ABISASD were more likely to lack palmitoleic acid (16:1 ),2(1 , A / =46) = 3.9, p=0.049, observed in 43.5% of controls, but absent in 87.0% of future ASD. Conversely, the molar percentage of palmitic acid (16:0) in stool was 42.6% higher in ABISASD (27.4±2.2% ABISASD; 15.8±16.5% controls, p = 0.056), intriguing since it was lower in the cord serum.

[0134] Integration of Questionnaire and Biomarker data revealed bacterial connections with early-emerging Gl and mood issues, environmental factors, and HLA genotype.

[0135] Next, microbial interactions associated with early symptoms, risk factors, and observed metabolomic differences was explored.

[0136] Early-emerging Gl and mood issues. Proteobacteria were 2.0-2.2 times higher in children with 'Gl problems' at age 2.5 (q=0.011 , q=0.048). Coprococcus and Slackia isoflavoniconvertens exhibited a protective trend with Iog2 fold-change (FC) ranging from 4.22-6.10, inversely correlating with 'Gl problems' and 'Mood and unrest' clusters.Akkermansia muciniphila and Alistipes fingoldii inversely correlated with 'Mood and unrest' symptoms, and Coprococcus eutactus negatively with the 'Gl problems' cluster.Adlercreutzia equolifaciens, Erysipelotrichaceae UCG-003 spp., and Roseburia were higher, and Robinsoniella peoriensis and Megasphaera micronuciformis lower, in infants without future stomachache or diarrhea.

[0137] Similar to associations with mood at 2.5 years, Akkermansia, Adlercreutzia, Coprococcus, and Roseburia species were more abundant in infants without mood or Gl symptoms at age five. Akkermansia muciniphila (FC=2.9), Blautia obeum (FC=3.0), and Turicibacter sanguinis (FC=1 .6) were inversely associated with general irritation or cranky mood. Coprococcus comes and Coprococcus eutactus inversely correlated with both ‘Stomach pain’ cluster and mood symptoms (FCs=2.2-2.7). Other taxa inversely correlatedwith both clusters included Anaerostipes caccae, Adlercreutzia equolifaciens, and Roseburia inulinivorans.

[0138] Environmental factors. Next, differentially abundant bacteria in the context of early-life factors carrying decreased (log2FC's >2; Figure 4B; Table S3) or increased risk of future NDs (Figure 7B; Table S3) was investigated. Protective early-life factors correlated with bacteria higher in controls, while those higher in ABISND correlated with C-section, and antibiotic use, infection, and infant diet. Akkermansia muciniphila, Bacteroides ovatus, Phascolarctobacterium faecium, Alistipes, and Bifidobacterium breve were higher in infants whose mother did not smoke during pregnancy. Higher Coprococcus comes, Adlercreutzia equolifaciens, Lacticaseibacillus, and Bifidobacterium were associated with lower psychosocial vulnerability scores. Only five taxa associated with controls were higher in infants never exposed to antibiotics, including Ruminococcus CAG-352, Clostridia UCG-014, and Coprococcus eutactus. Coprococcus euctatus was also higher in infants who never had otitis, as were Coprococcus comes and Ruminococcus gauvreauii. Bacteroides eggerthii and ASV-81 Bifidobacterium breve were higher in infants who never had gastroenteritis. Lacticaseibacillus and members of Anaeroglobus were higher in infants breastfed at least five months.

[0139] Several taxa were associated with risk factors. More abundant in those delivered by C-section were Megamonas funiformis, Parabacteroides merdae, Prevotella 9, Sutterella, ASVs-47, 77 and -155 Bacteroides, ASV-86 Enterobacter. ASV-77 Bacteroides sp. was more abundant in infants with the highest psychosocial vulnerability scores, while ASV-25 Bacteroides fragilis was more abundant in those whose mother smoked during pregnancy. Anaerostipes caccae, Oscillospiraceae UCG-002 spp., and Clostridium sensu stricto 1 neonatale appeared largely driven by diet in the first year, negatively associated with frequent consumption of fried potatoes / fries and chips, especially (FC’s=4.0-5.0). Anaerostipes caccae and Sutterella wadsworthensis were higher in abundance in infants who exclusively breastfed for the shortest period (one to four months).

[0140] HLA genotype. Bacteria associated with controls were more abundant in individuals without DR4-DQ8 or DR3-DQ2. DR3-DQ2 / DR4-DQ8 heterozygotes showed higher Acidaminococaccaeae and Christensenellaceae. Detailed species-level analysis (Figure 5E) revealed significant differences, with Adlercreutzia equolifaciens (FC=4.2), Phascolarctobacterium faecium (FC=2.4), Blautia obeum (FC=2.0), and Coprococcus comes (FC=1 .0) among the notable species. Similarly, in DR4-DQ8 homozygotes, lower abundances of bacteria associated with controls were observed, including Roseburia species, Phascolarctobacterium faecium, Coprococcus eutactus and Coprococcus comes, Alistipes putredinis, Alistipes finegoldii, and Adlercreutzia equolifaciens.

[0141] Integration of Questionnaire and Biomarker data also connected significant differences in bacterial prevalence with antibiotic use and chemical exposures.

[0142] Two Klebsiella michiganensis strains (ASVs-120 and 318) were more prevalent in ABISmuitipie (48.3%) and ABISASD (43.6%) versus controls (21 .7%), particularly common in controls with frequent antibiotic use (27.1%). Infants with either ASV had higher odds of developing ASD or multiple NDs (OR=2.8 [95% Cl, 1.46-5.30], p = .0019, OR=3.4 [95% Cl, 1 .71-6.44], p = .0004, respectively), which increased when comparing infants breastfeeding at least eight months with minimal antibiotic use (none or 1 -2 regimens). The presence of either ASV was associated with a 3.60-times [1 .60-8.08, 95% Cl] higher risk of comorbid ASD / ADHD, further heightened by exposure to antibiotics (OR=5.52, [95% Cl, 1.74-17.54]) or nicotine / alcohol (OR=5.69, [95% Cl, 2.20-14.71]). Presence of both ASVs increased the risk of comorbid ASD / ADHD (23.1% ABISASD, 24.1% ABISmuitipie, 8.3% controls). Salmonella- related enteric bacteria were notably higher in future NDs (21% comorbid ASD / ADHD vs. 3% controls, p = 1 .07e-7; Figure 5G). While Akkermansia was found in 48.7% of controls, it was only present in 25.0% of future comorbid ASD / ADHD (OR=2.84, 1 .12-7.20 95% Cl, p=0.027) and 28.2% of future ASD (OR=2.41 , 1 .19-4.88 95% Cl, p=0.014). These correlations held in logistic regression considering parental education and maternal smoking during pregnancy (adjusted ORs=3.2 and 0.55, respectively). Environmental exposures and genetic factors further increased risk.

[0143] In exploring the link between early otitis and ND outcomes, Citrobacter, Coprococcus, and Phascolarctobacterium exhibited prominent disparities, even after multiple discovery corrections. Infants with Citrobacter were 2.35 [1 .12-4.94, 95% Cl] times more likely to be ABISND with early otitis than a control without otitis (p=0.0235; Figure 3A). Conversely, Coprococcus was present in only 10% of future ND with otitis but in 34% of controls without otitis (Figure 3A). Further, Coprococcus was observed in only 15.7% of ABISND overall, with 78.6% of these children having experienced otitis in their first year. Adjustments for variables including mode of delivery, prenatal antibiotics and smoking exposures, and breastfeeding confirmed these associations (Figure 3A, 3C-3E).

[0144] Similarly, infants with Citrobacter showed a 3.99 [1 .66-9.61 , 95% Cl] times higher likelihood of being ABISND with early otitis than a control without otitis (p=0.002; Figure 3A). In the future ND group, 52% of those with past otitis had Citrobacter, compared to only 31% of those without infection (OR=2.71 [1 .08-6.80], p=0.0336). Among controls, otitis in the first year corresponded to lower Coprococcus prevalence (17% versus 28%). Interestingly, while Citrobacter was found in 61% of ABISND who had otitis in the first year, but only in 28-34% of ABIScontroi (Figure 3E). Furthermore, Phascolarctobacterium (Figure 3D) and Coprococcus(Figure 3F) abundances were significantly higher in those without otitis (pmean=0.60 versus Mmean=0. 1 0; |Jmean=0.34 versus pmean=0.1 0, respectively).

[0145] Integration of Biomarker metabolomic and microbiome data revealed significant associations with triglycerides and polar metabolites at birth, as well as vitamin and neurotransmitter precusors in infants’ stool.

[0146] Cord serum. To establish a connection between microbes and the cord serum metabolome, polar metabolites and relative abundances of taxa most consistently associated with NDs were examined using correlation analysis across all samples with cord serum and microbiome data (n=114). ASV-81 Bifidobacterium breve correlated with linolenic acid, oleic acid, and UDCA, while ASV-17 Anaerostipes hadrus correlated with 3-carboxy-4- methyl-5-propyl-2-furanpropanoate. PFDA and zealarone, higher in the cord serum of future ASD, positively correlated with Citrobacter (p=0.007) and ASV-86 Enterobacter sp. (p= 0.016), respectively. Lacticaseibacillus correlated with eleven of the fifteen triglycerides (TG) higher in concentration in controls, and ASV-81 Bifidobacterium breve with four. While TG(46:2) / (16:1 / 14:0 / 16:1) levels positively correlated with ASV-31 Bifidobacterium breve, they also correlated with ASV-120 and ASV-318 Enterobacter sp., which were higher in future ASD. Roseburia inulinivorans negatively correlated with four triglycerides.

[0147] Infant stool metabolites. Stool metabolome profiles of infants with and without future ASD were similarly analyzed. Total 16S rRNA copies of Akkermansia and ASV-88 Akkermansia muciniphila positively correlated with stool metabolites including glutarate, arabinose, picolinic acid, L-phenylalanine, L-isoleucine, L-serine, leucine, and serotonin and catecholamine precursors D-tryptophan and L-tyrosine. Coprococcus positively correlated with riboflavin, xanthine, uracil, arabinose, thymine, and 4-hydroxybenzoate, and negatively correlated with 3-isopropylmalate and dehydroascorbate, the latter also more highly expressed in future ASD.

[0148] To link metabolomic and microbial differences derived from stool, machine learning was used to extract the top twenty metabolites in the negative-ion mode predicting bacterial abundance for five of the most salient genera in this investigation, and correlations with the relative abundance determined Bifidobacterium, Roseburia, Faecalibacterium, Akkermansia, Coprococcus). Many positive correlations were observed with Bifidobacterium, including 2-deoxy-D-galactose-fructose, 3,4-hydroxy-phenyllactate, 3-hydroxyphenylacetate, N-acetyl-D-galactosamine, L-arabitol, 2-deoxy-D-glucose, and 6-deoxy-L-galactose, while xanothosine and mono-ethyl-malonate negative correlated. L-arginine positively correlated with Roseburia, while malonate negatively correlated. Shikimate negatively correlated with both Faecalibacterium and Coprococcus. Faecalibacterium was also positively correlatedwith uridine and negatively with 3-suflino-L-alanine, taurine, 5-hydroxyindoleacetate, creatine, L-carnitine. Akkermansia positively correlated with 3,4-dihydroxy-phenyl-propionic acid.

[0149] This study has followed a birth cohort for over twenty years to find factors associated with neurodevelopmental disorder (ND) diagnosis. Detailed, early-life longitudinal questionnaires captured infection and antibiotic events, stress, prenatal factors, family history, and more. Biomarkers including cord serum metabolome and lipidome, HLA genotype, infant microbiota, and stool metabolome were assessed. Among the 16,440 Swedish children followed across time, 1 ,197 developed an ND. Significant associations emerged for future ND diagnosis in general and for specific ND subtypes, spanning intellectual disability, speech disorder, attention deficit hyperactivity disorder, and autism.

[0150] Additional analysis:

[0151] Building on the findings above, long-read Oxford Nanopore metagenomic sequencing were conducted on an expanded cohort (7 samples from children with future autism, 12 from controls, totaling 3,405,226 reads). This enhanced resolution allows for greater taxonomic and functional accuracy, reinforcing our initial observations. We also sequenced samples not only collected at one year of age, but also near two years, building on the original results.

[0152] Functional Metagenomics: Key Differences in Autism vs. Controls

[0153] Statistically significant functional differences were observed between cases of ASD and controls. Genes elevated in future autism were associated with (A1 ) antibiotic resistance and stress adaptation, aligning with previous findings. Additionally, we observed (A2) imbalances in neurotransmitter and amino acid metabolism, including disruptions in arginine, tryptophan, valine, and polyamine pathways, which may alter brain signaling. Further differences included (A3) disruptions in DNA repair and energy metabolism, as well as (A4) microbial virulence, biofilm formation, and the gut barrier, as further evidence of gut dysbiosis that may impact immune and neurodevelopmental pathways. Those genes include: K21829 (D-arginine utilization repressor, p = 0.002); K00118 (glucose-fructose oxidoreductase [EC:1.1.99.28], p = 0.002); K10794 (D-proline reductase (dithiol) PrdB [EC:1 .21.4.1], p = 0.003); K04748 (nitric oxide reductase NorQ protein, p = 0.005); K11686 (chromosome-anchoring protein RacA, p = 0.008); K10795 (D-proline reductase (dithiol)- stabilizing protein PrdD, p = 0.009); K10796 (D-proline reductase (dithiol)-stabilizing protein PrdE, p = 0.010); K12988 (alpha-1 ,3-rhamnosyltransferase [EC:2.4.1 .-], p = 0.011); K14647 (minor extracellular serine protease Vpr [EC:3.4.21 .-], p = 0.012); K10793 (D-proline reductase (dithiol) PrdA [EC:1 .21 .4.1], p = 0.013); K10122 (fructooligosaccharide transportsystem permease protein, p = 0.013); K02086 (DNA replication protein, p = 0.016); K10121 (fructooligosaccharide transport system permease protein, (p = 0.022); K19545 (lincosamide nucleotidyltransferase A / C / D / E, p = 0.022); K01905 (acetate— CoA ligase (ADP-forming) subunit alpha [EC:6.2.1 .13], p = 0.028); K11161 (retinol dehydrogenase 13 [EC:1 .1 .1 .300], p = 0.028); K00712 (poly(glycerol-phosphate) alpha-glucosyltransferase [EC:2.4.1 .52], p = 0.038); K05568 (multicomponent Na+:H+ antiporter subunit D, p = 0.042); K10120 (fructooligosaccharide transport system substrate-binding protein, p = 0.042); K13282 (cyanophycinase [EC:3.4.15.6], p = 0.043); K16123 (tyrocidine synthetase II, p = 0.043); K15261 (poly [ADP-ribose] polymerase 10 / 14 / 15 [EC:2.4.2.30], p = 0.047); K01655 (homocitrate synthase [EC:2.3.3.14], p = 0.016); K06285 (transcription attenuation protein (tryptophan RNA-binding attenuator protein), p = 0.016); K18902 (multidrug efflux pump, p = 0.016); K16095 (bacitracin synthase 3, p = 0.027); K20447 (nicotinate dehydrogenase large molybdopterin subunit [EC:1 .17.1 .5], p = 0.008).

[0154] (A1) Bacitracin synthase 3 & tyrocidine synthetase II are involved in bacterial peptide antibiotic production, where elevated levels may reflective selective pressure from antibiotics or a dysbiosis that favors antimicrobial resistant strains. The elevated presence of multidrug efflux pumps indicates bacterial adaptation to stressors, which could include antibiotics (as suggested in our original application) or metabolic byproducts, which would collectively impact gut microbiome stability. Increased bacterial poly [ADP-ribose] polymerases (PARPs 10 / 14 / 15, EC:2.4.2.30) suggest heightened DNA repair activity, which could further indicate microbial adaptation to oxidative stress. The elevated level of antibiotic resistance genes (e.g., lincosamide nucleotidyltransferases) further indicates selective pressures, potentially in response to past antibiotic exposure or host immune responses.

[0155] (A2) Homocitrate synthase (EC:2.3.3.14) plays a role in lysine biosynthesis, which is essential for glutamate and GABA production — two neurotransmitters critical for brain function. The increased abundance of transcription attenuation proteins related to tryptophan metabolism suggests alterations in serotonin biosynthesis. Elevated levels of the gene for retinol dehydrogenase 13 [EC:1 .1 .1 .300] suggests increased conversion of retinol (vitamin A) to retinal, which could lead to imbalanced homeostasis of vitamin A, which is necessary for plasticity and brain development. Retinoic acid controls early neurogenesis in mouse models, where impairments disrupt the balance in direct and indirect neurogenesis, especially in the cortex144.

[0156] (A3) Significant elevation in genes like glucose-fructose oxidoreductase (EC:1 .1 .99.28) and DNA replication proteins suggest a higher bacterial turnover rate, possibly favoring certain microbial populations over others. Elevated levels of nicotinate dehydrogenase large molybdopterin subunit [EC:1 .17.1.5] may demonstrate increasedbacterial conversion of niacin (vitamin B3), which could potentially alter the availability of cofactor NAD+ that supports sirtuins and poly(ADP-ribose) polymerases (PARPs) regulating neuronal function and DNA repair145. Both niacin and serotonin are synthesized by tryptophan, where a sufficient supply is necessary for both. Deficiencies in niacin are implicated in depression and other neurological disorders146.

[0157] (A4) Elevated levels of alpha-1 ,3-rhamnosyltransferase (EC:2.4.1 .-) and poly(glycerol-phosphate) alpha-glucosyltransferase (EC:2.4.1.52) suggest increased bacterial exopolysaccharide production, which enhances microbial adherence and biofilm stability. Biofilm-associated bacteria include Citrobacter species4, such as those elevated in our sample with future NDs. The increased activity of extracellular serine proteases suggests potential disruption of gut barrier function, leading to increased gut permeability or "leaky gut."

[0158] In contrast, genes that were more abundant in controls contributed to key metabolic and adaptive processes. These included (C1) enhanced energy metabolism, particularly fatty acid oxidation and aldehyde detoxification, as well as (C2) microbial adaptability, specifically mechanisms enabling responses to environmental stressors. Moreover, genes linked to (C3) amino acid and neurotransmitter metabolism, including polyamine synthesis and branched-chain amino acid (BCAA) metabolism, were more prevalent in controls, potentially supporting neurotransmitter balance involving glutamate, GABA, and dopamine. Those genes include: K00205 (4Fe-4S ferredoxin, p = 0.019); K07480 (insertion element IS1 protein InsB, p = 0.025); K02167 (TetR / AcrR family transcriptional regulator, transcriptional repressor of bet genes, p = 0.038); K01581 (ornithine decarboxylase [EC:4.1.1.17], p = 0.047); K18244 (acyl-CoA dehydrogenase [EC:1 .3.99.-], p = 0.028); K00129 (aldehyde dehydrogenase (NAD(P)+) [EC:1 .2.1.5], p = 0.032); K00835 (valine-pyruvate aminotransferase [EC:2.6.1 .66], p = 0.042).

[0159] (C1 ) The 4Fe-4S ferredoxin, found at higher levels in controls, plays a central role in cellular energy metabolism by facilitating electron transfer in pathways like oxidative phosphorylation. The higher level of sequences corresponding to this gene in control samples suggests more efficient energy production and a well-maintained redox state in individuals with a healthy microbial profile. Aldehyde dehydrogenase (NAD(P)+) [EC:1 .2.1.5], also elevated in controls, plays a critical role in detoxifying harmful byproducts of metabolism (aldehydes). By reducing toxic buildup, this supports homeostasis and facilitates protection against oxidative stress. The higher expression in controls suggests more efficient detoxification mechanisms, where lower expression in future cases could be indicative of increasing susceptibility to environmental toxins, potentially impacting inflammation and oxidative stress. Notably, our past work identified higher levels ofperfluoroalkyl substances at birth in those who later develop autism, further supporting the notion that these children may potentially benefit from the formulations described herein.

[0160] (C2) In contrast to the findings described earlier on antibiotic resistance genes being more readily expressed in the autism samples, more abundant in controls was the TetR / AcrR family transcriptional regulator, responsible for repressing bet genes that regulate cellular osmotic stress responses and methylation. This is important as this gene contributes to microbial adaptability, especially under environmental stress conditions. The higher expression in controls suggests a more resilient microbiome that can remain stable despite changes in the environment, while lower levels in individuals with future autism may reflect a less adaptable microbial community, which aligns with our previous findings regarding repeated otitis (and likely antibiotic use) seen in children with future neurodevelopmental conditions.

[0161] (C3) Ornithine decarboxylase [EC:4.1 .1 .17] and valine-pyruvate aminotransferase [EC:2.6.1 .66] are both essential genes for amino acid metabolism, needed for polyamines essential for synaptic plasticity and branched-chain amino acids (BCAAs), including valine, leucine, and isoleucine, that are fundamental in neurotransmitter regulation. It has been shown that l-valine increases macrophage phagocytosis, especially valinearginine in combination, killing bacteria with multi-drug resistance149. Higher expression levels of these genes in controls suggest an efficient system for amino acid and neurotransmitter metabolism, reinforcing the primary evidence of dysbiosis submitted previously. A balanced neurotransmitter profile is crucial for healthy brain development.

[0162] These findings suggest that children who later develop autism may harbor a less adaptable, metabolically constrained gut microbiome, rendering it more vulnerable to environmental stressors such as antibiotic exposure.

[0163] Taxonomic Trends: Reinforcing Key Players in Autism Risk and Protection

[0164] New analyses from metagenomic sequencing further support our prior findings, particularly regarding key microbial taxa. Akkermansia muciniphila, a potential protective probiotic candidate, was significantly lower in future autism cases, with a mean relative abundance of 1.59% (median 0.68%) compared to 9.76% (median 7.44%) in controls. Notably, 58.3% of controls harbored A. muciniphila at >5% relative abundance, whereas no autism sample met this threshold.

[0165] Conversely, Citrobacter portucalensis152, a multidrug-resistant pathogen, was present in 57.1% of future autism cases but detected in only one control sample (8.3%). This aligns with its known associations with antibiotic resistance and gut dysbiosis. Previously, short-read sequencing limited our ability to resolve strain-level differences, preventingidentification of the specific Citrobacter strain driving these disparities. With long-read sequencing, we now pinpoint C. portucalensis as a likely contributor to the observed functional differences. Citrobacter protucalensis is a pathogen resistant to many antibiotics, including fluoroquinolones, p-lactams, aminoglycosides, sulfonamides, and cephalosporins153. The pathogen has received notable attention within the One Health framework concerning human-animal-environment surveillance and has a broad resistome, spanning not only antibiotics, but also pesticides, disinfectant, and even heavy metals154.

[0166] Cord Serum Proteomics: Early Markers of Autism Risk

[0167] We previously identified several important risk factors for autism, as well as other neurodevelopmental disorders, most notably smoking exposures and repeated use of antibiotics155. In a pilot of metabolomic associations, we found that perfluoroalkyl substances were detected more readily in newborns late diagnosed with autism155. Our preliminary data generated since that publication (n=1 ,204, including 286 controls and 156 with future autism) has revealed significant differences in the inflammasome between groups at birth (Figure 4A and Figure 4B), with markers connected to environmental factors during pregnancy, especially stomach flu of the mother. Notably, the Hexamethylene bis-acetamide inducible 1 (HEXIM1 ) transcriptional regulator is significantly elevated at birth, but especially in males who were diagnosed with autism by age 12, underscoring potential sex differences as the effect was only significant after FDR in males. Nuclear migration protein (NUDC) is also higher, whereas CCL28 (associated with antimicrobial activity) and NTF3 (with important roles in plasticity, myelination) are among the notable proteins with much lower expression in these children at birth. These proteins may further identify the at-risk populations, where immunity and neural development are most affected. HEXIM1 is an RNA polymerase II transcription inhibitor, aligning with long-gene dysregulation linked to autism156. NUDC plays roles in neurogenesis and migration and is associated with dynein (where mutations have been linked to autism), suggesting altered neurogenesis and hyperconnectivity or compensatory responses. Thus, a child having elevated levels of HEXIM1 and / or NUDC in cord blood compared to healthy controls would be identified as a child for being at risk for developing autism.

[0168] References:1 . Thomas, et al., (2015). Pediatrics 135, e994-1001 .2. Fayyad, et al., (2007). Br J Psychiatry 190, 402-409.3. Posner, et al., (2020). The Lancet 395, 450-462.4. Black, et al., (2015). 2012. NCHS Data Brief, 1-8.5. Norbury, et al., (2016). Journal of Child Psychology and Psychiatry 57, 1247-1257.6. Hsiao, et al. (2013). Cell 155, 1451 -1463.7. Berding, et al., (2016). Nutrition Reviews 74, 723-736.8. Mullins, et al., (2016). Neuron 89, 1131-1 156.9. Meltzer, et al., (2017). Neuropsychopharmacology 42, 284-298.10. Willfors, et al., (2017). Transl Psychiatry 7, e1014.1 1 . Mayer, et al., (2014). Bioessays 36, 933-939.12. McElhanon, et al., (2014). Pediatrics 133, 872-883.13. Doshi-Velez, et al., (2014). Pediatrics 133, e54-63.14. Sampson, et al., (2015). Cell Host Microbe 17, 565-576.15. Marchesi, et al. (2016). Gut 65, 330-339.16. Cryan, et al., (2019). Nature Microbiology 4, 552.17. Edmiston, et al., (2017). Biol Psychiatry 81 , 383-390.18. Belkaid, et al., (2014). Cell 157, 121-141.19. Honda, et al., (2016). Nature 535, 75-84.20. Wu, et al., (2012). Gut Microbes 3, 4-14.21. Kamada, et al., (2013). Nat Rev Immunol 13, 321-335.22. Byrne, et al., (2015). Int J Obes 39, 1331-1338.23. Lyte, M. (2013). PLoS Pathog. 9, e1003726.24. Foster, et al., (2016). Gut Microbiota and Brain Function: An Evolving Field in Neuroscience. Int. J. Neuropsychopharmacol. 19. 10.1093 / ijnp / pyv114.25. Cryan, et al. (2019). Physiol Rev 99, 1877-2013.26. Gheorghe, et al., (2019). Curr Opin Pharmacol 48, 137-145.27. Sgritta, et al., (2019). Neuron 101 , 246-259.e6.28. Buffington, et al., (2016). Cell 165, 1762-1775.29. Kang, et al., (2017). Microbiome 5, 10.30. Kang, et al., (2019). Sci Rep 9, 5821 .31 . Vuong, et al., (2017). Biol Psychiatry 81 , 41 1-423.32. Bull-Larsen, et al., (2019). Nutrients 11 , E2805.33. Sordillo, et al., (2019). JAMA Netw Open 2, e190905.34. Carlson, et al., (2018). Biol. Psychiatry 83, 148-159.35. Laue, et al., (2020). Sci Rep 10, 15515.36. Lacorte, et al., (2019). Front Neurol 10, 727.37. Meyer, et al., (2007). Neuroscientist 13, 241-256.38. Fleisch, et al., (2017). Metabolism 72, 18-26.39. Graignic-Philippe, et al., (2014). Neuroscience & Biobehavioral Reviews 43, 137— 162.40. Harris, et al., (2022). Biomolecules 12, 1333.41 . Pfefferle, et al., (2010). Journal of Allergy and Clinical Immunology 125, 108-115.e3.42. Sengpiel, et al., (2013). BMC Med 1 1 , 42.43. Keil, et al., (2010). Epidemiology 21 , 805-808.44. Molloy, et al., (2006). J Autism Dev Disord 36, 317-324.45. Hughes, et al., (2018). Front Cell Neurosci 12, 405.46. Aly, et al., (2006). Proceedings of the National Academy of Sciences 103, 14074- 14079.47. Shetty, et al., (2017). FEMS Microbiology Reviews 41 , 182-199.48. Baron-Cohen, et al., (2011 ). PLoS Biol 9, e1001081 .49. Orr, et al., (2013). Journal of Neurochemistry 127, 378-393.50. Orr, et al., (2013). Prostaglandins, Leukotrienes and Essential Fatty Acids 88, 97- 103.51. Bazinet, et al., (2014). Nat Rev Neurosci 15, 771-785.52. Castillo, et al., (2012). Neuron 76, 70-81 .53. Kuipers, et al., (2012). Prostaglandins, Leukotrienes and Essential Fatty Acids 86, 13-20.54. Cunnane, et al., (2009). Progress in Lipid Research 48, 239-256.55. McNamara, et al., (2007). Biological Psychiatry 62, 17-24.56. Keely, et al., (2019). American Journal of Physiology-Gastrointestinal and Liver Physiology 317, G872-G881 .57. Abdelkader, et al., (2016). Mol Neurobiol 53, 810-817.58. Keene, et al., (2002). Proc Natl Acad Sci U S A 99, 10671-10676.59. Rodrigues, et al., (2003). Proc Natl Acad Sci U S A 100, 6087-6092.60. Parry, et al., (2010). Clin Neuropharmacol 33, 17-21 .61. Christian, N.P. (2024). In Encyclopedia of Toxicology (Fourth Edition), P. Wexler, ed. (Academic Press), pp. 747-756.62. Omoike, et al., (2021 ). Environ Res 196, 1 10361 .63. Fenton, et al., (2021 ). Environ Toxicol Chem 40, 606-630.64. de Magistris, et al., (2010). J Pediatr Gastroenterol Nutr 51 , 418-424.65. D’Eufemia, et al., (1996). Acta Paediatr 85, 1076-1079.66. Luo, et al., (2022). npj Biofilms Microbiomes 8, 1-1 1.67. Engevik, et al., (2019). Front Microbiol 10, 2305.68. Louis et al., (2017). Environ Microbiol 19, 29-41.69. Derrien, et al., (2004). Int J Syst Evol Microbiol 54, 1469-1476.70. Reunanen et al., (2015). Appl Environ Microbiol 81 , 3655-3662.71. Rodrigues, et al., (2022). Front Immunol 13, 934695.72. Wang, et al., (2020). Molecular Immunology 125, 172-177.73. Bloemendaal, et al., (2018). eNeuro 5, ENEURO.0035-17.2018.74. Young, S.N. (2007). J Psychiatry Neurosci 32, 224.75. Duncan, et al., (2002). Appl Environ Microbiol 68, 5186-5190.76. Tamanai-Shacoori, et al., (2017). Future Microbiol 12, 157-170.77. Sarkar, et al., (2020). Biological Reviews 95, 1131-1166.78. Barcenilla, et al., (2000). Appl Environ Microbiol 66, 1654-1661.79. Louis, P., and Flint, H.J. (2009). FEMS Microbiol Lett 294, 1-8.80. Noto, et al., (2014). The Journal of Maternal-Fetal & Neonatal Medicine 27, 46-52.81 . Liu, et al., (2022). Cell Discov 8, 1-22.82. Jiang, et al., Cell Cycle 21 , 1519-1531 .83. Ma, et al., (2019). Frontiers in Cellular and Infection Microbiology 9.84. Martinez, et al., (2013). PLOS ONE 8, e69621 .85. Claus, et al., (2011 ). mBio 2, e00271 -10.86. Johnson, et al., (2019). Metabolites 9, 78.87. Johnson, et al., (2020). Plant Foods Hum Nutr 75, 512-517.88. Yap, et al., (2021 ). Cell 184, 5916-5931 ,e17.89. Yang, et al., (2018). Pneumonia (Nathan) 10, 10.90. Moroishi, et al., (2022). Common Med 2, 1-8.91 . Rai, et al., (2012). PLOS ONE 7, e38893.92. Kanina, et al., (2023). Transl Psychiatry 13, 1-8.93. Gumusoglu, et al., (2017). Brain Behav Immun 65, 274-283.94. Coker, et al., (2020). BJOG 127, 217-227.95. Zheng, et al., (2020). Cell Res 30, 492-506.96. Olivares, et al., (2015). Gut 64, 406-417.97. Torres, et al., (2012). Autism Res Treat 2012, 959073.98. Warren, et al., (1996). J Neuroimmunol 67, 97-102.99. Hall, et al., (2023). BMJ Open 13, e067682.100. Pepperell, et al., (2002). Antimicrob Agents Chemother 46, 3555-3560.101. Liu, et al., (2010). FEBS Lett 584, 911-916.102. Liu, et al., (2012). Inflamm Bowel Dis 18, 418-429.103. Kang, et al., (2018). Sci Rep 8, 847.104. Umar, S. (2012). Curr Colorectal Cancer Rep 8.105. Chen, et al., (2023). J Autism Dev Disord. 10.1007 / s10803-022-05829-z.106. Liu, et al., (2019). Front Cell Neurosci 13, 7.107. Zhang, et al., (2018). Front Microbiol 9, 1716.108. van der Schoor, et al., (2004). Gut 53, 38-43.109. Hallen, et al., (2013). Amino Acids 45, 1249-1272.1 10. Wu, et al., (2004). J Nutr Biochem 15, 442-451 .1 11. Park, et al., (2022). Transduct Target Ther 7, 1-23.1 12. Badurdeen, et al., (2015). Pediatr Res 77, 290-297.1 13. Polycarpou, et al., (2013). Journal of Parenteral and Enteral Nutrition 37, 617-622.1 14. Amin, et al., (2002). The Journal of Pediatrics 140, 425-431.1 15. Niinisto, et al., (2017). Diabetologia 60, 1223-1233.1 16. J et al., (2021 ). Molecular neurobiology 58.1 17. Laske, et al., (2022). Front Neurosci 16, 792996.1 18. De Filippis, et al., (2019). Cell Host & Microbe 25, 444-453. e3.1 19. Bennett, et al., (1992). Journal of Inherited Metabolic Disease 15, 220-223.120. Cascio, et al., (2020). Mol Genet Genomic Med 8, e1036.121 . Adams, et al., (201 1 ). Nutr Metab (Lond) 8, 34.122. Shamberger, R.J. (201 1 ). J Am Coll Nutr 30, 348-353.123. Powers, H.J. (2003). The American Journal of Clinical Nutrition 77, 1352-1360.124. Marashly, E.T., and Bohlega, S.A. (2017). Front Neurol 8, 333.125. Jarrett, et al., (2022). The American Journal of Clinical Nutrition 1 16, 1767-1778.126. Russell, et al., (2019). Nat Commun 10, 1-12.127. Ludvigsson, et al., (2011 ). BMC Public Health 11 , 450.128. Bell, C.C. (1994). JAMA 272, 828-829.129. Liebisch, et al., (2020). Journal of Lipid Research 61 , 1539-1555.130. American Psychiatric Association (2013). Diagnostic and Statistical Manual of Mental Disorders Fifth Edition. (American Psychiatric Association)10.1176 / appi. books.9780890425596.131 . Caporaso, et al., (2010). Nature Methods 7, 335-336.132. Callahan, et al., (2016). Nature Methods 13, 581-583.133. Quast, et al., (2013). Nucleic Acids Research 41 , D590-D596.134. Altschul, et al., (1990). J Mol Biol 215, 403-410.135. Chamberlain, et al., (2019). Anal Bioanal Chem 411 , 4807-4818.136. Pluskal, et al., (2010). BMC Bioinformatics 1 1 , 395.137. Koelmel, et al., (2017). BMC Bioinformatics 18, 331 .138. Oksanen, et al., (2022). vegan: Community Ecology Package. Version 2.6-4.139. Love, et al., (2014). Genome Biology 15, 550.140. Benjamini, et al., (1995). Journal of the Royal Statistical Society. Series B (Methodological) 57, 289-300.141 . Ho, et al., (201 1 ). Journal of Statistical Software 42, 1 -28.142. Pang, et al., (2021 ). Nucleic Acids Research 49, W388-W396.143. Pang, et al., (2022). Nat Protoc 17, 1735-1761 .144. Haushalter, et al., (2017). Dev. Biol. 430, 124-41 .145. Covarrubias, et al., (2021 ). Nat Rev Mol Cell Biol. 22: 119-41 .146. Gasperi , et a., (2019). Int J Mol Sci. 20: 974.147. Jabeen, et al., (2023), Front. Antibiot., 2. DQI:10.3389 / frabi.2023.1276982.148. Hackert, et al., (1994), J. Bacteriol., 176:7391 -7394.149 Chen, et al., (2017), Front Immunol. 8. DQI:1 O.3389 / fimmu.2O17.00207.150. Xu et al., (2020), Front Microbiol.1 1 . DOI:10.3389 / fmicb.2020.00219.151. Rodrigues, et al., (2022), Front Immunol 2022; 13: 934695.152. Ribeiro, et al., (2017), Int J Syst Evol Microbiol., 67: 3513-7.153. Luo X, et al., (2022), Microbiol Spectr. 10: e02510-22.154. Sellera et al., (2022), Microbiol Spectr. 10: e01506-21 .155. Ahrens AP, et al., (2024), Cell 187: 1853-1873.e15.156. Zhao Y-T, et al., (2018), Genome Res 28: 933-42.

Claims

What is claimed is:1 . A method of reducing occurrence of a neurodevelopmental disorder (ND) in a subject at risk for developing an ND, the method comprising administering a composition comprising an equol-producing bacterium to the subject, or other gut-health-promoting bacterium mentioned in this application.

2. The method of claim 1 , wherein the equol-producing bacteria is Slackia isoflavoniconvertens, Coriobacteriaceae, Eggerthella, or Adlercreutzia equolifaciens.

3. The method of claim 1 , wherein the gut-health promoting bacteria include, but are not limited to, Akkermansia muciniphila, Roseburia hominis, Faecalibacterium prausnitzii, Erysipelotrichiaceae UCG-003 spp., Alistipes putredinis, Phascolarctobacterium, Turicibater, Anaerostipes caccae, Coprococcus, Acidaminococcales, or Bifidobacterium sp.

4. The method of any one of claims 1-3, wherein the ND is autism spectrum disorder, attention-deficit / hyperactivity disorder, communication disorder, or intellectual disability.

5. A method of reducing occurrence of a neurodevelopmental disorder (ND) in a subject at risk for developing an ND an undergoing treatment for otitis media, the method comprising administering a composition comprising Coprococcus to the subject.

6. The method of claim 4, wherein the Coprococcus is an amoxicillin-resistant strain of Coprococcus.

7. A method of increasing levels of equol in the gut (as measured in stool) in a subject at risk for developing ND, the method comprising administering a composition comprising an equol-producing bacterium to the subject.

8. The method of claim 6, wherein the equol-producing bacteria is Slackia isoflavoniconvertens, Coriobacteriaceae, Eggerthella, or Adlercreutzia equolifaciens.

9. The method of claim 6 or claim 7, wherein the ND is autism spectrum disorder, attention-deficit / hyperactivity disorder, communication disorder, or intellectual disability.

Images