Age-related microbiome metabolites alter RNA splicing and chromatin accessibility in the brain

Age-related microbiome metabolites alter RNA splicing and chromatin accessibility in the brain | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Age-related microbiome metabolites alter RNA splicing and chromatin accessibility in the brain View ORCID Profile Meenakshi Chakraborty , View ORCID Profile Sophia M. Shi , View ORCID Profile Imani E. Porter , Daniel J. Richard , View ORCID Profile Georgi K. Marinov , Ashley A. Moore , View ORCID Profile Jenna L. E. Blum , View ORCID Profile Aravind Natarajan , View ORCID Profile James W. Jahng , View ORCID Profile Joseph C. Wu , View ORCID Profile Sydney X. Lu , View ORCID Profile Shawn M. Davidson , View ORCID Profile William J. Greenleaf , Nay L. Saw , View ORCID Profile Mehrdad Shamloo , View ORCID Profile Anne Brunet , View ORCID Profile Tony Wyss-Coray , View ORCID Profile Ami S. Bhatt doi: https://doi.org/10.1101/2025.10.03.680371 Meenakshi Chakraborty 1 Department of Genetics, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Meenakshi Chakraborty Sophia M. Shi 2 Department of Chemistry, Stanford University , Stanford, CA, USA 3 Stanford Chemistry, Engineering and Medicine for Human Health (ChEM-H), Stanford University , Stanford, CA, USA 4 Department of Neurology and Neurological Sciences, Stanford University School of Medicine , Stanford, CA, USA 5 Wu Tsai Neurosciences Institute, Stanford University School of Medicine , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sophia M. Shi Imani E. Porter 1 Department of Genetics, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Imani E. Porter Daniel J. Richard 1 Department of Genetics, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Georgi K. Marinov 1 Department of Genetics, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Georgi K. Marinov Ashley A. Moore 6 Department of Medicine, Divisions of Hematology and Blood & Marrow Transplantation, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jenna L. E. Blum 7 Division of Pulmonary and Critical Care Medicine, Department of Medicine, The Feinberg School of Medicine, Northwestern University , Chicago, IL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jenna L. E. Blum Aravind Natarajan 6 Department of Medicine, Divisions of Hematology and Blood & Marrow Transplantation, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Aravind Natarajan James W. Jahng 8 Stanford Cardiovascular Institute, Stanford University School of Medicine , Stanford, CA, USA 9 Department of Medicine, Division of Cardiovascular Medicine, Stanford University School of Medicine , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for James W. Jahng Joseph C. Wu 8 Stanford Cardiovascular Institute, Stanford University School of Medicine , Stanford, CA, USA 9 Department of Medicine, Division of Cardiovascular Medicine, Stanford University School of Medicine , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Joseph C. Wu Sydney X. Lu 6 Department of Medicine, Divisions of Hematology and Blood & Marrow Transplantation, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sydney X. Lu Shawn M. Davidson 7 Division of Pulmonary and Critical Care Medicine, Department of Medicine, The Feinberg School of Medicine, Northwestern University , Chicago, IL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Shawn M. Davidson William J. Greenleaf 1 Department of Genetics, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for William J. Greenleaf Nay L. Saw 10 Stanford Behavioral and Functional Neuroscience Laboratory, Wu Tsai Neurosciences Institute, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Mehrdad Shamloo 10 Stanford Behavioral and Functional Neuroscience Laboratory, Wu Tsai Neurosciences Institute, Stanford University , Stanford, CA, USA 11 Department of Neurosurgery, School of Medicine, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Mehrdad Shamloo Anne Brunet 1 Department of Genetics, Stanford University , Stanford, CA, USA 5 Wu Tsai Neurosciences Institute, Stanford University School of Medicine , Stanford, CA, USA 12 Glenn Center for the Biology of Aging, Stanford University , Stanford, CA, USA 13 The Phil and Penny Knight Initiative for Brain Resilience, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Anne Brunet Tony Wyss-Coray 4 Department of Neurology and Neurological Sciences, Stanford University School of Medicine , Stanford, CA, USA 5 Wu Tsai Neurosciences Institute, Stanford University School of Medicine , Stanford, CA, USA 13 The Phil and Penny Knight Initiative for Brain Resilience, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Tony Wyss-Coray Ami S. Bhatt 1 Department of Genetics, Stanford University , Stanford, CA, USA 6 Department of Medicine, Divisions of Hematology and Blood & Marrow Transplantation, Stanford University , Stanford, CA, USA 13 The Phil and Penny Knight Initiative for Brain Resilience, Stanford University , Stanford, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Ami S. Bhatt For correspondence: asbhatt{at}stanford.edu Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract The gut microbiome generates diverse metabolites that can enter the bloodstream and alter host biology, including brain function. Hundreds of physiologically relevant, gut-brain signaling molecules likely exist; however, there has been no systematic, high-throughput effort to identify and validate them. Here, we integrate computational, in vitro , and in vivo approaches to pinpoint microbiome-derived metabolites whose blood levels change during aging, and that induce corresponding changes in the mouse brain. First, we mine large-scale metabolomics datasets from human cohorts (each n ≥ 1200) to identify 30 microbiome-associated metabolites whose blood levels change with age. We then screen this panel in an in vitro transcriptomic assay to identify metabolites that perturb genes linked to age-related neurodegeneration. We then test four metabolites in an acute-exposure mouse model, and use multi-omic approaches to evaluate their impact on cellular functions in the brain. We confirm the known neurodegeneration-promoting effects of trimethylamine N-oxide (TMAO), including mitochondrial dysfunction, and further discover its disruptive impact on the pathways of glycolysis, GABAergic signaling, and RNA splicing. Additionally, we identify glycodeoxycholic acid (GDCA), a microbiome-derived secondary bile acid, as a potent regulator of chromatin accessibility and suppressor of genes that protect the brain from age-related, neurodegeneration-promoting insults. GDCA also acutely reduces mobility. In summary, we present a scalable framework for linking microbiome metabolites to host pathologies, and apply it to identify microbial metabolites that induce molecular changes related to neurodegeneration. Introduction Aging is the greatest risk factor for many neurodegenerative diseases. Remarkably, age-related conditions that often precede these diseases, such as mild cognitive impairment (MCI), can be reversible 1 , 2 . Thus, identifying potentially modifiable age-related factors that promote neurodegeneration could alter and perhaps improve the natural history of disease initiation and progression. One physiological compartment that changes with aging is the gut microbiome 3 – 5 , likely due to factors such as increased permeability of the gut lining, increased medication usage, immunosenescence, and accumulated exposure to environmental toxins. In light of growing interest in the microbiota-gut-brain axis 6 , 7 , a natural question arises: does “microbiome aging” contribute to “brain aging” and associated pathologies? Fecal microbiota transfers between young and aged mice - which can induce or reverse neuroinflammation and cognitive decline - suggest that it does 8 – 10 . However, the microbiome-related molecules and underlying molecular mechanisms that drive these phenotypes remain largely uncharacterized in both animals and humans, which limits translation of this knowledge to the clinic. This gap in the gut-brain field is partly due to the underutilization of in vitro systems, which can help create a pipeline from hypothesis-generating data analyses to in vivo validation. The microbiome is known to substantially impact the levels of blood metabolites, with hundreds of metabolites differing in abundance between germ-free mice and their conventionally raised counterparts 11 , 12 . This is significant because blood metabolites enable the proper functioning of host organs; when their levels change, this can have profound effects on physiology 13 – 15 . We therefore hypothesized that microbiome aging affects brain aging by altering the levels of bioactive blood metabolites. These metabolites could impact the blood-brain barrier - a key regulator of brain homeostasis whose functionality declines with age 16 - and perhaps even cross it to act directly on neurons and glia. To evaluate this hypothesis, we first leveraged public datasets 12 , 17 – 23 to identify a panel of 30 blood metabolites whose levels: 1) change with human aging, and 2) are significantly associated with the presence and/or composition of the gut microbiome. We then screened these candidates in vitro for their ability to alter the expression of genes in pathways related to age-related neurodegeneration, and validated four of the most interesting metabolites in vivo through multi-omic analyses of the mouse brain. Our results confirm and expand the neurodegeneration-promoting effects of the microbiome-derived metabolite trimethylamine N-oxide (TMAO), including novel effects on RNA splicing. In addition, we find that glycodeoxycholic acid (GDCA), a microbiome-derived bile acid, alters chromatin accessibility and silences many genes that maintain brain homeostasis after stress and injury. Finally, in behavioral experiments, we find that GDCA, whose blood levels are correlated with neurodegeneration and cognitive impairment in multiple human cohorts 24 – 28 , acutely reduces mobility. Together, our findings on TMAO and GDCA, our curated panel of age- and microbiome-associated blood metabolites, and the integrative framework we present - combining computational, in vitro , and in vivo approaches - provide a foundation for substantial advances in identifying microbiome-derived molecules that play key roles in host physiology. Results Identifying microbiome-associated metabolites whose blood levels change with age There have been extensive efforts to characterize the specific metabolomic changes that occur with aging 17 . In addition, recent work has identified metabolites associated with microbiome composition and activity 11 , 12 , 21 – 23 . We leveraged these data to build a panel of metabolites for medium-throughput in vitro testing ( Fig. 1 ; Supplementary Table 1; Methods). Download figure Open in new tab Fig. 1. Workflow used to define a panel of 30 microbiome-associated metabolites whose blood levels change with age. First, metabolites whose blood levels changed with age in large cohort(s) were overlapped with a curated list of metabolites that are producible by microbes, either alone or via host-microbe cometabolism. Metabolites were retained for the final panel if they met at least one of two criteria: (1) showed significantly different blood levels in germ-free vs. conventionally raised mice (based on Lai et al. , 2021 12 ), and/or (2) ranked in the top 20% for the proportion of variance in blood levels “explained” by the gut microbiome (metric from the GUTSY Atlas 22 ). After excluding alanine and uridine (see Results), the final panel comprised 30 metabolites. Specifically, we first referenced a recent review on the metabolomics of aging 17 to identify three large cohort studies 18 – 20 that each analyzed blood samples from at least 1,200 participants. Pooling data across these studies yielded hundreds of metabolites whose blood levels changed with age in at least one study. Separately, we curated a list of human-associated metabolites that have been reported to be produced by microbes, either alone or via host-microbe “cometabolism” (where both host and microbial metabolism contribute to the final product). Intersecting this curated list with the age-associated metabolites yielded 79 unique overlapping metabolites. For experimental feasibility, we further narrowed the list, prioritizing metabolites most strongly linked to the gut microbiome. We used two key resources: (1) a recent study comparing blood metabolite levels between germ-free and conventionally raised mice 12 ; and (2) the GUTSY Atlas 22 , which reported the proportion of variance in blood metabolite levels “explained” by the gut microbiome for 1,168 metabolites based on paired stool-blood data for over 8,000 humans. Metabolites were retained for the final panel if they met at least one of two criteria: (1) showed significantly different blood levels between germ-free and conventionally raised mice 12 ; (2) ranked in the top 20% of blood metabolites for the GUTSY variance-explained metric. We eliminated alanine because it is expected to have minimal impact on cellular processes 29 , and uridine because its blood levels increased with age in one study 18 but decreased with age in another 20 . The final panel comprised 30 metabolites, 26 of which increased with age and 4 of which decreased with age (Supplementary Table 1). The metabolites encompassed a broad spectrum of categories, spanning six super classes, thirteen classes, and sixteen sub classes within the Human Metabolome Database (HMDB) 30 taxonomy (Fig. S1). In vitro screening reveals that microbiome-associated metabolites perturb genes and pathways linked to age-related neurodegeneration While in vivo validation remains the gold standard for assessing physiological relevance, we began with in vitro screening, as testing dozens of metabolites in vivo for potential effects on the brain would not have been practically feasible. We chose human brain endothelial cells as our model system because they line the brain’s blood vessels and are the first brain cell type encountered by blood metabolites. In fact, many blood metabolites never reach other brain cell types, since endothelial cells are a key component of the blood-brain barrier (BBB), which selectively regulates entry into the brain parenchyma. Importantly, even metabolites that do not cross the BBB could significantly impact brain function, since BBB decline is a defining feature of brain aging and neurodegeneration 16 . We chose to use hCMEC/D3 cells, an immortalized human brain endothelial cell line widely used in BBB studies 31 . To conduct the screen, each metabolite was added individually to low-passage hCMEC/D3 cells in triplicate, alongside matched vehicle controls containing the corresponding solvent in the media ( Fig. 2A ; Supplementary Table 3). After a 3-hour incubation (chosen over longer timepoints to minimize metabolite clearance), cells were lysed in TRIzol for downstream RNA extraction and sequencing. Each metabolite was added at a representative physiological concentration, primarily based on its listed blood concentrations in HMDB (Methods; Supplementary Table 1). In addition to vehicle controls, we included lipopolysaccharide (LPS) controls in each sequencing batch, given the known effects of LPS on human brain endothelial cells 32 . As expected, LPS induced the expression of genes in KEGG pathways associated with inflammation, such as “TNF signaling pathway” and “NOD-like receptor signaling pathway” ( Fig. 2B ; Fig. S2; Supplementary Table 2). Download figure Open in new tab Fig. 2. In vitro screening reveals that microbiome-associated metabolites perturb genes and pathways linked to age-associated neurodegenerative disease. (A) Schematic illustrating the setup of the in vitro transcriptomic screen. The full panel of 30 metabolites was distributed across four sequencing batches, with each metabolite included in one batch. Within each batch, at least n =3 wells were treated with each assigned metabolite, its respective vehicle control, or the LPS control (see Methods and Supplementary Table 3). (B) Top enriched KEGG pathways for the genes significantly upregulated by the LPS control in the first sequencing batch. The analogous results for batches #2-4 are presented in Fig. S2. The non-specific “KEGG root term” was filtered out. (C) Number of differentially expressed genes (DEGs) per metabolite in the KEGG pathways for Alzheimer’s Disease (AD) and Parkinson’s Disease (PD). Metabolites impacting at least 5 unique AD/PD genes are shown. (D) Identity and characteristics of the four metabolites chosen for subsequent in vivo validation of their effects on the brain. IAA = indoleacetic acid, GDCA = glycodeoxycholic acid, TMAO = trimethylamine N-oxide, PA = palmitic acid, AD = Alzheimer’s Disease, PD = Parkinson’s Disease, GF = germ-free, CSF = cerebrospinal fluid. GUTSY refers to the GUTSY Atlas 22 , which reports microbiome-attributable variance in blood levels for 1,168 blood metabolites. Structure diagrams were produced with rdkit 47 . Relevant studies (e.g., those reporting metabolite detection in CSF or brain) are cited in the text. The number of differentially expressed genes (DEGs), relative to vehicle controls, varied considerably across the 30 test metabolites, reflecting substantial differences in the magnitude of their transcriptional impact (Fig. S3; Supplementary Tables 4-5). The median number of DEGs was 45.5, with an interquartile range (IQR) of 18.5 to 359.5, and a total range of 3 to 6,419 genes affected. Twelve metabolites perturbed ≥5 genes in the KEGG pathways for “Alzheimer’s Disease” (AD) and/or “Parkinson’s Disease” (PD), which are the two most common age-related neurodegenerative disorders ( Fig. 2C ; Supplementary Table 6). The affected AD/PD genes converged on processes including energy metabolism, proteasome-mediated protein degradation, and inflammatory signaling, whose disruptions are central to neurodegenerative disease 33 . We prioritized four metabolites for in vivo validation of their effects on brain function: indoleacetic acid (IAA), glycodeoxycholic acid (GDCA), trimethylamine N-oxide (TMAO), and palmitic acid (PA). TMAO’s effects on the brain have been studied previously 34 , so we included it partially as a control and also to expand the scope of knowledge about its impact on the brain. All four chosen metabolites share the following characteristics ( Fig. 2D ). They (1) impact the expression of at least 5 unique genes in the AD and/or PD KEGG pathways in our transcriptomic screen ( Fig. 2C ), (2) increase in the blood with age (Supplementary Table 1), (3) have been associated with AD, PD, and/or impaired cognition in humans 24 – 28 , 34 – 39 , (4) are strongly associated with production or regulation by the gut microbiome 12 , 22 , 40 , 41 , and (5) have been detected in human cerebrospinal fluid and/or brain 42 – 46 . TMAO disrupts transcriptional homeostasis and RNA splicing in the mouse brain To interrogate the effects of each of the four chosen metabolites on cellular functions in the brain in vivo , we conducted bulk transcriptomics on mouse brain hemispheres following intravenous retro-orbital injection of each metabolite or vehicle (2-hour treatment; n =4 male mice per injection group; Fig. 3A ). Doses were chosen to reflect physiologically relevant exposures, by converting blood concentrations previously observed in seniors 26 , 28 , 48 – 55 to mouse-equivalent doses using well-established scaling guidelines 56 , 57 . This experiment design was informed by prior work on TMAO 58 , such that TMAO could serve as a positive control. The main modification was the use of intravenous retro-orbital injection - a well-established technique 16 , 59 – 61 -in place of intraperitoneal injection, to enable direct entry into systemic circulation and more precise control of blood metabolite levels. Download figure Open in new tab Fig. 3. The four metabolites selected for in vivo experiments impact brain gene expression to markedly different extents. (A) Schematic illustrating the setup of the in vivo transcriptomics experiment; n =4 C57BL/6 male mice were injected with each metabolite or vehicle. (B) Volcano plots showing the results of differential gene expression analysis between PA-and IAA-treated brains and their vehicle controls. (C) Volcano plots showing the results of differential gene expression analysis between TMAO- and GDCA-treated brains and their vehicle controls. In both (B) and (C) , the y-axis represents the -log10(adjusted p -value) from DESeq2, and the x-axis represents the log2FoldChange vs. vehicle control, after applying the recommended log2FoldChange shrinkage via lfcShrink 63 . Purple dots represent significantly upregulated genes, with an adjusted p ≤ 0.1 (DESeq2 default) and shrunken log2FoldChange ≥ 0.1. Teal dots represent significantly downregulated genes (adjusted p ≤ 0.1 and shrunken log2FoldChange ≤ -0.1). The y-axis for the GDCA volcano plot was capped at 30 to avoid distortion by extreme values; the non-capped, adjusted p -values for Apold1 and Ccn1 are 1.82E-108 and 3.50E-70, respectively (Supplementary Table 7). For TMAO, the chosen mouse dose was 1.8 mg/kg, the same as the original study 58 , approximating 30μM concentration in human blood. The remaining doses were 22.2 mg/kg for PA (∼100μM human), 1.5 mg/kg for IAA (∼10μM human), and 0.4 mg/kg for GDCA (∼1μM human). All human-equivalent concentrations were non-cytotoxic in vitro in hCMEC/D3 cells (Fig. S7). For PA - unlike for the other metabolites - the human-equivalent concentration used for the mouse experiment was reduced relative to the concentration used in the in vitro transcriptomic screen. This was because the original concentration of 500μM was cytotoxic to hCMEC/D3 cells (Fig. S7), likely explaining the disproportionately large number of DEGs observed for PA in vitro (Fig. S3). Overall, the cytotoxicity testing (Fig. S7) ensured that the mouse exposures reflected physiologically relevant human levels while avoiding human concentrations that are overtly cytotoxic. PA and IAA had negligible effects on gene expression in the mouse brain, with fewer than ten differentially expressed genes per metabolite ( Fig. 3B ; Supplementary Table 7). By contrast, TMAO and GDCA each had a significant transcriptional impact in the brain, modulating the expression of 2,305 and 179 genes, respectively ( Fig. 3C ; Supplementary Table 7). We first evaluated the TMAO results to verify concordance with the literature and also to derive additional insights about the impact of this microbiome metabolite on brain function. In agreement with prior literature 58 , we observed a coordinated downregulation of mitochondrially-encoded genes involved in oxidative phosphorylation, including mt-Nd1, mt-Nd2, mt-Nd3, mt-Co2, mt-Co3, mt-Cytb, mt-Atp6, and mt-Atp8 (Supplementary Table 7). However, gene set enrichment analysis (GSEA) based on hallmark gene sets revealed a more complex rewiring of bioenergetic pathways ( Fig. 4A ; Supplementary Table 8). Despite the clear downregulation of mitochondrial oxidative phosphorylation, the term “oxidative phosphorylation” was positively enriched in GSEA when comparing TMAO-treated brains to controls. This result was due to the upregulation of several nuclear-encoded genes involved in oxidative phosphorylation (e.g., Cox4i1, Cox6a1, Cox6b1, Cox7a2l, Ndufa7, Ndufa8, Ndufb2, Ndufb6, Ndufb8, Ndufc2 ), perhaps reflecting a compensatory response, similar to that observed in patients with macro-deletion of mtDNA 62 . Download figure Open in new tab Fig. 4. TMAO disrupts transcriptional homeostasis and RNA splicing in the mouse brain. (A) Selected significantly enriched terms from gene set enrichment analysis (GSEA) based on hallmark gene sets; full results in Supplementary Table 8. (B) The glycolysis pathway, annotated with the glycolytic genes upregulated by TMAO. * = genes whose products catalyze rate-limiting steps of glycolysis. Base figure adapted from Hu et al. , 2014 77 . (C) Selected significantly enriched terms from GSEA using the REACTOME pathway database; full results in Supplementary Table 8. (D) Percentage of marker genes in each brain cell class that are downregulated by TMAO. Neuron = neuronal lineage, Astro = astrocyte lineage and stem cells, Epend = ependymal cells, Vasc = vasculature cells, Olg = oligodendrocyte lineage, Immune = immune cells. (E) Results of differential splicing analysis comparing brains from TMAO-exposed and control mice. Each point represents a splicing event. The x-axis represents the inclusion level difference (ΔPSI) and the y-axis represents the -log10(FDR), capped at 15 for visual clarity. Events significantly enriched in control or TMAO are shown in purple and teal, respectively; non-significant events are shown in gray. To reduce visual clutter, points were filtered to those with -log10(FDR) > 0. RI = retained intron, A5SS = alternative 5’ splice site, SE = skipped exon. (F) Unified model of TMAO’s neurodegeneration-promoting effects on the brain. Asterisks indicate portions of the model that are novel based on insights derived from this study. TMAO induces mitochondrial dysfunction, leading to compensatory upregulation of glycolysis and increased production of reactive oxygen species (ROS). The resulting oxidative stress further promotes neuroinflammation. Perhaps as a result of the rewired bioenergetic state, TMAO also promotes synaptic dysfunction, including impairments in GABAergic transmission. In addition, it downregulates splicing factors, resulting in changes to splicing patterns. In addition, GSEA highlighted the term “glycolysis” as enriched ( Fig. 4A ). Indeed, eight glycolytic genes were upregulated by TMAO, including Hk1 and Pkm , which catalyze rate-limiting steps of glycolysis 64 ( Fig. 4B ; Supplementary Table 7). Furthermore, Slc2a1 (i.e., Glut1 ), a major glucose transporter in the brain that supports glycolytic flux, was the second most significantly upregulated gene by TMAO (adjusted p -value 5.62E-11; Fig. 3C and Supplementary Table 7). These findings agreed with data from our in vitro transcriptomic screen, where TMAO upregulated human glycolytic genes including HK1 , PKM , PFKP , GPI , and ALDOA (Supplementary Table 5). Overall, our results align with previous findings that TMAO results in mitochondrial dysfunction in the brain 34 . Furthermore, our findings demonstrate a previously unreported upregulation of the glycolytic pathway in the brain. Mitochondrial dysfunction can cause oxidative stress. Consistent with this, the “reactive oxygen species pathway” was significantly more highly expressed ( Fig. 4A ) in the brain transcriptome of TMAO-exposed animals vs. controls. In addition, Txnip - a known inducer of oxidative stress and activator of the inflammasome 65 - was significantly upregulated by TMAO (Supplementary Table 7). This finding is consistent with the most enriched GSEA term, “TNFα signaling via NF-кB” ( Fig. 4A ), which reflects inflammatory signaling. Together, these results support previous studies showing that TMAO promotes oxidative stress and inflammation, including in the brain 34 , 66 , 67 . GSEA using the REACTOME pathway database provided additional insights, including the positive enrichment of genes involved in synaptic transmission and GABAergic signaling, especially genes related to GABA B receptor activation ( Fig. 4C ; Supplementary Table 8). By contrast, we observed a marked downregulation of genes encoding GABA A receptors (Fig. S8). These results suggest that TMAO may shift inhibitory neural circuits from fast GABA A -mediated responses towards slower, GABA B -mediated responses. Although previous literature has demonstrated negative effects of TMAO on synaptic function 34 , this is the first observation, to our knowledge, of specific perturbation of GABAergic signaling pathways. To assess the extent of TMAO’s impact on neuronal cells vs. other brain cell types, we computed “marker genes” for six different brain cell classes, using a well-characterized single-cell dataset from the mouse brain 68 (Methods). TMAO downregulated over 15% of the marker genes for cells in the neuronal lineage, representing more than twice its impact on any of the other five cell classes ( Fig. 4D ). GSEA also highlighted the depletion of genes associated with “mRNA splicing” ( Fig. 4C ). Strikingly, several of the genes that were most significantly downregulated by TMAO (e.g., Tia1 , Srrm2 , Pnisr , and Rbm39 , see Fig. 3C and Supplementary Table 7) regulate alternative splicing 69 – 72 . This result was intriguing given the well-established link between changes in splicing patterns and the development of neurological disease 73 . Given the observed impact of TMAO on splicing-related factors, we sought to determine whether any genes were differentially spliced in TMAO-treated animals vs. controls. In agreement with the changes in splicing factor expression, the widely used rMATS-turbo software 74 nominated 126 differential splicing events between TMAO-treated and vehicle-treated brains, which we narrowed to 44 using conservative filters ( Fig. 4E ; Supplementary Table 9; Methods). Manual inspection in Integrative Genomics Viewer (IGV) revealed clear patterns consistent with differential splicing, including that of Kmt2b , a gene whose aberrant splicing has been linked to dystonia 75 , 76 , a movement disorder that can occur independently or as a symptom of Parkinson’s disease (Fig. S9). Kmt2b also exhibited differential expression in TMAO-vs. vehicle-treated brains (log2FoldChange -0.39, adjusted p 0.001; Supplementary Table 7), potentially as a consequence of its differential splicing. To our knowledge, this is the first demonstration of TMAO regulating alternative splicing patterns. Overall, our findings confirm 34 , 58 and broaden current understanding of TMAO’s neurodegeneration-promoting effects, which we summarize in a unified model ( Fig. 4F ). GDCA remodels chromatin and dramatically suppresses genes that protect the brain from stress and injury Like TMAO, GDCA is a well-known microbiome-derived metabolite. It is a secondary conjugated bile acid, formed when the host conjugates microbially produced DCA with glycine. Blood levels of GDCA have been associated with neurodegeneration and/or cognitive impairment in at least five studies 24 – 28 (summarized in Supplementary Table 10). However, GDCA’s brain-specific effects have not, to our knowledge, been experimentally evaluated in mammals. GDCA strongly suppressed the expression of many genes in the mouse brain ( Fig. 3C ; Supplementary Table 7). The most significantly downregulated gene was Apold1 (log2FoldChange -3.7, adjusted p -value 1.82E-108), a gene whose expression is critical for stroke recovery in mice 78 . Ccn1 , the second most downregulated gene (log2FoldChange -3.0, adjusted p -value 3.50E-70), similarly promotes wound healing and tissue repair 79 . This pattern extended beyond individual genes: many of the genes that were strongly downregulated by GDCA (e.g., Klf2 , Klf4 , Adamts1 , Akap12 , Maff , Plaur , Ddit4 , Atf3 , Btg2 ) are involved in cellular responses to injury and stress, including oxidative stress, hypoxia, DNA damage, and stroke 80 – 88 . These observations were supported by GSEA based on hallmark gene sets. GSEA demonstrated significant negative enrichment of the terms “TNFα signaling via NF-кB” and “hypoxia” ( Fig. 5A ), as well as nominal negative enrichment of “p53 pathway” (Supplementary Table 8), consistent with reduced activity of stress-response pathways. The results suggested marked disruption of endothelial cell homeostasis; for example, the simultaneous reductions of Klf2 and Klf4 alone would be expected to cause severe endothelial dysfunction 89 , 90 . However, given the use of bulk RNA-seq, it is unknown whether these decreases originated specifically from endothelial cells, or from other brain cell types. Download figure Open in new tab Fig. 5. GDCA suppresses stress-response genes and remodels chromatin in the brain. (A) All significantly enriched terms from gene set enrichment analysis (GSEA) based on hallmark gene sets; full results in Supplementary Table 8. (B) All significantly enriched terms from GSEA using the REACTOME pathway database; full results in Supplementary Table 8. (C) Volcano plot of differential chromatin accessibility between GDCA-treated and vehicle control mouse brains, based on ATAC-seq of flash-frozen right hemispheres from the initial RNA-seq cohort (see Methods). Y-axis: -log10(adjusted p-value) from DESeq2; x-axis: log2FoldChange vs. vehicle control. Purple dots represent genomic peaks with higher accessibility in brains from GDCA-exposed mice (adjusted p ≤ 0.1 and positive log2FoldChange). Teal dots represent genomic peaks with lower accessibility in brains from GDCA-exposed mice (adjusted p ≤ 0.1 and negative log2FoldChange). (D) Coverage heatmap for the 200 genomic peaks that were significantly less accessible in brains from GDCA-exposed mice, vs. control. Each row represents one peak. BAM files for individual brains within each condition were merged prior to visualization (Methods). CPM = counts per million. (E) Observed versus expected proportions of differentially accessible peaks overlapping various genomic features. Expected proportions were calculated based on the reference genome. Introns and promoters are significantly enriched among differential peaks, whereas intergenic regions are depleted. The symbol * indicates p < 0.05 and *** indicates p expected and lower-tailed when observed < expected). (F) Transcription factor (TF) motifs with the highest variability in accessibility across ATAC-seq samples, as quantified by chromVAR 96 . Dots represent point estimates of motif variability and horizontal lines indicate the corresponding 95% bootstrap confidence intervals. The color scale shows the significance of variability as -log10(adjusted p -value from chromVAR). To facilitate visualization, -log10(padj) values were capped at 200; the true padj for the variability of MA0007.4_Ar across samples is 0. GSEA using the REACTOME pathway database ( Fig. 5B ; Supplementary Table 8) demonstrated the positive enrichment of several terms related to chromatin dynamics, including “PKMTs methylate histone lysines,” “epigenetic regulation of gene expression,” and “chromatin organization.” Based on this result, we postulated that GDCA achieves its notable gene regulatory effects, including the strong silencing of many genes, by inducing changes in chromatin organization and accessibility. To test this hypothesis, we conducted bulk ATAC-Seq on the flash-frozen right brain hemispheres of the GDCA-and vehicle-treated mice. Consistent with our prediction that GDCA induces chromatin changes, ATAC-seq identified hundreds of genomic regions with differential accessibility between the brains of GDCA- and vehicle-treated mice ( Figs. 5C-D ; Fig. S10A-B; Supplementary Table 11). 200 of the 204 differentially accessible peaks (98%) were less accessible in the GDCA-exposed brains. Moreover, using HOMER to annotate each differential peak with its nearest gene, we found that six of the “nearest genes” were also differentially expressed in RNA-seq (Supplementary Table 11). This set of six genes included several known stress-response genes: Ddit4 85 , Errfi1 91 , Rhob 92 , and Sgk1 93 . For all six genes, the direction of change was concordant across datasets -decreased expression of the gene and decreased accessibility of the nearby peak - providing compelling evidence that the strong transcriptional changes induced by GDCA are driven, at least in part, by chromatin remodeling. We further annotated the differentially accessible peaks based on overlapping genomic features and compared the proportion of peaks in each category to the proportions expected from the reference genome. This analysis revealed enrichment of differential peaks in promoters and introns ( Fig. 5E ; Fig. S10C; Supplementary Table 11). Promoters and introns frequently harbor regulatory elements such as enhancers 94 ; therefore, the enrichment of these features among differential peaks further supports a direct link between GDCA-induced changes in chromatin accessibility and effects on gene expression. Consistent with this interpretation, 4 of the 5 most significantly differential intronic peaks fully overlapped elements annotated as distal enhancers in the ENCODE SCREEN database ( https://screen.wenglab.org/ ) 95 . Promoters and enhancers commonly contain motifs that bind transcription factors (TFs) critical for the regulation of gene expression. We therefore used chromVAR 96 to assess whether GDCA altered the accessibility of any TF motifs from the “CORE” collection of the well-established JASPAR 97 database. We found that TF motif accessibility patterns quantified by chromVAR clearly separated GDCA-exposed brains from controls (Fig. S11A). The TF motif with the highest variability across samples was MA0007.4, which is recognized by the Ar transcription factor that coordinates the cellular response to androgens ( Fig. 5F ; Supplementary Table 11). This Ar motif was consistently less accessible in the GDCA-exposed brains than in controls (Fig. S11B). A recent study showed that certain microbiome-derived bile acids, which share the sterol structure of androgens, can antagonize the human androgen receptor 98 . Thus, GDCA may block mouse Ar from nuclear translocation and binding to its target motifs, thereby reducing motif accessibility - since TF binding usually opens chromatin 99 - and suppressing androgen-responsive transcription. Consistent with this model, several of the genes strongly downregulated by GDCA (e.g., Adamts1 , Akap12 , Sgk1 ) belong to the “Androgen Response” hallmark gene set from MSigDB. GSEA likewise showed negative enrichment of this gene set (normalized enrichment score = -2.17; Supplementary Table 8), although this did not reach statistical significance after multiple testing correction (raw p = 0.05, adjusted p = 0.16), possibly reflecting a lag between certain chromatin changes and downstream transcriptional effects. GDCA acutely reduces mobility Growing evidence indicates that androgens support both cognitive and motor function 100 – 102 . In addition, increased chromatin accessibility can be beneficial for cognitive function 103 , in stark contrast to the accessibility reductions observed after GDCA treatment. Therefore, we hypothesized that GDCA impairs cognitive and motor performance and may also influence additional aspects of mouse behavior. To test this, we first administered GDCA for three days (0.4 mg/kg/day, intravenous retro-orbital injection) and assessed behavior ( Fig. 6A ). There were no detectable differences between GDCA- and vehicle-treated mice, as assessed in an activity chamber on day 4 or in a Y-maze on day 5 ( n =15 male mice per group; see Figs. 6C-E , 6G-H , and Supplementary Table 12). We reasoned that the administered GDCA would likely be cleared by the day of testing (e.g., via hepatic uptake 104 ), such that systemic levels were no longer elevated. Therefore, in a second mouse cohort, we evaluated the impact of acute exposure ( Fig. 6B ). Here, GDCA was administered via oral gavage, because retro-orbital injection, which is invasive, is not compatible with same-day behavioral testing. Gavage doses (10 or 50 mg/kg) were based on prior mouse studies evaluating GDCA’s impact on obesity and ovarian function 105 , 106 , and previous estimates that only a small percentage (5-10%) of bile acids escape the enterohepatic circulation 107 , 108 . Download figure Open in new tab Fig. 6. GDCA acutely reduces mobility. (A) Schematic depicting the initial behavioral experiments performed. There were n =15 mice per group, yielding n =30 mice total in this first behavioral cohort. (B) Schematic depicting the second set of behavioral experiments. There were n =15 mice per group, yielding n =45 mice total in this second behavioral cohort. Core body temperature was also measured in this cohort. (C) Total distance moved in the periphery of the activity chamber (cohort 1). Groups were compared using Welch’s two-sample, two-sided t -test. (D) Same as (C), but for total distance moved in the center of the activity chamber. (E) Percent alternation between maze arms over the 5-minute duration of the Y-maze test (cohort 1). An alternation occurs when, after exiting one maze arm, the mouse does not enter one of the two arms they just visited, but rather enters the novel arm. Alternation percentages are defined as 100% * (number of alternations) / (total number of opportunities to alternate). Statistics compare the alternation percentages within each group to the chance level (50% 113 , 114 ), using Welch’s one-sample, two-sided t -test. (F) Same as (E), but for cohort 2. One mouse from the G_50 group was excluded as it had less than three arm entries and thus had no opportunities to alternate. (G) Cumulative entries into the arms of the Y-maze (cohort 1). (H) Distance moved within the Y-maze, minute-by-minute (cohort 1). (I) Same as (G), but for cohort 2. (J) Same as (H), but for cohort 2. (K) Boxplot of the data from minutes 0-5 in panel (I). Groups were compared using one-way ANOVA with Dunnett’s post-hoc test. (L) Boxplot depicting the total distance traveled in the Y-maze by cohort 2, calculated as the sum of the values underlying panel (J). Statistics as in (K). (M) Core body temperature (°C) in cohort 2 measured at 0, 10, 30, and 60 min after gavage, six days after Y-maze testing. Bars show mean ± SEM. All n =15 mice per group were measured at time 0; n =5 mice per group were measured at each subsequent timepoint. Within each treatment group, time effects were assessed using one-way ANOVA; when significant, Dunnett’s posthoc test was also applied (statistics shown). In all panels, ns indicates not statistically significant, * indicates p ≤ 0.05, ** indicates p ≤ 0.01, *** indicates p ≤ 0.001, and **** indicates p ≤ 0.0001. In all boxplots, boxes represent the interquartile range (IQR), the horizontal line indicates the median and whiskers extend between (25 th percentile -1.5*IQR) and (75 th percentile + 1.5*IQR). One hour after administration of GDCA or vehicle, mice were tested in the Y-maze. Like vehicle-treated mice, GDCA-treated mice alternated between maze arms above chance levels ( Fig. 6F ), indicating no impairment in spatial working memory. However, GDCA reduced mobility in a dose-dependent manner, as evidenced by fewer entries into maze arms and lower total distance moved within the Y-maze ( Figs. 6I-L ; Supplementary Table 12). At 50 mg/kg GDCA, the mobility reduction was accompanied by a reduction in core body temperature ( Fig. 6B and 6M ). Total distance moved also trended towards a GDCA-induced reduction within the setup of cohort 1 ( p =0.054; Fig. 6H ; Supplementary Table 12). Although this cannot be formally established from the current data, the concurrent decreases in mobility and body temperature following oral GDCA administration suggest a centrally mediated response, as specific populations of neurons in the pre-optic area of the hypothalamus have been shown to jointly regulate locomotion and thermoregulation in mice 109 , 110 . In addition, elevated plasma GDCA has been associated with Parkinson’s Disease 26 , a disorder that primarily involves the central nervous system and is characterized by both impaired locomotion 111 and a reduced mesor (rhythm-adjusted mean) of core body temperature 112 . Discussion The gut microbiome produces and impacts the systemic concentration of hundreds of metabolites, each of which may profoundly influence host organs, including the brain. However, systematic, high-throughput efforts to identify metabolites relevant to gut-brain signaling are lacking, partly because testing hundreds of molecules in animal models - the most accurate system to assess brain effects - is impractical. In this study, we integrated computational, in vitro , and in vivo approaches to identify age-related, microbiome-derived blood metabolites that - at physiologically relevant levels - induce significant molecular changes in the mouse brain. Among other changes, we observed impacts on RNA splicing and chromatin accessibility, whose dysregulation is implicated in brain aging and in age-related neurodegeneration 73 , 115 , 116 . Furthermore, we found that a microbiome-derived secondary bile acid, GDCA, can reduce mobility, a function that declines with aging and neurodegenerative disease. These findings demonstrate that “microbiome aging” may substantially contribute to “brain aging” through impacts on the blood metabolome, and that targeting the microbiome and its metabolites may offer cognitive benefits during aging. The aminoxide TMAO has been extensively studied in the context of cardiovascular disease 13 , 117 , and there is a growing body of research on its impact on the brain, where it has been shown to induce mitochondrial dysfunction, oxidative stress, neuroinflammation, and reductions in synaptic plasticity 34 , 58 . We found that TMAO also upregulates the glycolytic pathway in the brain - perhaps to compensate for mitochondrial dysfunction - and impacts RNA splicing by inducing strong downregulation of splicing factors and driving dozens of differential splicing events. Although prior work has identified splicing changes in the brains of germ-free vs. conventionally raised animals 118 , to our knowledge, this is the first demonstration that specific microbiome-derived metabolites can alter host splicing patterns - whether in the brain or in any other host organ. GDCA may also affect splicing ( Fig. 5B ), raising questions about the relative impacts of each individual microbiome-derived blood metabolite and how these impacts might converge to influence normal brain function and the development of brain pathologies. Unlike TMAO, little is known about the effects of GDCA on the mammalian nervous system, with prior knowledge limited to its association with neurodegenerative disease and cognitive impairment in several human cohorts 24 – 28 . Here, we build upon this knowledge, finding that GDCA alters chromatin accessibility, which might be the mechanism by which it also silences genes required for stress and injury response in the mouse brain. Large microbiome perturbations, such as antibiotics and germ-free status, have been shown to impact the brain epigenome 119 ; however, previous metabolite-specific work has focused on the impact of short-chain fatty acids 120 , leaving open questions about the many other classes of microbial effectors, including bile acids. Our data suggest that GDCA substantially alters the brain transcriptome and epigenome, largely by antagonizing the host androgen receptor. This may suppress host androgen signaling, a pathway that is generally considered beneficial for cognitive and motor function 100 – 102 . This hypothesis aligns with our finding that GDCA exposure induces mobility deficits. However, further study - with varied dosing and testing paradigms - will be required to fully elucidate the cognitive and motor impacts of GDCA. Although our work clearly demonstrates that circulating microbiome-derived metabolites can induce significant molecular alterations in the brain, it has several limitations. First, we tested only short-term exposures. However, brain aging and neurodegeneration occur over long periods of time. It will thus be important to decipher the short- vs. long-term effects of metabolites of interest. In vitro , this may require determining the half-life of each metabolite and ensuring continuous exposure to the desired concentration. In vivo , metabolites may need to be administered via osmotic minipump or through the diet. To determine the most physiologic dosing regimens, pharmacokinetic (PK) studies - involving metabolomics at various timepoints - will be necessary. Indeed, even the limited effects of acute exposure to PA and IAA in our study ( Fig. 3B ) may simply reflect suboptimal bioavailability of the administered formulations, underscoring the value of PK studies. Another limitation is that, while our results show that TMAO increases the expression of glycolytic enzymes, both in vitro and in vivo , we did not definitively establish changes in glycolytic flux. This can be investigated using 13 C isotope tracing or similar approaches. Additionally, while we find interesting molecular phenotypes in adult mice, these results may or may not translate to aged animals. Future in vivo work in aged mice will be informative, in this regard, given that age may modify the brain’s response to each metabolite. Relatedly, age-related comorbidities such as diabetes and cardiovascular disease may impact host response to individual metabolites, highlighting the importance of testing metabolite effects in the relevant disease models. Finally, because our multi-omic analyses were conducted using bulk rather than single-cell methods, the cell-type-specific effects of TMAO and GDCA remain unclear. Despite these limitations, our work underscores that the gut microbiome, which is potentially modifiable, is a valuable therapeutic target in the context of brain aging. We identify specific age-associated microbiome metabolites that impact the transcriptional and epigenetic landscape of the brain, with likely consequences for brain function, as demonstrated by the observed impact of GDCA on locomotion. In addition, although we focused on brain aging in this work, the framework we describe -beginning with metabolomics data analysis, followed by in vitro transcriptomic screening, and concluding with in vivo studies of host organs and behavior - can be applied to identify the microbiome metabolites relevant to any host pathology of interest, providing a powerful tool for the field of microbe-host interactions. Methods Identifying microbiome-associated metabolites whose blood levels change with age To identify metabolites whose blood levels change with aging, we leveraged data from three large cohort studies 18 – 20 listed in a recent review 17 on the metabolomics of aging. From Menni et al . 18 , we retained all metabolites from Table S2, since only metabolites significantly associated with age were included in this table. From Dunn et al . 19 , we retained metabolites with p adj <= 0.05 under the test labeled “_age.g2” in Supplementary Material 2 (“AGE” tab). From Darst et al . 20 , we retained metabolites with p_age <= 0.05 in Table S1. Separately, we curated a list of 329 human-associated metabolites that have been reported to be produced by microbes and/or are the products of host-microbe cometabolism. To create the list, we first downloaded Table S3 from a recent study on interactions between human-associated metabolites and GPCRs 21 , and filtered to metabolites that included “Bacteria” in the “Source” column. After filtering, there were 320 metabolites. We then added the following nine metabolites known to be microbially producible and/or producible through host-microbe cometabolism 23 , 121 – 124 : acetic acid, butyric acid, indole-3-carboxaldehyde, indolelactic acid, enterolactone, imidazole propionate, p-cresol sulfate, hippuric acid and indoleacetic acid. Then, we intersected the age-associated metabolites from each study with the curated list of metabolites associated with microbial metabolism. The intersection method was different per study given the differing formats of the tables. Specifically, for Darst et al. , we were able to intersect based on HMDB ID. We also noted that there was one metabolite that matched on CAS number but not on HMDB ID, given that it was listed as “ornithine HCl” in our curated list of microbiome-associated metabolites, but simply “ornithine” in the Darst study. So, we added ornithine manually to the intersection output. For the Menni and Dunn datasets, which lacked standardized identifiers, we used fuzzy string matching (via the Python package fuzzywuzzy, v0.18.0) to identify metabolites with similar names, followed by manual curation to confirm true matches. We also added two intersecting metabolites that were missed by fuzzy matching -indoleacetate and hippurate. This process yielded 114 overlapping metabolites, 79 of which were unique based on HMDB ID. To narrow the list to metabolites most strongly associated with the gut microbiome, we leveraged two additional resources: (1) a recent study comparing blood metabolite levels between germ-free and conventionally raised mice 12 ; and (2) the GUTSY Atlas 22 , which reported the proportion of variance in blood metabolite levels “explained” by the gut microbiome for 1,168 metabolites based on paired stool-blood data for over 8,000 humans. Metabolites were retained for the final panel if they met at least one of two criteria: (1) showed significantly different blood levels between germ-free and conventionally raised mice 12 ; (2) ranked in the top 20% of blood metabolites for the GUTSY variance-explained metric. We eliminated alanine and uridine, the former due to its minimal expected impact on cellular processes 29 , and the latter because its blood levels increased with age in one study 18 but decreased with age in another 20 . The final table of 30 metabolites and their properties is contained in Supplementary Table 1. The table also specifies the exact product that was ordered for testing in the in vitro transcriptomic screen. Of note, at a later date, after conducting the in vitro RNA-seq experiments, we discovered that the p_age values in Table S1 of the Darst study had not been corrected for multiple hypothesis testing. However, even after we manually corrected for multiple hypothesis testing (using a False Discovery Rate approach in R v4.3.2), all the metabolites from the Darst study in our final panel maintained significance using an FDR threshold of padj ≤ 0.1, with all but cholate (padj = 0.065) and glycodeoxycholate (padj = 0.061) also having a padj ≤ 0.05. Cell culture Human cerebral microvascular endothelial cells (hCMEC/D3; Sigma, Cat #SCC066) were obtained as a gift from the lab of Prof. Mark Kay at Stanford University. Cell line authentication was conducted using the ATCC Human Cell STR Profiling Service at passage #9 in October 2023, prior to initiating the experiments described in this manuscript. Cultures were maintained in a 37°C, 5% CO 2 incubator at passage numbers ≤ 22, well below the limit of 35 established by the developers of the cell line 31 . The cell medium was prepared from the EGM2-MV BulletKit (Lonza, Catalog #CC-3202). The VEGF supplement was excluded from the medium based on literature precedent 58 , 122 and due to VEGF’s ability to impair the key barrier-forming properties of the cells 125 . Cells were passaged every two days and were maintained on tissue culture-treated plates. Cells were tested for mycoplasma contamination at six-month intervals during periods of active culture using the ATCC Universal Mycoplasma Detection Kit (ATCC, Cat #30-1012K). In vitro transcriptomic screen - experimental details The screen was conducted in four separate batches using 12-well plates. Prior to cell seeding, each well was coated with 0.5ml of a 1:20 dilution of rat tail collagen (Sigma, Catalog #08-115) in DPBS (Fisher, Catalog #MT21031CV). Collagen was incubated on the plates at 37°C for 1 hour and rinsed off twice with DPBS before cell seeding. Cells were seeded at passages #9-10, at a density of 100,000 cells per well. Seeded cells were grown for six days prior to treatment, based on prior literature 58 , 122 . The cell culture medium was refreshed every two days. Each 12-well plate was organized to include three wells for vehicle controls and nine wells for testing three metabolites, with three biological replicates per metabolite. All metabolites tested on a given plate were dissolved in the same solvent (H 2 O or DMSO) and vehicle controls included an equivalent volume (1%) of the solvent in the final media used for the assays. Because urate was insufficiently soluble in H 2 O or DMSO alone, the final treatment media for this metabolite included a small percentage (approximately 0.1%) of NaOH in addition to 1% H 2 O. Each batch consisted of three 12-well plates, which allowed room for testing nine metabolites per batch. To allow for inclusion of an LPS control condition per batch, eight candidate metabolites were tested per batch. Specifically, batches #2-4 tested 8 metabolites each and batch #1 tested 6, to cover the entire set of 30 metabolites (see Supplementary Table 3). Cells were treated with a metabolite, vehicle, or LPS for 3h before being harvested in TRIzol and stored at - 20°C for downstream RNA extraction and sequencing. For each metabolite, we determined a physiologically relevant concentration to add to cells (Supplementary Table 1), primarily based on information listed in the Human Metabolome Database (HMDB) as of November 2023. Specifically, for each metabolite, we searched for the metabolite in HMDB, clicked the “Concentrations” tab of the relevant entry, examined the concentrations listed for “normal” adult blood samples, and selected a concentration based on those listings. HMDB did not provide blood concentrations for four metabolites: glycodeoxycholate, imidazole propionate, methyl indole-3-acetate, and N-acetyl-L-alanine. For glycodeoxycholate, we chose the concentration based on Table S2 of Qing et al. , Schizophrenia 2022 126 . For imidazole propionate, a relevant physiological concentration was determined based on Fig. 1 of Molinaro et al. , Nature Communications 2020 121 . For methyl indole-3-acetate, we were not able to find literature on the exact concentrations in adult blood samples. We tested 1µM because it was between previously reported concentrations of the metabolite in adult urine 127 and the chosen test concentration of the related metabolite indole-3-acetate. For N-acetyl-L-alanine, we tested 3µM based on previous literature that included in vitro experiments with this metabolite (Yang et al. , Int J Mol Sci 2023; Section 2.2 128 ). RNA was extracted using the following protocol. Samples were thawed to room temperature and incubated for an additional five minutes after thaw. 200µl chloroform was then added per 1ml of TRIzol used; then, samples were briefly vortexed and incubated 2-3 minutes at room temperature, prior to a 15-minute centrifugation at 17,000 x g at 4°C. The aqueous phase was transferred to a new tube. Steps 2-7 of the “Part 1” quick-start protocol for the RNeasy Mini Kit (Qiagen, Catalog #74106) were then executed, skipping the on-column DNase digestion but including the optional dry spin. DNase digestion was conducted using TURBO DNase (Fisher Scientific, Catalog #AM2239). Then, the “RNA cleanup” section of the RNeasy Mini handbook was executed, using 2 minutes rather than 1 minute for the dry spin. Finally, an ethanol precipitation was conducted for all samples to clean up any ethanol contamination, based on the relevant protocol from Cold Spring Harbor Laboratories 129 . RNA samples were sent to Novogene for quality control, mRNA library preparation (poly A enrichment), and sequencing (NovaSeq 6000, 150bp paired end reads, 6Gb raw data per sample). In vitro transcriptomic screen - computational analysis Computational analysis of the in vitro RNAseq data was performed separately for each sequencing batch, as follows. First, FASTQ files from Novogene were trimmed to remove adapter and low-quality bases using TrimGalore v0.6.10 with the following parameters: -q 30 --fastqc --stringency 3 --max_n 15 --output_dir trimming_out --cores 2 --retain_unpaired –paired _1.fq.gz _2.fq.gz. Trimmed paired reads were pseudoaligned to the human reference transcriptome using kallisto v0.48.0 130 . The reference transcriptome was downloaded from Ensembl (release 110, Homo_sapiens.GRCh38.cdna.all.fa.gz) and an index was built using the kallisto index command. For pseudoalignment, the kallisto quant command was used with the -b 100 option, corresponding to 100 bootstraps. Samples were then grouped based on the solvent included in the media (DMSO or H 2 O), and a principal component analysis (PCA) was performed separately for each solvent group, as follows. First, transcript abundance tables generated by kallisto were normalized and filtered using the kallisto_table function from the sleuth R package (v0.30.2), using the options normalized = TRUE and use_filtered = TRUE. The resulting tables were log-transformed using the natural logarithm of (x+0.5) prior to PCA, which was conducted using the prcomp function in R v4.3.2. PCA plots were used to identify potential outlier samples within each group - defined as samples positioned far from: (1) the vast majority of samples in the plot, and (2) the other replicates within their treatment condition. Due to the subjective nature of visual outlier identification, samples were only excluded from downstream analyses if their removal was additionally supported by deviant quality control (QC) statistics from Novogene. Note: The PCA plots for each sequencing batch are displayed in Figs. S4 and S5, and the corresponding sample lists and QC statistics are contained in Supplementary Table 3. Based on the above criteria, the following 5 (out of 138) samples were excluded from downstream analysis. From batch 2, oxalate replicate #1 and carnitine replicate #2 were removed (supporting QC statistic: low %GC content compared to the other samples in the batch). From batch 3, glycodeoxycholate replicate #1 and quinolinate replicate #2 were removed (supporting QC statistic: low %GC content compared to other samples in the batch). From batch 4 - DMSO control replicate #5 was removed (supporting QC statistic: low RNA concentration compared to all other DMSO control samples). Following PCA and outlier removal, DESeq2 131 v1.42.1 was used to determine differentially expressed genes between each set of metabolite-treated samples and the corresponding vehicle controls in the sequencing batch, using the default significance cutoff of alpha=0.10. As specified in Supplementary Table 3, the number of vehicle controls per solvent per sequencing batch ranged from three to nine. Gene abundance values were determined from the per-transcript kallisto outputs using tximport (v1.31.0). When applicable, the corrplot R package (v0.95) was used to verify that all vehicle controls of the same solvent (e.g., 1% DMSO controls) on different plates could be grouped, based on strong correlations - Pearson’s r > 0.98 - of log-transformed gene abundances. As recommended by the DESeq2 developers, the function lfcShrink 63 was applied to conservatively shrink log2foldchanges, using the shrinkage estimator ‘apeglm.’ Additionally, to avoid overcounting the number of significant hits per metabolite, we filtered the DESeq2 output of differentially expressed genes to eliminate lines that were exact duplicates of previous lines besides the Ensembl ID field, which can occur when two different Ensembl gene IDs are mapped to the same physical gene. Finally, genes with an absolute shrunken log2 fold change ≤ 0.1 were not counted as DEGs, even if they met the significance threshold of adjusted p ≤ 0.1. The KEGG pathway enrichment analyses shown in Figs. 2B and S2 were conducted using g:Profiler ( https://biit.cs.ut.ee/gprofiler/gost ) 132 on August 17, 2025 using the default g:SCS significance threshold and a custom list of background genes - specifically, all genes that passed DESeq2’s internal filters in the corresponding DESeq2 analysis. For the analysis of DEGs in the Alzheimer and Parkinson’s KEGG pathways ( Fig. 2C ), the pathway gene lists were retrieved on July 19, 2025 from https://rest.kegg.jp/link/hsa/path:hsa05010 and https://rest.kegg.jp/link/hsa/path:hsa05012 , respectively. The numbers reported in Fig. 2C represent numbers of unique genes (based on Entrez ID). In vitro cytotoxicity testing hCMEC/D3 cells (passage #22) were seeded on Day 0 at a density of 100,000 cells per well in 12-well plates. Prior to seeding, each well was coated with 23μg rat tail collagen (Sigma, Cat. #08-115) in 1mL DPBS (Fisher, Catalog #MT21031CV) and rinsed twice with DPBS following a 10-minute incubation at room temperature. On Day 2, cells were treated for 3 hours with EGM2-MV BulletKit medium (Lonza, Catalog #CC-3202) excluding the VEGF supplement and supplemented with either: [1] metabolites at the concentrations shown in Fig. S7, [2] vehicle control (i.e., 1% DMSO or 1% H 2 O), or [3] 20% DMSO (cytotoxic control). Following treatment, cells were detached with 0.05% trypsin-EDTA (Fisher Scientific, Catalog #25-300-054), pelleted by centrifugation (100 x g, 5min, 4°C), and resuspended in 150μl serum-free, phenol red-free EGM2-MV medium (basal medium: Lonza, Catalog #CC-3129; supplements: Lonza EGM2-MV SingleQuots kit, Catalog #CC-4147, excluding FBS and VEGF). 30μl of Propidium Iodide ReadyFlow reagent (Thermo Fisher, Catalog #R37169) was added to the suspension, and fluorescence was measured using a NovoCyte Quanteon flow cytometer at the Stanford Shared FACS Facility. Animals Male C57BL/6 mice (3 months old) were obtained from Jackson Laboratories (stock # Jax 000664). All mice were kept on a 12 h-12 h light-dark cycle and provided ad libitum access to food and water. All animal care and procedures complied with the Animal Welfare Act and were in accordance with institutional guidelines and approved by the institutional administrative panel of laboratory animal care at Stanford University (PIs of animal protocols: T. Wyss Coray, M. Shamloo; protocol #s 29100 and 18466). Mouse brain transcriptomics after metabolite injection Guided by prior studies investigating TMAO and the microbiome metabolite p-Cresol sulfate 58 , 122 , we conducted bulk transcriptomics on mouse half-brains following a 2-hour exposure to each metabolite of interest or the corresponding vehicle control. As detailed in Figs. #2-3 and the accompanying text, four metabolites were tested in mice - palmitic acid (PA), indoleacetic acid (IAA), trimethylamine N-oxide (TMAO), and glycodeoxycholic acid (GDCA). Intravenous retro-orbital injection under isoflurane anesthesia was used as the administration route. Metabolite dosing was based on precedent from the study on TMAO that guided experimental design 58 . For TMAO (Cayman Chemicals, Item #17354, Batch #0522898-67), we used the same dose as the original study, 1.8 mg/kg. This dose was described as “calculated to approximate human circulating TMAO levels.” Thus, for the remainder of the metabolites, we examined the targeted metabolomics literature 26 , 28 , 48 – 55 to identify the range of circulating levels previously observed in human seniors, and calculated mouse-equivalent doses of observed human concentrations, using well-established guidelines 56 , 57 . PA (Sigma, Catalog #P0500-10G, Lot #0000253065) was administered at a final dose of 22.2 mg/kg (∼100µM total body dose). IAA (Sigma, Catalog #I2886-5G, Lot #0000249926) was administered at a final dose of 1.5 mg/kg (∼10µM total body dose). GDCA (Sigma, Catalog #361311-5GM, Lot #4045885) was administered at a final dose of 0.4 mg/kg (∼1µM total body dose). Two experiments were conducted in total. Experiment 1 contained 8 mice; 4 mice received 1.8 mg/kg TMAO in 100µl saline, and 4 mice received 100µl saline (vehicle control). Experiment 2 contained 20 mice, in two groups. Group 1 contained 8 mice; 4 mice received 0.4 mg/kg GDCA in 100µl saline, and 4 mice received 100µl saline (vehicle control). Group 2 contained 12 mice; 4 mice received 22.2 mg/kg PA in 100µl saline containing 6% DMSO, 4 mice received 1.5 mg/kg IAA in 100µl saline containing 6% DMSO, and 4 mice received 100µl saline also containing 6% DMSO (vehicle control). Note: unlike TMAO and GDCA, PA and IAA are only soluble in organic solvents such as DMSO, hence the use of DMSO above. For sacrifice, mice were anesthetized using 2.5% Avertin (v/v). As per the original TMAO study 58 , mice were transcardially perfused with 0.9% saline after sacrifice and prior to brain removal, in order to remove contaminating blood. Forceps used for brain extraction were pre- treated with RNaseZap (Thermo Fisher, Catalog #AM9780) and subsequently rinsed twice with RNase-free water to minimize the risk of RNA degradation. Brains were sagittally hemisected at the corpus callosum prior to storage. Left hemispheres were stored in 1ml RNAlater overnight at 4°C, and then dabbed on a Kimwipe to remove excess RNAlater before transfer to a fresh tube for storage at -80°C. Right hemispheres were either processed in the same manner or flash-frozen directly at -80°C. Specifically, for the TMAO-treated and corresponding vehicle control brains - which were part of the first experiment performed - the right hemispheres were processed identically to the left. In the second experiment, however, the right hemispheres were flash-frozen at -80°C without RNAlater incubation. RNA was extracted from frozen left hemispheres using a modified version of the protocol from the QIAWave RNA Mini Kit (Qiagen, Catalog #74536), as follows. A bead beater (Qiagen Tissue Lyser II) was used to homogenize the left hemisphere in a collection microtube (Qiagen Catalog #19560) containing 700μL of pre-chilled QIAzol (Qiagen, Catalog #79306) and stainless 5mm steel bead (Qiagen Catalog #69989). For bead beating, the tube was positioned in between cold metal blocks prechilled at -20°C. Samples were homogenized at 30Hz for five minutes, kept on ice for 1 minute, then homogenized at 30Hz for an additional 2 minutes. Following homogenization, the tube was incubated at room temperature for 3-5 minutes. 200µL was then set aside for RNA extraction in a DNA LoBind tube; the remaining volume was stored in a fresh DNA LoBind tube at -80°C. 500µL of additional QIAzol was added to the 200µL tube to make 700µL; then, all the liquid was transferred to a DNA LoBind tube containing 200µL chloroform. The tube was capped and vortexed for 15 seconds, then incubated at room temperature for 2-3 minutes and centrifuged at 12,000xg for 15min at 4°C. 150µL x 2 times (300µL total) of aqueous phase was then transferred to a DNA LoBind tube containing 300µL 70% ethanol. The tube was then inverted 10 times to mix, and briefly centrifuged. ∼600µl of sample was transferred to the RNeasy Mini spin column, then spun at 10,000xg for 30 seconds at room temperature. Flow-through was discarded, and 350µL RW1 was added to the column. The column was spun at 10,000xg for 30 seconds at room temperature, and the flow-through was again discarded. 500µL RPE was added to the column, and the column was spun at 10,000xg for 30 seconds at room temperature, after which flow-through was discarded. The preceding RPE wash and spin was then repeated for 2 minutes. The tube was then centrifuged at maximum speed for one minute at room temperature. The column was transferred to a new 1.5ml DNA LoBind tube, to which 50µL RNAse-free water was added. The lid was closed and the tube was incubated at room temperature for 1-2 minutes. For elution, the tube was spun for one minute at 10,000xg at room temperature. The same 50µL of eluate was used to repeat the previous step. RNA samples were sent to Novogene for quality control, mRNA library preparation (poly A enrichment), and sequencing (NovaSeq 6000, 150bp paired end reads, 6Gb raw data per sample). A sample list and QC statistics are included in Supplementary Table 3. Mouse brain transcriptomics - preliminary data analysis The following steps were performed separately for each experiment defined above. First, FASTQ files from Novogene were trimmed to remove adapter and low-quality bases using TrimGalore v0.6.10 with the following parameters: --fastqc --stringency 3 --max_n 15 --output_dir trimming_out --cores 2 --retain_unpaired --2colour 30 --paired _1.fq.gz _2.fq.gz. Trimmed paired-end reads were pseudoaligned to the standard mouse transcriptome index provided with the v1 release of kallisto 130 , using kallisto v0.50.1. For pseudoalignment, the kallisto quant command was used with the -b 100 option, corresponding to 100 bootstraps. Principal component analysis (PCA) was conducted using the following functions available in DESeq2 (v1.42.1): vst (setting blind = TRUE) and plotPCA. PCA revealed two key insights that guided subsequent decisions: (1) For experiment 1 - TMAO replicate #3 was positioned away from every other TMAO-treated sample along the first two components (Fig. S6A), necessitating further investigation of this sample prior to downstream analysis. (2) For experiment 2 - PCA revealed strong separation of samples by RNA extraction batch (Fig. S6B), necessitating the inclusion of this variable as a covariate in downstream differential expression analysis. TMAO replicate #3 was ultimately excluded from subsequent analyses, including DESeq2 differential expression analysis, based on three additional facts beyond its placement in PCA. First, TMAO replicate #3 exhibited high leverage in the initial DESeq2 run; i.e., with TMAO replicate #3 included, it showed the highest mean Cook’s distance, more than 1.5x that of any other sample, indicating a disproportionate influence on the linear model. Second, DESeq2 differential gene expression analysis comparing TMAO replicate #3 vs. all other samples revealed enrichment of blood markers (Hba-a1: log₂FC = 2.28, padj = 3.9 × 10E-4; Hbb-bt: log₂FC = 1.72, padj = 0.03; Supplementary Table 7), suggesting potential incomplete perfusion and/or other blood contamination. Third, excluding TMAO replicate #3 brought our differential expression results into closer agreement with those reported by Hoyles et al. 58 , the study that inspired the design of this experiment, and with the broader literature on TMAO, strengthening confidence in the results. DESeq2 131 v1.42.1 was used to determine differentially expressed genes between metabolite-treated samples and matched vehicle controls, using the default significance cutoff of alpha=0.10. For experiment 1, TMAO replicate #3 was excluded for the reasons detailed above, yielding n =3 TMAO brains and n =4 control brains for the final differential expression analysis. For experiment 2, all n =4 brains per condition were included in the final analysis, and the RNA extraction batch was included as a covariate, based on the PCA results described above (Fig. S6B). As recommended by the DESeq2 developers, the function lfcShrink 63 was applied to conservatively shrink log2foldchanges, using the shrinkage estimator ‘apeglm.’ All final DESeq2 differential gene expression results for PA, IAA, TMAO and GDCA are contained in Supplementary Table 7. Genes with an absolute shrunken log2foldchange ≤ 0.1 were not counted as DEGs, even if they met the significance threshold. All per-gene abundances for DESeq2 analysis were determined from the per-transcript kallisto outputs using tximport (v1.30.0). Mouse brain transcriptomics - gene set enrichment analysis Gene set enrichment analysis (GSEA) was conducted using the R package fgsea (v1.28.0). The relevant gene sets (hallmark gene sets and REACTOME pathways for mouse) were downloaded from https://www.gsea-msigdb.org/gsea/msigdb/mouse/collections.jsp on July 4, 2025. Enrichments with padj ≤ 0.05 were considered statistically significant. Mouse brain transcriptomics - marker gene analysis Marker genes for six different brain cell classes were determined using the following command in the R package Seurat (v5.3.0): FindAllMarkers(seurat_object, group.by = “ident”, slot = “data”, only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0.5). The resulting marker genes were further filtered to those with adjusted p -value ≤ 0.05. The Seurat object was constructed using single cell data from the mouse brain, specifically data from the 8 young mice (age-matched to those in our study) included in Ximerakis et al. , 2019 68 . Expression data, and metadata specifying the class of each cell, were downloaded from https://singlecell.broadinstitute.org/single_cell/study/SCP263/aging-mouse-brain on August 22, 2025. Mouse brain transcriptomics - differential splicing analysis Differential splicing events were identified using rMATS-turbo 74 (v4.1.1), as follows. First, STAR (v2.7.11b) was used to align trimmed paired-end reads to the mouse genome downloaded from GENCODE (Release M37, i.e., GRCm39). The following command was used: STAR –genomeDir --readFilesIn _1.trimmed.fq.gz _2.trimmed.fq.gz --outFileNamePrefix --outSAMunmapped Within -- outSAMattributes NH HI AS NM MD XS --twopassMode Basic --alignSJDBoverhangMin 1 --alignSJoverhangMin 8 --runThreadN 16 --outSAMtype BAM SortedByCoordinate -- outSAMstrandField intronMotif --readFilesCommand zcat. The resulting bam files were then run through rMATS-turbo using the following command: rmats.py --gtf --b1 --b2 --od --tmp -t paired - -libType fr-unstranded --readLength 150 --variable-read-length --nthread 10 --cstat 0.1 –novelSS --allow-clipping. Based on the “JCEC” output files, this command initially nominated 126 differential splicing events between TMAO-treated and saline-treated brains (FDR ≤ 0.05), which we further narrowed to 44 using the following four filters: (1) ≥ 5 read counts on average per group, (2) exclusion of events with average PSI 0.95 in both groups, (3) FDR ≤ 0.01, and (4) ΔPSI ≥ 0.10. The sashimi plot in Fig. S9A was generated using ggsashimi 133 (v1.1.5). Bulk ATAC-seq of brains of GDCA-exposed vs. vehicle control mice As described above under “Mouse brain transcriptomics after metabolite injection,” only the left hemisphere of each brain in these experiments underwent RNA-sequencing. The right hemispheres from GDCA-exposed and corresponding vehicle control mice were flash-frozen at -80°C, enabling ATAC-seq. Mouse brain nuclei were isolated following procedures established for processing mammalian brains 134 and then ATAC-seq was carried out as previously described 135 with some modifications. Briefly, dissected brain tissue was added to 1 mL nuclei extraction buffer (NEB: 250 mM Sucrose, 65 mM B-glycerol, 1x protein inhibitor, 25 mM KCl, 5 mM MgCl2, 20mM HEPES-KOH, 0.5% IGEPAL CA-630, 1 mM DTT, 0.2 mM Spermine, 0.5 mM Spermidine, 2-5% NGSerum) and incubated on ice for 10-15 minutes. then dounced 20 times with pestle A and another 20 times with pestle B. Another 300 µL of NEB was added and the mixture was transferred to a 2mL tube and incubated on a rotator at 4°C for 5 minutes, then filtered through a 70µm Flowmi strainer (Sigma, Catalog #BAH136800070) into a fresh 2mL tube. The volume was adjusted to 1,200 µL with more NEB if necessary and nuclei were lightly crosslinked by adding 6.52 µl of 37% FA, incubating for 4 minutes at room temperature, and quenching with 133 µL 2.5 M Glycine solution. Nuclei were finally isolated using density gradient separation, by adding 1,100 µL of O-50% solution (generated by mixing 18,751 µL O-60%/OptiPrep/iodaxinol solution, 3242 µL H20, 450 µL NGS, and 22.5 µL 1M DTT) in a 5mL Eppendorf tube, then creating a density gradient by adding 1,000 µL O-44% (8,800 µL O-50% plus 1,200 µL H 2 O) to the very bottom of the tube and 500 µL O-22% (2,200 µL O-50% plus 2,800 NEB buffer) in between the O-44% and the top layer with the nuclei. Tubes were then centrifuged for 30 minutes at 4°C with breaks off (deceleration 0, acceleration 5), and nuclei were extracted from the middle layer in the density gradient. After nuclei isolation, nuclei were counted and approximately 100,000 were used as input into each transposition reaction in four technical replicates for each individual mouse brain. Transposition was carried out in 50 µL transposition mix, consisting of 25 µL 2x TD buffer (20 mM Tris-HCl pH 7.6, 10 mM MgCl2, 20% Dimethyl Formamide), 22.5 µL nuclease-free H 2 O, and 2.5 µL loaded Tn5 transposase (produced and assembled following previously established protocols 136 by incubation in an Eppendorf ThermoMixer at 37°C at 1,000 rpm for 30 minutes. The transposition reactions were then stopped immediately by adding 150 µL reverse crosslinking buffer (1% SDS, 0.1M NaHCO3), and reverse crosslinking was carried out for 8 hours at 65°C. Transposed DNA was isolated using the MinElute PCR Purification Kit (Qiagen, Catalog #28004) by adding 600 µL PB Buffer and eluting in 20 µL EB Buffer heated to 65°C, then PCR amplified by mixing 20 µL transposed DNA, 2.5 µL i5 and 2.5 µL i7 PCR primers, and 25 µL 2x NEBNext High-Fidelity PCR Master Mix (New England Biolabs, Catalog #M0541L), with the following settings: 72°C for 5 minutes, 98°C for 30 seconds, followed by 10 cycles of 98°C for 10 seconds, 63°C for 30 seconds, and 72°C for 45 seconds. Final libraries were purified using the MinElute PCR Purification Kit (Qiagen, Catalog #28004), QC-ed using a Qubit 1X dsDNA kit (Thermo Fisher Scientific, Catalog #Q33230) and an Agilent TapeStation (High Sensitivity D1000 Screen Tape® Assay), and sequenced on a NextSeq 550 instrument in a 2×38 format. ATAC-seq - data analysis ATAC-seq data analysis was conducted largely as previously described 137 , with brief specifics provided below. Two replicates were sequenced per brain sample, yielding n =16 total replicates from n =8 mice, of which n =4 were GDCA-exposed mice and n =4 were the corresponding vehicle control mice (Supplementary Table 3). Forward and reverse reads from each replicate were first trimmed to a length of 36bp using the script PEFastqToTabDelimited.py available at the following Github repository: https://github.com/georgimarinov/GeorgiScripts . Trimmed reads were aligned to both the entire mouse genome (mm10) and to the mitochondrial portion alone (chrM) using bowtie (v1.0.1). The following parameters were used for alignment to the mitochondrial genome: -p 20 -v 2 -a -t –best --strata -q -X 1000 --sam --12. The following parameters were used for alignment to the entire genome: -p 20 -v 2 -k 2 -m 1 -t --best --strata -q -X 1000 --sam --12. The mitochondrial alignments were used to calculate the fraction of mitochondrial reads in each replicate (reported in Supplementary Table 3). For all subsequent analyses, the entire-genome alignments were used, after filtering out reads mapping to chrM. The remaining, non-mitochondrial reads were deduplicated using the MarkDuplicates tool from Picard Tools (v1.99) using the following parameters: VALIDATION_STRINGENCY=LENIENT ASSUME_SORTED=true REMOVE_DUPLICATES=true. The resulting deduplicated .bam files were indexed using samtools v0.1.18 and converted to .wig using the script makewigglefromBAM-NH.py from the Github repository linked above. Reads per million (RPM) was used as the normalization method. The .wig files were converted to .bigWig using wigToBigWig from UCSC Utils (version from 2017- 07-13). TSS enrichment scores reported in Supplementary Table 3 were calculated using the scripts signalAroundCoordinate-BW.py and ATACTSSscore.py , from the Github repository linked above. TSS coordinates were based on the UCSC mm10 refFlat gene annotation table (dated 2014-10-26). Peaks were called for each replicate using the callpeak function from MACS2 (v2.1.1) using the following parameters: -g mm -f BAMPE --to-large -p 1e-1 --keep-dup all --nomodel. Then, reproducible peaks between replicates were identified using the previously described irreproducible discovery rate (IDR) framework 137 . IDR v2.0.4.2 was used for this step. The required script BAMPseudoReps.py from the Github repository linked above was converted to Python 3 using the 2to3 Python program prior to use. The reproducible peaks between replicates per sample were then merged across samples using the createIterativeOverlapPeakSet.R script available at the following Github repository - https://github.com/corceslab/ATAC_IterativeOverlapPeakMerging - described in Grandi et al. , Nature Protocols 2022 138 . Blacklist regions provided to the script were downloaded from https://github.com/Boyle-Lab/Blacklist/tree/master/lists (file mm10-blacklist.v2.bed.gz ) on July 11, 2025. The spm was set to 0 because confident peaks had already been identified. Additional relevant parameters were: --suffix _summits.bed --blacklist mm10-blacklist.v2.bed --genome mm10 --rule “(n+1)/2” --extend 250. This step yielded a .bed file with the coordinates of the final set of merged peaks, which was converted to SAF format and used to create a read counts table (rows = merged peaks, columns = samples) using featureCounts (v2.0.6) as follows: featureCounts -p --countReadPairs -B -C -T 8 -F SAF -a All_samples_peaks.saf -o All_Samples.counts … Samples were retained for subsequent analysis if both replicates had TSS enrichment scores ≥ 10, the standard deemed as “acceptable” under ENCODE guidelines for mm10 ( https://www.encodeproject.org/atac-seq/ , last accessed July 28, 2025). As per Supplementary Table 3, 7 samples ( n =4 from GDCA-exposed mice, and n =3 from control mice) passed this filter. Differentially accessible peaks were determined using DESeq2 (v1.42.1) using the default significance cutoff of alpha=0.10. These peaks were further annotated using the annotatePeaks.pl script from HOMER, which was installed using the configureHomer.pl script downloaded from http://homer.ucsd.edu/homer/ on July 14, 2025. Exact command used: annotatePeaks.pl sig_peak_file_for_homer.txt mm10 -annStats annotation_stats.txt > annotated_sig_peak_file.txt. The strand in sig_peak_file_for_homer.txt was specified as “.” for all peaks. Coverage of differentially accessible peaks ( Fig. 5D ; Fig. S10A-B) was determined as follows. First, BAM files for individual brains within each condition were merged using samtools v1.21. The resulting two files (one for control brains and one for GDCA-exposed brains) were sorted and indexed with samtools v1.21, and subsequently converted to bigWig format using the bamCoverage function from deepTools v3.3.0, using a bin size of 100 bp and specifying counts per million (CPM) as the normalization method. Coverage matrices per condition were generated from each bigWig file using the signalAroundCoordinate-individual.py script available from https://github.com/georgimarinov/GeorgiScripts , using the previously specified parameters 137 , after conversion to Python 3 using the 2to3 Python program. chromVAR 96 analysis ( Figs. 5F and S11) was conducted using chromVAR 1.24.0, referencing the 0.99.6 version of the JASPAR 2024 database in R version 4.3.2. Behavioral testing and measurement of core body temperature Behavioral testing was conducted at the Stanford Behavioral and Functional Neuroscience Laboratory. Mice were acclimated to a reverse 12h-12h light-dark cycle for one week prior to initiation of experiments, ensuring that daytime testing coincided with their naturally active (dark) phase. For experiments with mouse cohort 1 (schematized in Fig. 6A ), mice were dosed for three consecutive days with GDCA (0.4 mg/kg) or saline vehicle ( n =15 mice per group, n =30 total). Intravenous retroorbital injection was used as the administration route, with isoflurane as the anesthetic. The total injection volume was 50µL. Behavioral testing was conducted after the final injection (activity chamber: one day after final injection, Y-maze: two days after final injection). Mice were observed for one hour in the activity chamber and for five minutes in the Y-maze. For experiments with mouse cohort 2 (schematized in Fig. 6B ), mice were administered 50 mg/kg GDCA, 10 mg/kg GDCA, or saline vehicle per os ( n =15 mice per group, n =45 total). Y- maze testing was conducted one hour post-dosing. Six days after Y-maze testing, all mice were redosed (keeping the same group assignments), and core body temperature was measured using a mouse rectal probe. All n =15 mice per group were measured at timepoint 0, prior to dosing; n =5 mice per group were measured at each subsequent timepoint (10min, 30min, and 60min post- dosing). Data availability All raw sequencing data are available under NCBI BioProject ID PRJNA1328687 139 . Code availability All software, parameters, and relevant details of custom code are described in the Methods section. Author contributions MC: conceptualization, investigation, methodology, data curation, formal analysis, visualization, writing - original draft, writing - review and editing. SMS: investigation, methodology. IEP: methodology. DJR: investigation, methodology. GKM: investigation, methodology, data curation, formal analysis. AAM: investigation, writing - review and editing. JLEB: investigation, methodology. AN: visualization. JWJ: methodology. JCW: supervision. SXL: methodology, supervision. SMD: methodology, supervision. WJG: supervision. NLS: investigation, methodology, data curation, formal analysis. MS: methodology, supervision. AB: funding acquisition, supervision. TWC: supervision. ASB: conceptualization, methodology, funding acquisition, supervision, writing - original draft, writing - review and editing. All authors read and approved the manuscript. Competing interests The authors declare no competing interests. Acknowledgements We thank Jeff Rubin, Ren Song, Ayan Mondal, Hongchao Guo, Michael Bassik, Betsy Mellins, and Mark Kay for guidance surrounding brain endothelial cell culture and experiments. Jeff Rubin and Mark Kay provided us with the hCMEC/D3 cell line. We thank Jakub Rajnuk for advice during the metabolite selection process, Anshul Kundaje for guidance on RNA-seq data analysis, Garrett Patrick for reviewing the manuscript during its preparation, Ariel Gulasch for assistance with in vitro transcriptomic screening, and Rudy Wycallis at the Stanford Shared FACS Facility for assistance with flow cytometry. Several non-data figures such as experiment timeline figures were created using BioRender.com . This work was funded, in part, by NIH R01AI148623 and R01AI143757, NCI R01 CA301727, a Stand Up 2 Cancer Grant, and the Stanford Wu Tsai Neurosciences Institute. M.C. was supported by the National Defense Science and Engineering Graduate Fellowship. I.E.P. was supported by the Stanford REACH PhD fellowship. S.M.S. was supported by a Stanford Bio-X Bowes Fellowship. Computing costs were supported, in part, by an NIH S10 Shared Instrumentation grant (1S10OD02014101). The Stanford Behavioral and Functional Neuroscience Laboratory is supported by the NIH S10 Shared Instrumentation for Animal Research (1S10OD030452-01). Footnotes Clarifications to text; Supplementary Table 10 updated References 1. ↵ Sachdev , P. S. et al. Factors predicting reversion from mild cognitive impairment to normal cognitive functioning: a population-based study . PLoS One 8 , e59649 ( 2013 ). OpenUrl CrossRef PubMed 2. ↵ Shimada , H. , Doi , T. , Lee , S. & Makizako , H . Reversible predictors of reversion from mild cognitive impairment to normal cognition: a 4-year longitudinal study . Alzheimers Res Ther 11 , 24 ( 2019 ). 3. ↵ Badal , V. D. et al. The Gut Microbiome, Aging, and Longevity: A Systematic Review . Nutrients 12 , ( 2020 ). 4. Pang , S. et al. Longevity of centenarians is reflected by the gut microbiome with youth- associated signatures . Nat Aging 3 , 436 – 449 ( 2023 ). OpenUrl PubMed 5. ↵ Boehme , M. , Guzzetta , K. E. , Wasén , C. & Cox , L. M . The gut microbiota is an emerging target for improving brain health during ageing . Gut Microbiome (Camb) 4 , ( 2023 ). 6. ↵ Cryan , J. F. et al. The Microbiota-Gut-Brain Axis . Physiological Reviews ( 2019 ) doi: 10.1152/physrev.00018.2018 . OpenUrl CrossRef PubMed 7. ↵ Loh , J. S. et al. Microbiota–gut–brain axis and its therapeutic applications in neurodegenerative diseases . Signal Transduction and Targeted Therapy 9 , 1 – 53 ( 2024 ). OpenUrl CrossRef 8. ↵ Parker , A. et al. Fecal microbiota transfer between young and aged mice reverses hallmarks of the aging gut, eye, and brain . Microbiome 10 , 1 – 25 ( 2022 ). OpenUrl CrossRef PubMed 9. D’Amato , A. et al. Faecal microbiota transplant from aged donor mice affects spatial learning and memory via modulating hippocampal synaptic plasticity- and neurotransmission-related proteins in young recipients . Microbiome 8 , 1 – 19 ( 2020 ). OpenUrl CrossRef PubMed 10. ↵ Boehme , M. et al. Microbiota from young mice counteracts selective age-associated behavioral deficits . Nat Aging 1 , 666 – 676 ( 2021 ). OpenUrl PubMed 11. ↵ Wikoff , W. R. et al. Metabolomics analysis reveals large effects of gut microflora on mammalian blood metabolites . Proceedings of the National Academy of Sciences 106 , 3698 – 3703 ( 2009 ). OpenUrl Abstract / FREE Full Text 12. ↵ Lai , Y. et al. High-coverage metabolomics uncovers microbiota-driven biochemical landscape of interorgan transport and gut-brain communication in mice . Nat Commun 12 , 6000 ( 2021 ). OpenUrl CrossRef PubMed 13. ↵ Wang , Z. et al. Gut flora metabolism of phosphatidylcholine promotes cardiovascular disease . Nature 472 , 57 – 63 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 14. Saeedi Saravi , S. S. , et al. Gut microbiota-dependent increase in phenylacetic acid induces endothelial cell senescence during aging . Nat Aging 5 , 1025 – 1045 ( 2025 ). OpenUrl PubMed 15. ↵ Knox , E. G. , Aburto , M. R. , Clarke , G. , Cryan , J. F. & O’Driscoll , C. M . The blood-brain barrier in aging and neurodegeneration . Molecular Psychiatry 27 , 2659 – 2673 ( 2022 ). OpenUrl CrossRef PubMed 16. ↵ Shi , S. M. et al. Glycocalyx dysregulation impairs blood–brain barrier in ageing and disease . Nature 639 , 985 – 994 ( 2025 ). OpenUrl PubMed 17. ↵ Panyard , D. J. , Yu , B. & Snyder , M. P . The metabolomics of human aging: Advances, challenges, and opportunities . Sci Adv 8 , eadd6155 ( 2022 ). OpenUrl CrossRef PubMed 18. ↵ Menni , C. et al. Metabolomic markers reveal novel pathways of ageing and early development in human populations . Int J Epidemiol 42 , 1111 – 1119 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 19. ↵ Dunn , W. B. et al. Molecular phenotyping of a UK population: defining the human serum metabolome . Metabolomics 11 , 9 – 26 ( 2015 ). OpenUrl CrossRef PubMed 20. ↵ Darst , B. F. , Koscik , R. L. , Hogan , K. J. , Johnson , S. C. & Engelman , C. D . Longitudinal plasma metabolomics of aging and sex . Aging (Albany NY) 11 , 1262 – 1282 ( 2019 ). OpenUrl PubMed 21. ↵ Chen , H. et al. Highly multiplexed bioactivity screening reveals human and microbiota metabolome-GPCRome interactions . Cell 186 , 3095 – 3110 .e19 ( 2023 ). OpenUrl CrossRef PubMed 22. ↵ Dekkers , K. F. et al. An online atlas of human plasma metabolite signatures of gut microbiome composition . Nat Commun 13 , 5370 ( 2022 ). OpenUrl CrossRef PubMed 23. ↵ Bishai , J. D. & Palm , N. W . Small Molecule Metabolites at the Host-Microbiota Interface . J Immunol 207 , 1725 – 1733 ( 2021 ). OpenUrl Abstract / FREE Full Text 24. ↵ MahmoudianDehkordi , S. et al. Altered bile acid profile associates with cognitive impairment in Alzheimer’s disease-An emerging role for gut microbiome . Alzheimers Dement 15 , 76 – 92 ( 2019 ). OpenUrl CrossRef PubMed 25. Low , D. Y. et al. Diet-Related Metabolites Associated with Cognitive Decline Revealed by Untargeted Metabolomics in a Prospective Cohort . Mol Nutr Food Res 63 , e1900177 ( 2019 ). OpenUrl PubMed 26. ↵ Chen , S.-J. et al. Alteration of Gut Microbial Metabolites in the Systemic Circulation of Patients with Parkinson’s Disease . J Parkinsons Dis 12 , 1219 – 1230 ( 2022 ). OpenUrl PubMed 27. Kijpaisalratana , N. et al. White matter hyperintensity severity modifies gut metabolite association with cognitive outcomes . J Prev Alzheimers Dis 12 , 100086 ( 2025 ). 28. ↵ Marksteiner , J. , Blasko , I. , Kemmler , G. , Koal , T. & Humpel , C . Bile acid quantification of 20 plasma metabolites identifies lithocholic acid as a putative biomarker in Alzheimer’s disease . Metabolomics 14 , 1 – 10 ( 2017 ). OpenUrl PubMed 29. ↵ Bröer , S. et al. Alanine metabolism, transport, and cycling in the brain . Journal of neurochemistry 102 , ( 2007 ). 30. ↵ Wishart , D. S. et al. HMDB 5.0: the Human Metabolome Database for 2022 . Nucleic Acids Res 50 , D622 – D631 ( 2021 ). OpenUrl CrossRef 31. ↵ Weksler , B. , Romero , I. A. & Couraud , P.-O . The hCMEC/D3 cell line as a model of the human blood brain barrier . Fluids Barriers CNS 10 , 16 ( 2013 ). 32. ↵ Sato , N. et al. Porphyromonas gingivalis lipopolysaccharide induces interleukin-6 and c-c motif chemokine ligand 2 expression in cultured hCMEC/D3 human brain microvascular endothelial cells . Gerodontology 39 , 139 – 147 ( 2022 ). OpenUrl PubMed 33. ↵ Wilson 3rd , D. M. , et al. Hallmarks of neurodegenerative diseases . Cell 186 , 693 – 714 ( 2023 ). OpenUrl CrossRef PubMed 34. ↵ Xie , H. et al. The Role of Gut Microbiota-Derived Trimethylamine N-Oxide in the Pathogenesis and Treatment of Mild Cognitive Impairment . International Journal of Molecular Sciences 26 , 1373 ( 2025 ). OpenUrl PubMed 35. Melo , H. M. et al. Palmitate Is Increased in the Cerebrospinal Fluid of Humans with Obesity and Induces Memory Impairment in Mice via Pro-inflammatory TNF-α . Cell Rep 30 , 2180 – 2194 .e8 ( 2020 ). OpenUrl CrossRef PubMed 36. Lin , C.-H. , Lin , Y.-N. , Lane , H.-Y. & Chen , C.-J . The identification of a potential plasma metabolite marker for Alzheimer’s disease by LC-MS untargeted metabolomics . J Chromatogr B Analyt Technol Biomed Life Sci 1222 , 123686 ( 2023 ). OpenUrl PubMed 37. Kalecký , K. , German , D. C. , Montillo , A. A. & Bottiglieri , T . Targeted Metabolomic Analysis in Alzheimer’s Disease Plasma and Brain Tissue in Non-Hispanic Whites . Journal of Alzheimer’s Disease ( 2022 ). 38. Chen , S.-J. et al. The Gut Metabolite Trimethylamine N-oxide Is Associated With Parkinson’s Disease Severity and Progression . Movement Disorders 35 , 2115 – 2116 ( 2020 ). OpenUrl PubMed 39. ↵ Xu , Z. et al. Plasma trimethylamine N-oxide, carnitine, choline, and betaine distinguish cognition impairment in Parkinson’s disease . Parkinsonism Relat Disord 135 , 107844 ( 2025 ). OpenUrl PubMed 40. ↵ Jonsson , A. L. et al. Impact of Gut Microbiota and Diet on the Development of Atherosclerosis in Apoe−/− Mice. Arteriosclerosis , Thrombosis, and Vascular Biology 38 , 2318 ( 2018 ). OpenUrl 41. ↵ Wei , W. et al. Psychological stress-induced microbial metabolite indole-3-acetate disrupts intestinal cell lineage commitment . Cell metabolism 36 , ( 2024 ). 42. ↵ Sankowski , B. et al. Higher cerebrospinal fluid to plasma ratio of p-cresol sulfate and indoxyl sulfate in patients with Parkinson’s disease . Clin Chim Acta 501 , 165 – 173 ( 2020 ). OpenUrl 43. Pilitsis , J. G. et al. Quantification of Free Fatty Acids in Human Cerebrospinal Fluid . Neurochemical Research 26 , 1265 – 1270 ( 2001 ). OpenUrl PubMed 44. Sarrias , M. J. , Cabré , P. , Martínez , E. & Artigas , F . Relationship between serotoninergic measures in blood and cerebrospinal fluid simultaneously obtained in humans . J Neurochem 54 , 783 – 786 ( 1990 ). OpenUrl PubMed Web of Science 45. Harnisch , L.-O. et al. Quantification of Bile Acids in Cerebrospinal Fluid: Results of an Observational Trial . Biomedicines 11 , 2947 ( 2023 ). OpenUrl PubMed 46. ↵ Wang , J. et al. Peripheral serum metabolomic profiles inform central cognitive impairment . Sci Rep 10 , 14059 ( 2020 ). 47. ↵ Bento , A. P. et al. An open source chemical structure curation pipeline using RDKit . J Cheminform 12 , 51 ( 2020 ). 48. ↵ Djousse , L. et al. Serum Individual Nonesterified Fatty Acids and Risk of Heart Failure in Older Adults . Cardiology 146 , 351 – 358 ( 2021 ). OpenUrl PubMed 49. Cui , Y. et al. Gas chromatography-mass spectrometry analysis of the free fatty acids in serum obtained from patients with Alzheimer’s disease . Biomed Mater Eng 26 Suppl 1 , S2165 – 77 ( 2015 ). OpenUrl 50. Dou , L. et al. The cardiovascular effect of the uremic solute indole-3 acetic acid . J Am Soc Nephrol 26 , 876 – 887 ( 2015 ). OpenUrl Abstract / FREE Full Text 51. Lin , Y.-T. et al. Indole-3 acetic acid increased risk of impaired cognitive function in patients receiving hemodialysis . Neurotoxicology 73 , 85 – 91 ( 2019 ). OpenUrl PubMed 52. Nho , K. et al. Altered bile acid profile in mild cognitive impairment and Alzheimer’s disease: Relationship to neuroimaging and CSF biomarkers . Alzheimers Dement 15 , 232 – 244 ( 2019 ). OpenUrl CrossRef PubMed 53. Bain , M. A. , Faull , R. , Fornasini , G. , Milne , R. W. & Evans , A. M . Accumulation of trimethylamine and trimethylamine-N -oxide in end-stage renal disease patients undergoing haemodialysis . Nephrol Dial Transplant 21 , 1300 – 1304 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 54. Suzuki , T. , Heaney , L. M. , Bhandari , S. S. , Jones , D. J. L. & Ng , L. L . Trimethylamine N-oxide and prognosis in acute heart failure . Heart 102 , 841 – 848 ( 2016 ). OpenUrl Abstract / FREE Full Text 55. ↵ Stubbs , J. R. et al. Serum Trimethylamine-N-Oxide is Elevated in CKD and Correlates with Coronary Atherosclerosis Burden . J Am Soc Nephrol 27 , 305 – 313 ( 2016 ). OpenUrl Abstract / FREE Full Text 56. ↵ U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER) . Guidance for Industry: Estimating the Maximum Safe Starting Dose in Initial Clinical Trials for Therapeutics in Adult Healthy Volunteers . fda.gov https://www.fda.gov/media/72309/download ( 2005 ). 57. ↵ Nair , A. B. & Jacob , S . A simple practice guide for dose conversion between animals and human . J Basic Clin Pharm 7 , 27 – 31 ( 2016 ). OpenUrl CrossRef PubMed 58. ↵ Hoyles , L. et al. Regulation of blood-brain barrier integrity by microbiome-associated methylamines and cognition by trimethylamine N-oxide . Microbiome 9 , 235 ( 2021 ). 59. ↵ De Miguel , Z. et al. Exercise plasma boosts memory and dampens brain inflammation via clusterin . Nature 600 , 494 – 499 ( 2021 ). OpenUrl CrossRef PubMed 60. Pornnoppadol , G. et al. Bispecific antibody shuttles targeting CD98hc mediate efficient and long-lived brain delivery of IgGs . Cell Chem Biol 31 , 361 – 372 .e8 ( 2024 ). OpenUrl PubMed 61. ↵ Yang , A. C. et al. Physiological blood-brain transport is impaired with age by a shift in transcytosis . Nature 583 , 425 – 430 ( 2020 ). OpenUrl CrossRef PubMed 62. ↵ Reinecke , F. , Smeitink , J. A. M. & van der Westhuizen , F. H . OXPHOS gene expression and control in mitochondrial disorders . Biochim Biophys Acta 1792 , 1113 – 1121 ( 2009 ). OpenUrl CrossRef PubMed Web of Science 63. ↵ Zhu , A. , Ibrahim , J. G. & Love , M. I . Heavy-tailed prior distributions for sequence count data: removing the noise and preserving large differences . Bioinformatics 35 , 2084 – 2092 ( 2019 ). OpenUrl CrossRef PubMed 64. ↵ Zuo , J. et al. Glycolysis Rate-Limiting Enzymes: Novel Potential Regulators of Rheumatoid Arthritis Pathogenesis . Frontiers in Immunology 12 , 779787 ( 2021 ). OpenUrl PubMed 65. ↵ Zhou , R. , Tardivel , A. , Thorens , B. , Choi , I. & Tschopp , J . Thioredoxin-interacting protein links oxidative stress to inflammasome activation . Nat Immunol 11 , 136 – 140 ( 2010 ). OpenUrl CrossRef PubMed Web of Science 66. ↵ Spector , R . Trimethylamine N-oxide induces inflammation and endothelial dysfunction in human umbilical vein endothelial cells via activating ROS-TXNIP-NLRP3 inflammasome . Biochemical and Biophysical Research Communications 481 , 63 – 70 ( 2016 ). OpenUrl CrossRef PubMed 67. ↵ Li , D. et al. Trimethylamine-N-oxide promotes brain aging and cognitive impairment in mice . Aging Cell 17 , e12768 ( 2018 ). OpenUrl CrossRef PubMed 68. ↵ Ximerakis , M. et al. Single-cell transcriptomic profiling of the aging mouse brain . Nat Neurosci 22 , 1696 – 1708 ( 2019 ). OpenUrl CrossRef PubMed 69. ↵ Del Gatto-Konczak , F. et al. The RNA-Binding Protein TIA-1 Is a Novel Mammalian Splicing Regulator Acting through Intron Sequences Adjacent to a 5′ Splice Site . Molecular and Cellular Biology 6287 – 6299 ( 2023 ). 70. Xu , S. et al. SRRM2 organizes splicing condensates to regulate alternative splicing . Nucleic Acids Res 50 , 8599 – 8614 ( 2022 ). OpenUrl CrossRef PubMed 71. Tagawa , T. et al. A virus-induced circular RNA maintains latent infection of Kaposi’s sarcoma herpesvirus . Proceedings of the National Academy of Sciences 120 , e2212864120 ( 2023 ). OpenUrl PubMed 72. ↵ Campagne , S. et al. Molecular basis of RNA-binding and autoregulation by the cancer-associated splicing factor RBM39 . Nature Communications 14 , 1 – 18 ( 2023 ). OpenUrl PubMed 73. ↵ Tollervey , J. R. et al. Analysis of alternative splicing associated with aging and neurodegeneration in the human brain . Genome Res 21 , 1572 – 1582 ( 2011 ). OpenUrl Abstract / FREE Full Text 74. ↵ Wang , Y. et al. rMATS-turbo: an efficient and flexible computational tool for alternative splicing analysis of large-scale RNA-seq data . Nat Protoc 19 , 1083 – 1104 ( 2024 ). OpenUrl CrossRef PubMed 75. ↵ Grosz , B. R. et al. A novel synonymous KMT2B variant in a patient with dystonia causes aberrant splicing . Molecular Genetics & Genomic Medicine 10 , e1923 ( 2022 ). OpenUrl 76. ↵ Zech , M. , et al. Haploinsufficiency of KMT2B, Encoding the Lysine-Specific Histone Methyltransferase 2B, Results in Early-Onset Generalized Dystonia . The American Journal of Human Genetics 99 , 1377 – 1387 ( 2016 ). OpenUrl CrossRef PubMed 77. ↵ Hu , Z.-Y. , Xiao , L. , Bode , A. M. , Dong , Z. & Cao , Y . Glycolytic genes in cancer cells are more than glucose metabolic regulators . J Mol Med (Berl) 92 , 837 – 845 ( 2014 ). OpenUrl PubMed 78. ↵ Fan , Z. et al. The vascular gene Apold1 is dispensable for normal development but controls angiogenesis under pathological conditions . Angiogenesis 26 , 385 – 407 ( 2023 ). OpenUrl CrossRef PubMed 79. ↵ Kim , K.-H. , Won , J. H. , Cheng , N. & Lau , L. F . The matricellular protein CCN1 in tissue injury repair . J Cell Commun Signal 12 , 273 – 279 ( 2018 ). OpenUrl CrossRef PubMed 80. ↵ Shi , H. et al. Kruppel-like factor 2 protects against ischemic stroke by regulating endothelial blood brain barrier function . Am J Physiol Heart Circ Physiol 304 , H796 – 805 ( 2013 ). OpenUrl CrossRef PubMed 81. Hatipoglu , O. F. et al. ADAMTS1 is a unique hypoxic early response gene expressed by endothelial cells . J Biol Chem 284 , 16325 – 16333 ( 2009 ). OpenUrl Abstract / FREE Full Text 82. Seo , J. H. et al. AKAP12 Supports Blood-Brain Barrier Integrity against Ischemic Stroke . Int J Mol Sci 21 , ( 2020 ). 83. Katsuoka , F. et al. Genetic evidence that small maf proteins are essential for the activation of antioxidant response element-dependent genes . Mol Cell Biol 25 , 8044 – 8051 ( 2005 ). OpenUrl Abstract / FREE Full Text 84. Smith , H. W. & Marshall , C. J . Regulation of cell signalling by uPAR . Nature Reviews Molecular Cell Biology 11 , 23 – 36 ( 2010 ). OpenUrl CrossRef PubMed Web of Science 85. ↵ Foltyn , M. et al. The physiological mTOR complex 1 inhibitor DDIT4 mediates therapy resistance in glioblastoma . Br J Cancer 120 , 481 – 487 ( 2019 ). OpenUrl CrossRef PubMed 86. Hai , T. , Wolfgang , C. D. , Marsee , D. K. , Allen , A. E. & Sivaprasad , U . ATF3 and stress responses . Gene Expr 7 , 321 – 335 ( 1999 ). OpenUrl PubMed Web of Science 87. Yuniati , L. , Scheijen , B. , van der Meer , L. T. & van Leeuwen , F. N . Tumor suppressors BTG1 and BTG2: Beyond growth control . Journal of Cellular Physiology 234 , 5379 – 5389 ( 2019 ). OpenUrl CrossRef PubMed 88. ↵ Huang , T. , Yin , J. , Ren , S. ‘e & Zhang , X. Protective effects of KLF4 on blood-brain barrier and oxidative stress after cerebral ischemia-reperfusion in rats through the Nrf2/Trx1 pathway . Cytokine 169 , 156288 ( 2023 ). OpenUrl PubMed 89. ↵ Sangwung , P. et al. KLF2 and KLF4 control endothelial identity and vascular integrity . JCI Insight 2 , e91700 ( 2017 ). OpenUrl 90. ↵ Sayed , N. et al. Clinical trial in a dish using iPSCs shows lovastatin improves endothelial dysfunction and cellular cross-talk in LMNA cardiomyopathy . Sci Transl Med 12 , ( 2020 ). 91. ↵ Xu , D. & Li , C . Gene 33/Mig6/ERRFI1, an Adapter Protein with Complex Functions in Cell Biology and Human Diseases . Cells 10 , 1574 ( 2021 ). OpenUrl 92. ↵ Fritz , G. & Kaina , B . Transcriptional activation of the small GTPase gene rhoB by genotoxic stress is regulated via a CCAAT element . Nucleic Acids Res 29 , 792 – 798 ( 2001 ). OpenUrl CrossRef PubMed Web of Science 93. ↵ Luca , F. et al. Adaptive Variation Regulates the Expression of the Human SGK1 Gene in Response to Stress . PLOS Genetics 5 , e1000489 ( 2009 ). OpenUrl 94. ↵ Meng , F. et al. Genomic editing of intronic enhancers unveils their role in fine-tuning tissue-specific gene expression in Arabidopsis thaliana . Plant Cell 33 , 1997 – 2014 ( 2021 ). OpenUrl CrossRef PubMed 95. ↵ ENCODE Project Consortium et al. Expanded encyclopaedias of DNA elements in the human and mouse genomes . Nature 583 , 699 – 710 ( 2020 ). OpenUrl CrossRef PubMed 96. ↵ Schep , A. N. , Wu , B. , Buenrostro , J. D. & Greenleaf , W. J . chromVAR: inferring transcription-factor-associated accessibility from single-cell epigenomic data . Nat Methods 14 , 975 – 978 ( 2017 ). OpenUrl CrossRef PubMed 97. ↵ Rauluseviciute , I. et al. JASPAR 2024: 20th anniversary of the open-access database of transcription factor binding profiles . Nucleic Acids Res 52 , D174 – D182 ( 2024 ). OpenUrl CrossRef PubMed 98. ↵ Jin , W. , et al. Microbiota-derived bile acids antagonize the host androgen receptor and drive anti-tumor immunity . Cell 188 , 2336 – 2353 .e38 ( 2025 ). OpenUrl CrossRef PubMed 99. ↵ Klemm , S. L. , Shipony , Z. & Greenleaf , W. J . Chromatin accessibility and the regulatory epigenome . Nature Reviews Genetics 20 , 207 – 220 ( 2019 ). OpenUrl CrossRef PubMed 100. ↵ Janowsky , J. S . The role of androgens in cognition and brain aging in men . Neuroscience 138 , 1015 – 1020 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 101. Jardí , F. et al. Testosterone boosts physical activity in male mice via dopaminergic pathways . Sci Rep 8 , 957 ( 2018 ). OpenUrl PubMed 102. ↵ Davey , R. A. et al. Androgen Action via the Androgen Receptor in Neurons Within the Brain Positively Regulates Muscle Mass in Male Mice . Endocrinology 158 , 3684 – 3695 ( 2017 ). OpenUrl PubMed 103. ↵ Gräff , J. & Tsai , L.-H . The potential of HDAC inhibitors as cognitive enhancers . Annu Rev Pharmacol Toxicol 53 , 311 – 330 ( 2013 ). OpenUrl CrossRef PubMed 104. ↵ Slijepcevic , D. et al. Hepatic uptake of conjugated bile acids is mediated by both sodium taurocholate cotransporting polypeptide and organic anion transporting polypeptides and modulated by intestinal sensing of plasma bile acid levels in mice . Hepatology 66 , 1631 – 1643 ( 2017 ). OpenUrl CrossRef PubMed 105. ↵ Zhao , S. et al. Post sleeve gastrectomy-enriched gut commensal Clostridia promotes secondary bile acid increase and weight loss . Gut Microbes 17 , 2462261 ( 2025 ). OpenUrl PubMed 106. ↵ Qi , X. et al. Gut microbiota–bile acid–interleukin-22 axis orchestrates polycystic ovary syndrome . Nature Medicine 25 , 1225 – 1233 ( 2019 ). OpenUrl CrossRef PubMed 107. ↵ Meessen , E. C. E. et al. Glycodeoxycholic Acid Inhibits Primary Bile Acid Synthesis With Minor Effects on Glucose and Lipid Homeostasis in Humans . J Clin Endocrinol Metab 110 , 1468 – 1477 ( 2025 ). OpenUrl PubMed 108. ↵ Monteiro-Cardoso , V. F. , Corlianò , M. & Singaraja , R. R . Bile Acids: A Communication Channel in the Gut-Brain Axis . Neuromolecular Med 23 , 99 – 117 ( 2021 ). OpenUrl CrossRef PubMed 109. ↵ Yu , S. et al. Glutamatergic Preoptic Area Neurons That Express Leptin Receptors Drive Temperature-Dependent Body Weight Homeostasis . J. Neurosci . 36 , 5034 – 5046 ( 2016 ). OpenUrl Abstract / FREE Full Text 110. ↵ Uchino , E. et al. Identification of hypothermia-inducing neurons in the preoptic area and activation of them by isoflurane anesthesia and central injection of adenosine . J Physiol Sci 74 , 33 ( 2024 ). 111. ↵ Taylor , T. N. , Greene , J. G. & Miller , G. W . Behavioral phenotyping of mouse models of Parkinson’s disease . Behavioural Brain Research 211 , 1 – 10 ( 2010 ). OpenUrl CrossRef PubMed Web of Science 112. ↵ Yalçin , M. , Grande , V. , Outeiro , T. F. & Relógio , A . Circadian clock dysfunction in Parkinson’s disease: mechanisms, consequences, and therapeutic strategy . NPJ Parkinsons Dis 11 , 213 ( 2025 ). 113. ↵ Kirshenbaum , G. S. , Dachtler , J. , Roder , J. C. & Clapcote , S. J . Characterization of cognitive deficits in mice with an alternating hemiplegia-linked mutation . Behav Neurosci 129 , 822 – 831 ( 2015 ). OpenUrl CrossRef PubMed 114. ↵ Heredia-López , F. J. et al. An automated Y-maze based on a reduced instruction set computer (RISC) microcontroller for the assessment of continuous spontaneous alternation in rats . Behav Res Methods 48 , 1631 – 1643 ( 2016 ). OpenUrl CrossRef PubMed 115. ↵ Bendl , J. et al. The three-dimensional landscape of cortical chromatin accessibility in Alzheimer’s disease . Nat Neurosci 25 , 1366 – 1378 ( 2022 ). OpenUrl CrossRef PubMed 116. ↵ Li , X. et al. Transcriptional and epigenetic decoding of the microglial aging process . Nature Aging 3 , 1288 – 1311 ( 2023 ). OpenUrl PubMed 117. ↵ Canyelles , M. et al. Gut Microbiota-Derived TMAO: A Causal Factor Promoting Atherosclerotic Cardiovascular Disease? Int J Mol Sci 24 , ( 2023 ). 118. ↵ Stilling , R. M. et al. Social interaction-induced activation of RNA splicing in the amygdala of microbiome-deficient mice . Elife 7 , ( 2018 ). 119. ↵ Thion , M. S. , et al. Microbiome Influences Prenatal and Adult Microglia in a Sex-Specific Manner . Cell 172 , 500 – 516 .e16 ( 2018 ). OpenUrl CrossRef PubMed 120. ↵ Kaur , H. , Singh , Y. , Singh , S. & Singh , R. B . Gut microbiome-mediated epigenetic regulation of brain disorder and application of machine learning for multi-omics data analysis . Genome 64 , 355 – 371 ( 2021 ). OpenUrl 121. ↵ Molinaro , A. et al. Imidazole propionate is increased in diabetes and associated with dietary patterns and altered microbial ecology . Nat Commun 11 , 5881 ( 2020 ). OpenUrl CrossRef PubMed 122. ↵ Shah , S. N. et al. Cerebrovascular damage caused by the gut microbe/host co-metabolite p-cresol sulfate is prevented by blockade of the EGF receptor . Gut Microbes 16 , 2431651 ( 2024 ). 123. Ticinesi , A. , Guerra , A. , Nouvenne , A. , Meschi , T. & Maggi , S . Disentangling the Complexity of Nutrition, Frailty and Gut Microbial Pathways during Aging: A Focus on Hippuric Acid . Nutrients 15 , ( 2023 ). 124. ↵ Li , M. et al. Gut microbiota metabolite indole-3-acetic acid maintains intestinal epithelial homeostasis through mucin sulfation . Gut microbes 16 , ( 2024 ). 125. ↵ Mayhan , W. G . VEGF increases permeability of the blood-brain barrier via a nitric oxide synthase/cGMP-dependent pathway . Am J Physiol 276 , C1148 – 53 ( 1999 ). OpenUrl 126. ↵ Qing , Y. et al. Targeted metabolomics reveals aberrant profiles of serum bile acids in patients with schizophrenia . Schizophrenia (Heidelb) 8 , 65 ( 2022 ). 127. ↵ Pavlova , T. et al. Urinary intermediates of tryptophan as indicators of the gut microbial metabolism . Anal Chim Acta 987 , 72 – 80 ( 2017 ). OpenUrl CrossRef PubMed 128. ↵ Yang , B. et al. HIV-Differentiated Metabolite N-Acetyl-L-Alanine Dysregulates Human Natural Killer Cell Responses to Mycobacterium tuberculosis Infection . International Journal of Molecular Sciences 24 , 7267 ( 2023 ). OpenUrl PubMed 129. ↵ Green , M. R. & Sambrook , J . Precipitation of RNA with Ethanol . Cold Spring Harb Protoc 2020 , 101717 ( 2020 ). OpenUrl PubMed 130. ↵ Bray , N. L. , Pimentel , H. , Melsted , P. & Pachter , L . Near-optimal probabilistic RNA-seq quantification . Nature Biotechnology 34 , 525 – 527 ( 2016 ). OpenUrl CrossRef PubMed 131. ↵ Love , M. I. , Huber , W. & Anders , S . Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2 . Genome Biol 15 , 550 ( 2014 ). 132. ↵ Raudvere , U. et al. g:Profiler: a web server for functional enrichment analysis and conversions of gene lists (2019 update) . Nucleic Acids Res 47 , W191 – W198 ( 2019 ). OpenUrl CrossRef PubMed 133. ↵ Garrido-Martín , D. , Palumbo , E. , Guigó , R. & Breschi , A. ggsashimi: Sashimi plot revised for browser-and annotation-independent splicing visualization . PLoS Comput Biol 14 , e1006360 ( 2018 ). OpenUrl CrossRef PubMed 134. ↵ Liu , B. B. et al. Dissecting regulatory syntax in human development with scalable multiomics and deep learning . Preprint at https://www.biorxiv.org/content/10.1101/2025.04.30.651381v1 ( 2025 ). 135. ↵ Marinov , G. K. , Shipony , Z. , Kundaje , A. & Greenleaf , W. J . Genome-Wide Mapping of Active Regulatory Elements Using ATAC-seq . Methods Mol Biol 2611 , 3 – 19 ( 2023 ). OpenUrl CrossRef PubMed 136. ↵ Picelli , S. et al. Tn5 transposase and tagmentation procedures for massively scaled sequencing projects . Genome Res 24 , 2033 – 2040 ( 2014 ). OpenUrl Abstract / FREE Full Text 137. ↵ Marinov , G. K. & Shipony , Z . Interrogating the Accessible Chromatin Landscape of Eukaryote Genomes Using ATAC-seq . Methods Mol Biol 2243 , 183 – 226 ( 2021 ). OpenUrl CrossRef PubMed 138. ↵ Grandi , F. C. , Modi , H. , Kampman , L. & Corces , M. R . Chromatin accessibility profiling by ATAC-seq . Nat Protoc 17 , 1518 – 1552 ( 2022 ). OpenUrl CrossRef PubMed 139. ↵ Systematic identification of age-related microbiome metabolites that impact the brain . National Center for Biotechnology Information . https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1328687 ( 2025 ). View the discussion thread. Back to top Previous Next Posted November 02, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Age-related microbiome metabolites alter RNA splicing and chromatin accessibility in the brain Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Age-related microbiome metabolites alter RNA splicing and chromatin accessibility in the brain Meenakshi Chakraborty , Sophia M. Shi , Imani E. Porter , Daniel J. Richard , Georgi K. Marinov , Ashley A. Moore , Jenna L. E. Blum , Aravind Natarajan , James W. Jahng , Joseph C. Wu , Sydney X. Lu , Shawn M. Davidson , William J. Greenleaf , Nay L. Saw , Mehrdad Shamloo , Anne Brunet , Tony Wyss-Coray , Ami S. Bhatt bioRxiv 2025.10.03.680371; doi: https://doi.org/10.1101/2025.10.03.680371 Share This Article: Copy Citation Tools Age-related microbiome metabolites alter RNA splicing and chromatin accessibility in the brain Meenakshi Chakraborty , Sophia M. Shi , Imani E. Porter , Daniel J. Richard , Georgi K. Marinov , Ashley A. Moore , Jenna L. E. Blum , Aravind Natarajan , James W. Jahng , Joseph C. Wu , Sydney X. Lu , Shawn M. Davidson , William J. Greenleaf , Nay L. Saw , Mehrdad Shamloo , Anne Brunet , Tony Wyss-Coray , Ami S. Bhatt bioRxiv 2025.10.03.680371; doi: https://doi.org/10.1101/2025.10.03.680371 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Neuroscience Subject Areas All Articles Animal Behavior and Cognition (7629) Biochemistry (17660) Bioengineering (13881) Bioinformatics (41909) Biophysics (21435) Cancer Biology (18576) Cell Biology (25479) Clinical Trials (138) Developmental Biology (13366) Ecology (19887) Epidemiology (2067) Evolutionary Biology (24301) Genetics (15598) Genomics (22482) Immunology (17726) Microbiology (40359) Molecular Biology (17162) Neuroscience (88529) Paleontology (666) Pathology (2830) Pharmacology and Toxicology (4820) Physiology (7636) Plant Biology (15125) Scientific Communication and Education (2044) Synthetic Biology (4290) Systems Biology (9817) Zoology (2269)