Gut microbes shape microglia and cognitive function during malnutrition

Abstract Fecal‐oral contamination promotes malnutrition pathology. Lasting consequences of early life malnutrition include cognitive impairment, but the underlying pathology and influence of gut microbes remain largely unknown. Here, we utilize an established murine model combining malnutrition and iterative exposure to fecal commensals (MAL‐BG). The MAL‐BG model was analyzed in comparison to malnourished (MAL mice) and healthy (CON mice) controls. Malnourished mice display poor spatial memory and learning plasticity, as well as altered microglia, non‐neuronal CNS cells that regulate neuroimmune responses and brain plasticity. Chronic fecal‐oral exposures shaped microglial morphology and transcriptional profile, promoting phagocytic features in MAL‐BG mice. Unexpectedly, these changes occurred independently from significant cytokine‐induced inflammation or blood–brain barrier (BBB) disruption, key gut‐brain pathways. Metabolomic profiling of the MAL‐BG cortex revealed altered polyunsaturated fatty acid (PUFA) profiles and systemic lipoxidative stress. In contrast, supplementation with an ω3 PUFA/antioxidant‐associated diet (PAO) mitigated cognitive deficits within the MAL‐BG model. These findings provide valued insight into the malnourished gut microbiota‐brain axis, highlighting PUFA metabolism as a potential therapeutic target.

Interdependent societal (e.g., political instability, socioeconomic status) and environmental factors not only promote poverty, but also contribute to the biological consequences of malnutrition (Black et al., 2013;Smith & Haddad, 2015).
Environmental factors range from climate change to poor sanitation, water quality, and/or hygiene practices Phalkey et al., 2015;Tofail et al., 2018). These latter conditions promote microbial fecal-oral contamination and subsequent gastrointestinal (GI) insult. Chronic GI dysbiosis and nutrient deficiency form a deadly cycle of deteriorating nutritional status and health, linked to childhood stunting and lasting immune, metabolic, and cognitive impairment within malnourished communities (Black et al., 2013;Di Giovanni et al., 2016;Guerrant et al., 2013). Despite continued progress in the pathology and treatment of malnutrition, the precise role of fecal-oral contamination remains largely unexplored.
We specifically assessed the impact of fecal-oral contamination and malnutrition on microglia, non-neuronal cells whose functionality and maturation are shaped by commensal microbes (Erny et al., 2015).
We report altered cognitive, microglial, and microbiome features in malnourished mice exposed to fecal-oral contamination. MAL-BG microglia exhibit impaired morphology, transcriptional profile, and phagocytic features. These microglial alterations occur independently from significant BBB deficits or elevated proinflammatory cytokines (TNF-α, IL-6) within the CNS. Metabolomic profiling, however, revealed shifts in polyunsaturated fatty acid (PUFA) metabolism and lipoxidative stress within the MAL-BG CNS. Moreover, supplementation with ω3 PUFAs and vitamin antioxidants (PAO diet) improved cognitive deficits in the MAL-BG model. Collectively, our results highlight dynamic microglial responses to commensal microbes and diet, identifying fatty acid metabolism as a potential gut-brain pathway informing neurocognitive consequences of childhood malnutrition.

| Mouse studies
Three-week-old female C57BL/6J mice (Jackson Laboratory, Bar Harbor, ME) were housed in the Modified Barrier Facility at the University of British Columbia on a 12-h light-dark cycle. On arrival, mice were randomized and housed in separate groups (3-5 per group, ventilated cages with wood chip bedding and enrichment). Mice were fed either a standard mouse chow (D09051102: Research Diets, New Brunswick, NJ; Supplemental File 1) or an isocaloric low-protein/fat diet (D14071001). While malnutrition refers to any deviation from healthy nutritional status, undernutrition specifies nutritional deficiency. Despite protein and fat deficiencies, the D14071001 diet has standard caloric content. To maintain consistent naming with earlier publications, mice fed this diet are referenced as malnourished or MAL Brown et al., 2015;Huus et al., 2020). Mice received ad libitum chow and water.
To assess microglia morphology and motility (Figure 2 and Figure S3), we utilized weaned male and female CX3CR1 +/EGFP mice ranging 3-6 weeks old at initiation of trial on C57BL/6 background (Jung et al., 2000), which were bred and housed at the Animal Resource Unit facility at the University of British Columbia. Apparent sex differences were not observed in the study. The microglia analyses for the PAO intervention trial utilized newly-weaned female C57BL/6 mice housed at the Modified Barrier Facility. All animal work was done in accordance with the Animal Care Committee at the University of British Columbia and the Canadian Council on Animal Care guidelines.

| MAL-BG model
To elicit MAL-BG features, a subgroup of mice on the low-protein diet was exposed to a cocktail of seven commensal bacteria as previously reported (Brown et al., 2015). Original bacterial strains were provided by Emma Allen-Vercoe (University of Guelph). Briefly, frozen stocks of bacterial cultures were plated on FAA in anaerobic conditions. Bacteria were mixed in a 1:1 ratio in sterile, reduced PBS for oral gavage (100 μl, 10 9 cells/ml). Three gavages were administered over a period of 5 days 2 weeks following initiation of the experimental diets. To control for gavage stress, non-MAL-BG mice received PBS gavages over the same period.

| PAO intervention
To assess whether MAL-BG features could be reversed, ω3 PUFAenriched, antioxidant-associated (PAO) diets with were developed.

| Novel object recognition test (NORT)
To assess exploratory behaviors, two identical objects were placed in the OFT box at opposing sides. Individual mice were allowed to freely explore objects for a 3 min habituation phase. Following a 3-h delay period, one object was replaced by a distinct, yet similarly sized object (novel object). Individual mice were returned to the OFT for a 3 min test phase. A blinded observer recorded interaction times. For this test, we defined mouse interaction as sniffing and/or placing the snout on the object.

| Light-dark test
Mice were placed in a 10.5 Â 34.5 cm 2 light-dark box (one third light zone, two third dark zone). Animals were allowed to freely explore for 3 min, the light-dark box was cleaned with 70% ethanol between use.

| Morris water maze
The Morris water maze (MWM) was utilized to assess learning and spatial memory in mice. Testing occurred at the University of British Columbia Modified Barrier Facility. Mice were tested in a pool $116 cm diameter (water temperature, 21-23 C). The testing arena was supplied with indirect light with the MWM pool surrounded by distal visual cues. A circular platform (11 cm diameter) was used as the goal platform (see Figure S2f). Fecal droppings were removed from the platform between trials. After testing, mice were gently dried and placed in a warming cage prior to returning to their home cage. Platform and animal start positions were randomly determined for each of the training and testing days. Certain mice in MWM tests were removed due to video recorder malfunction and/or prolonged swim refusal (≤3 mice/trial of 48 total mice [ Figure 1] or 32 total mice [ Figure 5]). All mouse removals were selected during blinded analyses.
2.4.5 | Visible platform training (1 day, four trials) Fixed mouse start position/variable platform location-in this habituation day, individual mice were given 60 s to locate the visible goal platform (opaque top, 1-1.5 cm above water). Mice that failed to climb the platform within 60 s were gently guided onto the goal platform. To promote spatial memory, mice were given a 30 s rest period on the platform between successive trials. All mice were able to recognize the platform by the end of training.
2.4.6 | Acquisition training (2 days, 12 trials) Variable mouse start position/fixed platform location-the goal platform was not visible (clear top, 0.5 cm submerged) during acquisition periods. Individual mice were given 60 s to locate the goal platform.
Mice that failed to locate the platform were gently guided onto the platform following the trial period. Mice were allowed to briefly rest on the platform for the first four consecutive trials of each day. In remaining trials, mice were immediately removed after locating the platform. Only the first four trials each day were recorded. PAO intervention mice underwent a maximum of four trials.
2.4.7 | Free swim 1 (1 day) Variable mouse start position/platform removed-mice were allowed to freely explore the pool during a 30 s probe run. PAO mice did not undergo any free swim trials.
2.4.8 | Acquisition training reversal (2 day, 12 trials) Variable mouse start position/fixed platform location-before training, the goal platform was moved to a different quadrant. Protocol follows initial acquisition training.
2.4.9 | Free swim 2 (1 day, 1 probe) Same protocol as initial free swim. Free swims occurred 24 h following the final acquisition trial.

| Ex vivo cytokine profiling
Following euthanasia, whole brain tissues were collected within individual Eppendorf tubes containing 1 ml dPBS and cOmplete™ EDTA-free Protease Inhibitor, prior to storage at À70/80 C or immediately homogenized using a Retsch MM 301 Mixer Mill or FastPrep ® -24 (MP Biomedicals) 2x for 1 min using tungsten beads. Collected blood was spun at 6000 g for 8 min to obtain sera. Cytokine measurements from tissue supernatant and sera were obtained with the BD Biosciences Cytometric Bead Array Mouse Inflammation Kit. Cytokine measurements from whole brain samples were normalized to tissue weight.
2.7 | Two-photon microscopy, time-lapse imaging, and lesion analysis Acute slices from CX3CR1 +/EGFP C57BL/6 mice were imaged immediately after recovery using a Coherent Chameleon Ultra II laser (modelocked pulse train at 80 MHz at 920 nm) with a Zeiss LSM 7 MP microscope and Zeiss 20x-W/1.0 NA objective. Green fluorescence was detected by a 520/60 nm filter (Chroma tech) and GaAsP photomultiplier tube (PMT; Zeiss LSM BiG). Images were acquired as a z-stack (zoom factor 2.8; 151.82 Â 151.82 μm xy scale, 8-line averaging) 18 μm thick, centered approximately 150 μm below the slice surface (2 μm slice interval) in the stratum radiatum region of the CA1 hippocampus. Following a 10-min baseline imaging period, a lesion was created by focusing the laser to the region of interest and scanning at 800 nm at 100% power for approximately 30 s. Microglial response to this lesion was then imaged for an additional 15 min using the same imaging parameters as baseline.
For motility analysis, baseline movies were maximum projected and loaded into a custom MATLAB program. This program quantifies the number of new pixels (additions) and number of removed pixels (retractions) across time as the Motility Index. To quantify the microglial response to lesion, a circular region of interest with a diameter of 30 μm was centered on the lesion response region, and the mean intensity was measured at each frame.

| 3D-morph and phagocytic cup quantification
EGFP is well preserved by the SNAPSHOT protocol, and these slices were ready to image immediately after 1-week incubation in clearing solution at 4 C. By two-photon microscopy, a z-stack at 1024 Â 1024 (zoom factor 1.5; 283.12 Â 283.12 μm xy scale, 16-line averaging) from 125-175 μm deep (2 μm slice interval) was acquired. Using these images, 3D-Morph MATLAB analysis was completed as previously reported (York, LeDue, et al., 2018) to quantify microglial morphologies. Before analysis, all images were processed by background subtraction in Fiji, and all treatment rounds were batch processed using the same analysis parameters. From these morphological images, the number of phagocytic cups was manually counted.
2.9 | Blood-brain barrier integrity: IgG and biocytin To investigate BBB permeability, 100 μl TMR Biocytin (AnaSpec AS-60658; reconstituted with sterile PBS; MW = 869 Da) was delivered by tail-vein injection to mice 20 min prior to cardiac perfusion. Following brain dissection and coronal slicing (300 μm thick by vibratome), tissues were imaged using a Zeiss Axio Zoom microscope with TMR emission filter settings. Fluorescence intensity was measured from slices spanning the entire rostral-caudal area of the brain. Mean intensity was compared across treatments.
As an additional permeability measure, slices were stained for anti-mouse IgG, which should not be present in the brain parenchyma.
For staining, thick slices were cleared (20% DMSO and 2% Triton X-100 in PBS) for 1 week, blocked in 4% normal goat serum overnight at room temperature, and incubated with Alexa Fluor 488 goat antimouse IgG for 6 days at 4 C. After four, 1-h washes in PBS at room temperature, the tissue was imaged by two-photon microscopy using a 20x-W/1.0 NA objective and 5Â zoom factor. The mean fluorescence intensity was averaged across three separate images per slice, and compared between mice.

| RNA-SEQ analysis
Whole mouse cerebra were stored on ice in RPMI growth media prior to tissue dissociation. Tissues were dissociated via the Adult  (Zerbino et al., 2018) using STAR v. 2.6.1d, followed by read-count generation using HTSeq v. 0.11.2 (Anders et al., 2015). Differential gene expression was estimated with DeSEQ2 v. 3.9 with further pathway analyses conducted using the ReactomePA pipeline, as described (Love et al., 2014;Yu & He, 2016). Analyses were conducted with R (v. 3.5.1). Code provided in attached Glia_RMarkdown (Supplementary information). Raw and processed data files were deposited to the NCBI gene expression omnibus (GEO): GEO accession GSE138182.

| qPCR
RT-qPCR analysis was performed using QuantiTect SYBR Green PCR Master Mix (Qiagen) from ileum or whole brain (cortical) tissue using the following primers, Ctsd CCCGCCTTCTGTATCTGTGT) based on established PCR protocols (Brown et al., 2015). Gapdh and Hprt provided an endogenous control for gut (Tjp1) and microglial genes of interest, respectively, and were used for normalization. ddCT calculations provided relative expression to control samples. Following staining, cells were washed twice and fixed in a 1:1 solution of supplemented dPBS À/À : 4% paraformaldehyde overnight at 4 C.

| Flow cytometry
After fixation, cells were resuspended in supplemented dPBS À/À and enumerated via flow cytometry (BD LSR II with 561 laser). Microglia populations were identified as CD11b high /CD45 low . Subsequent data was analyzed using FlowJo software (v. 10.5.3).

| Metabolomics
Mouse hippocampal tissues were collected for untargeted reversedphased ultrahigh performance liquid chromatography-Fourier transform mass spectrometry (RP-UPLC-FTMS) metabolomics analysis. Tissue samples were kept in dry ice prior to storage at À70/80 C. Metabolomics were completed by TMIC (The Metabolomics Innovation Centre).

| Metabolite extraction
Each mouse hippocampal sample in an Eppendorf tube was mixed with water; 5 μl per mg of the tissue, and two 4-mm metal balls were added. The tissue was homogenized on a MM 400 mill mixer at a vibrating frequency of 30 Hz for 1 min twice. After 5-s spin-down, a mixture of methanol-chloroform (4:1) was added, at 25 μl per mg tissue, to each tube. The sample was homogenized again for metabolite extraction using the same setup for 1 min twice, followed by sonication in an ice-water bath for 5 min. The tube was centrifuged at 15,000 rpm at 10 C for 20 min. The clear supernatant was transferred to a 1.5-ml Eppendorf tube. A 60-μl aliquot from each sample was dried down inside the same nitrogen evaporator and the residue was reconstituted in 40 μl of 80% methanol. 10 μl was injected for RP-UPLC-FTMS. Two rounds of sample injections were made, with positive-and negative-ion detection, respectively.

| RP-UPLC-FTMS analysis
A Dionex Ultimate 3000 UHPLC system coupled to a Thermo LTQ-Orbitrap Velos Pro mass spectrometer, equipped with electrospray ionization (ESI) source, was used. RP-UPLC-FTMS runs were carried out with a Waters BEH C8 column (2.1 Â 50 mm 2 , 1.7 μm) for chromatographic separations. The mobile phase was (A) 0.01% formic acid in water and (B) 0.01% formic acid in acetonitrile-isopropanol (1:1). The elution gradient was 5%-50% B in 5 min; 50%-100% B in 15 min; and 100% B for 2 min before column equilibration for 4 min between injections. The column flow was 400 μl/min while the column temperature was 60 C. For relative quantitation, the MS instrument was run in the survey scan mode with FTMS detection at a mass resolution of 60,000 full width at half maximum at m/z 400. The mass scan range was m/z 80-1800, with a reference lock-mass for real-time calibration. Two UPLC-FTMS datasets were acquired for each sample, one with positiveion detection and the other with negative-ion detection. LC-MS/MS data was also acquired from each sample set with collision-induced dissociation at different levels of normalized collision energy.

| Data processing
Each LC-FTMS dataset was respectively processed with XCMS (https://xcmsonline.scripps.edu/) in R for peak detection and two rounds of retention time (RT) shift correction, peak grouping, and peak alignment. Mass de-isotoping and removal of chemical and electronic background peaks were performed with manual interventions. The output of data processing is the pairs of m/z (mass-to-charge ration), RT (min), and LC-MS peak areas of the detected metabolites or metabolite features across the samples for each set.

| Metabolomics analyses
To assign the metabolite candidates of any potential biomarkers, the measured m/z's were searched against metabolome databases, includ-  (Chong & Xia, 2018) with the following parameters: mass tolerance = 0.0003, RT tolerance = 30, data was normalized against pooled CON, data transformation = log-transformation, and data scaling = auto data scaling. Over-representation analysis determined over-represented pathways in each group. Analyses were conducted based on previous analyses (Brown et al., 2015). A one-way ANOVA was used to determine significant changes across groups (Padj < 0.05; fold change >2).

| Fatty acid preparation
Total lipids from homogenates (50-150 mg) tissue were extracted with chloroform/methanol (2:1 vol/vol; 3Â) in the presence of 0.01% butylated hydroxytoluene. The resulting chloroform phase was evaporated under nitrogen. After lipid extraction, fatty acyl groups were analyzed as methyl esters derivatives by gas chromatography (GC).
Briefly, fatty acids were transesterified at 75 C for 90 min through incubation in 2 ml of 5% methanolic. The resulting fatty acid methyl esters (FAMEs) were extracted by adding 2 ml of n-pentane and 1 ml of saturated NaCl solution. The separated n-pentane phase was evaporated under nitrogen and dissolved in 80 μl of carbon disulfide. Two μl of sample were used in GC analysis.

| GC method
Analyses were conducted using the GC System 7890A/Series Injector 7683B (Agilent, Barcelona, Spain) and a flame ionization detector/DBWAX capillary column (30 m length Â 0.25 mm [inner diameter] Â 0.20 μm [film thickness]). The injections were performed with the splitless mode at 220 C. The flow rate of carrier gas (helium 99.99%) was maintained at a constant rate of 1.8 ml/min. The column temperature was held at 145 C for 5 min, increased by 2 C/min to 245 C for 50 min, and held at 245 C for 10 min with a post-run of 250 C for 10 min.

| Data analysis
Identification of the 25 FAMEs was made with authenticated standards (Larodan Fine Chemicals, Malmö, Sweden). Results were expressed as %mol and then normalized to CON. The fatty acid profile detected, identified, and quantified represents more than 95% of the total chromatogram. The following fatty acid indexes were calculated: PUFAs from ω3 and ω6 series (PUFA ω3 and PUFA ω6) and a proinflammatory index (ω6/ω3):(PUFA ω6/PUFA ω3).
Samples containing 0.5 mg of protein were delipidated as The samples were hydrolyzed at 155 C for 30 min in 1 ml of 6N HCl, and then concentrated by speed-vac. The N,O-trifluoroacetyl methyl ester derivatives (TFAMEs) of the protein hydrolysate were prepared as previously described (Knecht et al., 1991). Briefly, hydro-

| Microbiome analyses
Fecal samples were collected from mice and kept in À70 C prior to isolation. Fecal DNA was released by boiling sample suspensions for 15 min at 100 C. Library preparation for 16S rRNA sequencing was then performed by Microbiome Insights according to a standardized pipeline (https://microbiomeinsights.com/itag-microbiome-analysis/).
Briefly, PCR amplification of the 16S rRNA gene was performed using barcoded primers against the V4 region (Kozich, Schloss et al, 2013 To specifically quantify E. coli abundance, qPCR of extracted fecal DNA was performed using Enterobacteriaceae-specific primers (F: CATTGACGTTACCCGCAGAAGAAGC, R: CTCTACGAGACTCA AGCTTGC), as described in (Brown et al., 2015). Universal bacterial 16S primers were used for normalization (F: ACTCCTACGGGAGG CAGCAGT, R: ATTACCGCGGCTGCTGGC). The relative abundance of Enterobacteriaceae was calculated using the formula 2^-(X-Y) where X = the mean Ct for the universal 16S reaction and Y = the mean Ct for the Enterobacteriaceae reaction. qPCR was performed on an Applied Biosystems 7500 machine, with QuantiTect SYBR Green PCR Kits (QIAGEN 204143), in 10 ul reaction volumes using 2 ul template DNA.

| Statistical analysis
Datasets generated/analyzed in this project were deposited in appropriate online repositories. Any additional data that support the findings of this study are available from the corresponding author upon reasonable request. Statistical analyses were performed using GraphPad PRISM. Statistical significance was given as ****p < .0001, ***p < .001, **p < .01, *p < .05, and Padj = FDR correction. Analyses are expressed as the mean with SEM unless otherwise stated.

| Behavior and cognitive plasticity altered in MAL-BG mice
Upon weaning, C57BL/6J mice were randomized onto a standard diet (CON mice) or an isocaloric malnourished diet (MAL and MAL-BG mice; Figure 1a). MAL-BG mice received three gavages containing E.
coli/Bacteroidales, a bacterial cocktail designed to model fecal-oral contamination, a prevalent driver of gut dysbiosis and long-term cognitive deficits among undernourished communities (Brown et al., 2015;Guerrant et al., 2013;Kau et al., 2015;Vonaesch et al., 2018). After 4 weeks on the malnourished diet, MAL and MAL-BG mice exhibit modest growth faltering and reduced tail length, a proxy for murine stunting ( Figure S1a).
Our lab previously demonstrated that E. coli and Bacteroidales exposures fail to trigger growth deficits and gut dysbiosis in the absence of malnutrition (CON-BG model), likely due to lack of robust E. coli/Bacteroidales colonization in mice fed a healthy diet (Brown et al., 2015); growth and fecal Enterobacteriaceae relative abundance measurements repeated in Figure S1b,c. In contrast, MAL-BG mice exhibit striking gut dysbiosis, impaired gastrointestinal hostmicrobe interactions, intestinal barrier deficits, and altered gut metabolomes (Brown et al., 2015;Huus et al., 2020). As E. coli/ Bacteroidales bacterial gavage fails to robustly colonize healthy control mice, we were not able to appropriately assess whether these fecal exposures influenced brain and behavior alterations in the absence of dietary malnutrition. Consequently, we focused further analyses on malnutrition models.
We next assessed whether the MAL and MAL-BG models exhibit behavioral and/or cognitive impairments, neurologic consequences of early-life malnutrition and co-occurring fecal-oral contamination (Black et al., 2013;Investigators, 2018). CON mice provided a healthy control for murine tests.
The OFT measures locomotion and exploration (Gould et al., 2009). Increased aversion to the central open field zone (OFZ) connotes anxiety-like behavior in rodents (Gould et al., 2009). MAL-BG mice spent more time within the OFZ (Figure 1b,c), displaying increased exploration compared to either CON or MAL groups (F 2,57 = 6. 878, p = .0021). Immobility (resting) within the OFZ, but not total OFT immobility, increased among MAL-BG mice, supporting an absence of OFZ-induced anxiety ( Figure S2a). These results were not shaped by gross locomotion deficits as total distance traveled in the OFT was comparable across all groups ( Figure S2b). In addition, CON, MAL, and MAL-BG mice displayed comparable behavior within the light-dark box, an established murine anxiety test (Bourin & Hascoët, 2003), further supporting altered exploration, rather than anxiety-like activity ( Figure S2c).
MAL-BG mice also exhibited distinct exploratory patterns during the NORT. NORT not only measures novelty exploration, but also assessed short-term memory performance (Leger et al., 2013). During a brief familiarization period, individual mice freely explored an arena with two identical objects. After familiarization, one object was replaced with a similarly sized, but distinctly "novel" object ( Figure S2d).
Individual mice were returned to the disinfected arena after several hours for the recall period. As rodents typically exhibit novelty preference, decreased exploration of the novel item indicates impaired novel object recognition. All groups exhibited novelty preference (novel:old exploration ratio > 1; t 47 = 6.779, p < .0001) and comparable total exploration time ( Figure S2d,e). Compared to CON and MAL counterparts, MAL-BG mice exhibited a modest, but not significant, increase in novel object interaction ( Figure S2d). Results from both the OFT and NORT suggest altered exploratory behavior in MAL-BG mice.
We further assessed spatial memory and cognition via the Morris water maze test (MWMT), which measures spatial learning, reference memory, and cognitive flexibility (Vorhees & Williams, 2006). Mice underwent two training periods (acquisition, reversal) to learn the location of a hidden platform (Figure 1d and Figure S2f). Average swim speeds were recorded during a 30 s free swim (no platform) 24 h after each training period. As MAL and MAL-BG mice displayed similar swimming capability to healthy controls ( Figure S2g), MWMT results were not influenced by altered physicality.
We observed comparable reference memory and spatial learning during acquisition training (Figure 1d). We next probed learning within the context of cognitive flexibility, placing the hidden platform within the opposite pool quadrant (reversal learning). Upon reversal, MAL and MAL-BG escape latencies (time to platform) increased, indicative of impaired learning flexibility (Vorhees & Williams, 2006). Learning deficits persisted, even broadened, across the reversal period (Day 1 R : F 2,45 = 2. 836, p = .0692; Day 2 R : F 2,42 = 5.205, p = .0096). Accidental platform discovery did not drive these findings, as reversal escape latencies during the initial trial were comparable across groups ( Figure S2h). CON mice rapidly learned the new location of the hidden platform, while malnourished mice persistently honed to the prior platform location, indicative of impaired memory extinction. By the final training day (Day 2 R ), both CON and MAL mice significantly eliminated prior platform entries (CON: paired t-test = 3.753, p = .0024; MAL: paired t-test = 4.250, p = .0007). Indeed, over half of CON mice never entered the prior platform area and these mice rapidly located the hidden platform (Figure 1f and Figure S1i). In contrast, MAL-BG mice did not exhibit significant memory extinction (Figure 1e,f and Figure S1i), as measured by prior platform entry from Day 1 R to Day 2 R (MAL-BG: paired t-test = 1.237, p = 0 .2381), exhibiting marked cognitive inflexibility (Mills et al., 2014).

| Diet and gut microbes influence microglial morphology and function
Microglia modulate brain plasticity and maintain CNS homeostasis.
Indeed, alterations in microglial morphology and phagocytic capacity inform memory plasticity (Wang et al., 2020). Like peripheral macrophages, microglia are highly responsive to environmental changes within the brain, regulating and responding to inflammatory and metabolic shifts, as well as gut microbial alterations (Erny et al., 2015;Wu et al., 2015;. Mature microglia exhibit a range of phenotypes from quiescent "resting" (ramified morphology with extended processes) to "activated" (amoeboid morphology with retracted processes). Consequently, morphology provides a valuable indicator for microglial activation and broad functionality (Karperien et al., 2013). While aberrant microglia contribute to various neuropathologies, undernourished microglia remain largely unstudied.
Using two-photon microscopy, we assessed microglial morphology and motility within the hippocampus (CA1 region) of CX3CR1 +/ EGFP mice on a C57BL/6J background (Jung et al., 2000). To assess microglial morphology, we utilized 3DMorph (York, LeDue, et al., 2018), which provided semi-automatic, multidimensional measurements across four independent experiments. We observed Altered volume/territorial volume, as observed in malnourished microglia, affects the area of brain continually surveyed by microglia.
These morphological alterations were of sufficient interest to further characterize microglia, specifically microglial motility and transcriptional profile.
CON, MAL, and MAL-BG mice exhibit comparable motility as measured by process additions and retractions across time (Figure 2d and Figure S3b). We then assessed microglial surveillance in the context of acute hippocampal insult. Damaged and apoptotic cells trigger F I G U R E 2 Gut microbes modulate microglial morphology, but not motility, during malnutrition. (a) Microglia cells counts within the CA1 hippocampal region of CX3CR1 +/EGFP , data pooled from three experiments with data normalized to the CON group of each experiment, n = 9 CON, 9 MAL, and 9 MAL-BG. (b) CA1 hippocampal images with representative CON, MAL, and MAL-BG microglia (inset). (c) Microglial morphology was quantified from four separate experiments using 3DMorph software and normalized to the CON group of each experiment, n = 13 CON, 13 MAL, and 12 MAL-BG. (d) To assess whether morphological alterations affect motility, we examined microglial process additions and retractions over 10 min. Representative images from motility assays: yellow = static, red = process addition, green = process retraction. We observed no striking differences in CON, MAL, and MAL-BG microglial motility, as quantified by process addition or retraction motility indices, n = 9/group (see also Figure S3b). Results from a/d are a subset of mice from data described in c. Graphs indicate mean and SEM with statistical significance determined by one-way ANOVA with post hoc Tukey's test; n.s., non-significant rapid microglial responses, with microglial processes cordoning off injured tissue (Davalos et al., 2005). Complement-dependent microglial phagocytosis shapes diverse brain functions, from continued modulation of neural plasticity and memory via synaptic pruning, to engulfment of noxious stimuli during neuroimmune responses (Wang et al., 2020;Wu et al., 2015;. To identify the scope of MAL-BG alterations and examine how fecal-oral contamination contributes to microglial alterations and putative phagocytic function, we assessed key gut microbiota-brain pathways, namely inflammation, barrier integrity, and neurometabolism (Bauer et al., 2016(Bauer et al., , 2019.

| Inflammation and BBB integrity unaltered in MAL-BG model
Co-occurring malnutrition and fecal-oral contamination often present with systemic comorbidities including immune dysregulation and lowgrade inflammation, processes linked to gut dysbiosis (Brown et al., 2015;Crane et al., 2015;Di Giovanni et al., 2016). To assess whether MAL and MAL-BG microglia respond to neuroinflammation, we examined CNS and peripheral (sera) inflammation. MAL and MAL-BG mice exhibited low levels of TNF-α or IL-6 proinflammatory cytokines, comparable to CON counterparts ( Figure S5a,b).
To specifically address microglial-mediated inflammatory responses, we measured expression of key immune receptors by flow cytometry (CD11b high /F480 high /CD45 low population; Figure S5c). CD86, MHC II and toll-like receptor (TLR) 4 contribute to immunoregulation and pathogen responses and microglial upregulation of these receptors has been established during neuroinflammatory conditions (Schetters et al., 2018;Wang et al., 2019). The frequency and geometric mean fluorescence intensity (gMFI) of CD86, MHC II, and TLR 4 were comparable across dietary conditions ( Figure S5d). Collectively, these findings suggest that the morphological and transcriptional profile observed in MAL-BG microglia is distinct from classical features of inflammatory microglial activation.
We previously reported that fecal-oral contamination influences small intestinal permeability in malnourished mice (Brown et al., 2015), (see also Figure S5e). In addition to regulating the enteric barrier, gut microbes have been linked to the development and maintenance of the CNS analog-the BBB (Braniste et al., 2014). BBB permeability was measured by IgG immunostaining and tetramethylrhodamine biocytin (biocytin-TMR) permeability across the neural vasculature. All groups exhibited low levels of interstitial IgG, indicative of BBB integrity (Readnower et al., 2010) ( Figure S5f). As further validation, we measured biocytin intensity within the brain following biocytin-TMR tail vein injection. CNS endothelial cells lack vitamin transporter Slc5a6 required for expected biocytin transport, though significant BBB deficits enable biocytin-TMR CNS distribution (Knowland et al., 2014). We observed no difference in biocytin-TMR intensity throughout cortical tissue in CON, MAL, and MAL-BG mice ( Figure S5g-i). We note that BBB integrity was measured at our standard endpoint (14 day following bacterial gavage). These findings do not exclude the possibility of transitory BBB deficits at earlier timepoints. Moreover, these methods may not capture more subtle BBB deficits. Markers of transient BBB disruption and/or pathogen-induced proinflammatory cytokine elevation, notably peripheral immune cell infiltration, were not observed in CON, MAL, and MAL-BG brains (see macrophage population, Figure S4a) (Schetters et al., 2018). Collectively, these results support broad BBB maintenance in the malnourished models.
We then assessed neurometabolism, specifically targeting the hippocampus, a critical region for cognitive function and spatial memory and the site of microglial morphology analyses (Mills et al., 2014).
This method identified and relatively quantified >6300 unique metabolite features with 25 differentially abundant hits (one-way ANOVA Fischer's LSD, Padj < 0.05), far fewer compared to the previously reported small intestine metabolome (Brown et al., 2015), likely highlighting CNS resilience against malnutrition. Indeed, the BBB and extensive energy requirements contribute to a distinct, and energetically resilient, metabolomic brain profile (Camandola & Mattson, 2017;Qi et al., 2020).
These findings suggest that MAL-BG microglia both respond and contribute to CNS oxidative pathways.
Aberrant oxidation has long been considered both a driver and consequence of early-life malnutrition (Khaled, 1994;Manary et al., 2000;Preidis et al., 2014 (Barrera et al., 2016;Kwiecien et al., 2014;Pamplona et al., 2005). To assess intestinal oxidative stress, we measured ROS in ex vivo epithelial cells harvested from the CON, MAL, or MAL-BG small intestine. Following E. coli/Bacteroidales exposure, the average ROS levels roughly doubled within the MAL-BG gut ( Figure 4f).
As fecal-oral contamination promoted intestinal dysbiosis, we also assessed the malnourished gut microbiota. PCA of unweighted UniFrac distances generated from 16S rRNA sequencing revealed marked shifts in the MAL and MAL-BG fecal microbiota, with samples clustering by bacterial exposure, then diet (Figure 4g and Figure S7a).
Notably, the MAL-BG microbiota exhibited a significant increase in the relative abundance of select bacterial gavage members, verifying robust microbial exposure ( Figure S7b). To further explore how fecal commensals modulate the malnourished microbiome, we assessed predictive functionality of the MAL and MAL-BG microbiota with PIC-RUSt. The majority of top differentially abundant pathways, following FDR correction, included amino acid metabolism and nucleic acid biosynthesis/scavenging pathways ( Figure S7c and Supplemental File 3b pathways responsive to diet and oxidative strain (Huang et al., 2017;Ishii et al., 2018;Mar et al., 2016;Ni et al., 2019).
Long-term consequences of malnutrition include brain and behavioral deficits, linked to impaired gut-brain interactions (Black et al., 2013;Guerrant et al., 2013;Tofail et al., 2018). MAL-BG results suggest that fecal-oral contamination exacerbates impaired PUFA metabolism and lipoxidative stress during malnutrition. Indeed, acute undernutrition promotes systemic fatty acid oxidation within pediatric, malnourished cohorts at risk for intestinal infection (Bartz et al., 2014;Semba et al., 2017). We hypothesize that abnormal PUFA processing and subsequent lipoxidative stress likely contribute to aberrant microglial activity, ultimately affecting behavior and cognition.

| Dietary intervention shapes MAL-BG brain and PUFA profiles
To We chose to assess the impact of PAO fortification in the MAL-BG model in C57BL/6 mice, as these mice exhibit significant cognitive deficits, a tragic consequence of chronic undernutrition (Black et al., 2013). CON mice provided a healthy control. For concise naming, malnourished mice fed a PAO diet are subsequently labeled MBG-PAO (MAL-BG + PAO diet). Here, we report analyses conducted on CON, CON-PAO, MAL-BG, and MBG-PAO models, model set-up reported in Figure S8a. As appropriate learning involves microglia phagocytic processes (Wang et al., 2020;Wu et al., 2015), we next assessed microglia morphology. We observed no significant differences in healthy and malnourished C57BL/6 microglial cell volume, territory volume, or endpoints ( Figure  , an ω3 PUFA fish oil. To further corroborate PAO fortification, we quantified ω6/ω3 PUFA ratios within another fatty organ highly sensitive to dietary shifts-the liver Naudí et al., 2013). As expected. PAO-fed mice displayed significant reduction in liver ω6/ω3 PUFA ratios compared to healthy and malnourished counterparts ( Figure S8i). Moreover, brain ω3/liver ω3 ratios

| DISCUSSION
Despite intervention efforts, global malnutrition is expected to increase in response to interdependent disruptions from COVID-19 and climate change (Littlejohn & Finlay, 2021;Osendarp et al., 2021;Phalkey et al., 2015). Many undernourished communities will experience lack of appropriate sanitation and chronic fecal-oral contamination (Black et al., 2013;Humphrey, 2009). The pathways and pathologies informing malnourished gut-brain interactions remain greatly unexplored. Here, we report a model of malnutrition that reflects prevalent dietary (i.e., fat/protein deficiency) and environmental (fecal-oral contamination) conditions (Black et al., 2013;Guerrant et al., 2013;Littlejohn & Finlay, 2021). In addition to growth stunting and gut dysbiosis, MAL-BG mice displayed cognitive deficits, notably impaired learning plasticity in the MWMT. These changes were accompanied by microglial and metabolomic shifts, distinct from both healthy (CON) and malnourished-only (MAL) mice. Notably, the MAL-BG microglial transcriptome contained increased expression of complement genes linked to phagocytic processes (e.g., C1qa) and hippocampal MAL-BG microglia presented with increased phagocytic features.
These findings support a surge of ongoing research highlighting gut-glia interactions.
Indeed, we specifically explored microglia due to (1) emerging work demonstrating the role of microglia on learning plasticity and (2) studies linking gut-brain interactions with microglia function.
In our studies, MAL-BG mice persistently honed to the prior MWM platform, displaying impaired memory extinction and poor learning plasticity (Figure 1d- reported that commensal gut microbes influence microglia maturation and function utilizing germ-free models. Germ-free mice exhibited an immature microglial phenotype characterized by increased volume, process length, and process complexity (branching). Researchers identified bacterial-derived short-chain fatty acids (SCFAs) as a critical microbial regulator of microglial maturation and immune function (Erny et al., 2015).  (Brown et al., 2015). As diet-dependent metabolic shifts also shape microglial function (Madore et al., 2020;Valdearcos et al., 2017), we profiled the hippocampal metabolome. Previous metabolomic profiling of the malnourished small intestine revealed marked alteration of amino acid and lipid metabolism (Brown et al., 2015), broadly matching reported shifts in the serum metabolome of children treated for malnutrition (Di Giovanni et al., 2016). Similar to the intestinal metabolomic profile (Brown et al., 2015), the MAL and MAL-BG hippocampal metabolome exhibits perturbed lipid metabolism, particularly PUFA pathways.
As noted, PUFAs serve as essential phospholipid components, participating in lipid signaling, inflammatory regulation, neurodevelopment, and microglial-dependent synaptic plasticity within the brain (Bazinet & Layé, 2014;Madore et al., 2020). While malnutrition shaped ω6 and ω3 PUFA metabolism, overall PUFA levels were comparable across CON, MAL, and MAL-BG mice. These findings highlight persistent PUFA maintenance within the brain, a metabolic resilience not observed in the liver, which displayed consistent, dietary-induced reduction of PUFA members . Both MAL-BG mice and MAL mice displayed similar PUFA profiles, suggesting that PUFA metabolism was largely driven by diet rather than microbial exposure.
The MAL-BG brain, however, exhibited significant elevation of oxidative stress markers, including MDAL (PUFA-specific lipoxidation biomarker), highlighting a putative role for aberrant PUFA metabolism and lipoxidation in malnutrition neuropathology.
To assess reversibility of MAL-BG features, we developed PAO Unlike initial oxidative profiling, mice in the PAO trial underwent MWM testing. The MWMT is a multi-day study involving repeated stress exposures (e.g., water avoidance, body temperature fluctuations). Behavioral test stressors have a profound impact on microglial morphology and activation (Hinwood et al., 2012;Walkera et al., 2013), as well as systemic oxidative responses (Seo et al., 2012).
PAO microglial analyses were conducted on C57BL/6 mice, 4 days following MWMT. Whether MWMT-induced stress influenced baseline controls remains unknown. Comparable oxidative profiling may also reflect modest PAO antioxidant capacity. Importantly, the PAO diets replaced ω6-rich oils, but were not enriched with ω3 PUFAs, in order to maintain caloric equivalence across experimental diets. Further ω3 PUFAs-enrichment and/or increasing antioxidant content and/or chemically blocking lipoxidative pathways (e.g., ω6 oxidative blockade with non-steroidal anti-inflammatory drugs) may reveal altered oxidative patterns and/or microglial phagocytic features not present with PAO fortification (Tapiero et al., 2002). Finally, mice fed PAO diets generally ate less and grew smaller than respective dietary counterparts ( Figure S8a-c). While PAO diets were not designed to target growth deficits, reduced dietary intake and subsequent anthropometric shortcomings likely contributed to modest intervention outcomes.
Malnutrition and fecal-oral contamination frequently coexist (Black et al., 2013;Guerrant et al., 2013), and research assessing gutbrain interactions within this framework are critical to improving understanding of cognitive consequences of early-life malnutrition.
This work supports research linking enteropathogenic burden and microbial dysbiosis with impaired cognitive development (Black et al., 2013;Investigators, 2018). While neurocognitive consequences of childhood malnutrition are unquestionably shaped by societal, economical, and political factors (Smith & Haddad, 2015;Webb et al., 2018), our findings demonstrate that poor diet and specific gut microbes may trigger neuropathologies independent of these "external" influences.
The altered gut-brain axis has emerged as a critical regulator of CNS function and, by extension, a potent therapeutic target (Bauer et al., 2019). In our malnourished model, dietary PAO intervention was sufficient to reverse impaired learning plasticity in malnourished mice exposed to repeated fecal-oral contamination. These findings may support recent efforts to enrich ω6-based therapeutic foods with ω3 PUFAs in the context of severe acute malnutrition . Largescale WASH (water, sanitation, and hygiene) and SHINE (Sanitation, Hygiene, Infant Nutrition Efficacy) interventions have yielded modest reductions of fecal-oral contaminants in malnourished communities without linear growth (stunting) benefits (Pickering et al., 2019). Combining diet-and microbial-targeted interventions, however, modestly improved early cognitive measures in a Bangladeshi WASH trial (Tofail et al., 2018), while diets designed to benefit microbiome development increased plasma biomarkers of neurodevelopment in children with persistent moderate acute malnutrition (Gehrig et al., 2019). Whether interventions targeting fecal-oral microbial exposures will robustly improve learning deficits associated with childhood malnutrition remains to be determined (Aboud & Yousafzai, 2018).
Despite concerted intervention efforts, global malnutrition persists and remains a silent specter during these unprecedented times (Osendarp et al., 2021). The MAL-BG model reveals that specific gut bacteria and malnutrition contribute to cognitive deficits, likely involving microglial and PUFA metabolic shifts. We anticipate that these findings will provide valued insight into dynamic gut microbiota-brain interactions and inform intervention strategies to mitigate lasting consequences of childhood malnutrition.

ACKNOWLEDGMENTS
We are grateful to Finlay Lab members for their guidance and feedback throughout project and manuscript development.

CONFLICT OF INTEREST
The authors declare no competing financial interests.