Fatty acid metabolism changes in association with neurobehavioral deficits in animal models of fetal alcohol spectrum disorders

Fetal alcohol spectrum disorders (FASD) show various behavioral problems due to prenatal alcohol exposure (PAE). Our previous study found significant changes in gene expressions linked to fatty acid metabolism in the brain of the PAE mouse model. Given the importance of fatty acids in normal brain functions and the contributions to neurodegenerative diseases, we hypothesized that the fatty acids changed by PAE contribute to neurobehavioral deficits in FASD. This study found an increase of palmitic acid and arachidonic acid in phospholipid compositions in the cerebral cortex of PAE at postnatal day 30. The increase of palmitic acid was consistent with the increase of the producing enzyme, fatty acid synthase (Fasn). The decrease of 26:6 fatty acid was also found in phospholipid. It is consistent with the increase of the Elongation of very long chain fatty acids protein 4 (ELOVL4) which uses 26:6 as a substrate for making very long chain fatty acids. However, there was no increase in the elongated products. Rather, we found an accumulation of the lipid droplets (LDs) in the PAE brain, suggesting changes in fatty acid metabolism that lead to the accumulation of excessive fatty acids. Although metabolic measurements, including plasma triglyceride level, were not affected by PAE, the abundance of fatty acid-related gut microbiota was altered. Interestingly, multi-omics association analysis revealed a potential contribution of the altered gut microbiota, primarily Ruminococcaceae that produces short chain fatty acid, to LD formation in the PAE brain and the behavioral problems, suggesting that the gut microbiome could serve as a tool to facilitate uncovering the brain pathophysiology of FASD and a potential target to mitigate neurobehavioral problems.

shown to associate negatively with cognition in adolescents (Meo et al., 2019) and positively with the 54 risk of developing dementia such as Alzheimer's disease (AD) in adults (Gustafson et al., 2003). organelles that store excess fatty acids to protect cells from lipid toxicity (Cohen, 2018), and the 60 biogenesis and degradation of LDs are tightly coupled with cellular metabolism to maintain homeostatic 61 lipid levels (Rambold et al., 2015). 62 Alcohol interacts directly with fatty acid to produce fatty acid ethyl ester, thereby identified as a 63 biomarker of maternal alcohol drinking (Himes et al., 2015). Furthermore, in pregnant mothers who drink 64 alcohol, fatty acid composition in their plasma was different between mothers with offspring showing 65 abnormal development and those showing normal development (Sowell et al., 2020). In animals exposed 66 to alcohol throughout gestation, docosahexaenoic acid (DHA), an n-3 fatty acid, was reduced in 67 phospholipid collected from the postnatal hippocampus (Kim, 2008;Wen & Kim, 2004). In another study, 68 administration of DHA between P11 and P20 improved social behavior deficits of PAE rats (Wellmann et 69 in control animals ( Figure 1C). In addition, the intracellular distribution pattern of ELOVL4 in the 118 neurons was not altered by PAE ( Figure 1B Representative images for immunolabeling of ELOVL4 (red) and DAPI (blue). Scale bar = 10µm. (C) 127 Representative images of coronal section from control cortex demonstrate that ELOVL4 expression (red) 128 is detected in NeuN-positive neurons (green). Scale bar = 25µm. Graphs represent mean ± SEM. Each dot 129 represents an individual animal. 130         The fatty acid composition of phosphatidylethanolamine in the cell membrane is altered in PAE 175 motor cortex at the juvenile stage 176 We then examined the fatty acid compositions in the cell membrane in the motor cortex at P30 by 177 analyzing fatty acid in phospholipid using liquid chromatography tandem mass spectrometry. to alter energy metabolism and is associated with non-alcoholic fatty liver diseases and obesity (van der 185 Veen et al., 2017). Therefore, we examined the PC/PE ratio. Similar to another study (Choi et al., 2018), 186 the PC/PE ratio indicated a higher amount of PE than PC in cortices of both control and PAE groups. 187 However, their ratio was not altered by PAE (Figure 2 -figure supplement 1C). 188 In the comparisons of each fatty acid molecular species, a significant increase was found only in      The Multi reaction monitoring (MRM) was used to detect very long chain fatty acids that 231 ELOVL4 specifically produces in both PC and PE. ELOVL4 is involved in elongating both SFA and 232 PUFA by adding 2 additional carbons to synthesize very long chain fatty acids that are longer than 28 233 carbons (C28); thus, any of the fatty acids that are in between C26 and C36 are thought to be substrates

Accumulation of lipid droplets (LDs) in PAE brain 246
The results described above suggested that ELOVL4-produced very long chain fatty acids are 247 metabolized in PAE motor cortex, or potentially sequestered away to prevent lipid toxicity. LDs store 248 To detect LDs, we used oil red o (ORO), which stains neutral triglycerides and lipids in the brain. 255 We first confirmed ORO staining with a 12-month-old mouse as LDs accumulate in aging brains 256  We then examined LD accumulation in the motor cortex and several brain regions of PAE and 261 control mice at P30, as depicted in Figure 3A. The number of LD accumulating cells was significantly 262 increased in the motor cortex of PAE mice compared with that of control mice ( Figure 3B, C). Notably, a 263 significant increase of LD included cells was also observed in other brain regions such as the striatum, 264 CA3, and dentate gyrus of the hippocampus ( Figure 3B). Although statistically not significant, the  Only at P20 (and P30 in Figure 3B), but not at P15 or P25, the differential accumulations of LDs were 279 interplay of the gut microbiome and neuronal signaling, called gut-brain axis, is mediated by bidirectional 300 molecular mediators, including bioactive lipids that can modulate the gut-brain axis (Baptista et al., 2020). 301 Therefore, we examined an association between the changes in neurobehavioral phenotype and gut 302

microbiota. 303
As shown in the experimental timeline in Figure 4A, following collections of fecal pellets for 16S 304 ribosomal RNA (rRNA) sequencing, animals were placed on an accelerated rotarod to assess their motor 305 learning by conducting 3 trials per day for two consecutive days. The next day, animals were placed in an 306 elevated plus maze (EPM) to assess anxiety. The behavioral test results showed that PAE mice have both 307 motor learning deficit and anxiety phenotype that were demonstrated by a significant reduction in the 308  The 16s rRNA sequencing data was analyzed using Mothur (Schloss et al., 2009). After 318 alignment and mapping with the SILVA v138 reference database, the operational taxonomic unit (OTUs) 319 were clustered at 97% identity threshold to sort reads into each gut bacteria. OTUs with less than 4 read 320 counts were removed from each sample, and then bacteria with its OTU counts showing less than 10 % 321 prevalence in entire samples were removed from the analysis. After those filtrations, the final numbers of 322 the total read counts were similar between samples, ranging from 81765 to 82281 (  compared to that of control mice (p=0.00030926) as well as differential microbial communities (beta 340 diversity). Alpha diversity was compared using Chao1 at the genus level. Graph represents a boxplot with 341 25th, median, and 75th percentiles. Whiskers represent 9 and 91 percentiles, and outside of whiskers 342 represent outliers. Each dot represents an individual animal, and the black diamond represents the mean. 343 Beta diversity analysis at the genus level shows that PAE and control mice have differential microbial          data types while discriminating between multiple phenotypic groups. The neurobehavioral phenotypes, 458 brain LD, and microbiome results obtained from an animal in either PAE or control group were used for 459 the analysis. As shown in Figure 5, the motor learning index showed a strong negative correlation with 460 the total brain LD accumulations. Among the brain regions, the piriform cortex was one of the brain 461 regions we saw a strong correlation with anxiety measures. The piriform cortex is mainly known for the 462 odor processing region, but it also receives input from the basolateral amygdala, a brain region important 463 for anxiety, to form a cortical circuit to shape responses to the threatening stimuli (East et al., 2021). 464 There were stronger correlations with LD accumulation and neurobehavioral measures in the 465 microbiota that are known to be associated with fatty acid synthesis and metabolisms, such as 466  (Figure 2, Figure 2 -figure supplement 2). We also applied MRM to quantify 488 the very long chain fatty acid species in PAE and FASD research for the first time. Despite the increased 489 expression of ELOVL4 in the cortex ( Figure 1A), an enzyme that elongates the C26, there were no 490 changes in C28 and the longer chain fatty acids ( Figure 2B). However, the reduction of C26:6 in PE 491 ( Figure 2B) suggested a possibility that, despite increased usage of C26:6 by ELOVL4, the produced very 492 long chain fatty acids are broken down immediately after the elongation; therefore, no significant increase 493 was observed in phospholipid. 494 We also found an increase in palmitic and arachidonic acids in PE in PAE mice (Figure 2A). Of 495 note, previous studies showed an increase of arachidonic acids in maternal plasma from mothers who 496 consumed alcohol during pregnancy and had children with lower cognitive test scores compared to those 497 mothers who consumed alcohol but had children with the normal scores (Sowell et al., 2020). In the same 498 study, palmitic acids in maternal plasma were also positively correlated with alcohol intake. 499 No changes in the amount of the fatty acids that are longer than C26 and simultaneous reduction 500 of C26 in the plasma membrane of the PAE brain cells indicated a possibility that excessively elongated 501 very long chain fatty acids are being metabolized and sequestered by forming LDs to prevent lipidopathy. 502 In fact, we observed the increase of LDs in various brain regions in PAE mice (Figure 3). Another 503 possible explanation as to why there was no increase of C26 observed despite the increase of ELOVL4 in 504 the PAE cortex might be due to the specific subcellular localization of these very long chain fatty acids: Mohammad et al., 2020). There were slight differences in the levels of accumulation between brain 510 regions; however, all of the brain regions showed an increase of LDs in PAE, suggesting brain-wide 511 issues. The variability in LD accumulation levels could be due to the difference in lipid composition 512 among different brain regions, where the prefrontal cortex and motor cortex show similar lipid 513 composition in the membrane but are different from other brain regions such as the hippocampus, 514 striatum, and cerebellum (Fitzner et al., 2020). In addition, behavioral measurements also showed 515 correlations with the LD accumulations in the brain regions that are known to be involved in motor 516 learning and anxiety ( Figure 5). Interestingly, the LDs appeared as transient pathological features that are 517 likely to be affected by diet during postnatal development. In our observation, the accumulation of LDs 518 was dynamic during the weaning time (Figure 3 -figure supplement 2). 519 In the functional annotation to changes in the microbial profiles in PAE animals, we found an 520 increase in the fatty acid synthesis pathway but a reduction in fatty acid degradation, suggesting a 521 possible systemic increase in amounts of fatty acids in PAE mice ( Figure 4I). This prediction was 522 consistent with the lipid-related brain pathology and changes in fatty acid contents in membrane 523 phospholipids in the brain (Figure 2). Further functional analysis of microbiome revealed that the 524 metabolic pathway of linoleic acid, which is a precursor fatty acid for arachidonic acid (Whelan & 525 Fritsche, 2013), was also increased in PAE mice ( Figure 4I). This change might be associated with the 526 increase of the arachidonic acid in PAE brains (Figure 2A). The primary and secondary bile acid 527 synthesis were also predicted to be changed in PAE's microbiome profiles ( Figure 4I). This finding was 528 interesting because the changes in bile acid biogenesis were observed from the biosignature of 529 microbiome-derived metabolites in both alcohol-exposed dams and their fetuses (Virdee et al., 2021). 530 Collectively, these biosignatures predicted from changes in gut microbial compositions suggest that the 531 systemic changes in fatty acid biogenesis and metabolism and that those may be linked to the changes in 532 PAE brains. 533 Association analysis between neurobehavior, lipid-associated brain pathology, and gut 534 microbiome revealed the strong correlations of microbiota that are involved in fatty-acid biogenesis and 535 metabolism with brain pathology and the behavior ( Figure 5). Among those microorganisms, Unclassified 536 Ruminococcaceae showed the strongest associations with both brain pathology and behavior. 537 Ruminococcaceae are found in both human and mouse gut microbiome, but the abundance is higher in 538 humans than in mice (Nguyen et al., 2015). Interestingly, the abundance of Ruminococcaceae in the gut is 539 5) also showed an intriguing possibility that the gut microbiota might directly affect the lipid pathogenesis 549 without changing metabolism in the entire body at a significant level. An exciting hypothetical 550 mechanism is that metabolites of gut microbiota affect vagus nerve signaling (Fülling et al., 2019) and 551 that vagus nerve stimulation directly affects lipid composition in various brain regions, including the 552 striatum and motor cortex (Surowka et al., 2015), which could be mediated by the gut-brain axis (Baptista 553 et al., 2020). with neurobehavioral problems or lipid-associated brain pathology. This may be due to milder phenotypes 559 after acute PAE compared to a chronic PAE model. However, importantly, our model exhibits various 560 neurobehavioral problems, including motor learning deficits and anxiety, and disruption in fatty acid 561 related metabolism in brain and gut. Therefore, our results suggest that the effects of gut microbiota may 562 profoundly and directly affect the brain pathology and function in PAE. The test was performed as previously described (Mohammad et al., 2020). Briefly, the mice were placed 578 on a rotating bar, and the length of time that they could retain their balance during acceleration of rotation 579 to a max speed of 80 rpm in 5 min was recorded. The testing phase consisted of 2 consecutive days with 580 three trials per day. Each trial was at least 15 minutes apart and was terminated when the mouse fell off, 581 made one complete rotation without walking on the rotating rod, or reached maximum speed after the 5-582 min session. The motor learning index was calculated by averaging the difference in terminal speed of the 583 two consecutive trials. 584 585

Elevated plus maze test (EPM) 586
The maze is a grey plus-shaped apparatus with two open arms and two closed arms linked by a central 587 platform. Mice were individually put in the center of the maze facing an open arm and allowed to explore 588 the maze for 300 seconds. A video was recorded during the experiment and analyzed with MouBeat 589 ImageJ Plugin as per the user guide as previously done (Hwang & Hashimoto-Torii, 2022). 590 591 Immunohistochemistry 592 Mice were deeply anesthetized with isoflurane (Henry Schein, Melville, NY) and perfused transcardially 593 with 10 ml of ice-cold PBS followed by 10 ml of chilled 4% paraformaldehyde (PFA). The brains were 594 removed and immerse-fixed in 4% PFA at 4 °C overnight. Then incubated in 10% and 30% sucrose in 595 PBS for 24 hr sequentially at 4 °C and embedded in the OCT compound (cat# 4583; VWR, Randor, PA). 596 Coronal sections were cut at 20 or 50µm on a cryostat (CM3050S; Leica, Buffalo Grove, IL). 597 Free-floating staining was performed with 50µm thick brain sections. Briefly, antigen retrieval was 598 For NeuN staining, sections were incubated for 2hr with HRP-conjugated anti-mouse IgG diluted at 1:300 609 and followed by 1 hr incubation of TSA plus Cyanine-2 (1:300; cat#NEL745001KT, Akoya Biosciences). 610 Sections were counterstained with DAPI and mounted on slides with CC/Mount mounting medium (cat# 611 C9368; Sigma, St. Louis, MO). Images were acquired using a confocal microscope (FV1000; Olympus 612 Center Valley, PA). All images were analyzed with ImageJ using the Cell Counter tool. 613

Oil Red O (ORO) staining 614
20µm thick brain sections were prepared as described above. For staining, ORO stock solution 615 was prepared by dissolving 0.05g of ORO powder (cat#O0625; Sigma) in 10mL of isopropanol. Then 60% 616 ORO working solution was prepared freshly with distilled water and filtered before use. Brain slices were 617 incubated in 60% ORO solution for 10 min, washed thoroughly with distilled water, and incubated for 15 618 min with hematoxylin for counterstaining. Then the slices were mounted on slides with CC/Mount 619 mounting medium, and images were acquired with an Olympus VS120 microscope. Brightness and 620 contrast were adjusted with CellSens, and ORO-positive cells were manually counted with ImageJ Cell 621 Counter tool. 622 623

Lipidomics of Phospholipid Fatty Acids 624
Motor cortical regions were dissected from both hemispheres of P30 mice and snap-frozen in liquid 625 nitrogen. Samples were stored at -80°C and shipped to Lipid Analysis Core at Emory University for 626 phospholipid fatty acid quantification. Lipids were extracted from the cortical tissues using Bligh and 627 Dyer method (Bligh & Dyer, 1959), and extracted lipids were directly loaded onto the mass spectrometer 628 (https://www.microbiomeanalyst.ca/) for downstream analysis after removing low counts less than 4 and 656 prevalence in samples less than 10%. Filtered data were scaled by total sum scaling, and alpha and beta 657 diversity were determined by Chao1 and Bray-Curtis Index, respectively. Differential abundance analysis 658 was carried out using EdgeR. To perform functional profiles analysis, representative OTU FASTA 659 sequences and OTU count tables were retrieved from Mothur and analyzed with "Tax4fun2" package in results of behavioral tests (accelerated rotarod and EPM tests). We found 8 control and 12 PAE mice that 669 meet this criterion. Centered log ratio (Clr) transformed microbiome and Z-score transformed the number 670 of LDs accumulated in the brain, and behavioral measurements were used as the input data. The data 671 integration was carried out following the manual. Briefly, a matrix model was designed to connect all 672 datasets with a link strength at 1 between two components were used to generate the final DIABLO 673 model. The circos plot shows the correlation strength and directionality between variables of different 674 types of our datasets with a correlation coefficient cut off > |0.6|. 675 676

Statistical analysis 677
All histological data were acquired from the defined subregions of the brain. All groups consisted of mice for EPM analysis (Figure 4). All the immunohistological analysis was done group-blinded by the 685 investigator. Plasma triglycerides, blood glucose, and body weight measurements were performed 686 unblinded (Figure 4 -figure supplement 1). All of the statistical analysis was carried out with GraphPad 687 Prism 7.01. We performed the D'Agostino-Pearson to test the normality of data. For the data that passed 688 the normality test, Student's t-tests or one-way or two-way ANOVA was used. Post hoc Tukey's or 689 Bonferroni's test was done as described in the figure legends. Simple main effects were reported when 690 there was a statistically significant interaction between independent variables by two-way ANOVA. 691 Pearson's or Spearman's correlation coefficient calculation was done for normally or non-normally 692 distributed data, respectively. P values of less than 0.05 were considered statistically significant. 693

Data availability 694
All original 16s rRNA sequencing raw data haven been deposited to the National Library of Medicine 695 Bioproject under accession ID PRJNA842719. Data generated or analyzed in this study are included in 696 the manuscript and Source Data 1 file. 697