A lipidome landscape of aging in mice

Understanding the molecular mechanisms associated with aging is important to improve the longevity of healthy individuals. Herein, we performed untargeted lipidomics to elucidate the relationship between aging and lipid metabolism in 13 mouse tissues at various life stages (2, 12, 19, and 24 months), in which sex and microbiome dependencies (speci�c pathogen free/germ free) were investigated. By analyzing 2,817 unique molecular pro�le data of 121 lipid subclasses, we characterized the common and tissue-specic lipidome changes associated with aging. For example, bis(monoacylglycerol)phosphate containing polyunsaturated fatty acids was enriched in various organs during aging, while other phospholipids containing saturated and monounsaturated fatty acids were decreased. In addition, we discovered an age-dependent increase of sulfonolipid (SL), which is biosynthesized in Alistipes and Odoribacter genera. SL molecules were absent in germ-free mice. Furthermore, the molecules appeared to be translocated from the intestinal lumen to various tissues such as the liver, kidney, and spleen. In the kidney, the associations of glycolipids such as galactosyl ceramides (GalCer), galabinosyl ceramides (Gal2Cer), trihexosyl ceramides (Hex3Cer), and mono and digalactosyl diacylglycerols were found in male mice, in which two lipid classes Gal2Cer and Hex3Cer were signi�cantly enriched in aged mice. Integrated analysis of the kidney transcriptome revealed uridine diphosphate galactosyltransferase UGT8a, alkylglycerone phosphate synthase, and fatty acyl-CoA reductase 1 as potential enzymes responsible for male-specic glycolipid biosynthesis in vivo, which would be relevant to sex dependency in kidney diseases. Our lipidome results will become an important resource, led to the understanding of the molecular mechanisms underlying cellular senescence and age-related diseases.

Version of Record: A version of this preprint was published at Nature Aging on April 12th, 2024.See the published version at https://doi.org/10.1038/s43587-024-00610-6.

Introduction
Lipids are diverse and multifunctional in living organisms and serve as signaling molecules, energy storage molecules, and cellular membranes 1 .These functions involve thousands of lipid species in mammalian cells, and the dysregulation of lipid metabolism is associated with various diseases such as atherosclerosis 2 , cancer 3 , non-alcoholic steatohepatitis 4 , and chronic kidney disease 5 .Undoubtedly, the risk of chronic diseases is related to aging processes, and the understanding of changes in lipid metabolism is likely to illuminate the biological mechanisms associated with aging [6][7][8][9] .
To date, the potential link between aging and lipid metabolism has been studied.For example, an increased activity of sphingomyelinase that convert sphingomyelin to ceramide results in accumulation of ceramides in the elderly patients in a human cohort study 10 .Additionally, low serum ceramide levels are associated with a reduced risk of dementia and Alzheimer's disease in elderly women 11 .The biosynthetic pathway and regulation of neutral lipids are also linked to longevity.De ciency of diacylglyceride acyltransferase 1 improves longevity and leanness in female mice, where age-onset obesity and in ammation in white adipose tissue are protected 12 .Furthermore, the ratio of saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), and polyunsaturated fatty acids (PUFA) is associated with the susceptibility of the cellular membrane to peroxidation.The accumulation of reactive oxygen species during aging leads to lipid peroxidation, which damages cellular membrane components 13 .
Nevertheless, the relationship between fatty acid properties and longevity is tissue-speci c, sex-linked, and species dependent.For example, lipidomic studies investigating human erythrocyte and lymphocyte membranes showed that centenarians have more PUFAs than that of the other aged group 14,15 , while the concentration of phospholipids containing docosahexaenoic acid in longest-living rodents (naked molerats) is lower than that of mice 16 .In addition, recent studies have shown an association between aging and microbiota, in which lipid metabolism is highly affected by the composition of the microbiome and food intake conditions [17][18][19][20][21] .These results strongly indicate the importance of understanding the metabolic changes in diverse lipid molecules in various mammalian tissues under sex-and microbiotadependent conditions.
Untargeted lipidomics has great potential to reveal the diversity of lipidomes, covering approximately 1000 lipid molecules per single specimen 22,23 .Although an untargeted approach results in higher false positive annotations compared to the targeted approach, the false discovery rate (FDR) can now be well controlled using current state-of-the-art computational mass spectrometry techniques 23 .Therefore, we applied untargeted lipidomics using liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) to elucidate the changes in lipidomes in tissues during aging (Fig. 1a).
Thirteen biological samples (kidney, liver, lung, spleen, muscle, bone marrow, plasma, skin from ear, skin from back, mesentery, small intestine, large intestine, and feces) were examined in this study.The lipidome was investigated in males and females of different ages (2, 12, 19, and 24 months) in the presence (speci c pathogen-free; SPF) or absence of microbiota (germ-free; GF).Although there are several studies that have applied a lipidomic technique to speci c tissues [24][25][26][27][28][29] (e.g., a study encompassing the metabolome atlas of brain tissue during aging 30 ), the current study is the rst to comprehensively investigate the lipidome of multiple organs with differences in sex and the microbiome.Furthermore, a multi-omics approach using transcriptome and lipidome data was used to reveal the relationship between gene expression and the lipid pro le of the kidney.
The signi cance of our study is that it offers comprehensive lipidomic changes associated with aging in mice from four different backgrounds: male and SPF (male/SPF), male and GF (male/GF), female and SPF (female/SPF), and female and GF (female/GF).Furthermore, to examine the details of the mouse lipidome, we updated our software program MS-DIAL 4 23 , which is a MS data processing environment in which 17 lipid subclasses containing several gangliosides, sterols, and glycolipids were newly introduced into the lipid annotation platform, covering 136 lipid subclasses that contain microbiota-and tissue-speci c lipid molecules.We used the platform of untargeted lipidomics to reveal that certain common lipid changes are not affected by male/female and SPF/GF differences, e.g., accumulation of PUFA containing complex lipids, such as bis(monoacylglycerol)phosphate (BMP).Moreover, we discovered that the microbiota-derived sulfonolipid (SL) was associated with aging and that the molecule appeared to be translocated to various organs, while most of the microbiota lipids were only detected in the fecal matter.Furthermore, male-speci c accumulation of glycolipids, including galactosyl ceramide (GalCer) and monogalactosyldiacylglycerol (MGDG), was observed in the kidney, and a uridine diphosphate (UDP) galactosyltransferase, UGT8a, was considered responsible for the metabolic regulation.

Mouse experiment
All animal experiments were performed in accordance with the ethical protocol approved by the RIKEN Center for Integrative Medical Sciences (2019-015(2)).Four-week-old male and female C57BL/6N mice were purchased from CLEA (Tokyo, Japan).The GF mice were housed in GF isolators at the animal facility at RIKEN.All the mice were fed the chow of AIN-93M (CLEA Japan, Tokyo, Japan).The kidney, liver, lung, spleen, skeletal muscle, bone marrow, plasma, ear skin, back skin, mesentery, small intestine, large intestine, and feces were harvested.Tissues were frozen immediately after dissection and stored at −80°C until lipid extraction.The details of the time-course experiments are described in Supplementary Table 1.

Lipid extraction
Lipid extraction was performed according to a previously reported method, in which a mixed solvent containing methanol, chloroform, and water (MeOH:CHCl 3 :H 2 O, 1:2:0.2,v/v/v) was utilized.Details of the solvent volumes and internal standards used in this study are described in Supplementary Tables 2 and  3. Brie y, the tissues were homogenized using a multi-bead shocker (YASUI KIKAI, Japan) with a metal cone (YASUI KIKAI, Japan) at 2500 rpm for 15 s × 2, and MeOH was added to the homogenate according to the tissue weight.After the solvent was homogenized again at 2000 rpm for 10 s, the appropriate amount of MeOH (<100 μL) solving 1-5 mg tissue weight was transferred to a 2-mL glass tube.After the solvent was added to 175 μL of MeOH, 25 μL of MeOH containing internal standards was added to the solvent, and the mixture was sonicated for 2 min.For bone marrow cells and plasma samples, 175 μL of MeOH containing 10 6 cells and 20 μL, respectively, were prepared without the homogenization process.
After the solvent was incubated for 2 h on ice, 100 μL of chloroform was added, and the mixture was sonicated for 2 min.After the solvent was incubated for 1 h at room temperature (set to 20°C), 20 μL of Milli-Q water was added, and the mixture was vortexed for 10 s.After 10 min of incubation at room temperature, the solvent was centrifuged at 2000×g for 10 min at 20°C, and the supernatant was transferred to an LC-MS vial (Agilent Technologies).
The MS detection of lipids was performed using a quadrupole/time-of-ight MS (TripleTOF 6600; SCIEX, Framingham, MA, USA).All analyses were performed in high-resolution mode in MS1 (~35,000 full width at half maximum (FWHM)) at high sensitivity mode (~20,000 FWHM) in the MS2.Data-dependent MS/MS acquisition was performed.The parameters were as follows: MS1 and MS2 mass ranges, m/z 70-1250; MS1 accumulation time, 200 ms; MS2 accumulation time, 70 ms; collision energy, +40/-42 eV; collision energy spread, 15 eV.Mass calibration was automatically performed using an APCIpositive/negative calibration solution via a calibration delivery system.All other parameters have been described in a previous report 31 .

MS-DIAL analysis
The in silico reference spectra of 17 lipid subclasses were created according to a previously described method 23 .The MS/MS spectra of gangliosides were experimentally con rmed by analyzing authentic standard compounds.For the other lipid subclasses, MS/MS spectra were obtained from the literature (Supplementary Table 4).A decision tree (rule-based) algorithm considering the existence of diagnostic fragment ions for lipid annotation was implemented.The expected RTs were predicted using our XGBoost-based machine learning methodology, which has been described in a previous report 23 .The summary of fragment ion annotations is also available at our website (http://prime.psc.riken.jp/compms/msdial/lipidnomenclature.html).
The version 4.20 algorithm of MS-DIAL was utilized for the analysis.The data processing parameters for minimum amplitude (for peak picking) and RT tolerance (for lipid annotation and peak alignment) were set to 500, 2.0 min, and 0.1 min, respectively.Other parameters are listed in Supplementary Table 5. RT ltering for lipid annotation was not utilized for HILIC lipidomics data.The annotated results were curated manually.The representative adduct forms used for lipid quanti cation are shown in Supplementary Table 6.The lipid molecules were semi-quanti ed using internal standards according to a previously described method: the value of lipid quanti cation was provided in pmol/mg tissue, pmol/μL plasma, or pmol/10 6 cells.Pairs of lipid subclasses and internal standards are described in Supplementary Table 6.The lipidome results are available in Supplementary Data 1.

Transcriptome analysis
Tissue samples were homogenized in 700 µL of TRIzol reagent (Thermo Fisher Scienti c, USA) per 50 mg of tissue in a glass tube.The supernatant was collected after centrifugation of the cell lysate for 10 min at 12000 × g at 4°C.After adding 100 µL of chloroform to 500 µL of the supernatant, the mixture was incubated for 3 min at room temperature (set to 22°C) and centrifuged for 15 min at 12000 × g at 4°C.Then, 120 μL of 70% ethanol (w/w) was added to 120 µL of the supernatant and suspended.Total RNAs was extracted from the mixture using the RNeasy Plus Mini Kit (Qiagen, Hilden, Germany), according to the manufacturer's protocol.Quality checks were performed using a bioanalyzer (Agilent Technologies, USA) in accordance with the Agilent RNA 6000 Nano Assay protocol.RNA integrity number of the samples was con rmed to be 6.9-9.1 (with an average of 8.6).Subsequently, the library was synthesized using the NEB Next Ultra RNA Library Prep Kit for next-generation sequencing.The pooled libraries were sequenced on an Illumina HiSeq 2500, and 50 bp single-end reads were generated.The reads were aligned using STAR software onto a transcript reference based on UCSC_mm10 (mouse reference).After the genes in which the minimum value of read count is less than 10 or the gene name contains "rik" were excluded, differentially expressed genes were calculated using the R/Bioconductor package, DESeq2 (1.28.1).The transcriptome data is available in Supplementary Data 2.

Statistical analysis
Statistical analysis and data visualization were performed using the R language environment (version 4.1.2).

Results
The optimal amounts of plasma (~20 μL), cells (~10 6 cells), and tissues (~5 mg) were subjected to lipid extraction followed by LC-MS/MS analysis (Fig. 1b).Lipid annotation in MS-DIAL was performed using the MS/MS characterization system of lipids and 2.0 min RT tolerance to the predicted RT database.The annotation results were manually curated.After the peak alignment was performed by the RT tolerance of 0.1 min and m/z tolerance of 0.015 Da, the semi-quanti cation values were calculated by the internal standards (see Materials and Methods).The molecules in plasma (408) and kidney (~ 961) were characterized per biospecimen.In total, 2,817 unique molecules from 121 lipid subclasses were annotated and used for data interpretation.
Principal component analysis (PCA) using auto-scaled values was performed to examine the major variances of the lipidomic datasets (Figure 2).Individuals of each group were mapped at a topology similar to that of the PCA score plots, indicating that our lipidomics output obtained from a long experimental procedure from mouse bleeding to MS data analysis is valuable for further statistical analyses.We observed that the largest variances, i.e., at the rst PC (PC1), in the kidney and feces re ected the sex and microbiome backgrounds, respectively.The results of the fecal lipidome changes indicated that many lipid molecules are biosynthesized and/or degraded by intestinal microbiota.However, the SPF and GF differences were not re ected signi cantly in PCA by the host lipidomes in the small and large intestines.Furthermore, the plasma lipidome of young GF mice was clearly distinguished from the others, in which triglycerides (TG) were enriched in young GF mice according to the loading values (Supplementary Figure 1).A previous study investigating serum, liver, and adipose tissue of young mice (aged 12-14 weeks) reported that the serum TG level was lower in SPF mice than in GF mice, while the TG lipids were increased in the liver and adipose tissue of SPF mice 32 .Our results are consistent with the previous observations (Figure 3).Moreover, sex-speci c lipidome differences were clearly observed in the kidneys at PC1.In addition, age-dependent lipid changes were re ected in the PCA score plots of several tissues, such as the kidney, liver, mesentery, plasma, bone marrow, and skin.We also observed a larger individual difference in aged mice (1.5-2 years old mice), indicating that many biological and environmental factors accumulate during aging.
In contrast to unsupervised mathematical analysis such as PCA, the supervised chemometrics approach is useful for extracting lipidome changes with aging.We applied orthogonal partial least squares regression (OPLS-R) to investigate lipid changes during aging (Figure 3).Importantly, we used the entire lipidome data, which contained four different backgrounds (SPF/male, GF/male, SPF/female, and GF/female) as the variables for OPLS-R.The Q 2 values calculated by 6-fold cross validations indicated that the prediction of age states is possible using the information of the lipid metabolome.Among the tissues examined, the Q 2 values in the kidney, lung, mesentery, muscle, spleen, ear skin, and large intestine exceeded 0.70, indicating that a large proportion of lipids was associated with aging in these organs.Of these, the highest Q 2 value was observed in the kidneys.We investigated the variable importance for prediction (VIP) values, which indicate the important metabolites used for OPLS-R model creation.The top 10 VIPs, whose chromatogram peak shape of metabolites was of good quality (for quanti cation), were used for the interpretation of lipid changes.Decreases in phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylserine (PS), phosphatidylinositol (PI), diacylglycerol (DG), and TG containing SFA and MUFA were observed in the lungs, mesentery, muscle, ear skin, large intestine, and plasma.In contrast, these lipid classes containing PUFAs, particularly 20:4 and 22:6, were increased in the liver, lung, muscle, bone marrow, and small intestine.An increase in the PUFA/MUFA ratio during aging has been reported in human and Caenorhabditis elegans 9 , indicating that our aging-lipidomics study investigated the changes in lipids that have previously been unknown.BMP-containing PUFAs clearly increased with aging in the kidney, liver, muscle, spleen, and small intestine.Because BMP lipids are enriched in the late endosome and lysosome and promote acidic sphingomyelinase activity, the BMP lipid signature may explain the accumulation of ceramides during aging in mammalian organisms 33,34 .
Increases in cholesteryl esters (CE) have also been observed in several tissues, such as the kidney, lung, spleen, and large intestine.Furthermore, decrease in ceramides, epidermal acyl ceramides, SHexCer (sulfatide), gangliosides (NGcGM3, GD3), and cardiolipin (CL) were observed, which may be linked to the functional defect of cell organelles such as mitochondria, endoplasmic reticulum, Golgi body, and endosome systems.Importantly, these common or tissue-speci c lipid changes can be considered as common metabolic signatures associated with aging because the OPLS-R model was constructed using the entire lipidome data that contained all mice groups examined in this study.
We next explored age-dependent lipid changes in uenced by sex and the gut microbiota (Fig. 4a).The quantitative value of each lipid subclass was calculated by summing the quantitative values of molecules annotated by the same lipid subclasses.The fecal lipidome indicated that the pro les of many lipid subclasses were affected by the gut microbiota, and the orange circle indicates that the lipid subclass was enriched in SPF mice compared to GF mice.Greater than four-fold increases in unconjugated bile acid (BA); acyl diacylglyceryl glucuronide (ADGGA); several ceramide subtypes (Cer_ADS, Cer_AP, Cer_BDS, and Cer_EBDS); glycolipids such as MGDG, digalactosyl DG (DGDG), and NGcGM3; fatty acid esters of hydroxy fatty acid (FAHFA), PE and phosphatidylglycerol (PG), ceramide PE and PI (PE_Cer and PI_Cer); SL; and N-acyl glycine, glycylserine, and ornithine (NAGly, NAGlySer, NAOrn) were observed in SPF mice than in GF mice.In contrast, glycine-and taurine-conjugated BAs were substantially elevated in GF mice owing to the absence of intestinal bacteria responsible for BA deconjugation.We also observed sex-speci c lipid pro les in several lipid subclasses, such as TG, lyso-PG, and glycolipids in the fecal lipidome.Interestingly, sex speci city in lipid metabolites was substantially observed in the kidney, where the glycolipids containing dihexosyl Cer (Hex2Cer), trihexosyl Cer (Hex3Cer), MGDG, DGDG, and the alkylacyl types (Ether MGDG and Ether DGDG) were enriched in male mice, and the percentage, i.e., the value of male/(male + female) was >80%.Furthermore, Hex2Cer and Hex3Cer levels in the kidney increased with aging.The major acyl chain properties of these glycosylceramides were 18:1;O2/16:0, 18:2;O2/16:0, 18:1;O2/24:0, and 18:2;O2/24:0, where the sphingobase moiety is considered sphingosine (18:1(4E);1OH,3OH) and sphingadienine (18:2(4E,14Z);1OH,3OH) according to a previous study 35 .Interestingly, MGDG and DGDG molecules, known as plant thylakoid membrane lipids, were detected in the kidney.Although such glycosyldiacylglycerols are ubiquitous as a minor component of the brain and other nervous tissues 36,37 and are enriched in oligodendrocytes, our results indicated that both diacyl-and alkylacyl-type glycosyldiacylglycerols were present in the kidney, and the molecules were enriched in male mice and not affected by the gut microbiome.Furthermore, we found that SLs, known as microbiota-derived lipids, were detected in peripheral tissues, whereas most of other microbiota-derived lipids were not detected (Fig. 4b and 4c).The major component of SL was SL 17:0/17:0;O in peripheral tissues, which increased with aging.Two bacterial genera, Alistipes and Odoribacter, are producers of SL.Several bioactivities of SL have been reported, including antagonistic action on von Willebrand factor receptors, cytotoxicity in cancer cells, inhibition of DNA polymerase, and anti-in ammatory effect 38 .Overall, our aging lipidome atlas clearly demonstrates lipid diversity and its unique pro les during aging.Of these, we believe that it is important to investigate the mechanism of kidney lipidome changes, because the kidney lipidome showed a characteristic pro le depending on sex, the presence of intestinal bacteria, and a strong agedependent variation.
We performed RNA-seq analysis of the kidney tissue of the same mice used for lipidome pro ling.According to the PCA score plot, male and female transcriptome data were separated by PC1 with a contribution ratio of 57.3% (Fig. 5a).This result indicates that kidney gene expression is highly affected by sex, as supported by previous studies 39,40 .Similar tendency was observed in the volcano plot of male and female comparisons where 601 and 1,102 genes were up and downregulated, respectively, with the condition of >0.5 log 2 fold change and <0.05 adjusted p-value (Fig. 5b).Contrastingly, only 18 and one gene were recognized as up and downregulated genes, respectively, between SPF and GF groups, indicating less effect of microbiome on the transcriptome in the kidney.Between aged (24 months) and young (two months) mice, we detected 32 upregulated and 100 downregulated genes.Previous studies have reported that aldehyde dehydrogenase 1 family member A1 (ALDH1A1), a key enzyme of retinoid metabolism, is one of the most prominently upregulated proteins and genes in aged kidney 41 .This result suggests that our kidney transcriptome data are comparable to those in previous literature and will provide useful information for subsequent data interpretation using sex, aging, and microbiotadependent transcriptome pro les.
Gene ontology (GO) analysis using the signi cant genes described above was performed by g:Pro ler 42 , which provided the GO terms of molecular function, biological process, and cellular component, in addition to the pathway enrichments in the KEGG and Reactome databases (Fig. 5c).According to the top ve enriched ontology terms for each source, we observed the enrichment of terms related to stress and immune responses in aged mice owing to the signi cant increase in gene expression, including vascular cell adhesion molecule 1, a chaperone protein resistance to inhibitors of cholinesterase 3, poly(ADPribose) polymerase family member 3, solute carrier family 11 member 1, tumor necrosis factor receptor superfamily member 1 B, phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit delta, and ALDHs (see Source data).Moreover, we observed many terms related to "metabolism" including lipidrelated metabolisms when upregulated genes in male mice were applied to g:Pro ler application.The genes included many of UDP glycosyltransferases (UGTs), cytochrome P450s (CYPs), acyl-coenzyme A (CoA) oxidases, peroxisomal trans-2-enoyl-CoA reductase, phytanoyl-CoA 2-hydroxylase, ALDHs, etc.In contrast, the terms related to the kidney's main functions, including the renin-angiotensin system, drug metabolism, and regulated calcium reabsorption, were enriched in female mice.They include glutathione S-transferases, the kallikrein gene family, and serine proteases for proteins related to blood pressure reduction.The severity of acute and chronic kidney injury is generally higher in males and lower in females 43 , and these differences in gene expression may contribute to the phenotypic differences in the kidney.
Furthermore, we investigated the transcription factors (TFs) and histone modi cations (HMs) that would regulate the signi cant changes in males versus females and in aged versus young mice using ChIP-Atlas 44 (Fig. 5d).Interestingly, signal transducer and activator of transcription 3 (STAT3), vitamin D receptor (VDR), androgen receptor (AR), and hepatocyte nuclear factor 4-alpha (HNF4a) were commonly detected as signi cant TFs between males and females and between aged and young mice.HNF4a is the main TF for drug-metabolizing enzymes in the kidney 45 and regulates peroxisome proliferator-activated receptor (PPAR)-α and γ, which are highly associated with lipid metabolism 46 .Sex differences in renal metabolism are mediated by testosterone, involving AR-dependent signaling pathways in male, but not female, kidneys 47 .In addition, STAT3 activity increases with age in epithelial compartments of the renal cortex 48 .Furthermore, the AR N-terminal domain is activated by the cytokine interleukin 6, which is mediated through the STAT3 signal transduction pathway, and androgen-STAT3 activation might contribute to sex disparity in human simple renal cyst disease 48,49 .In addition, renal vitamin D metabolism is regulated by estradiol and testosterone, and age-dependent dysfunction of vitamin D metabolism, including de ciency of VDR activation, is also related to chronic kidney disease 50,51 .The TF results are further discussed along with the lipid pathway map in the discussion section.Furthermore, the prediction of HMs in the kidney suggested that three major active histone marks, acetylation to the 9 th and 27 th lysine residues on histone H3 (H3K9ac, H3K27ac) and trimethylation to the 4 th lysine residue on histone H3 (H3K4me3), were enriched in male and aged mice, indicating that these HMs may act as regulators of the unique TF pro les of the kidney.However, only three genes (ALDH1a7, CYP2D12, CYP2D13) and two genes (apelin receptor APLNR, NYN domain, and retroviral integrase containing NYNRIN) were commonly upregulated and downregulated in both male and aged mice, respectively (Fig. 5e).The results of transcriptome analysis, including TF predictions, indicate that many aging-and sexassociated genetic changes in the kidney are related to lipid metabolism.
Next, we performed an integrated analysis of the transcriptome and lipidome data of the kidney using weighted correlation network analysis (WGCNA) 52 to investigate the associations between gene expression and lipid metabolites during aging (Fig. 6a).In total, 16 and 24 clusters were established from lipidome and transcriptome data, respectively.As expected, half of the clusters were signi cantly associated with aging or sex differences.In contrast, only one lipid cluster (group 11) consisting of SL, bile acids, and phospholipids containing odd carbon chains was associated with the existence of the microbiome.We focused on lipid clusters 1, 4, 5, 8, and 15 located in the same cluster region of the lipid dendrogram, whose groups were highly associated with sex, aging, and gene clusters.The lipid pro le containing 619 molecules was applied to the lipid ontology enrichment analysis using the LION/web application (Fig. 6b) 53 .The ontology terms were evaluated by the analysis of variance values in four groups that contained M24/male, M24/female, M2/male, and M2/female SPF mice, where M24 and M2 denote 24-month-old (aged) and 2-month-old (young) mice, respectively.As a result, the terms related to glycolipid metabolism including "simple glc series," "dihexosylceramides," and "hexosylceramides" were signi cantly enriched.Furthermore, terms related to alkylacyl phospholipid metabolism, in addition to steryl esters, were characterized as signi cant ontologies.Therefore, we performed a pathway analysis of these lipid metabolisms, along with the pro les of related genes.
The lipid metabolism related to glycosyl ceramides was investigated using lipid and transcriptome pro les (Fig. 7a).Interestingly, the abundance of HexCer, Hex2Cer, Hex3Cer, and SHexCer was substantially higher in male mice than in female mice, and an age-dependent increase in Hex2Cer and Hex3Cer was observed.In fact, sex-speci c differences in Hex2Cer in the kidney have also been highlighted in a reference map of sphingolipids 54 .By exploring the gene expression pro les related to metabolism, we identi ed UGT8a, an enzyme that converts ceramide to GalCer, as the gene expression was signi cantly different between male and female mice.These results indicate that sex-dependent UGT8a expression may affect galactosyl ceramide metabolism by providing GalCer, sulfatide, and Gal2Cer in the kidneys with aging.Furthermore, our untargeted lipidomics characterized MGDG, DGDG, and their ether-linked types (ether MGDG and ether DGDG) in the kidney, whose pro les were also affected by sex differences (Fig. 7b).Although the enzymes responsible for MGDG and DGDG have not been fully identi ed in mammalian organisms, the expression pro les of UGT8a and other signi cantly changed UGT-related genes, such as UGT2b37, UGT2b38, and UGT2b5, offer novel insights for identifying the machinery responsible for glycosyl glycerolipid metabolism in animals (Fig. 7c).In fact, two previous studies reported that the overexpression of rat UGT8a in Chinese hamster ovary (CHO) cells increases MGDG lipid levels, which has been con rmed by thin layer chromatography methods 55,56 .Thus, our nding showing an association between UGT8a and MGDG pro les strongly supports the involvement of UGT8a in the biosynthesis of MGDG and other glycoglycerolipids in vivo.We also found age-dependent decreases in ether-bond glycerolipids, such as ether DG, ether TG, ether MGDG, ether PC, ether PI, and ether PS, and these metabolites were enriched in males.We identi ed two genes as candidates for explaining the lipid pro les: alkylglycerone phosphate synthase (AGPS), the enzyme that converts acyldihydroxyacetonephosphate (DHAP) to alkyl-DHAP, and fatty acyl-CoA reductase 1 (FAR1), the enzyme responsible for the reduction of fatty acids to fatty alcohols.As our kidney transcriptome data also showed signi cant changes in genes related to peroxisome metabolism, the changes in peroxisomerelated lipid metabolism during aging may contribute to our understanding of the molecular mechanisms of various kidney diseases.An age-dependent increase in CE, but not in free cholesterol (ST 27:1;O), was also observed where there was no sex dependency (Fig. 7d).
To examine the stereochemistry of glycosyl HexCer and Hex2Cer, we analyzed the kidney tissue using HILIC (Fig. 7e).As expected, galactosyl ceramide was substantially enriched in male mice, whereas the ion abundance of glucosyl ceramide was mostly the same in male and female mice.Interestingly, Gal2Cer chromatographic peaks were absent in female mice (at least under the limit of detection), and the ion abundance of Gal2Cer was substantially higher than that of lactosylceramide.Although the substrate of globotriaosylceramide (Gb3, described as Hex3Cer) is known as lactosylceramide, this result suggests that the substrate of Hex3Cer in the kidney is likely to be Gal2Cer.Overall, our integrated analysis clearly shows the unique regulatory mechanism of lipid pro le, which is associated with sexand age-dependent differences.
Based on the WGCNA results, only lipid cluster 11 was signi cantly associated with SPF/GF and aging parameters.The lipid cluster contained molecules that are likely to be microbiota-derived, including sterol sulfates, bile acids, SLs, and PC containing odd chains.Because the lipid cluster was highly associated with gene clusters 8 and 19, we further investigated the association between microbiota-derived lipid molecules and related genes (Fig. 8).Most of the lipid molecules contained in the cluster were signi cantly affected by the SPF/GF parameter; one sterol sulfate (ST 28:1;O;S), two SLs (SL 17:0;O/16:0 and SL 17:0;O/17:0;O), and one PC (PC 17:1_18:2) were recognized as prominent molecules associated with aging.Indoxyl sulfate is a typical urinary toxic metabolite that causes the progression of chronic kidney disease.Intestinal bacteria produce indole from tryptophan, and sulfotransferase 1A1 (SULT1A1) activity adds the sulfate moiety 57 .In contrast, steroid sulfotransferases such as SULT2B1b are responsible for producing cholesterol sulfate, which have various biological activities such as the inhibition of serine proteases (e.g., pancreatic trypsin and chymotrypsin) and PI3K activities 58 .Our results indicate that the production of sterol sulfate is also affected by the microbiome, and microbiota-speci c sulfolipids, including SLs, may contribute to homeostasis in the kidney.Furthermore, the genes contained in clusters 8 and 19 were highly associated with lipid metabolism.The association between the lipids in cluster 11 and the genes in clusters 8 and 19 has not yet been reported.In fact, some bile acids, including chenodeoxycholic acid, regulate a variety of metabolic functions by modulating nuclear receptors such as farnesoid X receptor (FXR), whose activation is important for reducing in ammation and oxidative stress in the kidney 59 .The bromodomain genes adjacent to the zinc nger domain, such as 2 B, centrosomal protein 250 kDa (CEP250), integrator complex subunit 1, transmembrane protein 199, microtubule-associated protein, RP/EB family, member 2, adaptor-related protein complex 2 mu 1 subunit, succinate dehydrogenase complex subunit C integral membrane protein 15 kDa, and tyrosyl-DNA phosphodiesterase 2 (TDP2), are predicted to be the target genes of the FXR TF 60 .Moreover, the associations of Midasin AAA ATPase 1, phosphatidylinositol transfer protein membrane associated 2 (PITPNM2), and BAF chromatin remodelling complex subunit 11b (BCL11b) with the microbiome have also been reported 61,62 .The age-dependent changes in PITPNM2, BCL11b, TDP2, and CEP250 were observed.Therefore, our results contribute to the understanding of the mechanism of host-microbiome interactions from the viewpoint of lipidomes.

Discussion
We demonstrated untargeted lipidomics for 13 biological samples, including tissues, cells, and plasma, with aging.These results provide the largest lipidome landscape to investigate metabolic changes associated with aging from the viewpoint of sex differences and commensal bacterial dependence.The FDR of MS-DIAL-based automatic lipid annotation was 2-3% with a RT ltering of 2.0 min 23 .We manually curated the annotated results by considering the basis of the equivalent carbon number model of lipids, in which the elution behavior of molecules in reverse phase LC depends on the length of acyl chains and the number of double bonds in lipids.In addition, our results highly compensate for recent reports encompassing the mouse brain metabolome changes associated with aging 30 .
The results of OPLS-R demonstrated that (1) the ratio of PUFA/MUFA in complex lipids increased in many organs, (2) the abundance of phospholipids containing MUFA decreased, (3) the abundance of TG and phospholipids containing PUFA increased, and (4) the most prominent change in the lipidome was found in the kidney with aging.Although similar results of PUFA/MUFA ratio have been reported in plasma and serum lipidomics studies, our results indicate that the tendency of PUFA and MUFA metabolism is a common phenomenon of aging in organs examined in this study.The unique enrichment of BMP-containing PUFAs is worth noting because only few reports have investigated the age dependency of BMP.BMP lipid is enriched in late endosomes and lysosomes, and it promotes the enzymatic activity of acidic sphingomyelinase (ASM) by direct binding 63 .Cellular senescence is associated with lysosomal homeostasis.Cytosolic pH change is associated with cellular senescence induced by lysosomal membrane damage 64 , and ASM activity is elevated with aging 65 .Therefore, the age-dependent increase in BMP may be related to lysosomal membrane homeostasis and subsequent cellular senescence in tissues.
We identi ed many unique changes in the lipidome during aging in a sex-or microbiomedependent manner.Microbiota-dependent lipid changes were observed in the feces, small intestine, and large intestine.Interestingly, microbiome-dependent changes in BA and SL were observed in remote organs, except for the bone marrow.It is well known that cleavage of conjugated moieties in bile acids is usually performed by the gut microbiota.Thus, the difference in the SPF/GF ratio between conjugated and unconjugated bile acids is likely to re ect the effect of the enzyme activities of the microbiome.The mechanism of the unique translocation of SL into various tissues is unclear when considering the fact that most of the other microbiota-derived lipids such as N-acyl amides, PI ceramides, and PE ceramides were not detected in remote organs.The genera Alistipes and Odoribacter produce SLs, whose molecules have bioactivities, such as antagonistic action on von Willebrand factor receptors, cytotoxicity in cancer cells, inhibitory functions of DNA polymerase, and anti-in ammatory effect 38 .Alistipes is relevant to dysbiosis and disease, and contrasting evidence of its pathogenicity has been reported; it has protective effects against liver brosis, colitis, and cardiovascular disease and is associated with colorectal cancer and mental signs of depression 66 .To investigate the mechanism in detail, the development of databases organizing the pair of commensal lipids and their producers is indispensable, along with metagenome sequences for multiple organs.Moreover, MS-imaging should provide new insights into the local distribution and biological importance of SL for the understanding of host and microbiome relationships.
Integrated analysis of the kidney transcriptome and lipidome revealed the relevance between gene expression and lipidome from the viewpoint of SPF/GF and male/female differences.Agedependent decreases in ether-linked glycerolipids, including ether PC, ether PI, ether PS, ether DG, and ether MGDG, were clearly observed in male kidneys, and their pro les correlated with the expression levels of AGPS and FAR1.Based on the pro le of PE species in which the alkyl and alkenyl types can be distinguished by the ESI(+)-MS/MS spectrum, the ether link often consists of an alkenyl (plasmalogen) type whose vinyl ether moiety has antioxidant activity 67 .The decrease of these metabolites in aging is considered the functional attenuation of cellular organelle peroxisomes as enriched in the GO analysis of male-speci c transcriptome data.In addition, our results indicate that galactosyl lipid metabolism in the kidney is highly correlated with the expression level of UGT8a.Although the machinery responsible for MGDG and DGDG biosynthesis in mammalian cells is not fully understood, recombinant expression of UGT8a in CHO cells tends to enrich the MGDG lipid 55, 56 , indicating that both galactosyl ceramides and galactosyl glycerides can be synthesized via UGT8a enzymatic activity.While Ugt8a knock-out mice exhibit pronounced tremor and progressive ataxia, no major morphological or functional defects are reported in the kidney 68,69 .The biological importance of MGDG in the kidney should be investigated in the future.
Furthermore, gene expression is regulated in the kidney by TFs including AR and STAT3 and active histone marks (H3K9ac, H3K27ac, and H3K4me3) according to the proposal of ChIP-Atlas.Of these, STAT3 was identi ed as the most signi cantly enriched TF in the ChIP-Atlas program, and human FAR1 and UGT8 are predicted to be target genes of STAT3 60 .However, the UGT8 gene is not predicted to be the target of other STAT family members.We further ascertained the consistency of the sequences recognized by STAT3 in humans and mice, whose matrix IDs of JASPAR are MA0144.2and MA0144.1,respectively 70 .Based on the sequence information of the promoter regions of human UGT8 and mouse UGT8a, a high-scoring STAT3-binding sequence was detected using the JASPAR program.These results indicate that the substantial changes in glycolipids between male and female mice may be explained by the age-and sex-dependent AR-STAT3 TF activity that regulates the expression pro le of UGT8a in the kidney.Moreover, age-dependent increases in Gal2Cer and Hex3Cer were observed.As observed in Fabry disease, which is caused by de ciency or decreased activity of α-galactosidase A (GLA), an intracellular lysosomal enzyme, the abnormal accumulation of Hex3Cer causes severe renal damage 71 .Given that the gene expression level of GLA did not change between males and females and between aged and young mice, the accumulation of these glycolipids may be due to activity increase of the biosynthetic enzymes, dysfunction of intracellular organelles such as lysosomes, or other mechanisms that maintain metabolic homeostasis.While a few changes in gene expression between SPF and GF were observed in the kidney, we identi ed several correlations between microbiota-derived lipids and the host transcriptome, the causality of which will be investigated in the future.
This study provides new insights into the potential link between lipid metabolism and aging.While the biological importance of lipid metabolites described in this study will be further investigated, other layers of omics data, such as proteomics and metagenomics, will be required to examine the harmonized mechanism of host and bacterial lipid metabolism.In particular, it is important to perform integrated analysis also in the liver and intestinal tract, since the functions of these organs are considered to be substantially in uenced by aging and intestinal microbiota 72 .It is also important to translate the results of mice to human to consider the extensibility of our ndings.Moreover, annotation of unknown spectra is needed because more than 200 unique and highly abundant peaks remain unknown in the present study.The open sharing of MS raw data will facilitate further annotation with the updates of computational mass, which is an active research eld in metabolomics and lipidomics 73,74 .Further data accumulation, biochemical validation, and informatics research will lead to a better understanding of the molecular mechanisms underlying cellular senescence and age-related chronic diseases, and will contribute to build a foundation for establishing a healthy and long-living society.

Figure 1 Summary
Figure 1

Figure 2 Results
Figure 2

Figure 3 Investigation
Figure 3

Figure 4 The
Figure 4

Figure 5 Data
Figure 5

Figure 7 The
Figure 7