The circulating lipidome is largely defined by sex descriptors in the GOLDN, GeneBank and the ADNI studies

Biological sex is one of the major anthropometric factors which influences physiology, metabolism and health status. We have investigated the effect of sexual dimorphism on the blood lipidome profile in three large population level studies - the Alzheimer’s disease neuroimaging initiative - ADNI (n =806), the GeneBank Functional Cardio-Metabolomics cohort (n= 1015) and the Genetics of Lipid lowering Drugs and Diet Network - GOLDN (n=422). In total, 355 unique lipids from 15 lipid classes were detected across all three studies using LC-MS. Sixty percent of these lipids differed between men and women in all three cohorts, and up to 87% of all lipids demonstrated sex differences in at least one cohort. ChemRICH enrichment statistics on lipid classes showed that phosphatidylcholines, phosphatidylethanolamines, phosphatidylinositols, ceramides, sphingomyelins and cholesterol esters were found at higher levels in female subjects while triacylglycerols and lysophosphatidylcholines were found at higher levels in male participants across the three cohorts. This strong sex effect on the blood lipidome suggests that specific regulatory mechanisms may exist that regulate lipid metabolism in a different manner between men and women. Cohort studies involving blood lipidomics should consider separate analyses for male and female participants instead of combined analyses treating sex as a confounding factor.

and propanoate metabolism [5]. Mittelstrass et al. argued that metabolomics analysis 54 in epidemiology should be stratified by sex and showed a strong sex effect in 3,300 55 participants in the Cooperative Health Research in the Region of Augsburg (KORA) cohort. 56 The study shows that up to 78% metabolites were under sex effect, including amino 57 We here report on the effect of sex descriptors on a comprehensive panel of 355 76 blood lipids from 15 lipid classes in three large cohorts with the largest comparison 77 to date with 2,243 subjects in total. These cohorts included the Genetics of Lipid 78 Lowering Drugs and Diet Network (GOLDN) (n = 422), GeneBank, Cleveland Clinic 79 (n= 1,015), and the Alzheimer's Disease Neuroimaging Initiative (ADNI) (n = 806). 80 We have used univariate statistics and chemical similarity enrichment analysis to 81 highlight the strong sex effect on the detected lipids. 82

Genetics of Lipid Lowering
Drugs and Diet Network Figure 1 shows the lipidomics data acquisition and processing workflow. All 90 lipidomics data were acquired using identical LC-MS instruments at the West Coast 91 Metabolomics Center, UC Davis. A set of 15 internal standards were added to each 92 sample which were used for retention time correction. 93 A total 355 unique lipid species covering 15 lipid classes were included in the final 94 dataset, after removing poorly detected and duplicate signals. Random forest-based 95 normalization method using the SERRF tool[12] removed technical variance to as 96 low as 2-6% relative standard deviation across all studies, using BioreclamationIVT 97  Supplementary Table S1). 99

Significantly associated individual lipids 100
Up to 87% of all lipids were found to be significantly different (p<0.05) between men 101 and women in at least one cohort study using the raw p-values of the Mann-Whitney 102 U test. More lipids were found to be specifically altered in ADNI cohort comparison 103 to the GOLDN and GeneBank cohort studies. 33% of all significantly altered lipids 104 were found to be common across all three studies ( Figure 2).

117
Among the top-25 of the most significantly different lipids, several sphingomyelin 118 lipids (SM) were, starting with the most significant lipid SM d32:2 were found at 119 consistently higher levels in women than in men across all cohorts (Table 2- Table S1. 124

Significantly associated lipid classes 125
Next, we performed a lipid class level analysis to find which chemical classes were 126 significantly higher in the female versus male comparison. We have utilized the 127 ChemRICH enrichment analysis method, which does not rely on a background 128 database for computing the set level statistics. Figure 3 shows the lipids classes 129 associated with differences between both sexes as ChemRICH impact plots. three cohorts. It seems that age was factor for some classesfor example 150 triacylglycerols were lower in women in the younger cohorts (GOLDN and the 151 GeneBank cohort) but not in the ADNI cohort study that included predominately 152 elderly subjects. Interestingly, free fatty acids showed a mixed direction as indicated 153 by purple color in the ChemRICH plot, so we investigated the degree of saturation 154 within this lipid class. We found that saturated fatty acids were higher in men while 155 unsaturated fatty acids were higher in women in all three cohorts. 156

Discussion 157
In comparison to earlier metabolomics studies, we have expanded the sexual 158 dimorphism analysis of blood lipidome with using a comprehensive panel and larger 159 studies. Several new lipids classes were found to be under strong impact of sexual 160 dimorphism. We have replicated the previous finding that levels of sphingomyelins 161 and phosphatidylcholines were elevated in women [9,10] while 162 lysophosphatidylcholines and acylcarnitines were found at higher concentrations in 163 men[6,11]. We also observed differences in triacylglycerol levels in the ADNI study 164 between women and men that could possibly be due to biological age[9], while these 165 differences was absent in comparatively younger participants of the GOLDN and the 166 GeneBank cohort . The most significant lipid clusters included SM, PC, TG, p-PC, 167 LPC and FA lipids in all three cohorts. We found new sex-regulated classes including 168 phosphatidylinositols, plasmalogens, and ceramides. Plasmalogen biosynthesis has 169 been linked with male fertility [13]. It has been previously shown that the hepatic ceramide biosynthesis is regulated by sex hormones, including testosterone [14]. 171 Differences in lipids between both sexes may suggest that sex specific remodeling 172 of lipid metabolism is a fundamental biological process and calls for further studies 173 to discover the underlying mechanisms that can create a basis for developing sex 174 specific disease prevention strategies. Maximum intensity values within the retention time window were used as the peak 323 height of the target compounds. Data were normalized using quality control pool 324 samples that were interjected between every 10 subject samples during data 325 acquisition. We employed the SERRF random forest machine learning algorithm [12] 326 to remove batch and drift effects in each cohort data set for each individual lipid.
Statistical analysis: Data were log transformed before statistical testing. The Mann-328 Whitney-U test was used to find raw significance values for each lipid between men 329 and women in each of the three cohorts. p-values from the Mann-Whitney-U test 330 were used as input for the chemical similarity enrichment analysis using the 331 ChemRICH software[24] to find statistical significance levels on the basis of lipid 332 classes. ChemRICH p-values were corrected for the false discovery rate. 333

Conclusions 334
Our study is the largest lipidomics study to report differences between male and 335 female participants. We have acquired LC-MS datasets on identical mass 336 spectrometers and have used machine learning methods-based signal correction 337 approach to remove technical variance and batch effects. Use of a database 338 independent lipid set enrichment analysis methods have identified a number of 339 specific lipid classes that were associated with differences between adult men and 340 women. Epidemiological studies focusing on these drastically different lipid classes 341 need to stratify the cohort data and interpret the results separately for male and 342 female participants.