Metabolic modeling of sex-specific tissue predicts mechanisms of differences in toxicological responses

Male subjects in animal and human studies are disproportionately used for toxicological testing. This discrepancy is evidenced in clinical medicine where females are more likely than males to experience liver-related adverse events in response to xenobiotics. While previous work has shown gene expression differences between the sexes, there is a lack of systems-level approaches to understand the direct clinical impact effect of these differences. Here, we integrate gene expression data with metabolic network models to characterize the impact of transcriptional changes of metabolic genes in the context of sex differences and drug treatment. We used Tasks Inferred from Differential Expression (TIDEs), a reaction-centric approach to analyzing differences in gene expression, to discover that androgen, ether lipid, glucocorticoid, tryptophan, and xenobiotic metabolism have more activity in the male liver, and serotonin, melatonin, pentose, glucuronate, and vitamin A metabolism have more activity in the female liver. When TIDEs is used to compare expression differences in treated and untreated hepatocytes, we see little response in those sex-altered subsystems, and the largest differences are in subsystems related to lipid metabolism. Finally, using sex-specific transcriptomic data, we create individual and averaged male and female liver models and find differences in the import of bile acids and salts. This result suggests that the sexually dimorphic behavior of the liver may be caused by differences in enterohepatic recirculation, and we suggest an investigation into sex-specific microbiome composition as an avenue of further research.


Introduction 59
Male subjects in both animal and human studies are disproportionately used for testing in 60 toxicology studies (Zucker & Beery 2010;Feldman et al., 2019). This discrepancy leads to 61 incorrect assumptions on female drug response as evidenced in the clinic where female patients 62 are more likely than males to experience liver-related adverse events in response to xenobiotics 63 such as acetaminophen, diclofenac, and isoniazid (O'Connor, Dargen, and Jones, 2003;Banks 64 et al. 1995;Ostapowicz et al., 2002). While gene expression differences between males and 65 females have been extensively studied (Yang et al., 2006;Yang et al. 2012;Lopes-Ramos et al., 66 2020), little is known about how these changes in expression contribute to functional changes 67 that result in this divergent clinical response.

68
Sexually differential metabolism is key to understanding these responses (Pannala et al., 2019).

69
Transcriptional profiling can provide insight into metabolic genes and how associated functions

85
Here, we present an analysis of male and female liver metabolism using tissue-specific and drug-86 specific gene expression data in the context of a human GEM. Liver-, kidney-, and brain-sourced 87 transcriptomics from the Gene Expression Omnibus (GEO) are used to establish general and 88 liver-specific sex differences in metabolism where the liver acts as the sexually dimorphic tissue 89 of interest, the kidney represents a similarly sexually dimorphic tissue, and the brain functions as . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted February 7, 2023. ; https://doi.org/10.1101/2023.02.07.527430 doi: bioRxiv preprint a known sexually monomorphic tissue (Yang et al., 2006). We also compare metabolism between

91
hepatocytes with and without exposure to drug using expression data from ToxicoDB (Nair et al.,

92
2020) with particular interest in those metabolic differences found between the male-and female-93 sourced liver tissue. We then use the male-and female-sourced liver gene expression data to 94 create individual and averaged male-and female-specific GEMs and use flux sampling to illustrate 95 differences in core metabolism and metabolites involved in enterohepatic recycling. Together,

96
these results suggests that sexually dimorphic adverse event frequency may be driven by 97 differences in the gut-liver axis.

100
Women experience liver-related adverse events more frequently than men 101 We used the Food and Drug Administration's Adverse Event Reporting System (AERS), a 102 collection of side-effect reports voluntarily sourced from health care professionals and the general 103 public, to quantify sex-specific adverse event frequency ( Figure 1). Each report includes 104 information on the event, the sex and age of the patient, other drugs being used, and the time of 105 the incident. We collected reports from this database using AERSMine (Sarangdhar et al., 2016),

106
an application designed to improve the accessibility of AERS. Reports related to liver dysfunction 107 were counted for each quarter from 2004 to 2021 and divided by the total amount of reports for 108 that quarter to account for increasing reporting trends ( Figure 1a). We found that not only did 109 more total reports exist for female patients, but that female patients are consistently reported to 110 experience liver-related adverse events more often than their male counterparts ( Figure 1b). This 111 result can be explained in part by a previous report of overall increased prescription drug 112 consumption in women (Hales, et al.,, 2019), so we next investigated sexual-dimorphic response 113 by drug. We compared the ratio of reports for each drug for each sex, only including drugs with 114 greater than 100,000 reports to ensure that the differential effect was robust and not due to small 115 sample size. Unexpectedly, more drugs were disproportionately affecting men, but those that 116 were reported more often in women tended to have a higher fraction of female-specific reports 117 ( Figure 1c). It is important to note, however, that not all drugs are used equally by both sexes. For 118 example, alendronic acid is primarily prescribed for osteoporosis and tenofovir disoproxil is used 119 as an HIV treatment, diseases with higher female and male incidence, respectively (Cawthon, 120 2011; CDC, 2019). Frequently-used pharmaceuticals exhibiting a spectrum of responses . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted February 7, 2023. ; https://doi.org/10. 1101/2023 indicates that sex is a relevant variable in determining hepatotoxicity, so understanding the sex-

122
based metabolic signature is an important aspect of this problem.

124
Transcriptome-informed metabolic model indicates a sex-and tissue-specific signature in 125 untreated tissue 126 Next, we sought to understand differences in functional liver metabolic networks between males 127 and females. To do this, we used microarray data from GEO to characterize gene expression 128 differences between male and female patients not experiencing toxicity. To understand sexual 129 dimorphic metabolism specific to the liver, kidney and brain tissue were also included in this 130 analysis as a comparison. While the kidney is understood to have sex-specific metabolic function, 131 the brain is considered a sexually monomorphic tissue with respect to gene expression (Yang et 132 al., 2006).

133
For each tissue type, differentially expressed genes (DEGs) between the sexes were calculated subsystems defined by Human1, we can assign weights to each reaction in the subsystem 141 dependent on the log fold change from the DEGs and the GPRs from the GEM to generate a task 142 score for that subsystem. This task score can then be compared to randomized weights for each 143 reaction to determine if the task score is significantly higher or lower (male-or female-biased) for 144 that subsystem.

145
Of the 135 subsystems defined by the GEM, 11 in the liver tissue, 23 in the kidney, and 0 in the 146 brain were found to be significantly different between the male and female patients. The gene 147 expression data for the male and female brain tissue did not have differences in TIDEs,

148
suggesting highly similar metabolic function of the sexes in this tissue type. The expression 149 profiles for the liver and kidney tissue, however, did have significant sex-biased TIDEs. Of the 150 sex-biased tasks, only two were found to be upregulated by the same sex between the liver and 151 kidney: Acylglyceride Metabolism and Steroid Metabolism pathways were upregulated in males.

152
Previous literature is consistent with increased very low-density lipoprotein and triglyceride . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted February 7, 2023. ; https://doi.org/10.1101/2023.02.07.527430 doi: bioRxiv preprint production in male livers (Palmisano et al., 2018) and a higher rate of steroid metabolite secretion 154 in men (Raven & Taylor, 1996). Of the other 8 subsystems significantly different in the liver, 6 are 155 only significant in the liver and 2 are biased in opposite sexes ( Figure 3a). Vitamin A storage in 156 rats has been shown to be higher in female livers and male kidneys (Booth, 1952), agreeing with 157 these results. Additionally, androgen receptors and glucocorticoid receptors have been shown to 158 be correlated in mouse livers (Kroon, Pereira, & Meijer, 2020), and rat liver tissue has previously 159 exhibited an increase in tryptophan 2,3 dioxygenase activity in response to glucocorticoids (Danesch et al., 1983), explaining how these three metabolic subsystems would all be 161 upregulated in male liver tissue. Tryptophan also acts as a precursor to serotonin and melatonin.

162
With more of the amino acid being used by male tryptophan 2,3 dioxygenase, an enzyme which 163 produces no products involved in serotonin and melatonin metabolism, less tryptophan is 164 available for serotonin and melatonin biosynthesis, explaining the relative increase in female 165 activity for this task. Previous research has also shown that ether lipid levels in female serum tend

206
When looking at variability within each sex, we see that reactions are frequently either a part of 207 the core metabolism or are present in very few of the models (Figure 4b). A similar distribution 208 can also be seen when comparing between sexes, where most reactions are found in near-equal

209
proportions in each sex with the second largest group being sex-specific reactions (Figure 4c).

210
Specifically, we found that the male and female models contained 51 and 246 unique reactions, 211 respectively, that are in at least 10% of the samples for each sex. These unique reactions tended 212 to group into specific pathways: purine metabolism in females (Table 1) and transport reactions 213 for males (Table 2). These groups of reactions suggest that while genes directly controlling these 214 subsystems do not result in significant differences in tasks, there are systemic differences in each 215 sex that contribute to differential metabolism.

216
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted February 7, 2023. ; https://doi.org/10.1101/2023.02.07.527430 doi: bioRxiv preprint 217 Averaged sex-specific models suggest differences in enterohepatic recycling which may 218 impact the hepatotoxic effects of drugs 219 To characterize broad differences between sexes, we used RIPTiDe with averaged gene

238
We present an analysis of the current landscape of female-biased adverse event reporting, the 239 sexually dimorphic profile of liver metabolism, the hepatocellular response to toxic 240 pharmaceuticals, and differences in sex-specific GEMs. We find that sex as a biological variable 241 is critical to describe the flow of metabolites through the hepatocyte and that there is little overlap 242 between those subsystems which exhibit sex differences and those which experience changes in 243 response to drug. Of the 135 subsystems defined in the Human1 metabolic network 244 reconstruction, 11 in the liver, 23 in the kidney, and 0 in the brain were found to differ based on 245 sex, suggesting that the kidney may have more sex-specific differences in functional metabolism 246 than other tissue types. This result suggests a different conclusion compared to previous literature 247 in mice (Yang et al., 2006) though this result may be an artifact of the available data for this 248 investigation. In the human liver dataset we used for our analysis, the participants are all classified . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted February 7, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 as obese and therefore may be experiencing liver-related health issues that dilute the effect of 250 sex, leading to potentially conflicting results.

251
In the liver, we find that xenobiotic metabolism was more active in untreated males, while pentose  (Briggs & Briggs, 1972). Additionally, women who take estrogen-based oral 277 contraceptives clear specific drugs more efficiently than men (Miners, Attwood, & Birkett, 1983).

278
These findings indicate that sex hormones may inhibit the reabsorption of drugs into the portal 279 vein, decreasing concentrations in the liver. The deconjugating microbiome of the patients may 280 also impact how bile acids and salts are recirculated, as previous work has shown that the gut 281 microbiome is sex-specific (Kim, 2022).
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted February 7, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 There are important factors to note when considering the results of this analysis. First, we assume 283 that gene expression levels are directly related to protein levels. There are other post-284 transcriptional and post-translational modifications that can alter protein abundance that are not 285 available with the given data (Vogel & Marcotte, 2012). Though this assumption is not unique to 286 this paper, it is nonetheless an important caveat when discussing its results. Additionally, our 287 analysis here only considers one tissue type at a time; the cross-talk between organs and organ 288 systems is a necessary consideration when evaluating toxicity as absorption, distribution, 289 metabolism, and excretion by different organs can impact liver injury (Talevi & Bellera, 2022).

290
Sex as a biological variable continues to be a relevant consideration for any biological study.

291
Many expression datasets do not have sex information recorded with the samples, decreasing 292 our pool of potential data to be modeled as well as the quality of the data. An understanding of 293 the necessity for sex as a variable will provide the foundation required to further the pursuit of 294 precision medicine and drug therapy.

297
Adverse event reporting 298 We used AERSMine (Sarangdhar, M. et al, 2016) to evaluate the difference in adverse event

299
reporting in the United States between the first quarter of 2004 to the third quarter of 2021. Only 300 those reports with age and sex information were used, and patients between the ages of 15 to 65 301 were considered. Adverse events were considered "liver-related" if labelled with one of the 302 following: "drug-induced liver injury", "hepatotoxicity", "hepatic enzyme increased", "hepatic and 303 hepatobiliary disorders nec", "liver function analyses", "hepatic and hepatobiliary disorders",

304
"hepatic failure and associated disorders", and "hepatic enzymes and function abnormalities".

305
Once compiled, the percentage of total reports was calculated for each quarter and compared 306 with a two-sided Mann-Whitney U test. Visualization was performed in R.

308
Gene expression analysis 309 Differential expression between male and female samples from data sets were found in the Gene

310
Expression Omnibus (Liver: GSE130991, Kidney: GSE36059, Brain: GSE5281). For those 311 datasets without information on sex, male was assigned to those samples that were in the top . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

322
TIDEs identifies all genes related to reactions in a user-defined "task" and assigns each reaction 323 a weight based on the log2 fold difference in the expression of those genes. For genes with an 324 "OR" gene protein rule, the highest fold difference will be used; for genes with an "AND" rule, the 325 lowest will be used. Each fold difference is then averaged for a given task, and this average 326 becomes its task score. This score is then compared to 1000 randomized task scores, calculated 327 using randomly chosen log fold difference weights from other tasks. Significance was decided if 328 p < 0.025 because it is a two-sided test. Tasks were defined as KEGG ortholog subsystems as 329 assigned by the model Human1. (1 referring to the reaction with the highest gene abundance) and the sum of fluxes is maximized.

339
The model is sampled using these constrained flux distributions. This technique was performed 340 with male and female liver data, with 804 (586 female/218 male) individual models and averaged 341 models for male and female created and sampled.

342
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  (3), 820-827.
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted February 7, 2023. ; https://doi.org/10.1101/2023.02.07.527430 doi: bioRxiv preprint