Activation of pregnane X receptor induces atherogenic lipids and PCSK9 by a SREBP2‐mediated mechanism

Many drugs and environmental contaminants induce hypercholesterolemia and promote the risk of atherosclerotic cardiovascular disease. We tested the hypothesis that pregnane X receptor (PXR), a xenobiotic‐sensing nuclear receptor, regulates the level of circulating atherogenic lipids in humans and utilized mouse experiments to identify the mechanisms involved.


| INTRODUCTION
Hypercholesterolemia, especially high level of low-density lipoprotein (LDL) cholesterol, is one of the major risk factors for atherosclerotic cardiovascular disease, the leading global cause of mortality (G. A. Roth et al., 2015). Both inherited and acquired factors may cause hypercholesterolemia (Garg & Simha, 2007). The most common secondary causes are thought to be obesity, unhealthy diet and sedentary lifestyle (Garg & Simha, 2007). In addition, foreign compounds may affect cholesterol status.
Pregnane X receptor (PXR, systematic name NR1I2), is a nuclear receptor activated by a large variety of structurally divergent exogenous ligands including drugs, natural extracts and environmental chemicals (Kliewer et al., 2002). Thus, PXR is considered to play a key role in the sensing of chemical environment. Originally, PXR was found to facilitate detoxification and excretion of foreign compounds (Kliewer et al., 1998). More recently, PXR has been shown to regulate lipid metabolism (Hakkola et al., 2016;Zhou, 2016). PXR activation induces hepatic lipogenic genes and leads to liver steatosis in several animal models and lipid accumulation in human liver cell models (Bitter et al., 2015;Gwag et al., 2019;Zhou et al., 2006). Moreover, some animal studies suggest that PXR activation induces What is already known • Exposure to different PXR ligands, such as several drugs and environmental contaminants, associates with hypercholesterolemia.
• Studies in experimental animals suggest that PXR activation induces hypercholesterolemia and accelerates atherosclerosis.

What this study adds
• Activation of PXR elevates LDL and total cholesterol in humans.
• PXR ligands increase hepatic cholesterol synthesis, plasma PCSK9 level and promote proteolytic activation of SREBP2.

What is the clinical significance
• PXR-activating drugs and environmental contaminants may pose a cardiovascular risk by increasing circulating atherogenic lipids. hypercholesterolemia and accelerates atherosclerosis Meng et al., 2019;Zhou et al., 2009). Due to its wide ligand acceptance, PXR could play a major role as the mediator of cholesterogenic effect of drugs and environmental chemicals.
Interference with any of the major phases of cholesterol homeostasis, that is, synthesis, absorption, reverse cholesterol transport and excretion, could result in hypercholesterolemia (van der Wulp et al., 2013). Furthermore, the discovery of proprotein convertase subtilisin/kexin type 9 (PCSK9) has considerably changed the view of the regulation of cholesterol homeostasis. PCSK9 is a central regulator of plasma cholesterol level by influencing the levels of plasma membrane LDL receptors in hepatocytes (Warden et al., 2019). The mechanisms of PXR action remain controversial and both increased intestinal absorption and increased hepatic synthesis through activation of squalene epoxidase have recently been suggested to mediate PXR-induced hypercholesterolemia in mice Meng et al., 2019).
To address the question if PXR activation affects human cholesterol homeostasis, we analysed serum metabolomics in controlled clinical studies investigating the cardiometabolic effects of rifampicin, the prototypic human PXR ligand, and showed metabolic signature indicating increased serum cholesterol and enhanced synthesis of cholesterol. We utilized high-fat diet challenged mice as a model to reveal the mechanism mediating PXR-induced alterations in cholesterol homeostasis and showed that PXR activation induces the cholesterol synthesis pathway, increases plasma PCSK9 and promotes sterol regulatory element-binding protein 2 (SREBP2) proteolytic activation.

| Study design of the clinical trials
Three previously performed clinical trials exploring metabolic effects of rifampicin were utilized in the current study. Rifa-1 (Rysä et al., 2013) and Rifa-BP (Hassani-Nezhad-Gashti et al., 2020) studies had a randomized placebo-controlled crossover design with at least a 4-week washout period. Rifa-2 study (Hukkanen, Rysä, et al., 2015) had a one-arm design with no control arm. Rifa-1 and Rifa-2 had an open design while Rifa-BP was single-blind with study personnel blinded. The subjects (Table S1) in all three studies were administered 600-mg rifampicin (Rimapen, Orion Inc., Espoo, Finland) once a day for a week. The studies were designed to explore the effects of PXR activation on glucose tolerance (Rifa-1, n = 12), incretin secretion (Rifa-2, n = 12) and blood pressure regulation (Rifa-BP, n = 22). The inclusion criteria were healthy volunteers with age between 18 and 40 years (45 years in Rifa-2). The body mass index (BMI) criterion was between 19-28 kgÁm À2 in Rifa-2 and 19-30 kgÁm À2 in Rifa-BP, whereas Rifa-1 did not have a BMI limit.
In Rifa-BP, inclusion criterion for systolic blood pressure was 95-140 mmHg, whereas the other studies did not employ blood pressure limits.
The exclusion criteria were any regular medication (hormonal intrauterine device was allowed), any major somatic or psychiatric morbidity (as judged by the study physician on the basis of history, physical examination and basic laboratory values), insensitivity to rifampicin, continuous use of soft contact lenses (rifampicin may colour), pregnancy or breast feeding, drug or alcohol abuse, history of difficult venipuncture and participation of any other medical study during the study or the past 1 month. Additionally, the diastolic blood pressure above 90 mmHg was an exclusion criterion in Rifa-BP. The original sample size calculations were targeted for fasting glucose and 24-hr blood pressure and not to metabolomics analyses. Thus, the combined Rifa-1 and Rifa-BP data set with n = 34 for fasting metabolomics and Rifa-2 data set with n = 12 for oral glucose tolerance test metabolomics present samples of convenience for the study of metabolome. The subjects visited the Internal Medicine Research Unit of Oulu University Hospital, Oulu, Finland as outpatients. After the first tablet was taken under the supervision of a study nurse, the participants were asked to take their daily doses at home between 4 and 8 p.m. at least 1 hour before and 2 hours after a meal. The details of the experimental protocols in Rifa-1 and Rifa-2 are described in the previous publications (Hukkanen, Rysä, et al., 2015;Rysä et al., 2013). In Rifa-2, the oral glucose tolerance test was performed on the morning of the first study day before rifampicin dosing and the second oral glucose tolerance test was performed on the morning of the eight study day. The oral glucose tolerance test time points were 0, 30, 60, 90 and 120 min. The Rifa-BP trial employed 24-hr ambulatory blood pressure measurements at the end of each 7-day rifampicin or placebo arm (Hassani-Nezhad-Gashti et al., 2020). In all three trials, the participants were asked to abstain from the use of alcohol, over-the-counter medications and dietary and herbal supplements during the study arms. Smoking and coffee drinking were allowed. The subjects consumed their regular diets during the study arms. In Rifa-BP trial, the subjects were also asked to abstain from the use of liquorice-containing products, salty snacks as well as energy drinks.
A written, informed consent was obtained from each study sub-

| Analytical methods of human serum samples
A high-throughput nuclear magnetic resonance (NMR) metabolomics platform was used for the quantification of serum lipid and metabolite measures that represent a broad molecular signature of systemic metabolism. The experimental setup allows for the simultaneous quantification of routine lipids, lipoprotein subclass distributions, fatty acids, as well as other low-molecular weight metabolites, such as amino acids and glycolysis-related metabolites and ketone bodies in absolute concentration units (Soininen et al., 2015). The spectroscopic and analytical characteristics of the platform have been detailed elsewhere (Inouye et al., 2010;Soininen et al., 2009;Würtz et al., 2016;Würtz & Soininen, 2020).
Additional standards here added during derivatization. n-Alkanes (c = 8 mgÁL À1 in hexane) were used for calculation of retention indexes and 4,4 0 -dibromooctafluorobiphenyl (c = 9.8 mgÁL À1 in hexane) was used as injection standard to control the quality of injection.
Data were normalized with internal standards. 4-β-OH-cholesterol was measured in Rifa-1 with a gas chromatography-mass spectrometry method as published previously (Hukkanen, Puurunen et al., 2015) and in Rifa-BP with a liquid chromatography-electrospray-high-resolution mass spectrometry method as described (Hautajärvi et al., 2018 obesity. After this feeding period, the high-fat diet-fed mice were allocated to three groups to yield as similar average weights for the groups as possible, but otherwise randomly. High-fat diet alone, highfat diet vehicle and high-fat diet pregnenolone-16α-carbonitrile groups contained 9, 8 and 8 mice, respectively. To prepare for potential loss of animals one extra mouse was subjected to high-fat diet treatment. However, no animals were lost during the treatment resulting in slightly unequal group sizes. Chow group contained the other litter mates, 10 mice. Group sizes were determined based on our prior experience with similar mouse studies. Treatment groups were administered 50 mgÁkg À1 of pregnenolone-16α-carbonitrile or vehicle (30% dimethyl sulfoxide in corn oil) by intraperitoneal injections for 4 days once a day. Mice were fasted 12 hr overnight, orally gavaged 2 gÁkg À1 of glucose and anaesthetized with midazolam/fentanyl-fluanisone. Two hours after glucose administration mice were killed with carbon dioxide and EDTA-plasma and tissues were collected and snap-frozen in liquid nitrogen.
PXR knockout (PXR-KO) mouse experiments: The PXR-KO mouse line was generated as described earlier and kindly provided by Dr. Wen Xie, University of Pittsburgh (Xie et al., 2000). The original mouse strain in the C57BL/6 J background was backcrossed six times to C57BL/6 N strain to comply with the subtrain of the wild-type mouse used. The male PXR-KO mice were subjected to an identical experiment as described above for the wild-type mice except that the high-fat diet-feeding lasted for 18 weeks. Six mice were allocated per group, as the strong effect of pregnenolone-16α-carbonitrile seen in the wild-type mice enabled group size reduction without losing statistical power.

| Materials
Pregnenolone 16α-carbonitrile (PCN), dimethyl sulfoxide (DMSO) and corn oil were purchased from Sigma-Aldrich. Rifampicin was a product of Orion Inc., Espoo, Finland. See section 2.6. for the agents used in immunoblotting and ELISA. See sections 2.7 and 2.8 for reagents and kits used in RNA extraction, qPCR and RNA sequencing.

| Lipid and bile acid analyses of mouse plasma and liver
Plasma and liver cholesterol and non-cholesterol sterols, lanosterol, zymostenol, desmosterol and lathosterol, and squalene (cholesterol precursors), and campesterol, sitosterol and avenasterol (plant sterols), and cholestanol (5α-saturated derivative of cholesterol), were analysed using gas-liquid chromatography (GLC) with a 50-m capillary column (Ultra 2, Agilent Technologies, Wilmington, DE) and flame ionization detection with 5α-cholestane as internal standard as described earlier (Miettinen et al., 1989). In case of the liver samples, they were carefully homogenized before saponification. The concentrations of the non-cholesterol sterols and squalene were adjusted to cholesterol of the same GLC run and expressed as ratios to cholesterol (102 μgÁmg À1 of cholesterol) in order to enable their comparison between samples with different cholesterol levels.
Total bile acids from plasma were measured with Mouse Total Bile Acids Assay kit (Crystal Chem).

| RNA extraction and qPCR
Total RNA from liver was extracted with RNAzol RT reagent (Sigma) according to manufacturer's protocol. One microgram of total RNA was reverse transcribed to cDNA with RevertAid Reverse Transcriptase (Thermo scientific) using random hexamer primers according to manufacturer's protocol. cDNA samples were diluted 1/10 with H 2 O. Quantitative PCR (qPCR) was performed with FastStart Universal SYBR Green Master Mix (Roche) in ABI 7300 thermal cycler (Applied Biosystems) or with PowerUp SYBR Green Master Mix (Thermo scientific) in QuantStudio 5 (Applied Biosystems). GAPDH and TBP were used as reference genes. All qPCR reactions were optimized with primer concentration to reach 95%-105% reaction efficiency. qPCR primers are listed in Table S2. mRNA fold changes were calculated with 2 -ΔCT method where fold change = 2 -ΔCt sample / 2 -ΔCt control sample .

| RNA sequencing
For liver RNA sequencing, three mice per group were randomly selected. Total liver RNA was DNase treated with RNase-Free DNase Set (Qiagen) coupled with RNeasy MinElute Cleanup kit (Qiagen).
RNA concentration was determined with Qubit using Qubit RNA BR Assay kit (Thermo Fisher Scientific) and RNA integrity numbers with Agilent 2100 bioanalyzer using Agilent RNA 6000 Nano Kit.
Ribosomal RNA was depleted, RNA fragmented and cDNA libraries prepared for sequencing with TruSeq Stranded Total RNA with Ribo-Zero Gold kit (Illumina). The sequencing run was performed with NextSeq550 (Illumina).
Pathway analyses were done with the Ingenuity Pathway Analysis software (Qiagen; RRID:SCR_008653) to determine the most enriched pathways and to predict upstream regulators associated with differentially expressed genes. For pathways and upstream regulators, p value of overlap was calculated using right-tailed Fisher's exact test.
The complete data sets are available at the NCBI's Gene Expression Omnibus (GEO; RRID:SCR_005012) database (accession number GSE136667). Gene expression profiling data comply with the MIAME (Minimum Information About a Microarray Experiment) standard.

| Data and statistical analysis
The data and statistical analysis comply with the recommendations on experimental design and analysis in pharmacology (Curtis et al., 2018).
after the Benjamini-Hochberg procedure was considered statistically significant in the fasting state metabolomics. However, both nominal and corrected p values are reported. In Rifa-2 oral glucose tolerance test AUC data set with n = 12, a higher false discovery rate of 0.1 (10%) was utilized. Correlations were determined with Pearson's correlation coefficient using GraphPad Prism and considered significant when p < .05.
The statistical analyses of animal experiments were done with GraphPad Prism. Differences between two study groups were compared with Student's two-tailed t-test and multiple group comparisons with one-way ANOVA and Tukey's post hoc test when the p value of F < .05 and there was no significant variance inhomogeneity. Statistical differences were considered significant when p < .05.

| Nomenclature of targets and ligands
Key protein targets and ligands in this article are hyperlinked to corresponding entries in the IUPHAR/BPS Guide to PHARMACOL-OGY http://www.guidetopharmacology.org and are permanently archived in the Concise Guide to PHARMACOLOGY 2019/20 (Alexander, Cidlowski, et al., 2019;Alexander, Fabbro, et al., 2019).

| Rifampicin affects fasting serum metabolomics by increasing IDL and LDL cholesterol and markers of cholesterol synthesis in humans
To test the hypothesis that activation of PXR affects cholesterol homeostasis in humans in vivo, we utilized two clinical studies, Rifa-1 and Rifa-BP, designed to investigate cardiometabolic effects of rifampicin, a prototypical human PXR activator (Chen & Raymond, 2006).
Thirty-four healthy volunteers were administered rifampicin or placebo daily for 1 week in a crossover setting and the serum samples were collected after 10-hr fasting. Rifampicin elevated serum 4-β-OHcholesterol, a marker of PXR target CYP3A4 activity, in all study subjects indicating compliance with the rifampicin dosing regimen ( Figure S1a).
Serum samples were analysed with an NMR metabolomics platform developed to assess systemic metabolism (Soininen et al., 2009;Würtz et al., 2017). The ratios of mean concentrations of all lipoprotein subclasses and their lipid components after the rifampicin study arm compared with the placebo study arm are presented in Figure 1.
Absolute mean values of all the metabolites measured after both study arms and p-values are presented in Table S3 (Fernandez et al., 2013). Atheroprotective apolipoprotein A1 (ApoA1) was increased whereas apolipoprotein B (ApoB) to ApoA1 ratio, a predictor of cardiac risk, was not affected (Ingelsson et al., 2007;Khuseyinova & Koenig, 2006). In addition, the total concentration of polyunsaturated omega-6 fatty acids was increased by rifampicin, due to an increase of linoleic acid.
Interestingly, serum concentrations of citrate and acetate, precursors of cholesterol and fatty acid synthesis, were decreased (Figure 2a), possibly due to increased cholesterol synthesis (Li et al., 2015;Yoshii et al., 2015). To further assess the effect of rifampicin on cholesterol synthesis, we determined the serum concentration of lathosterol, a cholesterol synthesis intermediate, from the Rifa-BP study samples.
In summary, the fasting serum metabolomics screen indicates that rifampicin increases serum IDL and LDL fractions. The increase of the synthesis intermediate lathosterol and decrease of cholesterol synthesis precursors, citrate and acetate, suggests that this effect is due to activation of cholesterol synthesis.

| Effect of rifampicin on cholesterol fractions during glucose challenge
Feeding may affect metabolic responses and a recent study utilizing metabolomics approach similar to this study reported the effect of glucose feeding . Moreover, we have reported the modifying effect of glucose feeding on PXR function in mouse liver (Hassani-Nezhad-Gashti et al., 2019). To analyse dynamics of the lipidome during glucose challenge in response to rifampicin, we utilized a third clinical study, Rifa-2, with oral glucose tolerance test (oral glucose tolerance test) in 12 healthy volunteers before and after 1-week rifampicin dosing.
According to the area under the curve (AUC) analysis, serum total cholesterol, total cholesterol in LDL and free cholesterol were elevated by rifampicin dosing ( Figure S2, Table S4). The observed effect was mainly due to the increased concentration already at the 0 time point, reflecting the fasting response, and no further change was observed during the oral glucose tolerance test. Similarly, concentrations of IDL and all sizes of LDL particles were significantly increased as was their total lipid, phospholipid, cholesterol, free cholesterol and esterified cholesterol component concentrations (Table S4). Again, the effects were mainly due to the increased fasting levels and the incremental AUC values were not affected by rifampicin (Table S4).
Overall, the effect of rifampicin was quite similar in the Rifa-2 study during oral glucose tolerance test compared with the Rifa-1 and Rifa-BP studies (fasting). However, there were some important differences, the AUC of concentrations of very small VLDL particles and their total lipid, phospholipid, cholesterol, free cholesterol and esterified cholesterol components, were increased by rifampicin, whereas F I G U R E 1 Effect of rifampicin on human serum lipoprotein fractions. Lipoprotein parameters are expressed as ratios of mean concentration after the study arms (rifampicin/placebo; n = 34). The results with statistically significant difference by nominal P value (P < .05) are indicated by asterisk and after the Hochberg correction (adjusted P < .05) by double asterisk. Abbreviations: P, particle concentration; L, total lipids; PL, phospholipids; C, total cholesterol; CE, cholesterol esters; FC, free cholesterol; TG, triglycerides; VLDL, very low-density lipoprotein; IDL, intermediate-density lipoprotein; LDL, low-density lipoprotein; HDL, high-density lipoprotein; XXL, chylomicrons and extremely large; XL, very large; L, large; M, medium; S, small; XS, very small. See also Table S3 and Figure S1b no changes were detected in the Rifa-1 and Rifa-BP studies. Also, AUC of remnant cholesterol was increased by rifampicin dosing (11% mean increase). Also notable was the increase in ApoB (mean increase 8%) and ratio of ApoB/ApoA AUCs after rifampicin. The serum concentration of ApoB48, the intestinal form of ApoB, was measured with ELISA to elucidate the role of intestine in the effect of rifampicin on ApoB. The fasting concentration and AUC of ApoB48 was significantly decreased by rifampicin dosing by 14% (2.5 ± 0.39 to 2.2 ± 0.51 μgÁml À1 ) and 11% (284 ± 57 to 254 ± 57 min Â μgÁml À1 ), respectively ( Figure S2b). Thus, the increase of total serum ApoB is possibly due to induced ApoB100 production in liver or altered kinetics of serum ApoB100.
The concentration of citrate decreased at the later oral glucose tolerance test timepoints in response to the rifampicin treatment ( Figure S2a) suggesting rifampicin induced increase in the citrate consumption.

| Pregnenolone-16α-carbonitrile treatment induces cholesterol synthesis in the livers of obese mice
To identify the mechanisms mediating the activation of cholesterol synthesis by PXR ligands, we utilized a mouse model.
As expected, high-fat diet increased plasma cholesterol compared with chow fed mice (Figure 4a). In addition, the liver cholesterol content was increased by the high-fat diet.
Pregnenolone-16α-carbonitrile treatment had no effect on plasma cholesterol. In contrast, the liver cholesterol level was considerably higher in the pregnenolone-16α-carbonitrile-treated, high-fat diet-fed mice compared with the vehicle-treated, high-fat diet-fed mice ( Figure 4a).
To further investigate the origin of the increased liver cholesterol, we quantified markers of cholesterol synthesis (Figure 4b) in the plasma and liver and markers of cholesterol absorption in the plasma (Björkhem et al., 1987;Miettinen et al., 1989). In line with the human results, lathosterol to cholesterol ratio was increased both in the plasma and livers of the pregnenolone-16α-carbonitrile-treated mice suggesting that pregnenolone-16α-carbonitrile has no effect on cholesterol absorption. A previous study reported that Niemann-Pick C1-like 1 (Npc1l1) and microsomal triglyceride transfer protein large subunit (Mttp) are intestinal PXR target genes, whose activation leads to increased cholesterol absorption ). In the current study, pregnenolone-16α-carbonitrile treatment increased the intestinal expression of Cyp3a11 but did not affect Npc1l1, Mttp or Apob expression ( Figure S3a,b). Instead, pregnenolone-16αcarbonitrile modestly induced Abcb1a and Abcg5 involved in cholesterol excretion to the intestinal lumen ( Figure S3c) (Temel & Brown, 2015).
Altogether, these results clearly indicate that the pregnenolone-16α-carbonitrile treatment induces hepatic cholesterol synthesis, especially the Kandutsch-Russell pathway, but has no increasing effect on cholesterol absorption. Synthesis of bile acids represents a major route of cholesterol disposal. However, the high-fat diet did not affect total plasma bile acid  (Li & Chiang, 2005), but rifampicin dosing did not affect human hepatic CYP7A1 expression in vivo (Marschall et al., 2005). In the current experiment in the obese mice, pregnenolone-16α-carbonitrile treatment did not lead to statistically significant repression of Cyp7a1 (Figure 4f). Furthermore, bile acids activate farnesoid X receptor (FXR), but the expression of hepatic FXR target genes was not altered by the pregnenolone-16α-carbonitrile treatment ( Figure S4a).

The Bloch pathway intermediate desmosterol is a liver X receptor
(LXR) ligand (Spann et al., 2012;Yang et al., 2006). In agreement with the lack of induction of the desmosterol levels by pregnenolone-16αcarbonitrile, the expression of hepatic LXR target genes was not altered in the liver ( Figure S4b,c,d).

| PXR activation induces the superpathway of cholesterol biosynthesis in the livers of high-fat dietfed mice
To further elucidate the mechanisms mediating the pregnenolone-

| Pregnenolone-16α-carbonitrile treatment activates SREBP2 pathway in the liver
To further characterize the mechanisms mediating the induction of genes in the cholesterol biosynthesis pathway by PXR activation, we utilized Ingenuity Pathway Analysis software to predict the putative upstream regulators. The prediction of the upstream regulators of the 442 genes differentially regulated by pregnenolone-16α-carbonitrile is based on activation z-scores, which take into account the direction of gene expression, i.e., induction or repression.
As expected, PXR was among the most activated predicted upstream regulators (Figure 6a). Furthermore, the five most activated F I G U R E 6 Pregnane X receptor (PXR) induces cholesterol synthesis and PCSK9 through the sterol regulatory element-binding protein 2 (SREBP2) pathway. (a) Top five upregulated and downregulated predicted liver upstream regulators in high-fat diet (HFD)-fed and pregnenolone-16α-carbonitrile (PCN)-treated mice compared with vehicle control based on liver RNAseq data, p values of overlap and activation z scores (n = 3/group). (b) Liver qPCR of cholesterol synthesis regulators in wild-type (n = 8/group) and PXR-KO mice (n = 6/group). (c) Immunoblotting of SREBP2 in the liver nuclear fraction (chow n = 10; HFD n = 9; HFD vehicle = 7; HFD PCN n = 8. HFD vehicle group lacked one sample due to shortness of material). (d) Immunoblotting of liver insulin-induced gene 1 (INSIG1) (chow n = 10; HFD n = 9; HFD vehicle n = 8; HFD PCN n = 8). (e) Immunoblotting of INSIG1-antibody-reactive protein X. (chow n = 10; HFD n = 9; HFD vehicle n = 8; HFD PCN n = 8). (f) qPCR measurement of Ldlr and Pcsk9 mRNAs in the wild-type (PXR-WT; n = 8/group) and the PXR knockout (PXR-KO; n = 6/group) mice livers. (g) Effect of PXR ligands on the PCSK9 level in the mouse and the human plasma (mouse n = 8/group; human n = 34). The values represent mean ± SD. *P < .05 wild-type but not in the PXR-KO mice (Figure 6b). Therefore, the mRNA expression data did not support the Ingenuity Pathway Analysis software predictions of the upstream regulators.
Western blot analyses were performed to further clarify the activation status of the SREBP2 pathway. SREBP2 is activated through proteolytic processing in the Golgi and subsequent transport of the activated cleavage product to the nucleus (Shimano & Sato, 2017).
Thus, the nuclear accumulation of SREBP2 reflects the activation of the pathway. Remarkably, the pregnenolone-16α-carbonitrile-treated mice had fivefold higher level of SREBP2 in the hepatic nuclear protein fraction than the vehicle control clearly indicating activation of the pathway (Figure 6c). Interestingly, the nuclear level of sterol regulatory element-binding protein 1 (SREBP1) or SREBP1 target genes were not induced by pregnenolone-16α-carbonitrile ( Figure S5a,b).

| PXR activation increases plasma PCSK9
Besides cholesterol synthesis genes, SREBP2 regulates genes involved in the hepatic LDL uptake: LDL receptor (Ldlr) and Pcsk9 (Lagace, 2014). The LDLR mediates the hepatic uptake of LDL and PCSK9 is the major negative regulator of LDLR. Both Ldlr and Pcsk9 mRNAs were induced by pregnenolone-16α-carbonitrile in the wildtype but not in the PXR-KO mice ( Figure 6f) and especially Pcsk9 was strongly induced, about fourfold. Moreover, pregnenolone-16αcarbonitrile increased mouse plasma PCSK9 more than sixfold compared with the vehicle-treated mice (Figure 6g).
To evaluate the translational significance of the PCSK9 finding, we measured the effect of rifampicin administration on PCSK9 serum levels in humans. Similar to the mice, PXR ligand induced the serum level of PCSK9 compared with the placebo arm of the study (Figure 6g).

| PXR deficiency alters the cholesterogenic gene response to high-fat diet
In agreement with the previous studies (Getz & Reardon, 2006), we observed an increase in the plasma cholesterol levels in response to high-fat diet. Because PXR deficiency was previously shown to protect mice from obesity and obesity-induced metabolic impairments (He et al., 2013), we investigated the effect of PXR-KO on the cholesterol homeostasis in response to high-fat diet. In contrast to the previous studies (He et al., 2013;Spruiell et al., 2014), PXR deficiency did not protect against the high-fat diet-induced obesity and the weight gain was similar in the PXR-KO and the wild-type mice (Figure 7a).
High-fat diet repressed the classical PXR target gene Cyp3a11 in the wild-type but not in the PXR-KO mice (Figure 7b) indicating that Cyp3a11 repression by the high-fat diet involves PXR and that the high-fat diet modifies PXR function. In RNA sequencing, high-fat diet had a rather modest effect on cholesterogenic genes in the wild-type mice and none of the members of the superpathway of cholesterol biosynthesis were differentially regulated. However, in the qPCR measurements farnesyl diphosphate synthetase gene (Fdps) was induced about twofold (Figure 7b). The Ldlr was induced 1.6-fold, whereas the Pcsk9 was not affected by the high-fat diet. Factors affecting SREBP2 signalling were not altered except Insig2, which was slightly repressed.
The Fdps and Ldlr induction, seen in the high-fat diet-fed wild-type mice, was abolished by the PXR deficiency ( Figure 7b). Thus, PXR appears to be involved in the regulation of the SREBP2 pathway even in the absence of an exogenous ligand.
High-fat diet increased plasma PCSK9 level 3.7-fold in the wildtype mice. The PXR deficiency abolished the effect of high-fat diet, however the PCSK9 level was higher in the PXR-KO mice than in the wild-type mice (Figure 7c).

| DISCUSSION
In the current study, we show that treatment with rifampicin, a human pregnane X receptor (PXR) ligand, raises serum IDL and LDL of all sizes and serum total, esterified and free cholesterol. Furthermore, rifampicin increases the lathosterol to cholesterol ratio suggesting upregulation of cholesterol synthesis. This harmful effect of rifampicin on serum parameters is compounded by increases in sphingomyelins and PCSK9, other risk factors for cardiovascular diseases (Schlitt et al., 2006;Seidah et al., 2014).
Previous studies in patients treated with PXR-activating drugs including several antiretroviral drugs and atypical antipsychotic quetiapine have shown increased cholesterol levels during treatment (de Leon et al., 2007;Estrada & Portilla, 2011). Furthermore, the antiepileptics with PXR-activating properties, that is, phenytoin, phenobarbital and carbamazepine, induce cholesterol levels, whereas valproic acid with no affinity to PXR does not (Eirís et al., 2000;Luoma et al., 1979;Müjgan Aynaci et al., 2001). Importantly, phenobarbital induces human hepatic HMGCR protein in vivo (Coyne et al., 1976). Cafestol, a PXR ligand found in unfiltered coffee, raises blood cholesterol and is a risk factor for cardiovascular diseases (Ricketts et al., 2007;Weusten-Van der Wouw et al., 1994). Additionally, PXR gene polymorphisms associate with plasma LDL cholesterol levels in humans (Lu et al., 2010). Despite mounting epidemiological evidence linking PXR to cholesterol, the controlled human clinical studies are scarce.
Effect of rifampicin on cholesterol synthesis has previously been studied in one small clinical trial with ten healthy volunteers (Lütjohann et al., 2004). Although the serum total cholesterol level was not affected, rifampicin treatment increased the lathosterol to cholesterol ratio suggesting increase in cholesterol synthesis in agreement with our current study. Furthermore, a study with daily cholesterol measurement for 30 days reported that 14-day rifampicin dosing led to 9.5% increase in total cholesterol with gradual decrease to baseline after stopping the treatment (Kasichayanula et al., 2014).
The effect was not statistically significant due to small sample size (n = 12) but it resembles our result (7% increase). A few small studies have observed no change in total cholesterol during rifampicin treatment (Feely et al., 1983;Ohnhaus et al., 1979). Our studies form the largest clinical study set investigating the effect of rifampicin on cholesterol homeostasis and the only one reporting systemic metabolomic changes.
Rifa-1 and Rifa-BP studies utilized in the current investigation had identical crossover setting. The oral glucose tolerance test study (Rifa-2) had a one-arm design with oral glucose tolerance test F I G U R E 7 Cholesterol synthesis genes and Pcsk9 are repressed in the pregnane X receptor (PXR)-KO mice on high-fat diet. (a) Effect of highfat diet (HFD) on weight gain and percentual weight gain in the wild-type (chow n = 10; HFD n = 25) and the PXR knockout mice (chow n = 6; HFD n = 18). (b) Effect of PXR deficiency on mRNA expression of PXR target genes and key genes controlling or being involved in cholesterol synthesis during HFD (WT, chow n = 10; HFD n = 9 and PXR-KO mice, n = 6/group). (c) Effect of HFD on plasma PCSK9 levels in the PXR-WT (chow n = 10; HFD n = 9) and the PXR-KO mice (n = 6/group). (d) Effect of HFD on plasma cholesterol levels in the PXR-WT (chow n = 10; HFD n = 9) and the PXR-KO (n = 6/group) mice. The values represent mean ± SD. *P < .05 performed before and after 1-week rifampicin dosing. All three studies were performed on healthy young individuals with no overlapping medications or confounding illnesses. The metabolomic screen was systematic in nature and differences were first and foremost seen in the cholesterol metabolism, as for example triglyceride levels remained unaltered. In agreement with the hypothesis that the changes in serum cholesterol by rifampicin are PXR dependent, serum 4-β-OH-cholesterol, a marker of PXR activation (Diczfalusy et al., 2009), correlated with increases in serum total cholesterol. Cholesterol in the blood originates from either the diet through intestinal absorption or from de novo synthesis occurring mostly in the liver. Rifampicin decreased citrate and acetate consumed in cholesterol synthesis and increased lathosterol to cholesterol ratio, an indicator of cholesterol synthesis suggesting that rifampicin mainly affects cholesterol biosynthesis. Although citrate and acetate are also consumed in other metabolic, PXR-regulated pathways such as fatty acid synthesis, lathosterol is a well-established marker of cholesterol synthesis (Björkhem et al., 1987;Hakkola et al., 2016). The levels of remnant cholesterol and ApoB were significantly elevated by rifampicin only in the Rifa-2 study. The elevation of ApoB (the protein component of VLDL, IDL and LDL) is of importance because cholesterol is secreted from the liver in VLDL particles that are metabolized to IDL and further to LDL (Laufs et al., 2019). Indeed, using high-fat diet-fed mouse as a model, we established that PXR activation causes induction of hepatic cholesterol synthesis, but it does not enhance cholesterol absorption.
Previously, one cholesterol biosynthesis gene, squalene epoxidase (Sqle), was shown to be a direct PXR target ). In the current study, we observed widespread induction of the genes in the cholesterol synthesis superpathway. Experiments in the PXR-KO mouse indicated that PXR was necessary for the pregnenolone-16αcarbonitrile-mediated induction of the cholesterol synthesis genes.
Although not formally excluded, we consider it unlikely that all the cholesterol synthesis genes upregulated by pregnenolone-16αcarbonitrile would be direct PXR targets. Instead, we observed activation of SREBP2, a master regulator of the cholesterol synthesis (Shimano & Sato, 2017).
The mRNA expression of Srebp2 was not affected by PXR activation. Indeed, the protein processing is the major step controlling SREBP2 activity. SREBP2 transcriptional activity requires processing in Golgi and proteolytic release of the active fragment entering nucleus. INSIG1 is an endoplasmic reticulum (ER)-retention membrane protein hindering SREBP2 transfer to Golgi. Sterol level regulates INSIG1 stability and subsequently SREBP2 transfer and processing. In a situation of low sterol content, INSIG1 is degraded while SREBP2 is cleaved and translocated to the nucleus to induce its target genes (Shimano & Sato, 2017 (Spann et al., 2012;Yang et al., 2006). Interestingly, we observed a shift towards Kandutsch-Russell pathway and no change in the desmosterol levels. This mechanism may also contribute to the lack of SREBP2 inhibition.
Another key finding of the current study is the increase in the circulating PCSK9 after PXR activation. Indeed, the same effect could be observed both in the mice and humans. It should be noted as a limitation to the study that the circulating PCSK9 level was measured in the mouse samples 2 hr after glucose treatment while the human measurements were made after fasting; however, the results were still in agreement. SREBP2 is the major regulator of Pcsk9 gene and the upregulation of Pcsk9 expression further substantiates the role of SREBP2 in the PXR-induced effects on cholesterol homeostasis (Lagace, 2014). The similar finding in the mice and humans supports the hypothesis that the SREBP2 pathway mediates the PXR-elicited alterations on cholesterol homeostasis in humans, too.
PCSK9 has been proven a very important player in cholesterol homeostasis. PCSK9 is strongly linked to the levels of LDL-cholesterol and to the incidence of cardiovascular diseases (Seidah et al., 2014). Therefore, it is plausible that together with the increased cholesterol synthesis, also the impaired hepatic LDL uptake due to increased PCSK9 contributes to the elevated LDL cholesterol. Figure 8 summarizes the mechanisms revealed in the current study and mediating the effect of PXR activation on cholesterol and lipoprotein homeostasis.
Statins decrease intracellular cholesterol content and, in response, activate SREBP2 to induce PCSK9 levels in the plasma. To fight against this harmful compensatory mechanism, drugs inhibiting PCSK9 are used to treat hypercholesterolemia in combination with statins (Seidah et al., 2014;Warden et al., 2019). A recent meta-analysis reported that lipophilic statins (atorvastatin, simvastatin) elevate PCSK9 levels more than hydrophilic statins (rosuvastatin, pravastatin) (Sahebkar et al., 2015). Because the lipophilic statins activate PXR (Howe et al., 2011), it can be speculated that PXR plays a role in the greater PCSK9 induction potential of the lipophilic statins. Rifampicin now joins the very short list of drugs known to elevate PCSK9 (statins and fibrates) with rifampicin being the most efficient (about 1.7-fold PCSK9 elevation) (Glerup et al., 2017). Several other drugs used in long-term treatments are PXR ligands (Hukkanen, 2012). Whether these PXR-activating drugs induce PCSK9 plasma levels in the clinical practice requires further investigation.
We also assessed the effect of PXR deficiency on high-fat dietinduced hypercholesterolemia and on the cholesterogenic mechanisms. Contrary to the previous reports (He et al., 2013;Spruiell et al., 2014), PXR deficiency did not protect against the high-fat dietinduced obesity. The reason for this discrepancy is unknown but may be related to mouse strains. The previous studies were performed in the C57BL/6 J strain, but we utilized the C57BL/6 N background instead. The two strains differ by a mutation in the NAD(P) transhydrogenase gene affecting glucose homeostasis (Toye et al., 2005) and they differ in weight gain during high-fat diet (Nicholson et al., 2010). Nevertheless, the unaltered weight gain allowed us to directly assess the effect of PXR deficiency on cholesterol metabolism without the confounding effect of the weight.
Although both high-fat diet alone and the pregnenolone-16αcarbonitrile treatment in the high-fat diet background increased cholesterol synthesis in the wild-type mice, the mechanisms appear to be different. The high-fat diet treatment induced the liver squalene level, whereas the pregnenolone-16α-carbonitrile treatment tended to decrease squalene. Furthermore, pregnenolone-16α-carbonitrile activated the Kandutsch-Russell cholesterol synthesis pathway, whereas high-fat diet alone did not. Importantly, high-fat diet did not increase HMGCR protein level or the nuclear SREBP2, although pregnenolone-16α-carbonitrile combined with high-fat diet stimulated strong induction of both. Curiously, the PXR-KO modified the gene response to high-fat diet. The induction of Fdps and Ldlr by high-fat diet was abolished; the mRNAs of Srebp2 as well as SREBP2 target genes Hmgcr, Cyp51 and Pcsk9 were repressed. These findings indicate that even in the absence of any exogenous ligand, PXR modulates the SREBP2 pathway. PXR deficiency abolished the high-fat diet-induced upregulation of plasma PCSK9 but was not sufficient to prevent upregulation of plasma cholesterol level suggesting that PXR inhibition may not be a viable approach to prevent diet-induced hypercholesterolemia.
In summary, we show for the first time in controlled clinical studies that PXR activation induces cholesterol synthesis and elevates LDL and total cholesterol in humans. In addition, we report that PXR activation increases circulating PCSK9. On the basis of the experimentation in murine models, we propose that PXR induces hepatic cholesterol biosynthesis and Pcsk9 expression through activation of SREBP2 by enhancing its proteolytic processing. Altogether, these results reveal a novel mechanism controlling human cholesterol and lipoprotein homeostasis and establish a molecular mechanism for druginduced hypercholesterolemia.
F I G U R E 8 Mechanism of pregnane X receptor (PXR) mediated regulation of cholesterol and lipoprotein homeostasis. PXR activation promotes sterol regulatory element-binding protein 2 (SREBP2) proteolytic activation resulting in transcriptional activation of SREBP2 target genes: (1) widespread induction of cholesterol synthesis genes including the rate limiting Hmgcr and (2) Pcsk9, regulating the amount of LDL receptor in the hepatocyte plasma membrane. PXR activation increases cholesterol synthesis through the Kandutsch-Russell pathway and not through the Bloch pathway. While increase of the cholesterol synthesis is expected to launch feedback inhibition through insulin-induced gene 1 (INSIG)-mediated inhibition of SREBP2 proteolytic activation, this negative feedback pathway appears not to function under PXR activation. Altogether, these mechanisms result in increased level of atherogenic lipids

DATA AVAILABILITY STATEMENT
The complete RNA sequencing data sets are available at the NCBI's Gene Expression Omnibus (GEO) database with accession number GSE136667. The other data that support the findings of this study are available from the corresponding author upon reasonable request.
Some data may not be made available because of privacy or ethical restrictions.