Background

Elevated plasma low-density lipoprotein cholesterol (LDLC) levels are a major risk factor for cardiovascular disease (CVD), the leading cause of death worldwide. Statins lower plasma LDLC levels and CVD risk by inhibiting cholesterol biosynthesis and stimulating receptor-mediated LDL uptake from the blood. Although statins are effective for primary and secondary prevention of atherosclerotic CVD [1], not everyone receives the same lipid-lowering benefit from statin therapy, and many individuals still experience CVD events while taking statins. Much of the inter-individual variation in statin efficacy remains unexplained to date.

Previous studies have identified some factors (e.g., age, smoking status, and race/ethnicity) associated with LDLC statin response [2]. In addition, genome-wide association studies (GWAS) have identified a handful of genetic loci associated with LDLC response [3,4,5,6], but together these only explain about 5% of LDLC response variation [5]. Though expanding sample sizes of participants with suitable pre- and on-drug phenotype data could bolster power for statin response GWAS, other approaches beyond GWAS could also be used to identify genetic factors that influence statin efficacy.

Statins elicit a strong transcriptional response via regulation of sterol response element binding factor 2 (SREBF2, aka SREBP2) [7] and other factors, so we and others have studied statin-induced gene expression changes to reveal novel genes involved in the modulation of clinical statin response [8,9,10,11]. To do this, we have used statin clinical trial participant-derived lymphoblastoid cell lines (LCLs) as an in vitro genetic model in which to study statin response transcriptomics. Here, we conducted a large-scale study correlating in vitro gene expression statin response with in vivo LDLC statin response to identify candidate statin efficacy modulators. We then validated a top candidate gene in vivo in a mouse model.

Methods

CAP study participants

This study used clinical and demographic information and cell lines derived from 426 participants in the Cholesterol and Pharmacogenetics (CAP) 40 mg/day 6-week simvastatin clinical trial [2] (ClinicalTrials.gov Identifier: NCT00451828). Written informed consent was obtained from all participants, and the study was approved by the institutional review boards of the University of California, San Francisco and the University of California, Los Angeles. Plasma lipids were measured in two baseline (pre-statin) blood samples and two on-statin blood samples collected 4 and 6 weeks after statin initiation, and the two values were averaged in each case to obtain a single baseline and on-statin value per participant. LDLC values were calculated from total cholesterol, HDL-cholesterol, and triglyceride levels using the Friedewald formula [12]. Characteristics of this CAP subset are shown in Table 1. In addition, lipoprotein particle profiles were measured and analyzed in the second baseline sample and the 6-week on-statin sample from most CAP participants (n = 803) using ion mobility (described below).

Table 1 Characteristics of 426 CAP participants with LCL RNA-seq data
Full size table

Lymphoblastoid cell line statin exposure

CAP lymphoblastoid cell lines (LCLs) were established by immortalizing blood-derived B lymphocytes through Epstein-Barr virus (EBV) transformation, cultured in RPMI 1640 with 10% FBS, and exposed to 2 µM activated simvastatin or sham control buffer for 24 h in batches ≤ 12 cell lines, similar to previously described [13]. RNA was extracted from homogenized cellular lysates and RNA integrity was measured on an Agilent Bioanalyzer, as previously described [14].

RNA-sequencing and analysis

PolyA-selected, strand-specific [15] paired-end RNA-seq libraries were prepared, and fragments were sequenced with 100–101 bp read lengths on Illumina machines at the University of Washington Northwest Genomics Center or the Baylor College of Medicine Human Genome Sequencing Center, as previously described [14, 16]. Sequences were aligned to the human (hg19) and EBV (NC_007605) genomes using Tophat v2.0.4 [17] with Ensembl v67 [18] and EBV [19] transcriptome annotations, and sequence fragments falling within annotated genes were counted using HTSeq [20] as previously described [14, 16]. DESeq2 was used to adjust for library size and to variance stabilize gene expression levels using a log2-based transformation [21], and statin-induced changes in gene expression levels were calculated by subtracting expression levels in control-treated LCLs from the statin-treated expression levels of the same cell line.

After adjusting CAP participants’ in vivo plasma LDLC statin responses [ln(on-statin LDLC) − ln(baseline LDLC)] for age, race/ethnicity, and smoking status using linear regression, the resulting residuals were correlated to statin-induced changes in gene expression in the corresponding LCLs using Spearman’s rank-order correlation. p values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) approach [22].

Mouse studies

All mice in this study were housed under a 12-h light/dark cycle and had free access to food and water except when fasted as noted. Mice were fed a general grain-based chow diet prior to study initiation (irradiated Teklad global 18% protein rodent diet from Envigo, product 2918). C57BL/6J mice carrying an N-ethyl-N-nitrosourea (ENU)-induced hypomorphic missense mutation (R1092W, aka “bloto”) in Zfp335 (available from JAX as stock #025853) were generated previously [23] and were backcrossed to C57BL/6J wild-type mice for many generations. For the first study, animals were fed the general chow diet until they were fasted for 4–6 h and sacrificed at 13 weeks of age. Blood was collected via cardiac puncture, and plasma was isolated. Plasma lipid levels and lipoprotein particle concentrations were measured using Liasys and ion mobility, respectively, described below. Growth curve analyses were performed using weekly weights obtained after weaning, and growth curves of mice with different Zfp335 genotypes were compared within sexes. Results were analyzed statistically and adjusted for multiple testing using the CGGC (Compare Groups of Growth Curves) method with 10,000 permutation tests [24, 25].

In our second study, mice were transitioned to the AIN-76A purified rodent diet (ResearchDiets product D10001i) at 13 weeks of age, which contains approximately 5% fat and no added cholesterol, and weighed weekly for the duration of the study. After mice were fed the control purified diet for 4 weeks, they were transitioned to a matched diet with the addition of 1 g simvastatin/kg diet (9.77% active; ResearchDiets product D14060903i) for 4 weeks. Plasma was obtained from mice at 15, 17, and 19 weeks of age using non-terminal retro-orbital bleeds after a 4–6-h fast. Mice were sacrificed at 21 weeks of age after a 4–6-h fast, and plasma was isolated from blood collected via cardiac puncture as in the first study.

Mouse studies were approved by the Children’s Hospital & Research Center Oakland and the University of California, San Francisco Institutional Animal Care and Use Committees.

Plasma lipid analysis

Total cholesterol (TC) and high-density lipoprotein cholesterol (HDLC) were measured in human and mouse plasma samples on an AMS Liasys 330 Clinical Chemistry Analyzer, similar to previously described [26, 27]. Mouse non-HDL cholesterol (non-HDLC) levels were calculated by subtracting HDL from total cholesterol levels. In the chow-fed mouse experiment, HDLC was not successfully measured in samples from one male heterozygote mouse, and 1 female Zfp335R1092W homozygote was a statistical (Tukey) outlier for HDLC (low) and non-HDLC (high) and was excluded from analyses. Plasma samples from 4 mice (1 male wild-type, 1 male heterozygote, 1 female heterozygote, and 1 male Zfp335R1092W homozygote) in the statin response experiment for which non-HDLC was calculated to be ≤ 0 and/or fasting total cholesterol levels were abnormally low (≤ 50 mg/dl) were Tukey outliers and were excluded from subsequent analyses. Additive relationships were tested for association using linear regression, and comparisons between wild-type and mutant homozygotes were conducted statistically using t-tests.

Ion mobility (IM) was used to measure human CAP participant and mouse plasma lipoprotein particle concentrations as a function of particle size, as previously described [28, 29]. Of the 25 male and 33 female mice for which IM data was generated before and on the statin diet, nine were statistically significant LDL particle concentration outliers (Tukey) and excluded from subsequent IM data analysis, leaving data from 22 male and 27 female mice. Specifically, five mice (1 female wild-type, 1 female heterozygote, 1 female Zfp335R1092W homozygote, and 2 male wild-type) had abnormally high pre-statin LDL particle concentrations, three mice (1 female wild-type and 2 female heterozygotes) had abnormally high on-statin LDL particle concentrations, and one (male wild-type) mouse had abnormally low on-statin LDL particle concentrations. Data was binned into lipoprotein subfraction categories based on particle diameter definitions previously determined using human studies and comparisons with other methods [29]. Two-way ANOVA with Tukey’s multiple comparisons test was used to compare statin-induced changes in LDL subfraction particle concentrations across mouse genotype categories.

Results

Transcriptomic statin response in LCLs

Statin induced a robust transcriptional response in the 426 CAP LCLs, in agreement with our previous studies [9, 14] (Fig. 1, Table S1). Of the 14,028 well-expressed human and EBV genes tested, the expression level of 9092 genes (65%) changed significantly with statin treatment (paired t-test, FDR-adjusted p < 0.0001, 4671 upregulated and 4421 downregulated), though the degree of change was modest for most genes. As expected, many of the most robustly upregulated genes were known SREBF2 target genes.

Fig. 1
figure 1

Volcano plot illustrating gene expression statin response in 426 CAP LCLs. Approximate log2 fold change is 2^(variance stabilized statin gene expression level − variance stabilized control gene expression level) calculated for each cell line and averaged across cell lines

Full size image

Correlation of in vitro gene expression statin response with in vivo LDLC statin response

To identify genes that may modulate plasma LDLC statin response, we correlated transcriptome-wide statin-induced changes in LCL gene expression in vitro with plasma LDLC statin responses of the corresponding CAP clinical trial participants from whom the LCLs were derived. At a false discovery rate of 5%, 147 of 13,955 human genes tested had statin-induced expression changes that were significantly correlated with in vivo LDLC statin response (Table S2). The top 10 genes, which all had an FDR-adjusted p < 0.03, are shown in Table 2.

Table 2 Genes with LCL expression changes most correlated to plasma LDLC statin response
Full size table

The top two genes, zinc finger protein 335 (ZNF335) and CCR4-NOT transcription complex subunit 3 (CNOT3), appeared to be most worthwhile for follow-up due to the significance of their correlation relationships relative to those of other genes after multiple testing adjustment. ZNF335 (aka NIF-1) has been shown to physically interact with two other members of the CCR4-NOT complex, CNOT6 and CNOT9 [30], suggesting that ZNF335 and CNOT3 may be functionally related. ZNF335 and CNOT3 also exhibited similar correlation directionality (Fig. 2). For both genes, statin decreased expression in most LCLs. In addition, the LCLs that exhibited greater statin-induced reductions in gene expression tended to be derived from participants who experienced larger statin-induced reductions in plasma LDLC in the CAP clinical trial (Fig. 2A and B). Though no other populations were available for replication of these findings, we did see good agreement between the two independent experimental batches [(batch 1: N = 149; ZNF335 rho = 0.275, p = 7.0 × 10−4; CNOT3 rho = 0.316, p = 8.6 × 10−5); (batch 2: N = 277; ZNF335 rho = 0.200, p = 8.0 × 10−4; CNOT3 rho = 0.154, p = 0.010)]. Notably, the statin-induced changes in CNOT3 and ZNF335 were themselves highly correlated (rho = 0.625, p = 1.4 × 10−47, Fig. 2C). Other genes in the CCR4-NOT complex did not have gene expression statin responses that were significantly correlated to plasma LDLC response after adjustment for multiple testing, and their non-significant correlations with plasma LDLC response were negative, displaying opposite directionality to the positive CNOT3 and ZNF335 correlations (Table S2).

Fig. 2
figure 2

Correlation of ZNF335 and CNOT3 LCL gene expression statin responses with in vivo plasma LDLC statin responses of the corresponding donors (A–B) and with each other (C)

Full size image

Plasma cholesterol levels in Zfp335 hypomorphic mice

In mice, knocking out Cnot3 or Zfp335 (the murine ZNF335 homolog) results in embryonic lethality [31, 32]. Previous studies have shown that Cnot3+/− (haplodeficient) mice have altered lipid and glucose metabolism [33], and liver-specific Cnot3 knockout (Cnot3LKO) mice have altered lipid metabolism, among other deficits [34]. While ZNF335 has been reported to play a role in neurogenesis [32], to date the effects of Zfp335 disruption on lipid and cholesterol metabolism have not previously been studied. We chose to investigate plasma cholesterol levels and statin response in available mice carrying an ENU-induced hypomorphic missense mutation in Zfp335 (R1092W aka bloto) that was previously studied in the context of T lymphocyte maturation [23].

We first compared fasting plasma cholesterol levels of 13-week-old Zfp335R1092W/R1092W, Zfp335R1092W/+, and wild-type mouse littermates fed a grain-based chow diet. In male mice, increasing copies of Zfp335R1092W mutations significantly reduced total cholesterol (TC) and HDL cholesterol (HDLC) levels (Fig. 3, additive p = 0.01 and p = 0.0002, respectively). As expected based on previous studies, female mice had substantially lower HDL (and total) cholesterol levels than male mice across all Zfp335 genotype categories. However, associations of Zfp335 genotype with plasma TC or HDLC were not observed in female mice, with sex*Zfp335 genotype interactions of p = 0.15 and p = 0.015, respectively. Since we observed a significant sex interaction for HDLC levels in the mice, we went back to our CAP participant data and observed a modest association of ZNF335 levels in control-treated LCLs with CAP baseline HDLC in men (p = 0.048) but no association in women, with a weak sex*ZNF335 interaction (p = 0.09, Fig. S1). The directionality of the correlation in men seems consistent with the male mice as well, since lower human LCL gene expression and more copies of the mouse hypomorphic Zfp335 allele would both be presumed to have less protein activity. There was no correlation of baseline HDLC with control-treated LCL ZNF335 levels in a sex-combined model, and there was no correlation of baseline LDLC with control-treated LCL ZNF335 levels in either sex subset or a sex-combined model.

Fig. 3
figure 3

13-week old chow-fed mouse plasma A total, B HDL, and C non-HDL cholesterol levels split by sex and Zfp335 genotype. Male sample sizes were N = 12, 15 (16 for TC), and 10, and female sample sizes were N = 17, 15, and 5 (6 for TC) for wild-type, heterozygotes, and homozygous Zfp335R1092W, respectively. Values are mean ± SD. *p < 0.05, ***p < 0.001

Full size image

In contrast to HDLC, there were not significant sex differences in the relationship between mouse Zfp335 genotype and non-HDL cholesterol (TC − HDLC). Zfp335 genotype was significantly correlated with non-HDLC in a sex-combined model (additive p = 0.04) with sex included as a covariate. Though not statistically significant, male mice with one or two copies of the Zfp335R1092W mutation had lower mean levels of non-HDLC compared to wild-type littermates. In female mice, the strongest relationship of genotype with cholesterol levels was for non-HDLC, with a trend for reduced levels with increasing numbers of Zfp335R1092W mutations (additive p = 0.08).

Weight and growth curves

Since we observed that Zfp335R1092W/R1092W mice appeared smaller on average than their littermates and this phenotype had not been documented for this mutant mouse strain before, we measured their growth curves from 4 to 11 weeks of age on a grain-based chow diet (Fig. S2). Both male (adj. p = 0.0036) and female (adj. p = 0.036) Zfp335R1092W/R1092W mice had significantly reduced growth curves compared to their wild-type littermates. For males (which had a larger sample size than females), Zfp335R1092W/+ animals had growth curves that were intermediate between Zfp335R1092W/R1092W and Zfp335+/+ males (heterozygote vs. Zfp335R1092W/R1092W adj. p = 0.0021, heterozygote vs. wild-type adj. p = 0.0498), with females trending similarly (adj. p = 0.098 for both comparisons).

LDL statin response in Zfp335 hypomorphic mice

We next wanted to determine if statin efficacy differed between wild-type mice and mice carrying Zfp335R1092W mutations. To do this, we first fed mice a defined “control” diet (AIN-76A) for 4 weeks, starting at 13 weeks of age. We then transitioned mice to a matched diet supplemented with simvastatin for 4 weeks. We first measured statin-induced changes in non-HDLC as a crude proxy for mouse LDLC statin response, though non-HDLC also includes cholesterol found in other lipoprotein particles. Pre-statin non-HDLC levels (the average of the 15- and 17-week samples; 2 and 4 weeks on AIN-76A diet) were associated with Zfp335 genotype in male animals alone (additive p = 0.022, Fig. 4A) and in a sex-combined model with sex included as a covariate (additive p = 0.029), with a similar but non-significant trend in female animals alone (Fig. 4B).

Fig. 4
figure 4

Plasma non-HDL cholesterol levels in A male and B female mice before and during statin diet feeding and C statin-induced changes in non-HDL cholesterol split by sex and Zfp335 genotype. Male sample sizes were N = 7, 15, and 4 and female sample sizes were N = 12, 11, and 5 for wild-type, heterozygotes, and homozygous Zfp335R1092W, respectively. Values are mean ± SD. *p < 0.05

Full size image

Non-HDLC levels were reduced after 4 weeks of statin treatment in wild-type mice of both sexes, with wild-type male mice experiencing greater (inter-sex comparison p < 0.05) statin-induced reductions in non-HDLC (Fig. 4A and C, − 43 ± 18%, paired t-test p = 0.0016, N = 7) than wild-type females (Fig. 4B and C, − 23 ± 19%, paired t-test p = 0.0020, N = 12). Female Zfp335R1092W/R1092W mice (N = 5) had a significantly blunted non-HDL cholesterol statin response compared to their wild-type littermates (Fig. 4C, p = 0.02, t-test), but this trend was not as strong in male animals.

Lipoprotein subfraction statin response

Non-HDL cholesterol is a broad category that includes cholesterol content from lipoprotein particles of various sizes and types. Since LDLC response was the phenotype for which we saw a correlation with ZNF335 gene expression statin response in humans, we next performed a more detailed analysis of lipoprotein particles in the LDL size range. We therefore compared the ion mobility-based lipoprotein profiles from wild-type mice before and after they were fed 4 weeks of statin diet (Fig. 5 and Fig. S3). As has been well documented previously, most mouse lipoproteins were in the HDL size range, with much lower concentrations in the size ranges spanning LDL and intermediate density lipoproteins (IDL). Since the lipoprotein subfraction size categorization we used is based on human data, it may not always accurately reflect the type(s) of particles in the corresponding size intervals in mice. In mice, particles in the human very small LDL (designated LDL IV) particle size range comprise a heterogeneous mixture of species overlapping very large HDL and containing APOE-only lipoproteins [27], so we focused our analyses on small (LDL IIIb [20.49–20.82 nm] and LDL IIIa [20.82–21.41 nm]), medium (LDL IIb [21.41–22 nm] and LDL IIa [22–22.46 nm]), and large (LDL I [22.46–23.33 nm]) LDL size ranges. Wild-type males were the only mice that exhibited robust statin-induced decreases in these LDL particles (Figs. 5A, 6, S3, and S4). For comparison, we also included pre-statin and on-statin human lipoprotein profiles from CAP clinical trial participants (Fig. S5). Similar to male mice, men experienced a significant reduction in lipoprotein particle concentrations in the medium and large LDL to small IDL size range after statin treatment, but unlike the sex difference in statin response in the mice, there were similar reductions in levels of these particles in women.

Fig. 5
figure 5

Statin-induced changes in wild-type A N = 5 male and B N = 11 female mouse lipoprotein profiles measured by ion mobility. Mouse profiles were measured before and after 4 weeks of simvastatin-containing diet. The size intervals designating the major lipoprotein subclasses are based on those defined in humans [29]

Full size image
Fig. 6
figure 6

Statin-induced changes in male mouse LDL particle concentrations as measured by ion mobility. Sample sizes were N = 5, 12, and 5 for wild-type, heterozygotes, and homozygous Zfp335.R1092W, respectively. Values are mean ± SD. Genotypes were compared using two-way ANOVA with Tukey’s multiple comparison test. **p < 0.01

Full size image

Male mice carrying one or two copies of the Zfp335R1092W mutation exhibited significantly blunted LDL statin responses compared to wild-type (Fig. 6, p = 0.0011 and p = 0.0010, respectively). Female mice of any genotype experienced minimal changes in LDL particle concentrations with statin treatment (Figs. 5B, S3, and S4B); the statin-induced reductions in non-HDLC in female wild-type and heterozygous mice (Fig. 4C) seemed largely driven by particles in the IDL size range instead (Figs. 5B, S3, and S4B). We also went back to investigate whether there was a significant sex interaction of the original association of LCL ZNF335 statin response with plasma LDLC statin response in CAP. This association was also stronger in men than women, though the interaction of sex with gene expression change was not significant (p = 0.28, Fig. S6).

Discussion

In this study, we show evidence that zinc finger protein 335 (ZNF335/Zfp335) is a novel modulator of LDLC statin response. In our in vitro experiments, we identify 147 human genes whose in vitro statin responses correlate significantly with the in vivo plasma LDLC statin responses of the corresponding donors, with ZNF335 and CNOT3 topping the list. In our subsequent in vivo experiments in mice using a Zfp335 hypomorphic missense (Zfp335R1092W) genetic model, we show for the first time that reduced ZFP335 function leads to reduced plasma cholesterol levels and a blunted LDL response to statin, with significant sex differences.

ZNF335 resides in the nucleus and regulates transcription, including enhancing transcriptional activation by nuclear hormone receptors [35, 36]. It can activate transcription by recruiting histone methyltransferase complexes to the promoters of genes [32]. ZNF335 is widely expressed across human tissues [37] and has diverse functions, including roles in neuronal development [32] and T-cell maturation [23, 38, 39]. Variants associated with blood HDL cholesterol, triglyceride, non-HDL cholesterol, and LDL cholesterol levels in GWAS are located in and near the ZNF335 gene [40]. Though the dominant contribution to this association signal is likely the nearby PLTP (phospholipid transfer protein) gene, which has well-established roles in lipoprotein metabolism [41], it is possible that ZNF335 also contributes to this signal. Here, we demonstrate for the first time that ZNF335 influences plasma cholesterol levels and statin response. Though the correlation between statin-induced changes in LCL ZNF335 gene expression with donor plasma LDL-C stain response is quite modest, it is still a significant contribution given that only about 5% of the variation in statin response has been explained by genetic factors to date [5]. Further investigation is needed to determine which, if any, genes affecting cholesterol metabolism may be regulated transcriptionally by ZNF335 and to determine the mechanism by which ZNF335 influences plasma cholesterol phenotypes.

ZNF335 physically interacts with members of the CCR4-NOT (carbon catabolite repression 4–negative on TATA-less) complex, including CNOT6 (aka CCR4) and CNOT9 (aka RCD1) [30]. The CCR4-NOT complex can regulate gene expression via several mechanisms, including transcription and mRNA deadenylation [42]. Though a physical interaction has not been reported between ZNF335 and the CCR4-NOT complex regulatory subunit CNOT3, here we observe strong correlations between their statin-induced gene expression changes, indicating they may be co-regulated by some of the same statin-responsive factors. ZNF335 and CNOT3 expression changes in LCLs also show similar strength and directionality of correlation with plasma LDLC statin responses of the corresponding statin clinical trial participants, so they may be functionally related, and it is plausible that they could be acting as components of the same complex. CNOT3 has been previously reported to have roles in regulation of lipid and glucose metabolism and to be differentially regulated in fasted and fed states based on studies in CNOT3 haplodeficient mice [33]. A functional link between ZNF335 and CNOT3 could be further explored in future studies, as could effects of conditional knockdowns of both genes on lipids and statin response.

As with Cnot3 [31], knockout of Zfp335 in mice is embryonic lethal [32]. Thus, to study the potential role of Zfp335 in cholesterol metabolism and statin response, here we use mice carrying a hypomorphic mutation (R1092W aka bloto) in Zfp335. Previous studies using Zfp335R1092W mice have shown that the R1092W mutation in the twelfth zinc finger of ZFP335 impairs its ability to bind a subset of its DNA targets [23] by modifying one of its two known DNA binding domains [43]. Previous studies of this genetic mouse model have focused on the role of ZFP335 in T-cell development, but here we also show that the mutation has an impact on body weight, blood cholesterol levels, and blood cholesterol statin response. It is interesting to note that for some of these phenotypes (e.g., body weight and HDL-cholesterol levels in male mice), heterozygotes display intermediate phenotypes typical of an additive relationship, while for other phenotypes, heterozygotes are phenotypically more similar to wild-type animals (e.g., non-HDL cholesterol statin response in female animals) or to Zfp335R1092W/R1092W homozygous mutant animals (e.g., LDL particle statin response in male animals). Perhaps ZFP335 acts as a component of distinct complexes to elicit some of these phenotypes, which could explain why the same genetic model is not common to them all, or perhaps the fact that the mutation is only a partial loss of function mutation plays a role.

We observe striking sex differences when studying the impact of the hypomorphic missense Zfp335 mutation on mouse HDL cholesterol levels and LDL statin response. Similar, though less significant, sex differences are also observed for the human LCL ZNF335 statin response correlation with plasma LDLC statin response and the correlation of LCL ZNF335 levels with plasma HDL-cholesterol levels, showing inter-species replication despite known differences in lipoprotein metabolism and profiles between the species. The differences in the impact of the Zfp335 mutation between sexes could be related to sex-specific differences in the hormonal milieu, given ZNF335’s role in nuclear hormone receptor signaling, but the hormone(s) involved and which of their downstream targets regulate cholesterol levels and statin response remain to be elucidated.

Sex differences are also observed in the lipoprotein profiles and lipoprotein response to statin in wild-type mice. Wild-type female mice have less plasma HDL cholesterol and total cholesterol than males (in agreement with other studies in the literature which have studied sexual dimorphism in mouse lipoprotein metabolism [44, 45]) and exhibit a blunted non-HDL cholesterol statin response compared to males. In this study, though male mice experience significant reductions in LDL with statin treatment, the statin-induced reduction in non-HDL cholesterol in female mice appears to be driven mainly by larger particles (including IDL) rather than LDL. In contrast, in humans, women experience LDL reductions quite similar to men.

This study illustrates the utility of clinical trial participant-derived cell lines in identifying novel genetic factors involved in drug response phenotypes. The correlation of in vitro drug-induced transcriptomic changes in cell lines to in vivo drug response phenotypes from the corresponding participants during the clinical trial can be a powerful approach, even in the transformed cell lines used here. Though LCLs may not always mirror what might be happening in hepatocytes, they express cholesterol biosynthesis enzymes (including HMGCR), LDLR, and many other key cholesterol metabolism genes and have been previously used to identify novel modulators of statin response that have been validated in human hepatocyte cell lines and/or in vivo [8, 10, 11]. The emergence of induced pluripotent stem cells as a model system holds additional promise in investigating genes that are not as universally expressed across cell types as the genes we identify here, since they can be differentiated into a variety of specialized cell types.

Conclusions

Here, we identify a novel role for ZNF335 in modifying circulating levels of LDL-cholesterol and LDL-cholesterol statin response by leveraging naturally occurring variation in ZNF335 expression in statin clinical trial participant-derived cell lines and a hypomorphic missense mouse model for ZNF335. Thus, it is possible that inter-individual variation in ZNF335 expression contributes to differences in cholesterol levels and statin response in humans.