PCSK7 gene variation bridges atherogenic dyslipidemia with hepatic inflammation in NAFLD patients

Synthase; TGF  , transforming growth factor beta; ACTA2, Alpha Actin-2; COL1A1, Collagen, type I, alpha 1; CPT1, Carnitine palmitoyltransferase I GSEA, Gene Set Enrichment Analysis; TNF  , Tumor necrosis Factor alpha; IL6, Interleukin 6; PPAR  , peroxisome proliferator-activated receptor alpha; lncRNA, long noncoding RNA; PA, palmitic acid; OA, oleic acid; qt-PCR; quantitative polymerase chain reaction.  ABSTRACT Dyslipidemia and altered iron metabolism are typical features of non-alcoholic fatty liver disease (NAFLD). Proprotein Convertase Subtilisin/Kexin Type 7 ( PCSK7 ) gene variation has been associated with circulating lipids and liver damage during iron overload. Aim of this study was to examine the impact of the PCSK7 rs236918 variant on NAFLD-related traits in 1,801 individuals from the Liver Biopsy Cohort (LBC), 500,000 from the UK Biobank Cohort (UKBBC), and 4,580 from the Dallas Heart Study (DHS). The minor PCSK7 rs236918 C allele was associated with higher triglycerides, aminotransferases and hepatic inflammation in the LBC (p<0.05) and with hypercholesterolemia and liver disease in the UKBBC. In the DHS, PCSK7 missense variants were associated with circulating lipids. PCSK7 was expressed in hepatocytes and its hepatic expression correlated with that of lipogenic genes (p<0.05). The rs236918 C allele was associated with upregulation of a new “intra-PCSK7” lnc-RNA predicted to interact with the protein, higher hepatic and circulating PCSK7 protein (p<0.01), and the latter correlated with triglycerides (p=0.04). In HepG2, PCSK7 deletion reduced lipogenesis, fat accumulation, inflammation, TGF  pathway activation and fibrogenesis. In conclusion , PCSK7 gene variation is associated with dyslipidemia and more severe liver disease in high risk individuals, likely by modulating PCSK7 expression/activity. PCSK7 variants on hepatic expression of the protein, inflammation and lipogenesis, as well as the phenotype induced by PCSK7 deletion in hepatocytes in vitro a correlation of PCSK7 levels with lipid metabolism in the liver of severe obese individuals. the expression of lipogenic and lipid oxidation genes reduced in PCSK7 HepG2 cells as compared to the wild-type counterpart. Consistent with the human genetic data, in PNPLA3 148M/M HepG2 cells, PCKS7 haploinsufficiency protected against fat accumulation following exposure to fatty acids. These data suggest that the reduced PCSK7 expression alters lipid metabolism in hepatocytes and support the notion that the lipid phenotype associated with PCSK7 variation is mediated by modulation of PCSK7 expression/activity. The impact of PCSK7 on the regulation of both lipogenic and lipolytic pathways may explain why carriage of the PCSK7 rs236918 C allele did not result in a uniform increase in the risk of fatty liver, but specifically in individuals at higher risk, e.g. in those carrying the PNPLA3 I148M risk variant which impairs the lipid catabolism (3). PCSK7 deletion was also associated with reduced TNF  and TGF  expression. The latter suggests the existence of an interaction between hepatic fat accumulation and PCSK7 in determining hepatic inflammation and activation of fibrogenesis, favored by the presence of the I148M PNPLA3 variant. The in vitro model also confirmed that decreased protein synthesis leads to reduced detectability of the PCSK7 protein in the supernatants, although further studies are necessary to prove that the PCSK7 protein is actively shed and secreted by hepatocytes. In conclusion, the present results suggest that the rs236918 PCSK7 variant links dyslipidemia with a towards more liver damage in NAFLD patients at higher risk. Several clues suggest that the mechanism may increased intracellular PCSK7 protein and its secretion from hepatocytes. hemizygous PCSK7 deletion lipogenesis hepatocytes protects fat data suggest that of PCSK7 synthesis may decrease circulating lipids, liver risk logistic regression models used test Biopsy


INTRODUCTION
Nonalcoholic fatty liver disease (NAFLD) is emerging as a leading cause of liver damage worldwide (1).
From both epidemiological and pathophysiological points of view, NAFLD is strongly intertwined with insulin resistance and its clinical correlates, including visceral obesity, hyperglycemia and atherogenic dyslipidemia (2). NAFLD has a strong genetic component and variants in proteins regulating hepatocellular lipid handling, including PNPLA3, TM6SF2 and MBOAT7 predispose to hepatic fat accumulation and the development of progressive nonalcoholic steatohepatitis (NASH) (3,4).
Dysregulation of iron metabolism represents another typical feature of NAFLD associated with more severe hepatic and cardio-metabolic damage (5). This so called "Dysmetabolic hyperferritinemia" is most frequently observed in patients with risk factors for iron accumulation (6)(7)(8), and tracks with increased expression of iron transporters induced by excess fatty acids (9,10), but the underlying mechanism remains largely unexplained.
We hypothesized that PCSK7 variation bridges dyslipidemia with liver damage and dysregulation of iron metabolism. The aim of this study was therefore to examine the impact of PCSK7 variation on liver damage and metabolic traits in individuals at risk of NAFLD and in the general population. To elucidate the mechanisms underlying the association, we examined the impact of rs236918 and other functional by guest, on April 28, 2019 www.jlr.org Downloaded from

The Liver Biopsy Cohort
Part of the Liver Biopsy cross-sectional Cohort (LBC) has previously been described (4,22). Briefly, a total of 1,801 adult individuals of European descent were consecutively enrolled from Italian and Finnish referral centers. Inclusion criteria were liver biopsy for suspected NASH or severe obesity, availability of DNA samples and clinical data. Individuals with increased alcohol intake (>30/20 g/day in males/females), and other causes of liver disease were excluded. The study conformed to Declaration of  Table 1. For the Milan outpatient service sub-cohort (n=332), systematic evaluation of circulating iron parameters (iron, transferrin, transferrin saturation (TS), ferritin), and histological hepatic iron staining were available.

Histological evaluation
In individuals included in the LBC, steatosis was graded based on the percentage of affected hepatocytes as 0: 0-5%, 1: 6-33%, 2: 34-66%, and 3: 67-100%. Disease activity was assessed according to the NAFLD Activity Score (NAS) with systematic evaluation of hepatocellular ballooning and lobular inflammation; fibrosis was also staged according to the recommendations of the NAFLD clinical research network (23). The scoring of liver biopsies was performed by independent expert pathologists unaware of patients' status and genotype, who trained on a common slide set (22,24). In the LBC cohort, 87% of patients had NAFLD, and 55% NASH.

UK Biobank cohort
The association of the rs236918 variant with 98 phenotypes related to metabolic disorders and liver disease was evaluated in the UK Biobank Cohort (UKBBC, Supplemental  (25). Adjusted p-values<0.1 were considered statistically significant.

The Dallas Heart Study
The Dallas Heart Study (DHS) is a multiethnic population-based probability sample of Dallas County residents. The study design and recruitment procedures have been previously described (26). The study was approved by the Institutional Review Board of the University of Texas Southwestern Medical Center and all individuals provided written informed consent. Plasma lipids were measured by standard enzymatic assays. Hepatic triglyceride content was measured by proton magnetic resonance spectroscopy ( 1 H-MRS) as previously described (27,28). The present analysis includes up to 4,580 individuals (52% African-Americans, 29.5% European-Americans, 16% Hispanics, and 3% of other ethnicities), with available DNA samples and laboratory data. Liver fat measurements were available in a subset of individuals (n=2,736). Clinical characteristics of DHS participants are shown in Supplemental Table 1.

Gene Expression in Human Primary Cells
RNA was extracted using Trizol reagent (Life Technologies, Carlsbad, CA). One g of total RNA was retro-transcribed by using the VILO random hexamers synthesis system (Life Technologies). Gene expression was evaluated by quantitative real time-PCR, performed by the 7500 Fast Thermocycler (Life Technologies) using the Taqman Universal PCR Master Mix (Life Technologies) and Taqman probes for human PCSK7. All reactions were performed in triplicate and data were normalized to beta-actin gene expression.

Gene and protein expression analysis in liver biopsies
PCSK7 gene expression was measured in percutaneous liver biopsies of a subset of 125 severely obese patients from the Milan cohort (Bariatric surgery cohort), whose clinical features are shown in Supplemental Table 5 (PCSK7 rs236918 GG: n=100; GC: n=22; CC: n=3). RNA was extracted from liver biopsies using RNeasy mini-kit (Qiagen Hulsterweg). RNA quality was assessed through Agilent 2100 Bioanalyzer and samples with RNA integrity numbers (RIN) greater than or equal to 7 were used for library preparation (Ribo-reduction libraries). RNA sequencing was performed in paired-end mode with a read length of 150nt using the Illumina HiSeq 4000 (Novogene, Hong Kong, China). The RNA sequencing detailed protocol and data analysis approach are described in the Supplementary materials and methods.
For protein analysis, 10 samples with different rs236918 genotypes (GG or GC (n=6) +CC (n=4) were analyzed, and all reactions were performed in duplicate. The following antibodies were used: rabbit monoclonal anti-PCSK7 (Cell Signaling, Beverly, MA) and goat polyclonal anti beta-actin (Santa Cruz Biotechnology, Santa Cruz, CA).

Linkage disequilibrium analysis
The detailed protocol and data analysis approach are described in the Supplementary materials and methods.

Histone modification analysis
In order to assess the association of genetic variation with histone modifications, E066 Adult Liver Chip-Seq data (p-value signal tracks) and core (15-state) Hidden Markov model chromatin state prediction data were obtained from the Roadmap project (30). The Gviz package was employed for data visualization (31).

De novo assembly and specific transcripts evaluation
The detailed protocol and data analysis approach are described in the Supplementary materials and methods.

PCSK7 gene deletion in HepG2 cells
We generated a PCSK7 knockout model by exploiting CRISPR-Cas9 technology in HepG2 hepatoma cells. We first generated an inducible stable Cas9 cell line by lentiviral transfection and clonal selection by blasticidin. We then transfected pGL3-U6-sgRNA-PGK-puromycin plasmid vector and a small guide RNA targeting within the first exon of PCSK7 gene to induce PCSK7 knockout. We screened transfected clones through a digestion reaction using T7 endonuclease enzyme, and identified a clone carrying a large deletion in heterozygous state (PCSK7 -/+ ), which was confirmed by Sanger sequencing (Supplemental Figure 1).

Statistical analysis
For descriptive statistics, continuous variables were presented as mean and standard deviation (for approximately normally distributed variables) or median and interquartile range (for highly skewed variables). Categorical variables were presented as number and percent. Genetic association was tested assuming an additive model. Analyses were performed using generalized linear models: linear regression models were fit to analyze continuous traits, logistic regression for binary traits (NASH, severe fibrosis stage F3-F4), and ordinal regression for ordinal traits (components of the NAFLD activity score: severity of steatosis, lobular inflammation and hepatocellular ballooning, stage of fibrosis). Models were adjusted for confounding factors (including recruitment center), as specified. Variables with skewed distributions were logarithmically transformed before entering the models.
Statistical analyses were carried out using the JMP 12.0 (SAS Institute, Cary, NC, USA) and R statistical analysis software version 3.3.2 (http://www.R-project.org/). p-values <0.05 were considered statistically significant.

PCSK7 rs236918 affects circulating lipids, iron and hepatic inflammation in the LBC
Clinical features of LBC patients stratified by rs236918 genotype are shown in Table 1. No differences in demographic and anthropometric features were found across rs236918 genotypes. In multivariableadjusted models, the minor C allele was associated with increased circulating triglycerides after adjustment for sex, age, body mass index (BMI), type 2 diabetes (T2D) and TM6SF2 E167K variant (beta 0.09, 95% c.i. 0.01-0.17; p=0.033). We did not find any association between rs236918 genotype and lowdensity lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), insulin resistance, or T2D.
We next evaluated the impact of PCSK7 rs236918 C allele on iron metabolism in a subset of patients (n=332). The C allele was associated with increased TS after adjustment for sex, age, BMI, T2D, and the hemochromatosis gene (HFE) p.C282Y and p.H63D variants (beta 4.8, 95% c.i. 0.3-9.3; p=0.035).
Notably, the C allele was nearly associated with increased ALT levels (p=0.06) and with higher AST (p=0.0008) and the effect was even more strong by using a recessive model (ALT, p=0.0048; AST, p=0.0002) ( Table 1). However, under a genotypic model, we have now reported that rs236918 GC genotype was associated with ALT and AST levels as compared to the reference GG genotype (p=0.054; p=0.0042 respectively) suggesting that the effect was not completely explained by a recessive inheritance models, but that also heterozygous carriers may have a slightly increased risk of liver damage (Supplemental Table 6).
The association of PCSK7 genotype with the spectrum of histological liver damage related to NAFLD in the LBC is shown in Table 2. In the overall cohort, the minor C allele was not associated with steatosis grade; it was independently associated with lobular inflammation (beta 0.22, 95% c.i. 0.02-0.42; p=0.028; Table 2 and Supplemental Figure 2), but not with ballooning and fibrosis. After stratification of patients for enrolment criteria, rs236918 was associated with steatosis grade (beta 0.34, 95% c.i. 0.03-0.64; p=0.03) and lobular inflammation (beta 0.36, 95% c.i. 0.05-0.66; p=0.02) in the Liver Clinic cohort, which is made up of individuals at high risk of progressive NAFLD, but not in bariatric subjects (Table   2). In addition, after stratification of patients for the presence of the I148M PNPLA3 variant, the rs236918 C allele was associated with steatosis grade (beta 0.26, 95% c.i. 0.003-0.51; p=0.04) and lobular inflammation (beta 0.37, 95% c.i. 0.10-0.62; p=0.007) in carriers of the PNPLA3 I148M risk variant, but not in non-carriers (Table 2). These data suggest that PCSK7 variation influences circulating triglycerides and hepatic inflammation, but the effect is at least partly independent of modulation of hepatic fat content.

PCSK7 rs236918 affects circulating cholesterol and possibly liver disease in UKBBC
The clinical phenotypes related to rs236918 variant in the UKBBC are shown in Table 3 and Supplemental Table 2. We observed a strong positive association between PCSK7 rs236918 C allele and circulating total cholesterol levels (p=8.6*10 -9 ), use of cholesterol lowering medications (p=1.5*10 -7 ), and the presence of cardiovascular complications (p=0.003). Notably, PCSK7 rs236918 variant was also nominally associated with an increased risk of "liver failure/cirrhosis" phenotype (p=0.013).

PCSK7 variants impact on lipid metabolism, but not steatosis in the DHS
To examine whether PCSK7 influences circulating lipids by altering hepatic lipid accumulation, we evaluated the impact of PCSK7 variation on hepatic triglyceride content, metabolic traits and liver damage in DHS participants ( Table 4). The minor C allele of rs236918 was not associated with hepatic triglycerides content (beta 0.007±0.034; p=0.83) or circulating triglycerides (beta -0.007±0.027; p=0.79).

Downloaded from
Hispanics) was associated with increased circulating triglycerides (mean difference +58.8 mg/dL, beta 0.3±0.1; p=0.035). Neither of these two variants was associated with hepatic fat content.
We did not observe an association between PCSK7 variants and serum iron levels in the whole DHS population. However, rs236918 was marginally associated with higher serum iron in a subgroup of atrisk individuals, carrying iron-increasing variants in HFE (p=0.049).
Taken together, these data suggest that the impact of PCSK7 gene variation on circulating lipids is not secondary to modulation of hepatic fat content. Two low-frequency variants (R504H and A102T) were identified with opposite impact on metabolic traits, with A102T having a larger, but directionally concordant, effect compared with rs236918 G>C in the LBC.

PCSK7 is highly expressed in hepatocytes and correlates with de novo lipogenesis
As a first step to elucidate the role of PCSK7 in lipid handling and liver disease, we examined PCSK7 mRNA expression in primary human hepatocytes and hepatic stellate cells, in peripheral lymphocytes, monocytes/macrophages, and in an endothelial cell line (HUVEC). PCSK7 expression was highest in primary hepatocytes, whereas it was low in the other cell types (p<0.05; Figure 1A). In a candidate gene approach conducted in in severely obese individuals, hepatic expression of PCSK7 correlated with FAS To further investigate the correlation between PCSK7 expression and metabolic pathways, a co-regulation analysis was performed in the whole transcriptome. Gene Set Enrichment Analysis (GSEA) highlighted an enrichment in PCSK7 co-expressed genes within pathways related to inflammation (e.g. TNFα signaling, interleukin-6 response), epithelial mesenchymal transition, and mitosis. Conversely, pathways involved in adipogenesis, fatty acid metabolism, and oxidative phosphorylation were inversely correlated with PCSK7 expression (Supplemental Figure 4A).
Finally, to dissect the transcriptome variability associated with the presence of the PCSK7 rs236918 variant, differential expression analysis was performed. Collectively, we found a significant modulation of inflammatory pathways and in genes involved in oxidative phosphorylation associated with the minor allele at risk (Supplemental Figure 4B).

Mechanism of association between rs236918 and PCSK7 function
To investigate whether the observed associations of rs236918 with metabolic traits were due to linkage with other common (MAF0.01) missense PCSK7 variants, we examined linkage disequilibrium patterns at the PCSK7 locus. The rs236918 was not in linkage disequilibrium with common PCSK7 coding variants (Supplemental Figure 5A, Supplemental Table 7), suggesting that the impact of rs236918 on the observed phenotypes is not mediated by linkage with other variants influencing PCSK7 protein sequence and activity.
Next, we investigated whether the rs236918 variant impacts on hepatic PCSK7 mRNA levels. The rs236918 variant was not associated with overall PCSK7 expression, even if it was marginally associated with the relative abundance of some alternative mRNA transcripts (Supplemental Figure 5B). However, evaluation of epigenomic data obtained by the Roadmap consortium from adult liver, highlighted high levels of H3K36 trimethylation (H3K36me3) and H3K27 acetylation (H3K27ac) at the rs236918 region ( Figure 2A). Consistently, chromatin state prediction data revealed the presence of two regulatory regions near the rs236918 locus. De novo assembly of the PCSK7 locus from transcriptomic data revealed the presence of a 1609 bp transcript mapping in the rs236918 genomic region (Chromosome 11, start: 117220122 GRCh38, 110790837 GRCh37, Figure 2A). This transcript did not overlap with other products of the PCSK7 gene and was predicted with good confidence (score 0.9, optimal cutoff 0.18) as the product of a nested gene. As the longest open reading frame in the novel transcript was 83 bp, not consistent with a protein coding mRNA, we characterized this transcript as a long noncoding RNA (lncRNA). This "intra-PCSK7" lncRNA was not predicted to interact with the PCSK7 main transcript due to lack of sequence similarity. However, the "intra-PCSK7" lncRNA was predicted to locate in the cytoplasm and to bind PCSK7 protein by two independent bioinformatic tools: lncPro (score 72.73) and catRAPID. The strongest interaction was predicted by the catRAPID fragments analysis to be located between the initial portion (aa 60-111) of PCSK7 protein and a RNA region ranging approximately from position 700 to 900 of the "intra-PCSK7" lncRNA ( Figure 2B, discriminative power 59%), with rs236918 located at position 772. Remarkably, we could directly confirm the expression of this "intra-PCSK7" lncRNA in the liver of severely obese individuals. The hepatic expression of the "intra-PCSK7" lncRNA was higher in carriers of the PCSK7 rs236918 variant (n=11), as compared to non-carriers (n=30) (p=0.017; Figure 2C). Therefore, we hypothesized that the modulation of "intra-PCSK7" lncRNA sequence by the rs236918 variant could affect protein synthesis or stability. Indeed, the presence of the risk C allele was associated with higher hepatic protein levels (p<0.01; Figure 3A). Furthermore, in a subset of patients at risk of NAFLD (n=72), carriers of C allele showed increased circulating PCSK7 protein concentration as compared to non-carriers (p=0.007). Importantly, circulating PCSK7 was correlated with triglycerides (beta 0.26, 95% c.i. 0.02-0.50; p=0.04; Figure 3B). These data suggest that the rs236918 C allele may predispose to liver damage and influence lipid handling by up-regulating PCSK7 protein synthesis and secretion, as in a gain-of-function variant, via a new mechanism encompassing the induction of a novel PCSK7-interacting lncRNA.

PCSK7 deletion in HepG2 cells reduced lipogenesis and TGF expression
As expected, PCSK7 mRNA and protein levels were reduced by about 50% in PCSK7 -/+ cells ( Figure 4A; p<0.05). In a candidate gene approach, and consistently with the human liver transcriptome data, the expression of SREBP1c and FAS was reduced in PCSK7 -/+ compared to wild type cells ( Figure 4B; p<0.01). Similarly, the mRNA levels of PPAR and CPT1, involved in beta-oxidation, were reduced in PCSK7 -/+ cells although the difference in CPT1 expression was not significant (p=0.01 and p=0.22 respectively; Figure 4B). As a result, PCSK7 -/+ cells showed lower fat accumulation after exposure to fatty acids ( Figure 4C). Notably, secretion of PCSK7 protein was reduced in PCSK7 -/+ compared to wild type cells, independently of the exposure to fatty acids (p<0.05, Figure 4D). Finally, PCSK7 haploinsufficiency led to reduced expression of genes involved in inflammation (TNF; p=0.04), but in particular of fibrogenesis in untreated cells (TGF p=0.008; Figure 4B). Free fatty acids exposure induced TGF expression in wild type cells, whereas the effect was abrogated in PCSK7 -/+ cells ( Figure   4B and Supplemental Figure 6A). Consistently, p(Thr276) SMAD4/SMAD4 and p(Thr8) SMAD2-3/SMAD2-3 ratios were increased in wild type cells treated with FFAs compared to PCSK7 -/+ cells. These data are suggestive of hampered TGF signaling activation in haploinsufficient cells (Supplemental Figure 6B).

DISCUSSION
In this study, we examined the impact of the PCSK7 rs236918 noncoding variant on metabolic phenotypes and liver damage associated with NAFLD in individuals at risk and from the general population, and compared its effect with that of other PCSK7 missense variants.
Previous studies revealed an association of low-frequency PCSK7 missense variants with circulating lipids (11)(12)(13), suggesting that altered PCSK7 function account for this phenotype, and the non-coding variant rs236918 has also been associated with liver damage in patients with hereditary hemochromatosis (20,21). Here, we found that the rs236918 C minor allele was associated with increased circulating triglycerides in NAFLD patients and with dyslipidemia in the UKBBC (12,13). In the LBC, although we could not test the association with circulating transferrin receptor levels, the rs236918 variant was related to increased TS, therefore confirming a possible role of PCSK7 in dysregulation of iron metabolism (20).
The rs236918 variant was also associated with increased aminotransferases, especially with AST levels, a marker of hepatic damage, and with more severe lobular inflammation. However, in the overall LBC, the variant did not have a significant impact on hepatic fat accumulation, suggesting the mechanism underlying the association with hepatic inflammation is at least partially independent of the total amount of hepatic fat. Notably, in NAFLD patients who carry the PNPLA3 I148M risk variant, the rs236918 variant was also associated with steatosis grade. It could be speculated that the PCSK7 variant causes an increase in circulating lipids by affecting lipogenesis, and in patients who carry the PNPLA3 I148M variant the effect translates into the development of a more severe steatosis.
The lack of association between PCSK7 genotype and fibrosis in the LBC despite more severe inflammation may be related to a lack of power. Indeed, in the well-powered UKBBC we detected a nominally significant association between the variant and the risk of advanced liver disease. On the other hand, in the DHS the minor rs236918 C allele was not associated with lipids, hepatic fat content and liver enzymes. This discrepancy may be accounted for by the smaller sample size and more limited power of the DHS as compared to the UKBBC, or by a smaller effect of the variant in the general population vs high-risk patients (32). However, the inclusion of individuals of non-European background in the DHS allowed us to evaluate the impact of two low-frequency missense variants in PCSK7 (A102T and R504H), associated with circulating lipid levels in opposite directions. As we now provide evidence that increased PCSK7 is associated with dyslipidemia, we could therefore speculate that the A107T variant similarly acts by increasing, while the other R504H variant by decreasing PCSK7 protein activity (7).
However, the impact of these mutations on specific PCSK7 substrates remains to be demonstrated, and neither of these variants was associated with hepatic triglycerides, in line with the notion that the effect of PCSK7 gene variation on circulating lipids is not secondary to modulation of hepatic fat content.
In the attempt to decipher the role of PCSK7 in lipid handling and liver disease, we showed that PCSK7 is more expressed in hepatocytes compared to other liver cells. Furthermore, hepatic expression of PCSK7 correlated with that of SREBP1c and FAS, involved in de novo lipogenesis and with TGF. In line with these results, TGF expression and activation of the dependent signaling pathway were hampered in PCSK7 haploinsufficient cells as compared to wild type cells following exposure to free fatty acids. It has been demonstrated that TGF-SMAD2/3 signaling pathway promotes lipid accumulation with induction of lipogenesis-related genes and suppression of beta-oxidation in hepatocytes (33). These results obtained in PCSK7 haploinsufficient cells are also in line with previous data obtained in zebrafish, showing that PCSK7 increases both the mRNA expression and proteolytic cleavage of TGF, the premier profibrogenic cytokine and activator of hepatic fibrogenesis (34). It has also been speculated that transferrin caused activation and upregulation of procollagen-type gene expression in hepatic stellate cells indicating a significant role of transferrin receptor in iron-mediated liver fibrogenesis (35). Further studies are required to clarify whether the association of rs236918 with regulation of iron metabolism (36) and liver damage progression (37) is mediated by modulation of TGF signaling.
Concerning the mechanism linking rs236918 with PCSK7 overall function, we showed that this variant is not in linkage with other protein-altering sequence variants that may affect PCSK7 activity. Furthermore, it may only marginally affect the relative abundance of alternative mRNA transcripts. Epigenomic analysis revealed high level of H3K36me3 and H3K27ac at the rs236918 region, which may indicate active gene transcription (38), as this histone modification is often found in the core regions of actively transcribed regions or alternatively may play a role in maintaining gene expression stability. In addition, high levels of H3K27ac are often found around active enhancers (39) suggesting that rs236918 may fall near or within a regulatory element active in the liver. Consistently, de novo assembly analysis revealed a putative lncRNA in the rs236918 locus, which was predicted to locate into the cytoplasm and to interact with the PCSK7 protein. Remarkably, we could directly detect the expression of this "intra-PCSK7" lncRNA in the liver of severely obese individuals, thereby confirming the in silico prediction. We found that the hepatic expression of the "intra-PCSK7" lncRNA was higher in carriers of the rs236918 variant compared to non-carriers. Although the causal relationship between the "intra-PCSK7" lncRNA and PCSK7 proteins expression levels, and the impact of the rs236918 variant on this process remain to be demonstrated, it could be speculated that the rs236918 C allele increases the PCSK7-"intra-PCKS7" lncRNA interaction, by enhancing PCSK7 stability and possibly by hampering its degradation.
Consistent with this scenario, the presence of the C risk allele was associated with higher hepatic levels and secretion of PCSK7, despite the lack of a change in PCSK7 mRNA levels. Even though we could not rule out a possibility that rs236918 impacts the expression of nearby genes in the region, several clues suggest that modulation of PCSK7 expression is the underlying mechanism. Indeed, circulating levels of PCSK7 correlated with serum triglycerides. This is also consistent with the fact that the R504H loss-offunction variant protects from dyslipidemia (7), while the rs236918 variant, which we showed here is associated with increased PCSK7 levels, worsens the lipid profile. By analogy, the A102T variant, which is also associated with increased lipid levels, could be regarded as a gain-of-function mutation. A possible mechanism is represented by PCSK7-mediated cleavage of angiopoietin-like protein 4 (AGPTL4), which then acquires the ability to inhibit lipoprotein lipase thus impairing triglycerides clearance. Unfortunately, we could not assess activated ANGPTL4 and peripheral LPL expression in our cohort. On the other hand, we provided evidence suggesting that PCSK7 may also be involved in the regulation of de novo lipogenesis and lipid oxidation in a cell-autonomous fashion in hepatocytes. Indeed, human transcriptome by guest, on April 28, 2019 www.jlr.org Downloaded from data showed a correlation of PCSK7 levels with lipid metabolism in the liver of severe obese individuals.
Furthermore, the expression of lipogenic and lipid oxidation genes was reduced in PCSK7 +/-HepG2 cells as compared to the wild-type counterpart. Consistent with the human genetic data, in PNPLA3 148M/M HepG2 cells, PCKS7 haploinsufficiency protected against fat accumulation following exposure to fatty acids. These data suggest that the reduced PCSK7 expression alters lipid metabolism in hepatocytes and support the notion that the lipid phenotype associated with PCSK7 variation is mediated by modulation of PCSK7 expression/activity. The impact of PCSK7 on the regulation of both lipogenic and lipolytic pathways may explain why carriage of the PCSK7 rs236918 C allele did not result in a uniform increase in the risk of fatty liver, but specifically in individuals at higher risk, e.g. in those carrying the PNPLA3 I148M risk variant which impairs the lipid catabolism (3). PCSK7 deletion was also associated with reduced TNF and TGF expression. The latter suggests the existence of an interaction between hepatic fat accumulation and PCSK7 in determining hepatic inflammation and activation of fibrogenesis, favored by the presence of the I148M PNPLA3 variant. The in vitro model also confirmed that decreased protein synthesis leads to reduced detectability of the PCSK7 protein in the supernatants, although further studies are necessary to prove that the PCSK7 protein is actively shed and secreted by hepatocytes.
In conclusion, the present results suggest that the rs236918 PCSK7 variant links dyslipidemia with a tendency towards more liver damage in NAFLD patients at higher risk. Several clues suggest that the mechanism may involve increased intracellular PCSK7 protein and its secretion from hepatocytes.
Furthermore, hemizygous PCSK7 deletion decreases lipogenesis in hepatocytes and protects from fat accumulation. These data suggest that inhibition of PCSK7 synthesis may decrease circulating lipids, leading also to improvement of liver damage in high risk individuals.      Betas are shown in SD units of the outcome per each additional allele. P-values were determined using linear regression models adjusted for age, sex, 4 principal components or ancestry, BMI, and study visit. Analysis of triglyceride levels was in addition adjusted for T2D. Abbreviations: na -not available. The p-value for rs142953140 in subgroup analyses could not be calculated because there were fewer than 2 carriers of R504H in the corresponding subgroup. Mechanism of association between rs236918 and PCSK7 function. Correlation between circulating PCSK7 and serum triglycerides (C).