PSD3 downregulation confers protection against fatty liver disease

PSD3 sequence variation reduces FLD susceptibility in humans

The study design is shown in Supplementary Fig. 1. To identify previously unknown genetic loci affecting liver fat content, we used a candidate gene approach in the DHS, a population-based sample study with measurement of liver fat content by proton magnetic spectroscopy in 2,736 participants. Specifically, we selected all the tag single-nucleotide polymorphisms (SNPs) identified by a previous genome-wide study on circulating triglyceride levels¹⁸ (Supplementary Table 1). All missense and nonsense variants in the genes identified by the tag SNPs were then tested for association with liver fat content in the DHS¹⁴ using linear regression analysis under an additive genetic model adjusted for age, gender and the top four principal components of ancestry. For tag SNPs that were intergenic, missense and nonsense variants in both genes flanking the tag SNP were examined.

We identified three missense variants in three genes present in Europeans (minor allele frequency (MAF) >5% in Europeans from the 1000 Genome Project, Database of Single Nucleotide Polymorphisms (dbSNP), current build 154, released 21 April 2020) that were nominally associated (P < 0.050) with liver fat content in the DHS (Table 1).

Among these, we found two variants robustly associated with FLD and circulating triglycerides, namely the Glucokinase regulator (GCKR) rs1260326 (P = 0.007) and Transmembrane 6 Superfamily Member 2 (TM6SF2) rs58542926 (P = 5.7 × 10⁻⁸)^15,24. Moreover, we found a variant in the PSD3 gene, rs71519934, associated with lower liver fat content (P = 0.049). The DHS consists of three ethnic groups, namely European American, African American and Hispanic American. We next examined differences in clinical and anthropometric parameters, lipoprotein levels and liver fat content stratified by PSD3 rs71519934 genotype and by these ethnic groups (Supplementary Table 2). In the entire DHS there was no difference in these selected traits among the three genotypes, except for liver fat content. However, when stratified by ethnicity, there was a reduction in circulating total cholesterol levels for the PSD3 rs71519934 genotype in European Americans.

The PSD3 rs71519934, in the dbSNP, is annotated as a threonine to leucine amino acid substitution at position 186 of the PSD3 protein (dbSNP, current build 154, released 21 April 2020; https://www.ncbi.nlm.nih.gov/snp/rs71519934). However, exome sequencing of participants from the UK Biobank revealed that the 186 threonine is the minor allele in Europeans (Supplementary Fig. 2).

To test whether the PSD3 rs71519934 is associated with protection against the entire spectrum of FLD, we examined this variant in the LBC, comprising n = 1,951 Europeans at high risk for FLD with liver biopsy available (Supplementary Table 3). In the LBC, the PSD3 rs71519934 minor allele (186T) was associated with lower prevalence of liver steatosis (P = 5.9 × 10⁻⁶), fibrosis (P = 0.006), inflammation (P = 9.9 × 10⁻⁷) and ballooning (P = 0.002) (Table 2) using binary logistic regression analysis under an additive genetic model adjusted for age, gender, body mass index (BMI), centre of recruitment and the PNPLA3 rs738409.

In addition, carriers of the 186T minor allele were protected against more severe liver steatosis (P = 3.3 × 10⁻⁷), inflammation (P = 1.6 × 10⁻⁷), ballooning (P = 0.001) and fibrosis (P = 0.001) (Fig. 1a–d). These results were virtually identical when further adjusted for other genetic (TM6SF2 rs58542926 (E167K), MBOAT7 rs641738 and GCKR rs1260326 (L446P)) and environmental (presence of diabetes) variables influencing FLD (Supplementary Table 4). Moreover, consistent with European Americans in the DHS, carriers of the PSD3 186T minor allele had lower circulating total and low-density lipoprotein (LDL) cholesterol levels (P = 1.4 × 10⁻⁶ and P = 0.001, respectively; Supplementary Table 5). Next, we tested for an interaction between the PSD3 and PNPLA3 variants in the LBC but found no interaction between these two genetic variants and liver disease (Supplementary Fig. 3).

To seek independent replication of the association between the PSD3 variant and liver fat content, we examined white British participants (n = 10,970) from the UK Biobank with liver fat content measured as magnetic resonance imaging (MRI)-derived proton density fat fraction (PDFF). In this population-based study, genetic data on the rs71519934 dinucleotide substitution were not available, therefore we used the rs7003060 which is in complete linkage disequilibrium (D′ = 1, r² = 1) with the rs71519934 in Europeans. We did not find any association between the PSD3 minor allele and lower liver fat content in the UK Biobank (Table 3).

However, because genetic variations affecting FLD have a robust gene–environment interaction, with BMI amplifying their effect²⁵, we examined the differences in liver fat content among the PSD3 rs7003060 genotype after stratification for severity of overweight/obesity as measured by BMI. Severely obese (BMI > 35) carriers of the PSD3 minor allele had a lower liver fat content (beta = −0.175, P = 0.02; Supplementary Fig. 4 and Table 3).

Considering these results, we further tested the association between the PSD3 variant and the protection against liver disease in an independent replication cohort of obese (BMI ≥ 30 kg m⁻²) central European participants at risk for liver disease²⁶. More specifically, we examined 674 adult individuals (mean age 45 ± 12 years), with a mean BMI of 46 ± 10 kg m⁻² from three central European centres (Supplementary Table 3) with liver biopsies available. We found that the PSD3 minor allele was associated with a lower prevalence of liver steatosis (P = 0.024), fibrosis (P = 0.049) and ballooning (P = 0.047), and with less severe fibrosis and ballooning (P = 0.040 and P = 0.048, respectively; Table 4). No differences in clinical or metabolic traits were detected in this cohort stratified by PSD3 genotype (Supplementary Table 5). Finally, we performed a meta-analysis of the LBC and the replication cohort from central Europe. With both fixed and random effect models, the PSD3 genetic association was stronger for all the traits examined, except for the presence and severity of inflammation where the association was attenuated using a random effect model (Supplementary Table 6).

PSD3 hepatic expression is increased in individuals with FLD

The PSD3 protein has 18 annotated isoforms (Ensembl release 75). To determine which isoform is the most abundant in human liver, we examined the transcriptome of liver biopsies from a subset of individuals (n = 77) from the LBC²⁷. We found that isoform-a (annotated as NP_056125 in NCBI and as 001 ENST00000327040 in Ensembl) had the highest expression level in the liver (Fig. 2a). Interestingly, the total hepatic PSD3 messenger RNA levels were higher in livers with FLD than in those without (Fig. 1e), whereas no differences in the N-acetyltransferase 2 (NAT2) mRNA level were observed (Fig. 1f), When we stratified individuals based on the PSD3 genotype, we found no difference in the mRNA expression level of either PSD3 or NAT2 (Fig. 2b,c).

PSD3 186T protects against intracellular lipid accumulation

To understand the mechanism(s) underlying the association between the rs71519934 minor allele and lower liver fat content, we examined primary human hepatocytes from donors carrying the 186L or 186T amino acid change in homozygosity. Consistent with the genetic association, human primary hepatocytes homozygous for the 186T allele cultured in two dimensions (2D) had a lower neutral lipid fat content (P = 0.007) as measured by Oil Red O (ORO) staining compared with homozygous 186L hepatocytes (Fig. 3a). To examine PSD3 levels in the two different genotypes, we generated an antibody specific for human PSD3 (custom antibody generation detailed in Methods). We incubated cells with different amount of oleic acid (OA) (0, 10 and 25 µM) and examined protein levels between the two genotypes in primary hepatocytes. Overall, the amount of PSD3 was elevated with increasing concentration of OA (Fig. 3b). However, for each OA concentration, PSD3 protein expression was lower in cells homozygous for the 186T allele. To understand the mechanism behind this association, we examined differentially expressed genes involved in lipid homoeostasis using RNA sequencing (RNA-seq). We found a robust reduction in genes involved in triglyceride synthesis and secretion and cholesterol biosynthesis, whereas the expression of PGC-1α involved in mitochondrial biogenesis was increased (Fig. 3c).

PSD3 downregulation reduces intracellular lipid accumulation

Because PSD3 expression was elevated in livers from individuals with FLD, we tested the hypothesis that PSD3 downregulation would result in a reduction in intracellular fat content by using siRNA in human primary hepatocytes from donors homozygous for either the 186L or 186T allele. Downregulation of PSD3 reduced the intracellular neutral lipid content in hepatocytes carrying either allele when cultured in 2D (Fig. 4a,b). However, PSD3 silencing reduced intracellular lipid levels only in primary hepatocytes carrying the 186L allele when cultured in a three-dimensional (3D) spheroid model (Fig. 4c,d).

PSD3 downregulation reduces triglyceride synthesis and ARF6 activation

To understand the molecular mechanism(s) underlying the genetic association, we examined immortalized hepatocytes from rat (McArdle, McA-RH7777) homozygous for 180T, which corresponds to 186T in human PSD3. In these cells, Psd3 downregulation resulted in a reduction in the intracellular neutral lipid content as measured by ORO staining. Additionally, Psd3 downregulation in McA-RH7777 cells resulted in lower triglyceride production, measured as de novo triglyceride synthesis, and as mRNA expression of genes involved in triglyceride synthesis. Moreover, very low-density lipoprotein secretion, measured as apolipoprotein B (Apo B) secretion, was also lower in cells transfected with Psd3 siRNA compared with scramble (SCR) control siRNA. No differences were detected in intracellular lipid utilization, measured as beta oxidation (Supplementary Fig. 5). To confirm our data, we measured the intracellular lipid accumulation and triglyceride synthesis using radiolabelled tracers in human hepatoma cells (Huh7 cells) homozygous for the PSD3 186L allele, and obtained virtually identical results as McA-RH7777 cells (Supplementary Fig. 6). Thus, downregulation of endogenously expressed Psd3 threonine or PSD3 leucine resulted in decreased intracellular lipid levels. PSD3 is a guanine nucleotide exchange factor activating ADP-ribosylation factor 6 (ARF6). To test whether PSD3 downregulation results in changes in ARF6 activation, PSD3 was downregulated in Huh7 cells and levels of activated ARF6 were measured using a GGA3 protein-binding domain (PBD) pull-down assay. PSD3 downregulation resulted in lower levels of activated ARF6 (ARF–GTP) (Fig. 5).

In immunohistochemistry analyses, PSD3 was found to be present at a low level in hepatocytes and there was no apparent difference in the levels between L186T heterozygous and 186L homozygous donors (Supplementary Fig. 7). Otherwise, sinusoidal endothelial cells and biliary ducts showed moderate positivity. Interestingly however, the sample from one heterozygous patient characterized by the lowest grade of steatosis also showed the weakest staining intensity for PSD3. Regarding ARF6, hepatocytes showed multifocally moderate cytoplasmic and membrane positivity. The staining for activated ARF6, especially in membranes, appeared more pronounced in samples homozygous for 186L (Supplementary Fig. 7). However, immunohistochemistry does not allow for the quantification of PSD3 or activated ARF6.

Psd3 downregulation protects against FLD in mice

Next, we downregulated liver Psd3 in vivo by administrating triantennary N-acetylgalactosamine (GalNAc)-conjugated Gen 2.5 antisense oligonucleotides (ASOs) in C57BL/6 mice. Mice were fed a non-alcoholic steatohepatitis (NASH)-inducing diet for a total of 50 weeks, and during the last 16 weeks, groups of mice were treated with Psd3 ASO, control ASO or saline. Psd3 ASO treatment decreased the liver Psd3 mRNA expression levels by 98% (Fig. 6a) and reduced the liver weight, total liver triglyceride content and plasma alanine transaminase (ALT) levels (Fig. 6b–d). Psd3 ASO treatment did not affect the body weight of the mice (Fig. 7a).

Moreover, Psd3 ASO treatment reduced the total liver free cholesterol, liver cholesteryl ester, plasma aspartate transaminase (AST), total and LDL cholesterol levels but did not change the plasma triglyceride or high-density lipoprotein (HDL) cholesterol levels (Fig. 7b–h). Histologically, Psd3 ASO treatment reduced the liver collagen 1a1 (Col1a1) protein and liver lipid droplet levels (Figs. 6e,f and 8). Next, to examine the severity of liver disease using the semiquantitative scoring developed by Kleiner et al²⁸, we performed ordinal regression analyses and found that Psd3 ASO treatment reduced the severity of steatosis and lobular inflammation scores, and the NAFLD activity score (NAS), although the liver fibrosis score did not change significantly (Fig. 6g–j). To gain insights into the mechanisms by which Psd3 downregulation results in lower hepatic fat content, we examined expression levels of genes involved in triglyceride synthesis. Consistent with the in vitro experiments, Psd3 downregulation in mice fed the NASH-inducing diet reduced the expression of genes involved in de novo lipogenesis (Fasn, Acc1. Scd1) (Fig. 7i–l). Psd3 downregulation also reduced hepatic expression levels of monocyte chemoattractant protein 1 (Ccl2), involved in monocyte infiltration and α-smooth muscle actin (Acta2), a marker of myofibroblast formation (Fig. 7m,n).

To confirm our in vivo data in another model, liver Psd3 levels were silenced using a GalNAc-conjugated Psd3 ASO in mice fed a choline-deficient high-fat diet (CD-HFD) for 12 weeks (Supplementary Figs. 8 and 9). As expected, the CD-HFD led to a reduction in body weight irrespective of the treatment. However, this reduction was recovered at the end of the treatment. Psd3 ASO treatment resulted in robust downregulation of liver Psd3 mRNA levels. Consistent with the diet-induced NASH model, Psd3 ASO treatment reduced the liver total amount of triglycerides and free cholesterol in the CD-HFD model (Supplementary Fig. 8). Moreover, Psd3 ASO treatment reduced the severity of liver steatosis (Supplementary Fig. 9). However, there were no differences observed in liver inflammation or Col1A1 levels between the Psd3 and control ASO groups in this model (Supplementary Fig. 8).

Selection of gene loci

The study design is shown in Supplementary Fig. 1.

In the discovery phase, 32 genetic tag SNPs previously associated with triglycerides as the main (n = 24) or secondary trait (n = 8) at a genome-wide significance level were selected¹⁸ (Supplementary Table 1). All missense and nonsense variants within 50 kb of the selected loci that were on the Illumina HumanExome BeadChip (including the tag SNPs when available) were then tested for their association with liver fat content in 2,736 participants in the DHS. Liver fat content was measured using proton magnetic resonance spectroscopy. When the tag SNP was intergenic, variants in both genes adjacent to the tag SNP were examined.

Identification of the PSD3 genetic variant

Among the genetic variants tested, three had a MAF >5% in Europeans from the 1000 Genome Project (dbSNP, Current Build 154, released 21 April 2020) and a nominally significant association with liver fat content in the DHS. Among these variants, TM6SF2 rs58542926 and GCKR rs1260326, two well-known genetic variants associated with the entire spectrum of FLD as well as with circulating triglycerides^15,24, were associated with increased liver fat content. The other variant, namely the rs71519934 (found as rs7003060 that is in complete linkage disequilibrium (D′ = 1, r² = 1) in the PSD3 gene, was associated with lower liver fat content.

In the LBC, the rs71519934 had an overall MAF of 33%. When we stratified the cohort based on the centre of recruitment, we found that the rs71519934 had a higher frequency in individuals from Finland (MAF 48% and 42% in the two subgroups from Finland) than in participants from Italy (MAF 26% and 28% in the two subgroups from Italy; Supplementary Table 8) reflecting reference data (Exome Aggregation Consortium database). The genotype frequencies were in HWE in three-quarters of subcohorts (Supplementary Table 8). The deviation in the Milan subcohort may be because this subcohort is a selected population of individuals at risk for liver disease.

Owing to these differences in the MAF, we included the centre of recruitment as a covariate in subsequent multivariate analyses. To understand the effect of the genetic variant, we examined the histological degree of the different components of liver disease, namely, steatosis (the histological measurement of liver fat content), inflammation, ballooning (an index of cell damage) and fibrosis (the main predictor of mortality related to liver disease).

Study cohorts

Genotyping

All participants from the DHS were previously genotyped for the variants using an Illumina Infinium HumanExome BeadChip^12,14.

Participants from the LBC and from the central European independent replication cohort were genotyped by TaqMan 5′ nuclease assays (Life Technologies). The allelic discrimination probe for PSD3 rs71519934 was not commercially available. A custom assay for this variant has been designed (supplementary material).

Nine individuals denoted as AC homozygotes (n = 3), CT homozygotes (n = 3) and AC/CT heterozygous (n = 3) by the TaqMan assay were Sanger sequenced with consistent results (Supplementary Fig. 2).

UK Biobank participants were genotyped using two highly similar UK BiLEVE or UK Biobank Axiom arrays (>95% overlap). Genotyped data were then imputed based on the 1000 Genomes Phase 3, UK10K haplotype and Haplotype Reference Consortium reference panels⁴⁷. Genotype data for the rs71519934 dinucleotide change were not available in the UK Biobank; the rs7003060 (identifying the first nucleotide change of the rs71519934) was among directly genotyped variants and has been used instead.

Gene expression analysis

In vitro 2D cell culture and quantification of intracellular fat

Cryopreserved primary human hepatocytes were purchased from BioIVT and LONZA. After genotyping, cells were identified as homozygous for 186L (BioIVT lot number: BGW-M00995 (male, aged 50 years, BMI 20.4)) and homozygous for 186T (LONZA lot number: HUM183001 (female, aged 20 years, BMI 28.9)). Rat hepatoma McArdle (McA-RH7777 cells, homozygotes for 180T that corresponds to 186T in human PSD3 according to the alignment of human NP_056125.3 and rat XP_017455908.1) were from ATCC and human hepatocytes Huh7 (PSD3 L186L) were from the JCRB cell bank.

In basal conditions, primary cells were cultured with media provided by the respective companies, McA-RH7777 cells were cultured in DMEM containing 10% foetal bovine serum (FBS), and Huh7 were cultured in DMEM (low glucose) containing 10% FBS.

In experimental conditions, 24 h after seeding, cells were transiently transfected with 30 nM SCR siRNA (AM4611; ThermoFisher Scientific), human PSD3 siRNA (mix of s23653, s23654 and s23655 for primary and Huh7 cells;, ThermoFisher Scientific), rat Psd3 siRNA (mix of s157480, s157481 and s157482 for McA-RH7777 cells; ThermoFisher Scientific) or human ARF6 siRNA (mix of s1565, s1566 and s1567 for Huh7 cells; ThermoFisher Scientific) by Lipofectamine 3000 transfection reagent (L3000-075; ThermoFisher Scientific) according to the manufacturer’s instructions, and grown in serum-free regular medium supplemented with 10 µM (primary cells), 50 µM (McA-RH7777) or 25 µM (Huh7) OA for 48 h.

Intracellular neutral fat content was visualized by ORO staining and images were acquired using an Axio KS 400 Imaging System and AxioVision v.4.8 software (Zeiss) at ×100 magnification. The ORO-stained area was quantified using Biopix iQ software v.2.3.1 (refs. ^52,53).

RNA-seq and differentially expressed genes comparison

Human primary hepatocytes from donors homozygous for the 186T or 186L allele were cultured in 2D in their respective media. Total RNA was extracted and its purity and integrity number were evaluated (supplementary material). Illumina sequencing and subsequent analysis was carried out by Novogene (UK) (supplementary material). Differential expression analysis between two conditions/groups (two biological replicates per condition) was performed using the DESeq2 R package (v.2_1.6.3). The resulting two-sided P values were adjusted using the Benjamini and Hochberg’s approach for controlling the false discovery rate (FDR). Genes with an adjusted P value <0.05 were assigned as differentially expressed.

3D liver spheroid formation and quantification of intracellular fat

Cryopreserved primary human hepatocytes from two different donors (as described above) were used to generate spheroids. Cells were seeded into ultralow attachment 96-well plates (Corning) (supplementary material)⁵⁴ and transfected with negative control SCR siRNA or PSD3 siRNA as described above. The plates were then centrifuged at 100g for 5 min. Once the cells have collected at the bottom, self-aggregation leads to the formation of spheroids. Spheroids were cultured for 7 days (supplementary material) and their ATP-based viability was determined using a CellTiter-Glo Luminescent Cell Viability Assay kit (Promega). Cellular ATP was normalized to spheroid volume. Images of the spheroids were taken using an Axio Vert.A1 inverted microscope (Carl Zeiss AG). For quantification of neutral fat, spheroids were fixed and stained as described previously⁵⁵. The ORO-stained area was normalized to the number of 4,6-diamidino-2-phenylindole-stained nuclei and quantified using Image J (v.1.52h, NIH).

PSD3 polyclonal antibody generation

Human PSD3 (Uniprot Q9NYI0-1) amino acids D527–T1048 were expressed in Sf21 cells in fusion with an N-terminal 6× His tag (separated by a flexible linker) from the pFastBac1 vector (ThermoFisher Scientific). His–PSD3 was purified under endotoxin-free conditions by immobilized metal affinity chromatography on a nickel column (GE Healthcare) and size-exclusion chromatography on a Superdex200 column (GE Healthcare) in 2× PBS buffer pH 7.4. Antibody generation was carried out by Antibody Applications. Purified His–PSD3 (D527–T1048) was used to immunize three male 22-week-old New Zealand white rabbits, using a standard 90-day protocol. Anti-PSD3 IgG from each animal was affinity purified from final bleed rabbit serum on an NHS-activated Sepharose 4 fast flow agarose column (Cytiva) coupled to His–PSD3. A lead antibody was validated via silencing of endogenously expressed PSD3 in human hepatoma HepaRG cells (Supplementary Fig. 10).

Immunoblotting

Proteins were separated by SDS–PAGE and transferred to nitrocellulose membranes in accordance with standard procedures. Protein lysates were prepared from cell extracts using M-PER protein extraction reagent (ThermoFisher Scientific) supplemented with 10% protease inhibitor (SIGMAFAST; Sigma-Aldrich). Primary antibodies used were: anti-PSD3 custom antibody (dilution 1:1,000, as described above), rabbit anti-calnexin (dilution 1:1,000; Sigma-Aldrich) and mouse monoclonal anti-Arf6 (dilution 1:1,000, provided with the Arf6 activation kit, Cell Biolabs Inc.). Blots were probed with primary antibodies followed by the appropriate horseradish peroxidase-conjugated secondary antibody (dilution 1:2,000) and developed using ECL substrate (Immobilon Western Chemiluminescent HRP Substrate; Merck Millipore).

ARF6 activation assay

A gamma adaptin ear-containing ARF binding 3 family protein-binding domain (GGA3 PBD) pull-down assay was performed using an ARF6 activation assay kit (STA-407-6, Cell Biolabs) according to the manufacturer’s protocol. Briefly, Huh7 cells were transfected with either SCR siRNA, PSD3 siRNA as described above, or with ARF6 siRNA (mix of s1565, s1566 and s1567, ThermoFisher Scientific) for 48 h (supplementary material). Activated ARF6 was then pulled down using GGA3 PBD agarose beads (supplementary material) and along with the corresponding total cell fraction proteins, was analysed by immunoblotting for ARF6 and calnexin.

PSD3 and active ARF6 immunohistochemistry

Immunostaining for PSD3 and active ARF6 was performed on formalin-fixed and paraffin-embedded liver specimens from eight patients with NAFLD from the LBC (n = 4 heterozygotes (L186T) and n = 4 homozygotes for the 186L allele of the PSD3 gene). The automated Ventana BenchMark ULTRA IHC/ISH Staining Module (Ventana) was used together with the Ultraview 3,3′-diaminobenzidine v.1 detection system (Ventana). Tissue sections (4 µm) were deparaffinized (at 72 °C) and incubated in EZPrep Volume Adjust (Ventana). At intervals, steps slides were washed with a cell conditioning solution, Cell Conditioner 1, Tris–EDTA-based buffer, pH 7.8. For analysis of PSD3, sections were first incubated with ultraviolet inhibitor blocking solution for 4 min at 36 °C, then with the protease 1 for 4 min at 36 °C, and finally with anti-PSD3 primary antibody (29-749, dilution 1:100; ProSci) for 1 h at room temperature. For analysis of ARF6, a heat-induced antigen retrieval protocol set for 56 min at 91 °C was carried out using a Tris–EDTA–boric acid, pH 8.4, buffer (Cell Conditioner 1). Sections were incubated with ultraviolet inhibitor blocking solution for 4 min and then with anti-activated ARF6 primary antibody (26918, dilution 1:200; NewEast Biosciences) for 1 h and 4 min at room temperature. For both analyses, incubation with horseradish peroxidase-linked secondary antibody (8 min) (1706515 for PSD3, dilution 1:100, and 170-6516 for active ARF6, dilution 1:150; both Bio-Rad) was followed by incubation with 3,3′-diaminobenzidine chromogen and substrate (8 min) and copper enhancer (4 min). Counterstain (haematoxylin) was applied for 12 min before 4-min incubation with bluing reagent.

Stained tissue sections were analysed under an optical microscope by a pathologist who was blinded to the PSD3 genotype.

De novo triglyceride synthesis

De novo triglyceride synthesis was analysed using ³H-glycerol⁵⁶. Briefly, McA-RH7777 and Huh7 cells were seeded in triplicate in six-well plates. Twenty-four hours after seeding, cells were transfected with Psd3 siRNA or SCR siRNA for 48 h in regular medium (supplementary material). Forty-eight hours after transfection, cells were incubated with ³H-glycerol (PerkinElmer) plus OA for 15, 30 or 60 min. Cell lysates were collected and lipids were extracted (supplementary material)⁵⁷, separated and quantified by thin-layer chromatography. Spots corresponding to triglycerides were visualized with iodine vapour and added to vials with scintillation fluid. Radioactivity was measured using a scintillation counter as disintegrations per min.

Apolipoprotein B secretion

Apo B secretion in vitro was measured by adapting the previously described protocol⁵³. Briefly, McA-RH7777 cells were transfected with Psd3 siRNA or SCR siRNA. Forty-eight hours after transfection, cells were labelled with ³⁵S Met/Cys (PerkinElmer) and chased for 5, 15, 30 or 60 min (supplementary material). Apo B was then immunoprecipitated (supplementary material) and subsequently separated on 3–8% gradient SDS–PAGE. The gel was then dried, exposed overnight on BAS-MS film and visualized using Phosphoimager (Fujifilm FLA-3000).

Beta oxidation

Beta oxidation was analysed using ³H-palmitate⁵⁸. Briefly, McA-RH7777 cells were transfected with Psd3 siRNA or SCR siRNA as described above. Forty-eight hours after transfection, cells were labelled with ³H-palmitate (PerkinElmer). Labelled palmitate from the cell media was then precipitated (supplementary material). The supernatant was collected and radioactivity was measured using a scintillation counter as disintegrations per min.

Liver Psd3 silencing in mice

Statistical analysis

For the DHS, the P values for associations between liver fat content and the target variants were calculated using linear regression analysis adjusted for age, gender and the four leading principal components of ancestry. An additive genetic model was used in all analyses.

For the LBC and for the central European independent replication cohort, the association between the PSD3 rs71519934 variant and liver disease was evaluated under an additive genetic model by binary logistic (prevalence of liver disease) or ordinal regression (liver histological features) analysis adjusted for age, gender, BMI, centre of recruitment and number of PNPLA3 I148M mutant alleles. All analyses were performed using IBM SPSS statistics v.27.

For the UK Biobank, liver PDFF was first rank-based inverse normal transformed, and then the association with PSD3 rs7003060 was examined using linear regression adjusted for age, gender, BMI, the first ten genomic principal components and array type under an additive genetic model using R v.3.6.1, MATLAB R2020b (academic license).

For descriptive statistics, data are shown as the mean and s.d. or the median and interquartile range (i.q.r.) as appropriate. Categorical traits are shown as numbers and proportions. For continuous traits, P values were calculated by linear regression under an additive genetic model unadjusted or adjusted for age, gender and BMI. Non-normally distributed traits were log-transformed before being entered into the model. For categorical traits, P values were calculated by the chi-squared test or by binary logistic regression adjusted for age, gender and BMI using IBM SPSS statistics v.27.

Differences in PSD3 and NAT2 expression levels in human tissues were evaluated by the Mann–Whitney non-parametric test (for comparisons between healthy individuals and those with FLD) or by linear regression analysis (for comparisons among genotypes).

Differential gene expression analysis in primary cells between two conditions/groups (two biological replicates per condition) was performed using the DESeq2 R package (v.2_1.6.3) as described above (‘RNA-seq and differentially expressed gene comparison’). The resulting two-sided P values were adjusted using the Benjamini and Hochberg’s approach for controlling the FDR.

For meta-analyses of the histological cohorts, an inverse variance meta-analysis of the two studies (LBC and central European replication cohort) was performed using package ‘meta’ with fixed and random effect models in R v.3.6.1 (ref. ⁶²).

For in vivo and in vitro studies, data are shown as the mean and s.d. P values were calculated by the Mann–Whitney non-parametric test (in vitro) or one-way analysis of variance (ANOVA) Kruskal–Wallis non-parametric test with Dunn’s correction for multiple comparisons (in vivo) using GraphPad Prism v.9. The severity scores of liver disease in the in vivo mouse studies were analysed by using ordinal regression analyses. All the reported P values are two-sided.

Ethical approval

Our research complies with the principles outlined in the Declaration of Helsinki. The specific board/committee and institution that approved each protocol are listed in the relevant section. Each subject provided written informed consent.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Peer review information

Nature Metabolism thanks Anna Alisi, Herbert Tilg and the other, anonymous, reviewer for their contribution to the peer review of this work. Alfredo Gimenez-Cassina was the Primary Handling Editor.