Secondary Logo

Journal Logo

ARTICLE: LIVER

The Protection Conferred by HSD17B13 rs72613567 Polymorphism on Risk of Steatohepatitis and Fibrosis May Be Limited to Selected Subgroups of Patients With NAFLD

Vilar-Gomez, Eduardo MD, PhD1; Pirola, Carlos J. PhD2; Sookoian, Silvia MD, PhD3; Wilson, Laura A. ScM4; Liang, Tiebing PhD1; Chalasani, Naga MD1

Author Information
Clinical and Translational Gastroenterology: September 2021 - Volume 12 - Issue 9 - p e00400
doi: 10.14309/ctg.0000000000000400

Abstract

INTRODUCTION

Nonalcoholic fatty liver disease (NAFLD) is a multifactorial disease in which genetic risk factors and associated comorbidities might play key roles in the development and the progression of this disease. Recent genome-wide and exome-wide association studies have identified 2 important variants—rs738409 in PNPLA3 (1) and rs72613567 in HSD17B13 (2)—linked to NAFLD and its progression. In 2018, Abul-Husn et al. (3) reported that a loss-of-function splice variant (rs72613567) of the HSD17B13 gene was significantly associated with a lower risk of NAFLD progression and chronic liver disease. Furthermore, the protective effect of the HSD17B13 rs72613567 A-allele against NAFLD, alcoholic liver disease, and hepatocellular carcinoma has been extensively confirmed in other studies including more diverse and heterogeneous populations (2,4–10).

HSD17B13 encodes for a liver-specific lipid droplet (LD)–associated protein (11) that seems to be implicated in hepatic-lipid biogenesis and metabolism. HSD17B13 expression is significantly upregulated in the liver of patients and mice with NAFLD (7,12). Most recently, 17β-HSD13 has been reported to regulate retinoid metabolism by acting as a retinol dehydrogenase (7). 17β-HSD13 has been grouped with the short-chain dehydrogenases/reductases family, of which some members play a key role in the final step of estrogen and androgen steroid metabolism (13). However, the role of 17β-HSD13 in the regulation of sex steroid metabolism is still unclear and remains to be elucidated (14).

PNPLA3 rs738409 is another genetic variant strongly implicated in the regulation of lipid metabolism and LD biology (1). The PNPLA3 rs738409 minor allele (G) is recognized as a major determinant in the progression of NAFLD to steatohepatitis, cirrhosis, and even hepatocellular carcinoma (15,16). PNPLA3 has a triglyceride hydrolase activity and acts as a lipase responsible for retinyl palmitate hydrolysis in hepatic stellate cells (HSCs) in humans. In the presence of PNPLA3-rs738409 variation, there is reduced activity of these enzymatic activities, leading to impairment of lipid catabolism, LDs remodeling, and very-low density lipoprotein secretion in hepatocytes with reduced fatty acid hydrolysis, increased triglyceride accumulation, and decreased very-low density lipoprotein secretion; all these events result in hepatic steatosis, inflammation, and fibrosis (1,17,18).

Little is known about the interplays between PNPLA3 and HSD17B13 and underlying clinical conditions and their impact on the development and progression of NAFLD. The knowledge of the biological interactions between genetic variants with opposite effects on the disease risk, including PNPLA3 rs738409 that confers about 3.2-fold greater risk per G-risk allele of inflammation and fibrosis (15) and HSD17B13 rs72613567 that associates with approximately 40% reduction in the odds of developing steatohepatitis per A-INS allele (2), may increase the predictive accuracy of genetic risk assessment. Similarly, understanding the impact of these interactions on clinical phenotypes is a critical step to identity patients at high risk of NAFLD progression in whom more precise and targeted therapies could halt the disease progression. Thus, our study was designed to explore how PNPLA3 rs738409 and associated conditions might modify or moderate the relationship between HSD17B13 rs72613567 and risk of steatohepatitis and fibrosis. To this end, we assessed gene–gene and gene–phenotype interactions.

MATERIAL AND METHODS

Study design and participants

This cross-sectional analysis comprised patients enrolled in the nonalcoholic steatohepatitis Clinical Research Network (NASH CRN) studies, including NAFLD Adult Databases 1 and 2 (NCT01030484), along with PIVENS (NCT00063622) and FLINT (NCT01265498) trials. Detailed information on patients' characteristics, including demographics, laboratory, and liver histology of NASH CRN studies, has been previously published (19–21). Some participants included in this analysis were part of the study by Ma et al. that investigated the association between HSD17B13 variants and liver histology in patients with NAFLD (7). In brief, NASH CRN is a multiethnic prospective cohort recruiting patients with biopsy-proven NAFLD from 9 clinical centers across the United States. Appropriate institutional review boards approved the NASH CRN study, and all participants provided written informed consent. Participants were 18 years or older and had a histological diagnosis of NAFLD (≥5% of hepatocytes containing macrovesicular fat) in absence of a significant history of alcohol intake (>14 or >21 standard drinks for women and men, respectively) in the previous 2 years of the study enrollment.

A total of 1,697 adults with histologically confirmed NAFLD were enrolled through December 2019. Of them, 1,153 (68%) non-Hispanic whites with detailed information for HSD17B13 rs72613567, PNPLA3 rs738409, alcohol intake, centrally scored liver histology severity, and clinical phenotypes were included. All laboratory, clinical, and histological data analyzed in this study were obtained within 6 months of the liver biopsy. A thorough description of our selected cohort of non-Hispanic whites has been previously published (22).

Clinical and laboratory assessment

Baseline examinations included a medical history interview, physical examination, and blood draw. Body mass index (BMI) was calculated as body weight in kilograms divided by height in square meters. Diabetes was defined as a fasting plasma glucose level ≥ 126 mg/dL, 2-hour oral glucose tolerance test glucose of ≥ 200 mg/dL, glycosylated hemoglobin level of 6.5% or higher, or current use of glucose-lowering therapy. Alcohol consumption was obtained through Alcohol Use Disorders Identification Test (23) and Lifetime Drinking History (24) questionnaires. Nonheavy drinking was classified as a history of alcohol intake ≤14 and ≤21 standard drinks per week for women and men, respectively, during the 2 years preceding study enrollment. Nondrinkers were those who did not report alcohol intake at any time or those who had a history of nonheavy alcohol intake but became abstainers in the last 2 years before study enrollment. Other covariates including lipid-related panel, alanine (ALT) and aspartate (AST) aminotransferases, and a history of tobacco use were also gathered for study purposes.

Histologic evaluations

As per protocol, all liver biopsies are reviewed in a blinded manner by the NASH CRN Pathology Committee and scored according to the NASH CRN Scoring System (25) as follows: steatosis (0–3), lobular inflammation (0–2), ballooning degeneration (0–2), and overall fibrosis (0–4). The diagnosis of steatohepatitis was based on the severity of individual lesions and their patterns of distribution and classified as follows: not steatohepatitis, borderline steatohepatitis, and definite steatohepatitis (26). For study purposes, clinically significant fibrosis was considered as the presence of a stage of fibrosis ≥2, whereas a diagnosis of steatohepatitis was established in the presence of either borderline or definite steatohepatitis. Our primary study outcomes were risk of fibrosis and steatohepatitis.

Genetic variants, genotyping, and quality control

We analyzed HSD17B13 rs72613567 and PNPLA3 rs738409 genetic variants because of their established association with liver histology severity. Some of the data used in our analysis were partly reported in earlier articles on the discovery of susceptibility genotypes (7,27,28). The DNA was prepared from blood drawn at baseline; HSD17B13 rs72613567 and PNPLA3 rs738409 were successfully genotyped in 100% of the DNA samples. All DNA samples were blindly duplicated to assess the reproducibility of genotypes. A reproducibility of 100% was obtained. Detailed information on rs738409 and rs72613567 genotyping can be found in the Supplementary Material (see Supplementary Digital Content 1, https://links.lww.com/CTG/A669).

Statistical analysis

Categorical variables were expressed as frequencies/percentages and compared by χ2 test or Cochran-Armitage trend test for ordered alternatives. Continuous variables were presented as mean ± SD and compared through t test or Mann-Whitney U test, depending on whether they were or not normally distributed, respectively. The one-way analysis of variance was used to determine whether there were any statistically significant differences between the means of HSD17B13 rs72613567 genotypes.

Either ordinal or binary logistic regression models were implemented to assess the association of genetic variants or underlying risk factors with histologic outcomes including 2 or more ordered categories, respectively. For genotypes, logistic regression models assumed either an additive (−/−), (−/A), (A/A) coded 0, 1, or 2, respectively, or a dominant (−/−) or (−/A) + (A/A) genetic model. Odds ratios (OR) and their confidence intervals (CIs) express the strength of the association of genetic variants or potential risk factors with histologic outcomes. All analyses were adjusted for well-known risk factors associated with NAFLD severity: age, sex, type 2 diabetes (T2D), BMI, PNPLA3 rs738409, and nonheavy alcohol intake.

We further investigated the effect of interactions of HSD17B13 rs72613567 with either PNPLA3 rs738409 or selected risk factors and its impact on histologic outcomes. First, we created multiplicative interaction terms between HSD17B13 rs72613567 and each risk factor (PNPLA3 rs738409, age, BMI, sex, and T2D) using logistic regression models and including significant fibrosis or steatohepatitis as binary outcomes. Second, we used the margins command to visualize and interpret adjusted predictions of histologic outcomes for each HSD17B13 rs72613567 genotype at different levels of each risk factor (29). Third, we performed a moderation analysis that examined whether our selected risk factors can strengthen, weaken, or reverse the nature of the relationship between HSD17B13 rs72613567 and histologic outcomes (30). Moderation analyses were performed in SPSS statistical software using model 1 of PROCESS macro (version 3.5), which includes HSD17B13 rs72613567 as the exposure variable, each risk factor as a moderator, and histologic scores as individual outcomes (31). Conditional effects (β coefficients) were determined using 95% bootstrap bias-corrected CIs based on 10,000 bootstrap samples (32,33). β coefficient weights provide an index of the magnitude of the effect size (31) between HSD17B13 rs72613567 and histologic outcomes, and it is considered significant if the upper and lower bounds of the 95% bias-corrected CIs do not contain zero (31). Moderation analyses were adjusted for age, sex, T2D, BMI, PNPLA3 rs738409, and nonheavy alcohol intake. Moderation analyses were initially performed under an additive genetic model of HSD17B13 rs72613567. Results based on this genetic model displayed a dose-response relationship between (−/A) and (A/A) and the risk of NAFLD severity. Thus, we also decided to examine the effect of HSD17B13 rs71623567 on histologic outcomes under a dominant genetic model. Further details are included in the Supplementary Material, (see Supplementary Digital Content 1, https://links.lww.com/CTG/A669, including conceptual diagrams of our moderation models.

Because moderation analyses showed that the effect of HSD17B13 rs72613567 on the risk of NAFLD severity was markedly different at selected strata of moderators, sensitivity analyses were conducted separately for men and women, diabetes (yes or no), BMI (<35 or ≥35), age (<45 or ≥45), and PNPLA3 rs738409 (CC or GC + GG), and all multivariable analysis were adjusted for nonheavy alcohol intake along with those covariates who were not tested as moderators itself. The cutoff points for age and BMI were selected through margins analysis, which showed that the risk of having steatohepatitis among individuals with HSD17B13 rs72613567 (−/−) or (−/A) + (A/A) genotypes were significantly different starting at a BMI of ≥35 (P = 0.009) or an age of 45 years or older (P = 0.048).

Because our data showed that HSD17B13 rs72613567 might have opposite effects regarding steatosis and fibrosis severity, we sought to investigate whether the effect of this variant on inflammation and fibrosis scores might be explained partly by a change in steatosis severity. To do so, a causal mediation analysis was performed through the CAUSALMED procedure (SAS Institute) (34). This is a regression-based approach that examines whether an explanatory variable (i.e., HSD17B13 rs72613567) exerts its effect on outcome (i.e., fibrosis severity) through direct (HSD17B13 rs72613567 itself) and/or indirect (i.e., steatosis) pathways. A more detailed explanation of causal mediation analysis can be found in the Supplementary Material (see Supplementary Digital Content 1, https://links.lww.com/CTG/A669).

To adjust for multiple testing, false discovery rate correction was performed using Benjamini-Hochberg procedure, and adjusted 95% CI of the estimates and Q values were calculated through qqvalue package (Stata software), with a q < 0.05 considered statistically significant (35,36). Statistical analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC), STATA SE version 16 (Stata, College Station, TX), or SPSS version 26 (Chicago, IL).

RESULTS

Baseline characteristics of the cohort based on HSD17B13 rs72613567 genotypes

A total of 1,153 non-Hispanic whites with biopsy-proven NAFLD were included after fulfilling our inclusion and exclusion criteria (see Supplemental Figure 1, Supplementary Digital Content 2, https://links.lww.com/CTG/A670). Among non-Hispanic whites, the genotype frequency of HSD17B13 rs72613567 was as follows: (−/−) (0.64), (−/A) (0.30), and (A/A) (0.6), whereas the minor allelic frequency (A-insertion) was 0.21. Associations between HSD17B13 rs72613567 genotypes and baseline characteristics are summarized in Table 1. Clinical, demographic, anthropometric, and PNPLA3 rs738409 features were evenly distributed across all HSD17B13 rs72613567 genotypes. Under an additive genetic model, the strength and the direction of the associations of HSD17B13 rs72613567 with aminotransferases levels and histologic scores severity was consistent with published data (2,3,6,7).

Table 1. - Baseline characteristics according to HSD17B13 rs72613567 genotypes
Variables (−/−) n = 737 (−/A) n = 348 (A/A) n = 68 P value
Age (yr) 51.13 ± 11.32 50.02 ± 11.74 50.01 ± 11.57 0.283
Sex (women), n (%) 468 (64) 213 (61) 46 (68) 0.986
Type 2 diabetes mellitus, n (%) 265 (36) 114 (33) 29 (43) 0.940
Hypertension, n (%) 446 (61) 198 (57) 41 (60) 0.452
Tobacco use, n (%) 62 (9) 22 (6) 5 (8) 0.330
Body mass index (kg/m2) 34.77 ± 6.45 34.87 ± 6.19 35.88 ± 7.14 0.399
Nonheavy drinkers, n (%) 277 (38) 134 (38) 22 (32) 0.704
PNPLA3 rs738409 (GC) + (GG), n (%) 505 (69) 232 (67) 39 (57) 0.100
Laboratory reports
 Triglycerides (mg/dL) 183.02 ± 180.98 191.65 ± 184.53 181.65 ± 94.46 0.745
 Total cholesterol (mg/dL) 192.56 ± 43.77 192.55 ± 44.18 189.53 ± 50.84 0.861
 HDL cholesterol (mg/dL) 43.64 ± 11.53 42.90 ± 11.96 42.32 ± 10.33 0.471
 LDL cholesterol (mg/dL) 115.82 ± 36.53 115.93 ± 38.51 113.80 ± 39.61 0.911
 HOMA-IR 7.08 ± 9.81 6.83 ± 9.26 7.33 ± 7.28 0.884
 Glycosylated hemoglobin (%) 6.20 ± 1.20 6.07 ± 1.00 6.14 ± 1.09 0.202
 Alanine aminotransferase (U/L) 74.27 ± 49.33 65.00 ± 43.05 49.19 ± 35.33 <0.001
 Aspartate aminotransferase (U/L) 53.93 ± 33.64 48.38 ± 30.91 38.65 ± 16.86 <0.001
Histology scores
 Steatosis, n (%) <0.01
 <5% 19 (2.6) 2 (0.6) 0 (0)
 5–33% 281 (38.1) 117 (33.6) 22 (32.4)
 33–66% 269 (36.5) 120 (34.5) 24 (35.2)
 >66% 168 (22.8) 109 (31.3) 22 (32.4)
Lobular inflammation, n (%) <0.01
 No foci 1 (0.1) 1 (0.3) 0 (0)
 <2 foci/200× 356 (48.3) 204 (58.6) 46 (67.6)
 2–4 foci/200× 295 (40) 110 (31.6) 14 (20.6)
 >4 foci/200× 85 (11.5) 33 (9.5) 8 (11.8)
 Ballooning, n (%) 0.037
 None 231 (31.3) 124 (35.6) 28 (41.2)
 Few 204 (27.7) 100 (28.8) 16 (23.5)
 Many 302 (41) 124 (35.6) 24 (35.3)
Fibrosis stages, n (%) 0.032
 0 154 (20.9) 86 (24.7) 18 (26.5)
 1 193 (26.2) 102 (29.3) 18 (26.5)
 2 154 (20.9) 71 (20.5) 13 (19)
 3 155 (21) 61 (17.5) 11 (16.2)
 4 81 (11) 28 (8) 8 (11.8)
Steatohepatitis, n (%) <0.01
 None 135 (18.3) 85 (24.4) 17 (25)
 Borderline steatohepatitis 144 (19.5) 67 (19.3) 16 (23.5)
 Definite steatohepatitis 458 (62.2) 196 (56.3) 35 (51.5)
 Steatohepatitis, n (%)a 602 (82) 263 (76) 51 (75) 0.018
 Significant fibrosis (stage ≥2), n (%) 390 (53) 160 (46) 32 (47) 0.047
HDL, high-density lipoprotein; HOMA-IR, homeostatic model assessment of insulin resistance; LDL, low-density lipoprotein; PNPLA3, patatin-like phospholipase domain-containing protein 3.
aIt represents a composite of borderline and definite steatohepatitis.

Adopting an additive genetic model of HSD17B13 rs72613567, the A-insertion allele was associated with increased risk of steatosis (OR: 1.35, P < 0.001) but decreased risks of lobular inflammation (OR: 0.69, P < 0.001), ballooning (OR: 0.80, P = 0.022), steatohepatitis (OR: 0.77, P = 0.010), and fibrosis (OR: 0.84, P = 0.050) (Table 2). Adopting a dominant genetic model, which includes 1 or 2 copies of allele A and it is justified by the low number of homozygous A/A, the associations of HSD17B13 rs72613567 with histology scores were similar to those found in the additive genetic model (Table 2).

Table 2. - Multivariable association of HSD17B13 rs72613567 or selected moderators with histology severity
Steatosis (0–3) Lobular inflammation (0–3) Ballooning (0–2)
Odds ratio (95% CI) P valuea Odds ratio (95% CI) P valuea Odds ratio (95% CI) P valuea
Exposure
HSD17B13 (additive model)
 rs72613567 (-/A) or (A/A)b 1.35 (1.13–1.61) <0.01 0.69 (0.57–0.84) <0.001 0.80 (0.67–0.97) 0.022
HSD17B13 (dominant model)
 rs72613567 (−/A) + (A/A)b 1.47 (1.17–1.84) <0.01 0.65 (0.51–0.83) <0.001 0.79 (0.63–0.99) 0.049
Potential moderators
 Age (yr) 0.97 (0.96–0.98) <0.001 0.99 (0.98–1.00) 0.116 1.01 (1.00–1.02) 0.023
 Body mass index (kg/m2) 0.99 (0.98–1.01) 0.290 0.98 (0.97–1.00) 0.225 1.03 (1.02–1.05) <0.001
 Sex (women) 1.20 (0.95–1.52) 0.122 1.60 (1.25–2.06) <0.001 1.54 (1.22–1.95) <0.001
 Type 2 diabetes (yes) 1.13 (0.89–1.43) 0.298 1.37 (1.06–1.75) 0.013 2.26 (1.77–2.89) <0.001
PNPLA3 (dominant model)
 rs738409 (GC) + (GG) 1.43 (1.14–1.80) 0.002 1.34 (1.05–1.72) 0.018 0.95 (0.75–1.20) 0.686
Steatohepatitis categories (0–2) Overall fibrosis (0–4) Significant fibrosis (stage ≥2)
Odds ratio (95% CI) P valuea Odds ratio (95% CI) P valuea Odds ratio (95% CI) P valuec
Exposure
HSD17B13 (additive model)
 rs72613567 (−/A) or (A/A)b 0.77 (0.64–0.94) 0.010 0.84 (0.70–1.00) 0.050 0.82 (0.67–1.01) 0.068
HSD17B13 (dominant model)
 rs72613567 (−/A) + (A/A)b 0.75 (0.59–0.96) 0.022 0.80 (0.64–0.99) 0.046 0.78 (0.60–1.01) 0.061
Potential moderators
 Age (yr) 1.01 (1.00–1.02) 0.011 1.03 (1.02–1.04) <0.001 1.04 (1.03–1.05) <0.001
 Body mass index (kg/m2) 1.03 (1.01–1.05) <0.01 1.03 (1.02–1.05) <0.001 1.04 (1.02–1.06) <0.001
 Sex (women) 1.20 (0.93–1.54) 0.153 0.99 (0.79–1.24) 0.988 0.99 (0.75–1.30) 0.963
 Type 2 diabetes (yes) 2.42 (1.84–3.19) <0.001 2.55 (2.02–3.22) <0.001 2.61 (1.98–3.44) <0.001
PNPLA3 (dominant model)
 rs738409 (GC) + (GG) 0.95 (0.73–1.23) 0.708 1.36 (1.09–1.71) <0.01 1.55 (1.18–2.03) <0.01
The following variables were included in all multivariable analyses: HSD17B13 rs72613567, nonheavy alcohol intake, age, sex, body mass index, T2D, and PNPLA3 rs738409. CI, confidence interval.
aP values were calculated by ordered logistic regression.
bThe reference genotype for comparison was (−/−).
cP values were calculated by binary logistic regression.

Notably, carriage of the HSD17B13 rs72613567 A-allele depicted opposite effects on fibrosis and steatosis severity (Table 1). Because steatosis and fibrosis scores were inversely correlated (r = −0.10, P < 0.001) in our data, we sought to investigate whether the protective effect of the rs72613567 A-allele on fibrosis was significantly mediated by a change in steatosis severity. Our causal mediation analysis confirmed that the rs72613567 A-allele might directly modify the risk of fibrosis severity (β coeff for total effect: −0.17, 95% CI: −0.31 to −0.02 and β coeff for direct effect: −0.17, 95% CI: −0.31 to −0.02 were similar) and not likely operate through steatosis (β coeff for indirect or steatosis-mediated effect: −0.0004, 95% CI: −0.02 to 0.02) (see Supplemental Table 1, Supplementary Digital Content 1, https://links.lww.com/CTG/A669).

HSD17B13 rs72613567 and risk of steatohepatitis and fibrosis at strata of PNPLA3 rs738409

First, we sought to confirm the direction and strength of the association between HSD17B13 rs72613567 (−/A) or (A/A) genotypes and histologic outcomes at strata of PNPLA3 rs738409 genotypes (CC or GC + GG). A moderation analysis under an additive genetic model of HSD17B13 rs72613567 showed that either HSD17B13 rs72613567 (−/A) or (A/A) genotype was associated with a lower risk of steatohepatitis and fibrosis among individuals with PNPLA3 rs738409-(CC) genotype. Nevertheless, this protection was not evident in patients with PNPLA3 rs738409-(GC + GG) genotypes (see Supplemental Table 2 and Supplemental Figure 2A, Supplementary Digital Contents 1 and 2, https://links.lww.com/CTG/A669 and https://links.lww.com/CTG/A671). Further moderation analyses performed under a dominant genetic model of HSD17B13 rs72613567 confirmed that the A-allele was associated with remarkable protective effects on risk of steatohepatitis (β coeff: −0.19, P = 0.019) and fibrosis (β coeff: −0.45, P < 0.001) among individuals with PNPLA3 rs738409-(CC) genotype; however, this protection was fully mitigated among those with PNPLA3 rs738409-(GC) + (GG) genotypes (Table 3).

Table 3. - Conditional effect of HSD17B13 rs72613567 (−/A) + (A/A) genotypes on risk of steatohepatitis and fibrosis at levels of moderators
Moderators Univariate analysis Multivariable analysisb
Conditional effecta Standard error 95% CI P value Conditional effecta Standard error 95% CI P value
Steatohepatitis categories (0–2)
PNPLA3 rs738409
 CC, n = 377 −0.21 0.08 −0.37 to −0.04 0.014 −0.19 0.08 −0.35 to −0.03 0.019
 GC + GG, n = 776 −0.09 0.06 −0.21 to 0.03 0.136 −0.08 0.05 −0.19 to 0.03 0.150
 BMI ≥35, n = 506 −0.18 0.06 −0.29 to −0.06 <0.01 −0.17 0.06 −0.28 to −0.06 <0.01
 BMI <35, n = 647 −0.03 0.09 −0.20 to 0.15 0.766 −0.003 0.09 −0.17 to 0.17 0.972
 Age ≥45, n = 838 −0.18 0.06 −0.30 to −0.07 <0.01 −0.18 0.06 −0.29 to −0.07 <0.01
 Age <45, n = 315 0.02 0.09 −0.16 to 0.20 0.845 0.02 0.09 −0.16 to 0.19 0.851
 Women, n = 727 −0.18 0.06 −0.30 to −0.06 <0.01 −0.18 0.06 −0.29 to −0.06 <0.01
 Men, n = 426 −0.04 0.08 −0.19 to 0.12 0.661 −0.03 0.08 −0.18 to −013 0.751
 T2D (yes), n = 408 −0.18 0.08 −0.34 to −0.02 0.026 −0.18 0.08 −0.34 to −0.03 0.022
 T2D (no), n = 745 −0.09 0.06 −0.21 to 0.03 0.130 −008 0.06 −0.20 to 0.03 0.145
Fibrosis (0–4)
PNPLA3 rs738409
 CC, n = 377 −0.50 0.13 −0.77 to −0.23 <0.001 −0.45 0.12 −0.70 to −0.20 <0.001
 GC + GG, n = 776 −0.03 0.09 −0.22 to 0.16 0.735 −0.008 0.09 −0.18 to 0.17 0.932
 BMI ≥35, n = 506 −0.28 0.09 −0.46 to −0.09 <0.01 −0.22 0.08 −0.40 to −0.05 0.010
 BMI <35, n = 647 −0.01 0.14 −0.30 to 0.27 0.921 0.002 0.13 −0.26 to 0.27 0.991
 Age ≥45, n = 838 −0.24 0.09 −0.42 to −0.06 0.010 −0.23 0.09 −0.40 to −0.06 <0.01
 Age <45, n = 315 −0.05 0.14 −0.34 to 0.23 0.711 −0.04 0.14 −0.31 to 0.24 0.799
 Women, n = 727 −0.19 0.09 −0.39 to 0.00 0.051 −0.17 0.09 −0.35 to 0.01 0.067
 Men, n = 426 −0.18 0.12 −0.43 to 0.07 0.162 −0.14 0.12 −0.37 to 0.10 0.262
 T2D (yes), n = 408 −0.33 0.12 −0.58 to −0.08 0.010 −0.32 0.12 −0.56 to −0.07 0.011
 T2D (no), n = 745 −0.10 0.09 −0.28 to 0.09 0.291 −0.07 0.09 −0.2 to 0.11 0.429
The reference genotype for all comparative analyses HSD17B13 rs72613567 (−/−).
BMI, body mass index; CI, confidence interval; T2D, type 2 diabetes.
aRepresents the effect of the exposure (HSD17B13 rs72613567 (−/A) + (A/A) genotypes) on outcomes at values of the moderator (i.e., PNPLA3 rs738409 genotypes (CC) or (GC) + (GG)).
bAnalysis adjusted for nonheavy alcohol intake and those covariates not tested as moderators.

Figure 1a shows the difference in the risk of significant fibrosis between individuals with HSD17B13 rs72615367 (−/A) + (A/A) and (−/−) genotypes at strata of PNPLA3 rs738409 genotypes. In the presence of PNPLA3 rs738409-(CC) genotype, the risk of significant fibrosis was significantly reduced among carriers of HSD17B13 rs72613567 A-allele as compared with homozygous for the A-allele deletion (−/−) (adjusted risk difference −15, 95% CI: −24 to −6; P < 0.001). By contrast, the risk of significant fibrosis was not significantly different between HSD17B13 rs72613567 A-allele carriers and homozygous (−/−) in patients with PNPLA3 rs738409-(CC) + (GG) genotypes (adjusted risk difference −0.7, 95% CI: −7 to 6; P = 0.821).

Figure 1.
Figure 1.:
Effect of HSD17B13 rs72613567 (dominant genetic model) on risk of significant fibrosis or steatohepatitis at levels of moderators. Analysis based on margins plots. (a) PNPLA3 rs738409 (CC, n = 377, and GC + GG, n = 776). (b) Body mass index (continuous). (c) Body mass index (<35, n = 647, and ≥ 35, n = 506). (d) Age (continuous). (e) Age (<45, n = 315, and ≥ 45, n = 838). (f) Sex (women, n = 727, and men, n = 426). (g) Type 2 diabetes (no, n = 745, and yes, n = 408). CI, confidence interval; RR, relative risk. The risk of steatohepatitis includes borderline along with definite steatohepatitis. Analysis adjusted for nonheavy alcohol intake and those covariates not tested as moderators. The reference for all comparative analyses is the HSD17B13 rs72613567 (−/−) genotype. Logistic regression models were used to compute risks of steatohepatitis or significant fibrosis, and margins command to create a visual display of results. Circles on the lines represent average adjusted risks of steatohepatitis or significant fibrosis according to HSD17B13 rs72613567 (−/−) and (−/A) + (A/A) genotypes. Risk difference indicates a change in the risk of having steatohepatitis or significant fibrosis for HSD17B13 rs72613567 (−/A) + (A/A) as compared with (−/−) at levels of moderators.

Figure 2 depicts the association between HSD17B13 rs72613567 and risk of steatohepatitis (Panel A) or fibrosis (Panel B) at strata of moderators while controlling for relevant confounders. The HSD17B13 rs72613567 A-allele lowers the risk of steatohepatitis (adjusted OR: 0.61, P = 0.024) and fibrosis (adjusted OR: 0.48, P < 0.001) in those with PNPLA3 rs738409-(CC) genotype. However, the protective effect of HSD17B13 rs72613567 A-allele on risk of steatohepatitis (adjusted OR: 0.82, P = 0.208) and fibrosis (adjusted OR: 1.00, P = 0.979) was no longer evident among individuals with PNPLA3 rs738409-(GC) + (GG) genotypes.

Figure 2.
Figure 2.:
Effect of HSD17B13 rs72613567 (dominant genetic model) on risk of steatohepatitis and fibrosis. Adjusted risk at strata of moderators' cutoffs. Results based on ordered logistic regression. (a) Risk of steatohepatitis (0–2). (b) Risk of fibrosis (0–4). BMI, body mass index, NASH, nonalcoholic steatohepatitis. Analysis adjusted for nonheavy alcohol intake and those covariates not tested as moderators. The reference for all comparative analyses was the HSD17B13 rs72613567 (−/−) genotype.

HSD17B13 rs72613567 and risk of steatohepatitis and fibrosis at strata of traditional risk factors

Moderation effects of BMI

We first sought to illustrate graphically the influence of a 2-way interaction term of HSD17B13 rs72613567 (−/A) + (A/A) or (−/−) with BMI on the risk of steatohepatitis while adjusting for relevant confounders (Figure 1b). As shown in this figure, higher BMI was associated with a greater risk of steatohepatitis among individuals with (−/−) genotype (Adj OR: 1.03, P = 0.017); however, BMI did not seem to impact significantly on risk of steatohepatitis in carriers of HSD17B13 rs72613567 A-allele (adjusted OR: 1.01, P = 0.144). Of note, the risk of steatohepatitis between (−/−) and (-/A) + (A/A) genotypes were significantly different in those who had a BMI of ≥35 (P < 0.05 for risk differences at BMI values of 35, 40, 45, and ≥50).

Under an additive genetic model of HSD17B13 rs72613567, both (−/A) and (A/A) genotypes were significantly linked to a lower risk of steatohepatitis in patients with BMI ≥35 (see Supplemental Table 3 and Supplemental Figure 2B, Supplementary Digital Contents 1 and 2, https://links.lww.com/CTG/A669 and https://links.lww.com/CTG/A672). However, the protection of HSD17B13 rs72613567 A-allele was no longer significant in patients with BMI <35. We replicated our moderation analysis using a dominant genetic model of HSD17B13 rs72613567 (Table 3). This analysis confirmed strong protection of the A-allele on steatohepatitis and fibrosis only in those with a BMI of ≥ 35. When compared with HSD17B13 rs72613567 (−/−) genotype, the presence of the A-allele was associated with a reduced risk of steatohepatitis only in patients with BMI ≥35 (adjusted risk difference: −10, 95% CI: −18 to −3; P < 0.01) (Figure 1c). Furthermore, the protective effect of A-allele remained significantly associated with lower risk of steatohepatitis (adjusted OR: 0.55, P < 0.01) and fibrosis (adjusted OR: 0.72, P = 0.050) but only in patients with BMI ≥ 35 (Figures 2a,b).

Moderation effects of age

As we did with the analysis of BMI, the effect of a two-way interaction term of HSD17B13 rs72613567 (−/A) + (A/A) and (−/−) with age on the risk of steatohepatitis was explored through margins analysis. Figure 1d displays a positive and significant association between age and risk of steatohepatitis among those with HSD17B13 rs72613567 (−/−) genotype (adjusted OR: 1.01, P = 0.050). However, the effect of age on the risk of steatohepatitis was significantly attenuated among individuals carrying the A-allele (adjusted OR: 1.005, P = 0.460). It is important to note that the risk of steatohepatitis between (−/−) and (−/A) + (A/A) genotypes started to be significantly different at an age cutoff of 45 years or older (all P values < 0.05). Thus, further sensitivity analyses were performed to better understand the effect of HSD17B13 rs72613567 on histological outcomes between patients aged younger than 45 years and 45 years or older.

Supplemental Table 4 and Supplemental Figure 2C, Supplementary Digital Contents 1 and 2, https://links.lww.com/CTG/A669 and https://links.lww.com/CTG/A673, show that protective effects of both rs72613567 (−/A) and (A/A) genotypes on the risk of steatohepatitis were only evident among those aged 45 years or older. Further moderation analyses under a dominant genetic model of HSD17B13 rs72613567 confirmed strong and significant protective effects of the A-allele on risk of steatohepatitis and fibrosis among those aged 45 years or older (Table 3).

Among patients aged 45 years or older, the risk of steatohepatitis was significantly lower (adjusted risk difference: −8, 95% CI: −14 to −2; P < 0.01) in carriers of the HSD17B13 rs72613567 A-allele than in those with (−/−) genotype. Among people younger than 45 years, the risk of steatohepatitis did not significantly differ between (−/−) and (−/A) + (A/A) genotypes (adjusted risk difference: −1, 95% CI: −10 to 8; P = 0.693) (Figure 1e). A further sensitivity analysis looking at the association between HSD17B13 rs72613567 and histologic outcomes at strata of age (<45 or ≥45) showed strong protection of the A-allele against risk of steatohepatitis (adjusted OR: 0.64, P < 0.01) and fibrosis (adjusted OR: 0.72, P = 0.011) only in those aged 45 years or older (Figures 2a,b).

Moderation effects of sex

To confirm whether sex might moderate the effect of HSD17B13 rs72613567 on histologic outcomes, we first tested the nature of this relationship under an additive genetic model (see Supplemental Table 5 and Supplemental Figure 2D, Supplementary Digital Contents 1 and 2, https://links.lww.com/CTG/A669 and https://links.lww.com/CTG/A674). Among women, those with HSD17B13 rs72613567 (−/A) or (A/A) genotypes had a significantly lower risk of steatohepatitis and fibrosis when compared with those with (−/−) genotype. Nonetheless, this protective effect was not significant among men. The same analysis performed under a dominant genetic model of HSD17B13 rs72613567 confirmed a notable protective effect of the A-allele on risk of steatohepatitis exclusively among women (Table 3). For example, the effect of HSD17B13 rs72613567 A-allele on risk of steatohepatitis was estimated to be −0.18 (P < 0.01) and −0.03 (P = 0.751) in women and men, respectively.

Figure 1f clearly shows that women with the A-allele had a reduced risk of steatohepatitis (adjusted risk difference: −9, 95% CI: −16 to −3; P < 0.01) compared with those with (−/−) genotype. Among men, the risk of steatohepatitis was not significantly different between (−/−) and (−/A) + (A/A) genotypes (adjusted risk difference: −0.6, 95% CI: −9 to 7; P = 0.885). Covariate-adjusted logistic regression analyses validated our previous analyses and confirmed that protection conferred by the A-allele on the risk of steatohepatitis (adjusted OR: 63, P < 0.01) was particularly stronger in women (Figure 2a).

Because a decline in estrogen has been implicated with NAFLD progression among postmenopausal women and given the potential association of 17β-HSD13 with sex hormone metabolism, we sought to explore whether the effect of HSD17B13 rs72613567 on the risk of steatohepatitis and fibrosis might differ between women aged younger than 51 or 51 years or older; the median age for reaching menopause among non-Hispanic white women in the United States is 51 years (37). Our moderation analyses based on a dominant genetic model of HSD17B13 rs72613567 showed stronger protective effects of the A-allele against risk of steatohepatitis in women aged 51 years or older (β coeff: −0.31, P < 0.001) vs younger than 51 years (β coeff: −0.02, P = 0.834). The protection of the A-allele on risk of steatohepatitis and fibrosis did not significantly differ among men aged younger than 51 years or 51 years or older (Table 4). Figure 3 shows the effect of HSD17B13 rs72613567 on the risk of steatohepatitis at different strata of age and sex. Among women, the risk of steatohepatitis started to be significantly different (P < 0.05) between (−/−) and (−/A) + (A/A) genotypes at age 50 years or older (Figure 3a). However, the risk of steatohepatitis did not differ between rs72613567 (−/−) and (-/A) + (A/A) genotypes across all age cutoffs among men (Figure 3b).

Table 4. - Conditional effect of HSD17B13 rs72613567 (−/A) + (A/A) genotypes on risk of steatohepatitis and fibrosis at levels of 2 moderators (sex and age <51 or ≥ 51 yr)
Moderators Univariate analysis Multivariable analysisb
Conditional effecta Standard error 95% CI P value Conditional effecta Standard error 95% CI P value
Steatohepatitis categories (0–2)
 Men/<51, n = 238 0.05 0.11 −0.16 to 0.26 0.619 0.06 0.10 −0.15 to 0.26 0.584
 Men/≥51, n = 188 −0.15 0.12 −0.39 to 0.08 0.190 −0.11 0.11 −0.34 to 0.11 0.327
 Women/<51, n = 286 −0.03 0.09 −0.21 to 0.16 0.756 −0.02 0.09 −0.20 to 0.16 0.834
 Women/≥51, n = 441 −0.28 0.08 −0.44 to −0.12 0.001 −0.31 0.08 −0.46 to −0.15 <0.001
Fibrosis (0–4)
 Men/<51, n = 238 −0.09 0.17 −0.43 to 0.24 0.567 −0.08 0.16 −0.40 to 0.24 0.624
 Men/≥51, n = 188 −0.31 0.19 −0.68 to 0.06 0.103 −0.22 0.18 −0.57 to 0.14 0.232
 Women/<51, n = 286 −0.15 0.15 −0.45 to 0.14 0.317 −0.11 0.14 −0.39 to 0.18 0.459
 Women/≥51, n = 441 −0.17 0.13 −0.42 to 0.08 0.196 −0.23 0.12 −0.47 to 0.01 0.062
The reference genotype for all comparative analyses HSD17B13 rs72613567 (−/−). CI, confidence interval.
aRepresents the effect of the exposure (HSD17B13 rs72613567 (−/A) + (A/A) genotypes) on outcomes at values of moderators.
bAnalysis adjusted for PNPLA3 rs738409, nonheavy alcohol intake, body mass index, and type 2 diabetes.

Figure 3.
Figure 3.:
Effect of HSD17B13 rs72613567 (dominant genetic model) on the risk of steatohepatitis at different cutoffs of age (years). Sensitivity analysis based on sex: (a) women*, (b) men†. *Risk of steatohepatitis started to be significantly different (P < 0.05) between (−/−) and (−/A) + (A/A) at age ≥50 years. †Risk of steatohepatitis was not significantly different (P > 0.05) between (−/−) and (−/A) + (A/A) across all age cutoffs.

Moderation effects of type 2 diabetes mellitus

Compared with HSD17B13 rs72613567 (−/−) genotype, risks of steatohepatitis in patients with (−/A) or (A/A) genotypes were significantly lower among patients with diabetes than without diabetes (see Supplemental Table 6 and Supplemental Figure 2E, Supplementary Digital Contents 1 and 2, https://links.lww.com/CTG/A669 and https://links.lww.com/CTG/A675). Under a dominant genetic model of HSD17B13 rs72613567, our moderation analyses confirmed that only individuals with diabetes and carrying the A-allele might be protected against steatohepatitis (β coeff: −0.18; P = 0.022) and fibrosis (β coeff: −0.32; P = 0.011). However, the protection of the A-allele was no longer evident among those without diabetes (Table 3).

Two further sensitivity analyses confirmed that protective effects of HSD17B13 rs72613567 A-allele on the risk of steatohepatitis and fibrosis were seen only in participants with diabetes. Among individuals with diabetes, the carriage of rs72613567 A-allele was associated with a reduced risk of steatohepatitis when compared with those with (−/−) genotype (adjusted risk difference: −11, 95% CI: −19 to −3; P < 0.01). However, risk of steatohepatitis was not significantly different between (−/A) + (A/A) and (−/−) genotypes (adjusted risk difference: −3, 95% CI: −10 to 3; P = 0.291) in those without diabetes (Figure 1g). The presence of the A-allele was only associated with lower risk of steatohepatitis (adjusted OR: 0.61, P = 0.039) and fibrosis (adjusted OR: 0.65, P = 0.023) among individuals with diabetes (Figures 2a,b).

The association between HSD17B13 rs72613567 and histologic scores remained significant in the overall population and those selected strata of traditional risk factors after adjusting for multiple testing (see Supplemental Tables 7 and 8, Supplementary Digital Content 1, https://links.lww.com/CTG/A669).

In a separate analysis, we explored the relationship between HSD17B13 rs72613567 and liver histology severity in 156 Hispanics, 106 non-Hispanic Asians, and 52 non-Hispanic blacks with NAFLD whose liver histology was centrally reviewed by the NASH CRN pathology subcommittee. We observed that HSD17B13 rs72613567 was positively associated with steatosis (P = 0.02) and inversely linked with fibrosis stages ≥ 2 (P = 0.049) in Hispanics but not in the other 2 racial groups (data not shown).

DISCUSSION

Two common polymorphisms in HSD17B13 (rs72613567) and PNPLA3 (rs738409) have been strongly implicated in the risk of progression to steatohepatitis and fibrosis, including end-stage of liver disease in patients with NAFLD and alcoholic liver disease (2–4,6–9). Both variants, HSD17B13 rs72613567 and PNPLA3 rs738409, have opposite effects on the risk of NAFLD progression. However, whether HSD17B13 rs72613567 or PNPLA3 rs738409 may neutralize or attenuate the effect of each other on the risk of steatohepatitis and fibrosis is unclear. Beyond this gene–gene interaction, how the relationship between HSD17B13 rs72613567 and other major determinants of the natural course of NAFLD including age, sex, BMI, and diabetes might impact NAFLD progression has not been clarified either. Thus, this study aimed to elucidate how PNPLA3 rs738409 and other clinical and demographic factors might modify the relationship between HSD17B13 rs72613567 and risk of steatohepatitis and fibrosis.

As shown in previous studies, our data confirmed that the HSD17B13 rs72613567 A-insertion allele protected against steatohepatitis and fibrosis in the analysis performed under additive and dominant genetic models (2,3,7). Although individuals carrying the HSD17B13 rs72613567 A-allele had lesser risk of steatohepatitis and fibrosis, this protective effect was significantly different between individuals with PNPLA3 rs738409 genotypes (GC + GG) and (CC). Of note, the beneficial effects of HSD17B13 rs72613567 A-allele on steatohepatitis and fibrosis were fully mitigated in patients carrying the PNPLA3 rs738409 G-allele. Further analyses showed that associations between the HSD17B13 rs72613567 A-allele and histology scores were also significantly different depending on age, BMI, sex, and diabetes status. The protective effects of HSD17B13 rs72613567 A-allele on risk of inflammation and fibrosis were particularly stronger in women, persons aged 45 years or older, individuals with diabetes, or those with BMI ≥ 35, even after adjusting for the other relevant confounders.

Noteworthy, the severity of steatosis was significantly higher in carriers of the HSD17B13 rs72613567 A-allele compared with noncarriers, a finding previously described (3,7). Because severity in fibrosis and steatosis were inversely correlated in our cohort and one can expect that advanced liver fibrosis is often accompanied by a reduction in hepatic fat to the point of complete fat loss (38), we sought to investigate whether the effect of HSD17B13 rs72613567 on fibrosis severity was mediated by a change in steatosis severity. Our causal mediation analysis showed a direct effect of the HSD17B13 rs72613567 A-allele on fibrosis severity. In other words, the protective effect of this HSD17B13 variation on fibrosis was not significantly mediated by a change in steatosis. In support of our findings, Ma et al. have suggested that 17β-HSD13 does not directly regulate hepatocyte lipid content (7).

The effect of PNPLA3 rs738409 and HSD17B13 rs72613567 interactions on ALT and AST levels has been previously explored in 2 large studies including patients with biopsy-proven NAFLD along with other etiologies of chronic liver disease. Abul-Husn et al., using a large cohort of European descent, identified strong interactions between HSD17B13 rs72613567 and PNPLA3 rs738409 in association with ALT and AST levels (3). They reported that the presence of HSD17B13 rs72613567 A-allele mitigated increases in ALT and AST levels among carriers of PNPLA3 rs738409 G-allele. RNA sequencing–based expression analysis revealed that HSD17B13 rs72613567 was associated with decreased PNPLA3 messenger RNA expression in an allele dose-dependent manner. However, this study did not report interaction effects between these 2 variants and their association with liver histology severity. Another study recently published, which included white patients with biopsy-confirmed NAFLD from the NASH-CRN observational database and PIVENS study and those seen at the NIH Clinical Center, did not find a significant interaction between PNPLA3 rs738409 and HSD17B13 variants, although authors stated that the study was not powered to detect such an interaction effect (7). Interestingly, in this study, as illustrated through their supplemental figure 6, they noticed that the effect of HSD17B13 rs72613567 A-allele on ballooning, lobular inflammation, and fibrosis was attenuated in patients with the PNPLA3 rs738409 G-allele (Figure 6 of Supplementary Material, Supplementary Digital Content 1, https://links.lww.com/CTG/A669 in the study by Ma et al. manuscript) (7).

Although this analysis included the same NAFLD population (NASH CRN) as in the study by Ma et al. (7), we included a larger proportion of non-Hispanic whites (n = 1,153), which may significantly increase the statistical power to detect statistically significant interactions or moderation effects between HSD17B13 rs72613567, PNPLA3 rs738409, and other phenotypic risk factors.

17β-HSD13 has been found to associate with LDs and is significantly upregulated in the liver of patients with NAFLD (2,3,7,12). Most recently, 17β-HSD13 has been reported to regulate retinoid metabolism by acting as a retinol dehydrogenase (7,39). Most of the retinol present in the body is stored as retinyl esters within LD in HSCs, and it may play a critical role in the activation of HSCs, which may lead to fibrosis (40). HSD17B13 rs72613567 is supposed to alter mRNA splicing that results in a truncated protein with absent or reduced enzymatic activity. This loss-of-function mutation in the HSD17B13 gene confers a strong protective effect on inflammation, fibrosis, cirrhosis, and even hepatocellular carcinoma (2,3,5–9,41).

On the other hand, PNPLA3 rs738409 has been associated with an increased risk of NAFLD and its severity (1,15). The PNPLA3 rs738409 has been linked to impaired hydrolysis of triglycerides and LD remodeling in hepatocytes, which may lead to hepatic steatosis (1). Besides, PNPLA3 might also participate in the metabolism of retinoids. It seems to act as a lipase responsible for retinyl-palmitate hydrolysis in HSCs in humans. In the presence of PNPLA3 rs738409 variation, there is reduced activity of retinyl palmitate lipase enzyme, resulting in liver fibrosis (17).

Interestingly, both PNPLA3 and 17β-HSDB13 are involved in LD remodeling and retinoid acid (RA) metabolism pathways in the liver, which might suggest biologically plausible ways for this gene–gene interaction, although either genetic variant may impact the functionality of liver-specific LD-associated proteins or RA metabolism in opposite directions. A better understanding of the pathogenic role of this gene–gene interaction needs to be addressed in further studies.

The role of 17β-HSD13 in the regulation of sex steroid metabolism is not fully understood in humans. Adam et al. (14) recently reported that knockout of HSD17B13 in mice does not affect reproductive performance and sex hormone concentrations. Unlike human studies, they noted that hepatic steatosis and inflammation were significantly increased in mice with reduced HSD17B13 expression (14). It is noteworthy that the nonprotective effect of HSD17B13 knockout was only studied in male mice. Thus, whether the knockout of HSD17B13 may induce similar or different histologic phenotypes in female and male mice remains to be elucidated.

In this present study, we observed that the protection conferred by the HSD17B13 rs72613567 A-allele on risk of steatohepatitis was only evident in women aged 51 years or older compared with those aged younger than 51 years, even after adjusting for relevant clinical confounders such as BMI and T2D. The exact mechanism by which HSD17B13 rs72613567 might protect against NAFLD severity women older than 51 years remains to be elucidated. Low serum concentrations of RA and/or carotenoids have been documented in obesity (42), metabolic syndrome (43), and T2D (44). Low levels of RA and carotenoids have also been implicated in the pathogenesis of NAFLD (45–47). RA seems to be an important regulator of HSCs activation, which is a major contributor to liver fibrosis (40). Recently, 17β-HSD13 has been reported to regulate retinoid metabolism by acting as a retinol dehydrogenase (7). Thus, there is a potential biologic link to explain partly the protective effect of HSD17B13 rs72613567 on inflammation and fibrosis in NAFLD. Patients with obesity and those with metabolic syndrome tend to have a higher risk of NAFLD severity (48). In women, menopause is associated with estrogen depletion, which has been linked to stellate cell activation and fibrogenesis (49). In addition, postmenopausal women are more likely to develop visceral obesity, insulin resistance, and other features of metabolic syndrome, including dyslipidemia, high blood pressure, and T2D; all of them are strongly associated with increased risk of NAFLD and its progression (50). The abovementioned associations between obesity, sex (particularly postmenopausal women), metabolic syndrome, RA and/or carotenoids levels, stellate cell function, and NAFLD severity may mechanistically support the hypothesis that HSD17B13 72613567 may confer its higher protection to patients at increased risk of NAFLD progression by likely impacting the retinol metabolism and LD remodeling, although further studies are necessary to establish whether HSD17B13 rs72613567 might also exert its protection by affecting sex steroid metabolism.

To our knowledge, this is the first attempt to provide novel insight into how the protective effects of HSD17B13 rs72613567 against steatohepatitis and fibrosis severity may vary across a wide subgroup of patients at higher risk of NAFLD progression. These findings are particularly important regarding risk predictions because patients aged 45 years and older, those with obesity or T2D and postmenopausal women have been consistently associated with an accelerated clinical course of the disease with worse clinical outcomes (48,50–52). This also supports the idea that the explanatory power of genetic variants could be enhanced by combining them with other sources of individual variation, such as demographic risk factors or comorbidities, as shown in this study. Thus, our findings may have notable implications: from a public health perspective, identifying the subset of individuals in whom a preventive or therapeutic intervention could have the most beneficial effect; and regarding risk estimates, improving the fit of the model or its predictive capability by the inclusion of interaction terms in multivariate models.

Gene–gene or gene–phenotype interactions add complexity to the already complex individual roles of gene and environmental factors in the pathogenesis of NAFLD. Unfortunately, our study was not initially designed to explore the molecular mechanisms to explain these complex interactions. Thus, further experimental studies may help to unravel the mechanisms whereby these complex interactions may occur. Our results are based on non-Hispanic whites; hence, replication of our findings in independent multiethnic cohorts would be required. Although the effect sizes of our associations seemed to be relatively large, stratified analysis performed in our analyses might reduce the power of the study, resulting in a nondifferential bias underestimating the real size effect. To minimize this limitation, all moderation analyses used the methodology recommended by Preacher and Hayes (32), which uses bootstrapping, a nonparametric resampling procedure, because it is considered the most powerful, most effective method to use with small samples, and the least vulnerable to type I error. Furthermore, associations between HSD17B13 rs72613567 and histologic scores remained significant in the overall population and those selected strata of traditional risk factors after performing multiple comparisons adjustments.

In conclusion, our study examined how the relationship between HSD17B13 rs72613567 and NAFLD severity is modified by genetic, demographic, and clinical factors. Although HSD17B13 rs72613567 remained significantly associated with lower risk of steatohepatitis and fibrosis in the whole cohort, this protective effect seemed to be significantly modified at different levels of other traditional risk factors with recognized association with NAFLD progression. The protection of the HSD17B13 rs72613567 A-allele was significantly mitigated in carriers of the PNPLA3 rs738409 G-allele. The major benefits of the HSD17B13 rs72613567 A-allele were seen in patients at increased risk of NAFLD progression, including women (particularly those aged 51 years or older), individuals with T2D and those with BMI ≥35 or aged 45 years or older. Our data support the idea that identification of gene–gene or gene–phenotype moderation effects should be at the forefront in attempts to understand the pathophysiology of NAFLD, construct more accurate risk prediction models, and identify subsets of the population that would obtain the greatest benefit from therapeutic interventions.

CONFLICTS OF INTEREST

Guarantor of the article: Naga Chalasani, MD, had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Specific author contributions: All authors made substantial contributions to the intellectual content of the paper and approved the final version of the manuscript. Conception and design: N.C. and E.V.-G. Acquisition of data: L.A.W., N.C., E.V.-G., and T.L. Analysis and interpretation of data: N.C., E.V.-G., S.S., C.J.P., and L.A.W. Drafting of the manuscript: N.C., E.V.-G., S.S., and C.J.P. Critical revision of the manuscript for intellectual content: N.C, E.V.-G., S.S., C.J.P., T.L., and L.A.W. Statistical analysis: E.V.-G., C.J.P. and S.S. Obtaining funding and supervision: N.C.

Financial support: This study was approved and funded by the Nonalcoholic Steatohepatitis Clinical Research Network as an Ancillary Study (NASH CRN AS # 93). The NASH CRN is supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (U01DK061718, U01DK061728, U01DK061731, U01DK061732, U01DK061734, U01DK061737, U01DK061738, U01DK061730, and U01DK061713). This study was also supported partly by David W Crabb Endowed Professorship funds to N.C.

Potential competing interests: There are none for this paper. For full disclosure, N.C. has ongoing paid consulting activities (or had in the preceding 12 months) with NuSirt, Abbvie, Afimmune (DS Biopharma), Allergan (Tobira), Madrigal, Coherus, Siemens, La Jolla, Foresite labs, and Genentech. These consulting activities are generally in the areas of nonalcoholic fatty liver disease and drug hepatotoxicity. N.C. receives research grant support from Exact Sciences, Intercept, and Galectin Therapeutics, where his institution receives the funding. Over the past decade, N.C. has served as a paid consultant to more than 35 pharmaceutical companies, and these outside activities have regularly been disclosed to his institutional authorities. All other authors have no reported conflicts.

Study Highlights

WHAT IS KNOWN

  • HSD17B13 rs72613567 A-insertion allele may confer protection against nonalcoholic steatohepatitis and its progression.
  • PNPLA3 rs738409, age, body mass index, sex, and type 2 diabetes mellitus are strongly associated with the risk of nonalcoholic fatty liver disease progression.
  • ✓ The effect of interactions between HSD17B13 rs72613567, PNPLA3 rs738409, and other phenotypic risk factors on risk of steatohepatitis and fibrosis is not well understood.

WHAT IS NEW HERE

  • ✓ The presence of PNPLA3 rs738409 G-allele mitigates the protective effect of HSD17B13 rs7261537 A-allele on risk of steatohepatitis and fibrosis.
  • ✓ The HSD17B13 rs72613567 A-allele might confer stronger protection against steatohepatitis and fibrosis among individuals who are aged ≥ 45 years, women, and have class ≥ 2 obesity or diabetes.
  • ✓ Among women, the protection of the HSD17B13 rs72613567 A-allele is stronger in those aged 51 years and older.

ACKNOWLEDGMENTS

The authors thank the Nonalcoholic Steatohepatitis Clinical Research Network (NASH CRN) investigators and the Ancillary Studies Committee for providing clinical samples and relevant data from the Nonalcoholic Fatty Liver Disease (NAFLD) Database, Nonalcoholic Fatty Liver Disease (NAFLD) Adult Database 2, PIVENS trial, and FLINT trial.

REFERENCES

1. Romeo S, Kozlitina J, Xing C, et al. Genetic variation in PNPLA3 confers susceptibility to nonalcoholic fatty liver disease. Nat Genet 2008;40:1461–5.
2. Pirola CJ, Garaycoechea M, Flichman D, et al. Splice variant rs72613567 prevents worst histologic outcomes in patients with nonalcoholic fatty liver disease. J Lipid Res 2019;60:176–85.
3. Abul-Husn NS, Cheng X, Li AH, et al. A protein-truncating HSD17B13 variant and protection from chronic liver disease. N Engl J Med 2018;378:1096–106.
4. Chen H, Zhang Y, Guo T, et al. Genetic variant rs72613567 of HSD17B13 gene reduces alcohol-related liver disease risk in Chinese Han population. Liver Int 2020;40:2194–202.
5. Chen J, Zhuo JY, Yang F, et al. 17-beta-hydroxysteroid dehydrogenase 13 inhibits the progression and recurrence of hepatocellular carcinoma. Hepatobiliary Pancreat Dis Int 2018;17:220–6.
6. Luukkonen PK, Tukiainen T, Juuti A, et al. Hydroxysteroid 17-β dehydrogenase 13 variant increases phospholipids and protects against fibrosis in nonalcoholic fatty liver disease. JCI Insight 2020;5:e132158.
7. Ma Y, Belyaeva OV, Brown PM, et al. 17-Beta hydroxysteroid dehydrogenase 13 is a hepatic retinol dehydrogenase associated with histological features of nonalcoholic fatty liver disease. Hepatology 2019;69:1504–19.
8. Stickel F, Lutz P, Buch S, et al. Genetic variation in HSD17B13 reduces the risk of developing cirrhosis and hepatocellular carcinoma in alcohol misusers. Hepatology 2020;72:88–102.
9. Yang J, Trépo E, Nahon P, et al. A 17-beta-hydroxysteroid dehydrogenase 13 variant protects from hepatocellular carcinoma development in alcoholic liver disease. Hepatology 2019;70:231–40.
10. Di Sessa A, Umano GR, Cirillo G, et al. The rs72613567: TA variant in the hydroxysteroid 17-beta dehydrogenase 13 gene reduces liver damage in obese children. J Pediatr Gastroenterol Nutr 2020;70:371–4.
11. Horiguchi Y, Araki M, Motojima K. 17beta-Hydroxysteroid dehydrogenase type 13 is a liver-specific lipid droplet-associated protein. Biochem Biophys Res Commun 2008;370:235–8.
12. Su W, Wang Y, Jia X, et al. Comparative proteomic study reveals 17β-HSD13 as a pathogenic protein in nonalcoholic fatty liver disease. Proc Natl Acad Sci U S A 2014;111:11437–42.
13. Moeller G, Adamski J. Multifunctionality of human 17beta-hydroxysteroid dehydrogenases. Mol Cell Endocrinol 2006;248:47–55.
14. Adam M, Heikelä H, Sobolewski C, et al. Hydroxysteroid (17β) dehydrogenase 13 deficiency triggers hepatic steatosis and inflammation in mice. Faseb j 2018;32:3434–47.
15. Sookoian S, Pirola CJ. Meta-analysis of the influence of I148M variant of patatin-like phospholipase domain containing 3 gene (PNPLA3) on the susceptibility and histological severity of nonalcoholic fatty liver disease. Hepatology 2011;53:1883–94.
16. Burza MA, Pirazzi C, Maglio C, et al. PNPLA3 I148M (rs738409) genetic variant is associated with hepatocellular carcinoma in obese individuals. Dig Liver Dis 2012;44:1037–41.
17. Pirazzi C, Valenti L, Motta BM, et al. PNPLA3 has retinyl-palmitate lipase activity in human hepatic stellate cells. Hum Mol Genet 2014;23:4077–85.
18. Sookoian S, Pirola CJ, Valenti L, et al. Genetic pathways in nonalcoholic fatty liver disease: Insights from systems biology. Hepatology 2020;72:330–46.
19. Neuschwander-Tetri BA, Clark JM, Bass NM, et al. Clinical, laboratory and histological associations in adults with nonalcoholic fatty liver disease. Hepatology 2010;52:913–24.
20. Neuschwander-Tetri BA, Loomba R, Sanyal AJ, et al. Farnesoid X nuclear receptor ligand obeticholic acid for non-cirrhotic, non-alcoholic steatohepatitis (FLINT): A multicentre, randomised, placebo-controlled trial. Lancet 2015;385:956–65.
21. Sanyal AJ, Chalasani N, Kowdley KV, et al. Pioglitazone, vitamin E, or placebo for nonalcoholic steatohepatitis. N Engl J Med 2010;362:1675–85.
22. Vilar-Gomez E, Sookoian S, Pirola CJ, et al. ADH1B *2 is associated with reduced severity of nonalcoholic fatty liver disease in adults, independent of alcohol consumption. Gastroenterology 2020;159:929–43.
23. Berner MM, Kriston L, Bentele M, et al. The alcohol use disorders identification test for detecting at-risk drinking: A systematic review and meta-analysis. J Stud Alcohol Drugs 2007;68:461–73.
24. Skinner HA, Allen BA. Alcohol dependence syndrome: Measurement and validation. J Abnorm Psychol 1982;91:199–209.
25. Kleiner DE, Brunt EM, Van Natta M, et al. Design and validation of a histological scoring system for nonalcoholic fatty liver disease. Hepatology 2005;41:1313–21.
26. Kleiner DE, Brunt EM. Nonalcoholic fatty liver disease: Pathologic patterns and biopsy evaluation in clinical research. Semin Liver Dis 2012;32:3–13.
27. Rotman Y, Koh C, Zmuda JM, et al. The association of genetic variability in patatin-like phospholipase domain-containing protein 3 (PNPLA3) with histological severity of nonalcoholic fatty liver disease. Hepatology 2010;52:894–903.
28. Vilar-Gomez E, Jose Pirola C, Sookoian S, et al. Impact of the association between PNPLA3 genetic variation and dietary intake on the risk of significant fibrosis in patients with NAFLD. Am J Gastroenterol 2021;116:994–1006.
29. Williams R. Using the margins command to estimate and interpret adjusted predictions and marginal effects. Stata J 2012;12:308–31.
30. Wu AD, Zumbo BD. Understanding and using mediators and moderators. Soc Indicators Res 2007;87:367.
31. Hayes AF. An index and test of linear moderated mediation. Multivariate Behav Res 2015;50:1–22.
32. Preacher KJ, Hayes AF. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behav Res Methods 2008;40:879–91.
33. Hayes AF. Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. Guilford Press: New York, NY, 2013.
34. Valeri L, Vanderweele TJ. Mediation analysis allowing for exposure-mediator interactions and causal interpretation: Theoretical assumptions and implementation with SAS and SPSS macros. Psychol Methods 2013;18:137–50.
35. Benjamini Y, Drai D, Elmer G, et al. Controlling the false discovery rate in behavior genetics research. Behav Brain Res 2001;125:279–84.
36. Newson RB. Frequentist Q-values for multiple-test procedures. Stata J 2010;10:568–84.
37. Gold EB, Crawford SL, Avis NE, et al. Factors related to age at natural menopause: Longitudinal analyses from SWAN. Am J Epidemiol 2013;178:70–83.
38. van der Poorten D, Samer CF, Ramezani-Moghadam M, et al. Hepatic fat loss in advanced nonalcoholic steatohepatitis: Are alterations in serum adiponectin the cause? Hepatology 2013;57:2180–8.
39. Belyaeva OV, Chang C, Berlett MC, et al. Evolutionary origins of retinoid active short-chain dehydrogenases/reductases of SDR16C family. Chem Biol Interact 2015;234:135–43.
40. Haaker MW, Vaandrager AB, Helms JB. Retinoids in health and disease: A role for hepatic stellate cells in affecting retinoid levels. Biochim Biophys Acta Mol Cell Biol Lipids 2020;1865:158674.
41. Anstee QM, Darlay R, Cockell S, et al. Genome-wide association study of non-alcoholic fatty liver and steatohepatitis in a histologically characterised cohort(☆). J Hepatol 2020;73:505–15.
42. Roust LR, DiBaise JK. Nutrient deficiencies prior to bariatric surgery. Curr Opin Clin Nutr Metab Care 2017;20:138–44.
43. Liu Y, Chen H, Mu D, et al. Circulating retinoic acid levels and the development of metabolic syndrome. J Clin Endocrinol Metab 2016;101:1686–92.
44. Hozawa A, Jacobs DR Jr, Steffes MW, et al. Associations of serum carotenoid concentrations with the development of diabetes and with insulin concentration: Interaction with smoking: The coronary artery risk development in young adults (CARDIA) study. Am J Epidemiol 2006;163:929–37.
45. Park SK, Lee HJ, Lee DH, et al. [Associations of non alcoholic fatty liver with the metabolic syndrome and serum carotenoids]. J Prev Med Public Health 2008;41:39–44. Ko.
46. Erhardt A, Stahl W, Sies H, et al. Plasma levels of vitamin E and carotenoids are decreased in patients with Nonalcoholic Steatohepatitis (NASH). Eur J Med Res 2011;16:76–8.
47. Liu Y, Chen H, Wang J, et al. Association of serum retinoic acid with hepatic steatosis and liver injury in nonalcoholic fatty liver disease. Am J Clin Nutr 2015;102:130–7.
48. Chalasani N, Younossi Z, Lavine JE, et al. The diagnosis and management of nonalcoholic fatty liver disease: Practice guidance from the American Association for the Study of Liver Diseases. Hepatology 2018;67:328–57.
49. Yasuda M, Shimizu I, Shiba M, et al. Suppressive effects of estradiol on dimethylnitrosamine-induced fibrosis of the liver in rats. Hepatology 1999;29:719–27.
50. Klair JS, Yang JD, Abdelmalek MF, et al. A longer duration of estrogen deficiency increases fibrosis risk among postmenopausal women with nonalcoholic fatty liver disease. Hepatology 2016;64:85–91.
51. Vilar-Gomez E, Calzadilla-Bertot L, Wong VW, et al. Type 2 diabetes and metformin use associate with outcomes of patients with nonalcoholic steatohepatitis-related, child-Pugh A cirrhosis. Clin Gastroenterol Hepatol 2021;19:136–45.e6.
52. Balakrishnan M, Patel P, Dunn-Valadez S, et al. Women have a lower risk of nonalcoholic fatty liver disease but a higher risk of progression vs men: A systematic review and meta-analysis. Clin Gastroenterol Hepatol 2021;19:61–71. e15.

Supplemental Digital Content

© 2021 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of The American College of Gastroenterology