Non-alcoholic fatty liver disease (NAFLD) is the most prevalent liver disease globally, affecting approximately 25% of the adult population as of 2016 (1), and its incidence continues to increase. NAFLD encompasses simple steatosis (fatty liver; NAFL) and the more severe nonalcoholic steatohepatitis (NASH), which is characterized by fat accumulation, inflammation, and hepatocellular injury. Hepatic fibrosis can develop in NAFLD, which can progress into cirrhosis and hepatocellular carcinoma (HCC) (2, 3). As of 2019, NASH was the underlying cause of liver failure in over a third of individuals awaiting liver transplant (4). There are currently no FDA approved treatments for any stage of NAFLD, including NASH (5), highlighting the critical need to identify therapeutic targets.

Given that NAFLD is heritable, with heritability estimates ranging 20%-70% (6), genetic mapping has been undertaken to identify causal genes with potential therapeutic implications. Initial NAFLD studies focused on selected candidate genes, but were limited by small sample size, a high rate of false positive associations due to cryptic population stratification, and reliance on prior knowledge for gene selection (7–9). With the advent of genome-wide association methods that could be applied at population scale, some of these limitations have been overcome, resulting in the unbiased, reproducible genetic discoveries that are detailed below.

In this mini-review, we focus on the phenotypic complexity of NAFLD, the challenges this poses to executing genetic association studies, and the progress made in identifying new putative targets over the past four years.

NAFLD is a continuum of disease with multiple pathophysiologies and is defined and diagnosed by variable, often incompatible, approaches. In this section, we provide an overview of this heterogeneity in pathogenesis and detection, focusing on how this impacts the interpretation of the genetic associations studies described below. This brief summary does not cover the full breadth of this field, so we refer the reader to other recent reviews for a comprehensive treatment of all diagnostic methods (10), noninvasive diagnostic modalities (11–13), biomarkers (14, 15), and elastography techniques (16).

Clinically, the spectrum of fatty liver disease encapsulated in NAFLD is defined in the absence of excess alcohol intake (5). The distinction between NAFL and NASH is most commonly differentiated by the absence (NAFL) or presence (NASH) of hepatocyte ballooning (17). Some studies further delineate phases between NAFL and NASH (18, 19), and between NASH and cirrhosis (20), highlighting the continuum of pathophysiology. For simplicity, this mini-review will anchor on three stages of NAFLD – NAFL, NASH, and cirrhosis as depicted in Figure 1 .

Epidemiological, familial aggregation, and twin studies over the past two decades have demonstrated a heritable component to NAFLD (72), strongly suggesting that genetic mapping approaches could be productively deployed to identify genes with therapeutic potential. As mentioned earlier, initial genetic investigations into NAFLD utilized candidate gene approaches, but the development of next-generation sequencing (NGS) and high-throughput genotyping arrays enabled more robust, unbiased methods of genetic mapping studies including genome-wide association studies (GWAS) and exome-wide association studies (EWAS) (73, 74). GWAS has successfully identified loci that are associated with risk for many complex diseases and traits using common variants ascertained from genotyping (75), whereas EWAS examines variants predominantly in the exonic (i.e. protein-coding) regions of the genome (76). With the decreasing cost of NGS, current studies can detect exonic variants through whole-exome sequencing (WES) (77). Recent expert reviews have summarized variants identified from NAFLD genetic association studies (6, 78, 79). Here, we build upon these publications by reviewing the literature from the past four years, highlighting the consequence of NAFLD phenotypic heterogeneity on genetic discovery, and quantifying the limits of current association studies to identify new genetic signals.

The variants identified through GWAS and EWAS are susceptible to the study design choices. The sample size, diversity within the cohort, and specificity of the associated trait, along with many other confounders, can all affect the results (99). A major cofounder is that sample size in current NAFLD studies is highly correlated to the measurement modality. On the spectrum of sample numbers, liver enzyme levels, which are commonly available as part of routine blood testing, are on the high end, and liver biopsies, which require a clinical indication and are difficult to perform in large numbers, are on the low end (63). Because statistical power to detect significant associations is directly dependent on sample size (100), studies using biopsies are often underpowered, while studies using serum concentrations are better powered but less informative for NAFLD stages.

In Figure 3 (left panel), we illustrate the statistical power of the largest liver biopsy GWAS to date (82), which included 1,483 NAFLD biopsied cases and 17,781 controls. Given its size, this study would be predicted to successfully replicate previously characterized loci, including PNPLA3, TM6SF2, HSD17B13, and GCKR, based on their respective frequencies in the population and effect sizes on NAFLD risk. These associations are indeed found with genome-wide significance (82). Furthermore, a GWAS of this size would be predicted to not detect MARC1 and LYPLAL1 with genome-wide significance, as those variants have smaller effect sizes. Again, this is reflected in the results: although the variant in MARC1 was associated with NAFLD with p < 6 x 10^-6, the association did not meet the genome-wide significance threshold (82).

Future studies with increased sample sizes of NAFLD individuals and balanced case/control designed studies may reveal novel genetic associations which studies are currently underpowered to detect and provide additional support for existing associations ( Figure 3 , right panel). With the generation of larger NAFLD case/control cohorts and increased application of WES, more rare variants with large biological effects can be identified, which would facilitate therapeutic targeting. Rare, loss-of-function variants that confer protection from disease in particular have shown promise as therapeutic targets, as exemplified by the successful development of PCSK9 inhibitors to treat atherogenic cardiovascular disease (101, 102).

Identifying causal genes is a major challenge to translating genetic association signals into biological and potentially therapeutic knowledge. The majority of variants identified from GWAS are located in non-coding genomic sequences distant from protein-coding genes (99). For example, a variant on chromosome 8 that lies in the intergenic region between IDO2 and TC1 was associated with NAFLD, but it is unclear which gene is driving the phenotype (82). Additionally, while it is standard practice to designate the nearest gene to a variant as the causal gene, this may not always be true. This caveat was showcased by the variant rs2075650 residing in an intron of TOMM40, which was found to be associated with steatosis. While most proximal to the TOMM40 coding sequences, conditional analysis showed that this variant association was driven by linkage disequilibrium with the previously identified APOE rs429358 (86). In contrast to GWAS, EWAS analyses almost entirely use variants in the exonic regions of the genome, which can make causal gene identification more straightforward as the variants likely alter the sequence of the encoded protein. However, to date, the findings of NAFLD EWAS are still limited. So far, only one study specific to NAFLD has harnessed rare protein-coding genetic variants from WES rather than genome- or exome- arrays (103), but additional large-scale studies of rare variants and their effects on NAFLD are beginning to emerge and find new signals such as MAST3 and IFI30 (104, 105).

As mentioned earlier, increasing sample size to power robust discovery is a current challenge in NAFLD gene discovery due in large part to inherent limitations in the scalability of liver biopsies. Some studies have employed creative methods to increase sample sizes for their NAFLD genetic studies with some indications of success. These techniques include NLP algorithms (81), machine learning applied to liver imaging (88, 104), and a multi-step approach of first identifying genetic signals with a widely available biomarker, such as liver enzyme levels, in population cohorts, and then examining only these identified signals with independent histological cohorts (80, 94) to decrease the multiple hypothesis correction burden. For example, Abul-Husn et al. first conducted a GWAS in almost 47,000 individuals to identify variants significantly associated with either ALT or AST levels (80). 13 of these variants were next replicated in an additional cohort (n=12,527), and then these top variants were identified within exome sequences (n=1,857 NAFLD cases and 29,928 controls) and tested for association with chronic liver disease. This reduced the statistical threshold for significance without increasing the false positive rate. From this targeted exome association analysis, the protein-truncating variant in HSD17B13 (rs72613567) was found to confer lower odds across all categories of liver disease and provide protection against liver fibrosis in an allele dose-dependent manner. This discovery then led to the development of ARO-HSD, a RNAi therapeutic that selectively targets HSD17B13 mRNA in hepatocytes, which has demonstrated improvements in NASH outcomes, as assessed by ALT, AST, and MRI-PDFF, in a Phase 1/2 clinical trial (106, 107).

In the serendipitous case of HSD17B13, the consequence of the top identified variant was protein-truncating and thus could be predicted with high confidence to confer loss of function in HSD17B13 without additional functional characterization. In order to provide analogous interpretations to genetic variants that do not have such clear cut functional effects without performing validation experiments, NAFLD genetic studies have utilized computational prediction tools (108), ClinVar reported pathogenicity predictions, and allele frequency cut-offs (109), to narrow the search space to actionable variants, but with limited success in the absence of mechanistic investigation. In summary, despite current limitations, genetic discoveries for NAFLD have demonstrated promise in therapeutic target identification. Future genetic investigations with increased sample size and focusing on different stages of NAFLD are likely to reveal new genes with therapeutic potential.

Efforts are underway to improve standardization in classification and diagnosis of NAFLD to enable translational research that can identify putative drug targets. In 2020, an expert consensus panel proposed a new set of diagnostic criteria for NAFLD (110) and renamed it metabolic associated fatty liver disease (MAFLD). MAFLD is diagnosed by the presence of hepatic steatosis (ascertained by imaging, biomarker panel or histology) and either type 2 diabetes (T2D) or overweight/obesity, or two of the following metabolic risk factors: waist circumference, blood pressure, serum triglycerides, low serum HDL, prediabetes, insulin resistance, and plasma high-sensitivity C-reactive protein level (111). Recent publications indicate that the MAFLD criteria performs better than the NAFLD definition at identifying patients with more severe presentations of disease (112–114), but the new terminology is still heavily debated (115). To date, one genetic association study has been performed using the MAFLD definition and recapitulated the known genetic associations with PNPLA3 rs738409 and TM6SF2 rs8542926 (116). This is promising as MAFLD diagnosis does not require biopsies and can be diagnosed from readily available clinical measurements. Nevertheless, further validation is required, especially for the SNPs associated with hepatic injury and fibrosis, which are not explicitly included in the MAFLD definition.

Other than the RNAi targeting HSD17B13 mRNA mentioned above, most therapeutics currently in clinical trials do not target genes identified from GWAS (63). One possible direction to identify novel, actionable targets for NAFLD from gene or exome wide associations would entail a combination of imaging and biomarkers for NAFLD diagnostic staging that could be broadly applied to hundreds of thousands of individuals in biobanks, as demonstrated by recent publications (84, 85). A specific pathophysiology of NAFLD, such as NASH defined by MRI-PDFF and cT1, should be selected to ensure that there is a sufficiently large cohort of cases for a well-powered study. Association analysis could then be performed to identify rare variants with large effect sizes associated with this classification of NASH. The variants could be further investigated by functional validation in molecular assays to find the causal genes, which would then be the targets of drug development.

XD and AM conceived the manuscript outline and figure concepts. XD and ND authored sections of the manuscript and created figures. XD, ND, and AM were involved in critical manuscript revision. All authors contributed to the article and approved the submitted version.

This work was supported by grants from the National Institute of Diabetes and Digestive and Kidney Diseases (1R03DK113328-01 and 1R01DK123422-01 to ARM), a UCSD/UCLA Pilot and Feasibility grant (P30 DK063491 to ARM), and a Ruth L. Kirschstein Institutional National Research Service Award T32 GM008666 from the National Institute of General Medical Sciences (to ND).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.