Genetic Risk for Hepatic Fat among an Ethnically Diverse Cohort of Youth: The Epoch Study

Abstract

Objectives:

To assess the importance of genetic and non-genetic risk factors contributing to hepatic fat content in a multiethnic population of youth.

Study design:

We investigated the relationship between genetic factors and hepatic fat fraction (HFF) in 347 children aged 12.5–19.5 years. We examined 5 SNPs previously associated with HFF and a weighted genetic risk score (GRS) and examined how these associations varied with ethnicity (Hispanic versus non-Hispanic white) and BMI category. We also compared how much variation in HFF was explained by genetic factors versus cardiometabolic factors (BMI z-score and HOMA-IR) or diet.

Results:

PNPLA3 rs738409 and the GRS were each associated with HFF among Hispanic (β=0.39; 95% confidence interval (CI): 0.16, 0.62; p-value=0.001; and β=0.20; 95% CI: 0.05, 0.34; P value = .007, respectively) but not NHW (β=0.04; 95% CI: −0.18, 0.26; p-value=0.696; and β=0.03; 95% CI: −0.09, 0.14; p-value=0.651, respectively) youth. Cardiometabolic risk factors explained more of the variation in HFF than genetic risk factors among non-lean Hispanic individuals (27.2% for cardiometabolic markers versus 6.4% for rs738409 and 4.3% for the GRS), and genetic risk factors were more important among lean individuals (2.7% for cardiometabolic markers versus 12.6% for rs738409 and 4.4% for GRS).

Conclusions:

Poor cardiometabolic health may be more important than genetic factors when predicting HFF in overweight and obese young populations. Genetic risk is an important contributor to pediatric HFF among lean Hispanics, but further studies are necessary to elucidate the strength of the association between genetic risk and HFF in NHW youth.

The prevalence of non-alcoholic fatty liver disease (NAFLD) has been increasing rapidly, both in adults and children, in conjunction with increases in obesity, a major risk factor for fatty liver.[1] Genetics play a role in NAFLD, and heritability estimates range from 20–70% depending on study design and ethnicity.[2] Previous genome-wide association studies (GWAS) in adults have identified numerous single nucleotide polymorphisms (SNPs) associated with NAFLD severity, liver fibrosis, and hepatic fat fraction (HFF). [3–9] Studies of pediatric populations have confirmed that many of these SNPs are associated with measures of fatty liver in children.[10]

There are racial and ethnic differences in NAFLD prevalence, with higher prevalence in Hispanics than in non-Hispanic whites or African Americans both in adults (45%, 33% and 24%, respectively) [11] and in children (11.8%, 8.6% and 1.5% respectively).[12] The primary genetic risk factor for NAFLD is the PNPLA3 rs738409 G allele, and its frequency is likewise highest among Hispanics, then NHWs, then AAs.[13] Pediatric obesity prevalence is also highest among Hispanic children, followed by AAs and then NHWs.[14] NAFLD is less common among normal weight individuals, but it does occur and genetics likely contribute towards its development.[15] Furthermore, epidemiological evidence suggests that nonalcoholic steatohepatitis (NASH), fibrosis and mortality do not differ significantly between lean and obese individuals with NAFLD.[15,16] A better understanding of similarities and differences in the genetic associations with HFF between adult and younger populations, as well as across racial / ethnic and weight groups, may offer additional pathophysiologic insight into the condition and improve opportunities for early risk profiling before disease has been fully established.

Understanding the preclinical stages of hepatic fat accumulation, rather than focusing exclusively on individuals with established NAFLD, may help improve our ability reverse this condition in young populations.

Methods

In the current study, we use a multi-ethnic cohort of youth (Median age =16.8; Range 12.6–19.6 years) from the Exploring Perinatal Outcomes among Children (EPOCH) study, in order to examine 5 SNPs (PNPLA3 rs738409, NCAN rs2228603, GCKR rs1260326, PPP1 R3B rs4240624, LYPLAL1 rs12137855) that have been previously linked with HFF in adults, and some of which have also been associated in pediatric populations. We also use these SNPs to build a weighted genetic risk score (GRS) for HFF as measured by magnetic resonance imaging (MRI), assess how the associations vary across race / ethnicity, and explore the role of obesity and other risk factors. EPOCH is a historical prospective study of over 600 mother/child pairs identified through the Kaiser Permanente of Colorado Perinatal database based on child exposure to gestational diabetes mellitus during singleton pregnancies. The cohort used in this analysis includes the subset of the children (N=347) with both a measure of HFF at an in-person study visit conducted at an average age of 16.7 years (SD=1.2 years), as well as genetic data. The study was approved both by the Colorado Multiple Institutional Review Board and Human Participant Protection Program. All participants provided written informed consent.

Data collection

Hepatic imaging was performed using a modification of the Dixon method by Hussain involving multi-breath-hold double gradient echo sequences.[17,18] HFF was calculated from the mean pixel signal intensity data for each flip angle acquisition. NAFLD is categorized based on HFF with 5%−33% defined as mild, 34%−66% as moderate, and >66% as severe.[19] Height was measured by SECA stadiometer, and weight was measured using an electronic SECA scale, as described previously.[20] Age- and sex-specific BMI z-scores were calculated using CDC reference standards,[21] and weight groups were defined using percentiles of BMI-for-age: underweight (defined as <5th percentile); normal weight (5th to 85th percentile); overweight (85th up to 95th percentile); obese (95th percentiles or above).[22] Blood samples were obtained at the EPOCH study visit after an overnight fast, and glucose, triglycerides, and alanine aminotransferase were measured, as described previously.[20] HOMA-IR [homeostasis model of insulin resistance: fasting glucose (mmol/l) × fasting insulin (μU/ml)/ 22.5] was used as a marker of insulin resistance. Maternal gestational diabetes status was physician-diagnosed using a standard two-step screening protocol[23] and ascertained from the Kaiser Permanente of Colorado Perinatal database, an electronic database linking the neonatal and perinatal medical record. Race/ethnicity was self-reported using 2000 US census definitions and categorized as Hispanic (any race), non-Hispanic white, non-Hispanic African American, and non-Hispanic other.

Data collection also involved completion of the Block Kid’s Food Questionnaire, a semiquantitative food frequency questionnaire (FFQ) developed for children aged 8 years and older, which assesses 83 food items consumed in the last week.[24] Participants reported the frequency of consumption of small, medium, or large portions of each food or beverage ranging from “1 day” to “every day.”

Genetic data

Peripheral venous blood was drawn from children at the study visit and stored at −80°C. DNA was extracted using the QIAamp kit (Qiagen, Germantown, MD). DNA samples were quantified and purity assessed using a NanoDrop spectrophotometer and a Qubit fluorometer (Thermo Scientific, Wilmington, DE). Samples were genotyped in two batches. The first batch of N=336 (N=223 in this study) was performed using the Illumina Infinium Omni2.5–8 v1.1 BeadChip. The second batch of N=140 (N=124 in this study) was performed using the Illumina Multi-Ethnic Global Array (MEGA) v1.0. Prior to any other quality control procedures, variants genotyped on the Omni2.5–8 v1.1 BeadChip were filtered to retain variants that are represented at the same chromosome and position on the most up-to-date version of the Omni 2.5 array (Omni2.5–8 v1.4). Beyond this initial filtering step, the same quality control procedures were applied to both the Omni 2.5 and MEGA data sets. Individuals with >5% missing genotypes and variants with >2% missing genotypes were excluded.

Principal components (PCs) for global ancestry and possible batch genotyping effects were calculated using variants that were directly genotyped and passed quality control on both BeadChips. We selected variants with a minor allele frequency (MAF) >5% and performed linkage disequilibrium (LD) pruning to retain a subset of independent variants with a maximum pairwise correlation of 0.2. All calculations were completed using PLINK 1.9 (https://www.coggenomics.org/plink/1.9).[25]

Genotypes in each dataset were aligned to the forward strand.[26] We then used the Michigan Imputation Server (v1.0.4)[27] to phase and impute missing genotypes in each data set. Datasets were imputed separately to maintain the intended genotyping backbone of each BeadChip.[28] Genotypes were phased using Eagle and imputation was completed using the 1000 Genomes Phase 3 (version 5) reference panel.[29,30] Imputed genotypes were modeled as dosage in association models.

SNP selection

We identified SNPs previously associated with a continuous measure of hepatic fat in adult or pediatric populations from previously published research. We included 5 SNPs with evidence of association in non-Hispanic whites (NHW) or Hispanic populations because these groups form the majority of our cohort (PNPLA3 rs738409, GCKR rs1260326, PPP1 R3B rs4240624, LYPLAL1 rs12137855, NCAN rs2228603).[3] Among the 5 chosen SNPs, 4 had previously reported beta estimates for regressions of HFF.[31] The exception was GCKR rs1260326, which is in high linkage disequilibrium with rs780094 (R²≥0.91). We chose to include rs1260326 rather than rs780094 because it had slightly higher risk allele prevalence in EPOCH (allele frequency of 1.1 versus 1.0), as well as a slightly stronger univariate association with NAFLD status (p=0.031 versus p-value=0.113). Although both GCKR SNPs have been repeatedly associated with NAFLD,[32] we were only able to find reported effect estimates for association with a continuous measure of HFF for rs780094, which we used in the weighted GRS. The GRS was calculated using standard methods as the weighted sum of the number of risk alleles at each of 5 loci, weighted by the previously reported effect size for association with HFF, and scaled so that each unit increase in GRS corresponds approximately to one additional risk allele.[5,31,33]

Dietary Patterns

We previously created dietary patterns from the FFQ data for this cohort. We consolidated the 83 items from the FFQ into 42 food groups based on their nutritional properties.[34] We then estimated total daily energy intake using the United States Department of Agriculture Food Composition Database[35] and adjusted each food group by total energy intake using the residuals method.[36] Using principal components analysis (PCA; PROC FACTOR in SAS), we consolidated the food groups into principal components (factors) and rotated them orthogonally to maintain non-correlation and facilitate interpretability. PROC FACTOR extracts as many factors as there are original variables—ie, the 42 food groups were converted into 42 factors, each of which represents a unique dietary pattern parameterized as a continuous, normally-distributed score that can be interpreted as the extent to which an individual’s diet resembles the combination of food groups within a given factor. We considered food groups with factor loadings ≥|0.30| to be a key contributor to a dietary pattern. Of the 42 factors, we retained the first two based on standard criteria of the Scree plot and eigenvalues >1[37], and interpretability of the dietary patterns. We refer to the patterns as “prudent,” characterized by high fruit and vegetable intake, and “Western,” characterized by high levels of fried foods and refined carbohydrates; more detail can be found in Table I (available at www.jpeds.com).

Statistical Analyses

We compared cohort demographic characteristics by race / ethnicity, using one-way ANOVA for continuous variables and chi-squared or Pearson exact tests for categorical variables. We defined NAFLD as HFF≥5%.[38]

We used linear regression models with inverse-normalized HFF as a function of each genetic risk factor of interest (SNPs or the GRS), controlling for age, age-squared, sex and the first 3 genetic PCs (see above) in order to account for potential confounding by genetic ancestry, experimental batch effects and any residual relatedness among participants. This is generally in line with the approach used in published analyses of HFF[5] with the exception that we chose not to control for alcohol intake because this was a pediatric cohort, and the reported alcohol intake estimated from FFQs was negligible (max of 0.35 grams; a standard drink contains 14 grams of alcohol). We then checked for an interaction between NHW race / ethnicity and each genetic risk factor in the subset of individuals who were either NHW or Hispanic (N=308). For any genetic risk factor with an interaction p-value≤0.05, we ran stratified models in NHWs and Hispanics. We considered p-values of 0.01 as significant according to a Bonferroni threshold to account for multiple comparisons (5 SNPs), and nominally significant p-values are also noted. All analyses were performed using R v3.5.0.[39]

In order to evaluate predictors of HFF and the added predictive value of SNPs, we performed sequential regression models. The first model included the control variables from the model described above: age, age-squared, and sex; then BMI z-score and HOMA-IR were added; and finally dietary patterns previously associated with HFF in this cohort (a “prudent” pattern characterized by high intake of fruits and vegetables, and a “Western” pattern characterized by high intake of fried foods and refined carbohydrates; Table 1)[34] were added. We then added the genetic risk factors significantly associated with HFF, rs738409 or the weighted GRS, to the final regression model. We estimated the changes in model fit using the adjusted R² values. In these regression models, we also considered the inclusion of exposure to diabetes in utero, maternal pre-pregnancy weight, physical activity measures and acanthosis nigricans as predictor variables, but these did not improve model fit.[40] We imputed missing covariates for BMI z-score (N=1) and HOMA-IR (N=7) using the R mice[41] package and age, sex, race / ethnicity, and non-missing BMI z-score and HOMA-IR as predictors. We ran these regressions stratified by ethnicity among all NHWs and Hispanics first in the overall cohort and then in the subset of normal weight individuals.

Table 1; online.

Composition of the two dietary patterns at age 12–19 years associated with hepatic fat fraction (HFF) among youth in EPOCH.

Food group	% Variance	Factor loading
Factor 1 (Prudent)	8.80%
Leafy greens		0.68
Vegetables		0.68
Fruit		0.58
Cruciferous vegetables		0.46
Nuts & seeds		0.46
Yogurt		0.44
Stir-fried vegetables		0.4
Salad dressing		0.38
Sugar-sweetened beverages		−0.35
Fast food		−0.35
Beef		−0.36
Factor 2 (Western)	5.80%
Fried potatoes		0.58
Ketchup		0.52
Beef		0.44
Fast food		0.42
Salad dressing		0.41
Fried packaged snacks		0.36
Cereal		−0.48

Results

The cohort of 347 individuals was predominantly of NHW ethnicity (Table 2; N=190, 54.8%); about one third were of Hispanic ethnicity (N=118, 34.0%), and the remaining individuals were African American (AA; N=32, 9.2%) or other races (mostly Asian, N=7, 2.0%). We saw significant differences in adiposity measures by race / ethnicity;[42] AAs had the highest BMI z-score, obesity prevalence, and subcutaneous fat (SFAT), followed by Hispanics and then NHWs. Waist circumference and visceral fat showed a slightly different pattern with non-significant differences by race / ethnicity, but the values were highest among Hispanics, then NHWs and AAs. NAFLD prevalence and HFF were highest in Hispanics. Genetic risk for NAFLD also varied with race / ethnicity, with AAs and other races having lower GRSs than Hispanics and NHWs. The frequency of the HFF risk alleles by race / ethnicity in EPOCH is comparable with that of prior studies (Table 3). Almost all individuals in the cohort had at least one of risk allele for the SNPs examined (distribution of SNPs shown in Table 4 [available at www.jpeds.com]).

Table 2.

Demographic characteristics and risk factors among the EPOCH population by race/ethnicity.

		NHW	Hispanic	AA	Other	P-value
N (%)		190 (54.8%)	118 (34.0%)	32 (9.2%)	7 (2.0%)
Age (years)		16.9 (1.1)	16.5 (1.3)	16.2 (1.1)	16.8 (0.4)	0.003
Male sex		92 (48.4)	65 (55.1)	15 (46.9)	3 (42.9)	0.648
Risk factors
GDM		44 (23.2)	17 (14.4)	4 (12.5)	0 (0.0)	0.097
Smoker		14 (7.4)	4 (3.4)	0 (0.0)	0 (0.0)	0.191
BMI z-score		0.23 (1.03)	0.50 (1.15)	0.74 (1.21)	−0.16 (0.71)	0.016
BMI percentile		56.9 (29.2)	62.1 (30.4)	67.7 (30.7)	45.2 (24.8)	0.096
BMI category	Normal	149 (78.4)	79 (66.9)	18 (56.2)	7 (100.0)	0.004
	Overweight	25 (13.2)	14 (11.9)	4 (12.5)	0 (0.0)
	Obese	16 (8.4)	25 (21.2)	10 (31.2)	0 (0.0)
Waist circumference		80.0 (10.7)	81.7 (15.5)	79.4 (11.7)	70.2 (5.5)	0.097
Visceral fat		32.4 (18.9)	34.4 (23.5)	26.9 (16.7)	17.8 (8.0)	0.071
Subcutaneous fat		176.2 (116.9)	220.0 (166.9)	248.9 (185.4)	114.1 (71.8)	0.004
HOMA-IR		3.5 (2.8)	3.8 (3.5)	4.0 (3.0)	2.8 (1.5)	0.649
Cholesterol		144.6 (28.3)	145.0 (26.5)	144.6 (29.4)	139.1 (28.8)	0.962
HDL		45.9 (8.7)	46.6 (11.5)	46.9 (9.1)	50.7 (6.4)	0.560
Dietary pattern factor scores	Prudent	0.18 (1.10)	−0.17 (0.92)	−0.17 (0.82)	−0.13 (0.39)	0.025
	Western	−0.02 (1.01)	−0.01 (1.03)	−0.04 (0.88)	−0.25 (0.47)	0.964
Hepatic outcomes
NAFLD		8 (4.2)	16 (13.6)	1 (3.1)	0 (0.0)	0.012
HFF		2.0 (1.4)	3.2 (4.8)	2.1 (1.2)	2.1 (1.2)	0.010

NHW: Non-Hispanic white; AA: African American; GDM: gestational diabetes mellitus; NAFLD: Non-alcholic fatty liver disease; HFF: Hepatic Fat Fraction

Table 3.

Allele frequency in EPOCH by race / ethnicity for SNPs previously associated with Hepatic fat fraction (HFF).

		EPOCH
Gene	SNP	NHW	Hispanic	AA	Other
PNPLA3	rs738409 (G)	0.23	0.41	0.11	0.21
NCAN	rs2228603 (T)	0.08	0.03	0.02	0
GCKR	rs1260326 (T)	0.42	0.41	0.2	0.35
PPP1 R3B	rs4240624 (A)	0.92	0.85	0.86	0.93
LYPLAL1	rs12137855 (C)	0.79	0.85	0.9	0.79

SNP: Single nucleotide polymorphism; NHW: Non-Hispanic white; AA: African American

Table 4; online.

Distribution of HFF risk SNPs among the EPOCH population by race/ethnicity. The table shows the number of individuals having at least one risk allele for each of the SNPs examined, and the distribution of the combination of SNPs.

Number of risk SNPs	Combindations of risk SNPs	All	AA	Hispani	NHW	Other
1	rs4240624 (A)	1	0	1	0	0
2		78	17	15	45	1
2	rs4240624 (A), rs12137855 (C)	69	17	14	38	0
2	rs1260326 (T), rs4240624 (A)	6	0	1	5	0
2	rs738409 (G), rs4240624 (A)	3	0	0	2	1
3		147	11	54	76	6
3	rs1260326 (T), rs4240624 (A), rs12137855 (C)	90	8	26	51	5
3	rs738409 (G), rs4240624 (A), rs12137855 (C)	44	3	25	15	1
3	rs2228603 (T), rs4240624 (A), rs12137855 (C)	6	0	0	6	0
3	rs738409 (G), rs1260326 (T), rs4240624 (A)	3	0	2	1	0
3	rs738409 (G), rs1260326 (T), rs12137855 (C)	1	0	1	0	0
3	rs738409 (G), rs2228603 (T), rs12137855 (C)	1	0	0	1	0
3	rs2228603 (T), rs1260326 (T), rs4240624 (A)	1	0	0	1	0
3	rs738409 (G), rs2228603 (T), rs4240624 (A)	1	0	0	1	0
4		110	3	45	62	0
4	rs738409 (G), rs1260326 (T), rs4240624 (A), rs12137855 (C)	92	3	41	48	0
4	rs2228603 (T), rs1260326 (T), rs4240624 (A), rs12137855 (C)	12	0	1	11	0
4	rs738409 (G), rs2228603 (T), rs4240624 (A), rs12137855 (C)	6	0	3	3	0
5	rs738409 (G), rs2228603 (T), rs1260326 (T), rs4240624 (A), rs12137855 (C)	11	1	3	7	0

SNP: Single nucleotide polymorphism; HFF: Hepatic fat fraction; NHW: Non-Hispanic white; AA: African American

Levels of HFF differed substantially by race / ethnicity (Figure 1; available at www.jpeds.com). There were 5 Hispanics who had HFF values over 10%, with a maximum of 38.2%; no other ethnic groups had an individual with HFF above 8.6%. We ran regression models of HFF, and the effect estimates were generally in line with prior studies, except for PPP1 R3B rs4240624, which had a much weaker association with HFF in EPOCH compared with prior studies (Figure 2, A). The only SNP showing a nominally significant association with HFF was PNPLA3 rs738409 (p=0.028). We evaluated the interaction between the SNPs of interest and NHW ethnicity among the subset of NHW or Hispanic individuals (N=308), and there was a significant interaction for PNPLA3 rs738409 and the GRS. These genetic risk factors showed substantially stronger effects among Hispanics than NHWs (Figure 2. B). PNPLA3 rs738409 had negligible effect estimates among NHWs, whereas both rs738409 and the weighted GRS met the Bonferroni threshold of p≤0.01 for significance (p≤0.0035) in Hispanics.

Figure 1;

online. Hepatic fat fraction (HFF) distribution by race / ethnicity. A common cutoff for diagnosis of NAFLD is 5% HFF, which is indicated by the dotted line. Hispanics showed much greater variation in HFF than other racial / ethnic groups in EPOCH, and they had the highest prevalence of NAFLD.

NHW: Non-Hispanic white; AA: African American

Figure 2.

Forest plot of estimates and 95% confidence intervals from linear regressions of Hepatic fat fraction (HFF) as a function of SNPs previously associated with HFF in adults (prior reported estimates shown for comparison) [31] and a weighted genetic risk score (GRS) composed of these SNPs (a). PNPLA3 rs738409 and the GRS showed a significant interaction with ethnicity; effect estimates stratified by Hispanic / non-Hispanic white (NHW) ethnicity are shown in b). (Estimates for African Americans and other races not estimated due to small sample size.) Both rs738409 and the GRS were significantly associated with HFF among Hispanics but not non-Hispanic whites (NHWs) in EPOCH. Control variables include age, age squared, sex, and the first 3 PCs for models not stratified by ethnicity. Nominally significant p-values (≤0.05) indicated by *; p-values meeting the Bonferroni threshold for significance (≤0.01) indicated by **.

SNP: Single nucleotide polymorphism

We ran sequential linear regression models stratified by ethnicity (Figure 3) in order to investigate the incremental proportion of variation in HFF explained by adding, in order, demographics, cardiometabolic markers, dietary information and genetic predictors (PNPLA3 rs738409 or the weighted GRS). The adjusted R² values (amount of variation explained in HFF) for the baseline demographic models improved substantially with the addition of cardiometabolic markers (increases of 27.2% for Hispanics and 7.0% for NHWs), and showed subsequent improvement with the further addition of dietary information only in NHWs (change of −1.0% for Hispanics and 5.2% for NHWs). The inclusion of genetic risk factors only improved the model among Hispanics (increase of 6.4% for rs738409; 4.3% for the GRS); among NHWs, the adjusted R² values decreased with the addition of genetic information.

Figure 3.

Plots of the adjusted R-squared values from sequential linear regressions of hepatic fat fraction (HFF) among Hispanics and non-Hispanic whites (NHWs) in EPOCH, stratified by ethnicity, in the overall cohort and then among normal weight individuals. Predictors included 1) demographics (age, age squared and sex), 2) demographics and cardiometabolic markers (BMI z-score and HOMA-IR), 3) demographics, cardiometabolic markers and diet, 4) demographics, cardiometabolic markers, diet and PNPLA3 rs738409, and 5) demographics, cardiometabolic markers, diet and weighted genetic risk score (GRS). Cardiometabolic markers explain substantially more variation in HFF than other risk factors in the overall cohort. Genetic risk (particularly rs738409) only improves the models among Hispanic individuals, and particularly among Normal weight Hispanics.

We were particularly interested in elevated HFF among normal weight individuals, five of whom met the diagnostic criteria for NAFLD. All 5 of these individuals had at least one risk allele in PNPLA3 rs738409. We ran the same sequential linear regressions of HFF described above among the normal weight individuals in the cohort (N=253), and the baseline models had very poor R² values that showed little or no improvement with the inclusion of cardiometabolic markers or diet (Figure 3). The addition of genetic information showed the same pattern as in the overall cohort by ethnicity: it greatly improved the R² of models among Hispanics – even more so than in the overall cohort with increases of 12.6% and 4.4% for rs738409 and the GRS, respectively. Adjusted R² values decreased among NHWs with the addition of genetic information. Although BMI z-scores were a statistically significant predictor of HFF in lean Hispanics, HOMA-IR was not. Neither genetic risk factors nor cardiometabolic markers were significantly associated with HFF in lean NHWs.

Discussion

In this multiethnic cohort of youth, we found that the SNP PNPLA3 rs738409 and a weighted GRS including 5 SNPs were significantly associated with HFF among Hispanic but not NHW individuals. Cardiometabolic risk factors explained substantially more variation in HFF than genetic risk factors in both Hispanics and NHWs. Many of the prior pediatric studies of hepatic fat have focused exclusively on obese individuals.[10] However, NAFLD affects normal weight individuals as well, in whom the condition is less likely to be diagnosed and may have clinically distinct features.[43] Our results suggest that genetic risk, particularly the rs738409 G allele, may be a particularly important risk factor for high HFF among normal weight individuals.

There are differences in NAFLD prevalence by race / ethnicity, as well as in the environmental and genetic risk factors for NAFLD.[1,13,16] Hispanic ethnicity is not precisely defined and encompasses individuals of diverse cultures and an admixture of genetic ancestry of Native American, European and African Ancestry, with the proportion of each varying regionally.[44] Studies have also shown that there are regional differences in the prevalence of suspected NAFLD among Hispanics.[45,46] One study found that metabolic syndrome (MetS) was more strongly associated with HFF in Mexican Americans and Non-Hispanic blacks than in NHWs.[47] Although we did not examine MetS, which is not straightforward to define in adolescents,[48] we likewise saw that HOMA-IR was significantly associated with HFF in Hispanics but not NHWs, though BMI z-score was associated with HFF in both ethnic groups. In contrast, another study found that HOMA-IR was a risk factor for NASH among NHW but not Hispanic adults with biopsy-proven NAFLD.[49]

In this study, we examined five SNPs, and only one of them, PPP1 R3B rs4240624, showed a notably lower effect estimate relative to a prior study of HFF in adults.[31] The relationship between this SNP and HFF and liver damage is controversial.[31,50,51] It was initially associated with HFF from computerized tomography scans but not with histologic NAFLD or NASH, and in recent work, this it was associated with protection against HFF and fibrosis in individuals at high risk of NAFLD.[51] It is possible that this SNP has different effects in different subpopulations, masking the association in this cohort, or that it may promote hepatic fat accumulation at older ages than represented in this cohort.[50]

Early onset NAFLD may differ from NAFLD in adults in that it is characterized by different patterns of steatosis, inflammation and cellular injury, and it may be more prone to progress towards more serious conditions, such as NASH, [52] which is among the leading indications for liver transplants.[53] One study found that almost half of children who are diagnosed with NAFLD have NASH at the time of diagnosis.[52] This could reflect the higher threshold needed for physicians to suspect NAFLD in youth, or it may reflect differences in disease presentation and progression in younger populations. Profiling genetic risk may help to target early prevention efforts for NAFLD. However, it is important to understand whether genetic risk substantially improves our ability to identify children with high levels of HFF. Our results suggest that genetic risk factors may show a stronger association with HFF among Hispanic than NHW youth. Cardiometabolic risk factors that are more routinely captured in routine clinical care, including BMI z-scores and HOMA-IR, show a stronger association with HFF in this cohort. However, genetic risk factors may be an important tool to identify risk for elevated HFF among normal weight youth.

There are major gaps in our understanding of NAFLD among lean individuals, including the contribution of genetic and other predisposing risk factors.[54] NAFLD in lean individuals is suspected to occur in metabolically unhealthy lean individuals.[43] Although this may be the case, HOMA-IR was not significantly associated with HFF among normal weight individuals in our cohort. We also found that in lean Hispanics, albeit a small sample, genetic risk factors explained a greater proportion of variation in HFF than cardiometabolic markers. Although it is possible that there are different genetic risk factors unique to NAFLD in lean individuals,[54] we confirmed that in lean Hispanic youth, the primary established genetic risk factor for NAFLD, rs738409, was strongly associated with HFF.

This study has limitations. The cohort is relatively small, limiting the power, particularly for less common SNPs. The Hispanics in this cohort had much higher prevalence of NAFLD than other racial / ethnic groups and much greater variation in HFF, which may have limited our relative ability to detect associations with HFF in NHWs. The weighted GRS was based on regression estimates from the Genetics of Obesity-related Liver Disease (GOLD) study,[31] which is comprised of individuals of European descent, and we applied this GRS to all of the individuals in this multiethnic cohort. Although factors like epigenetics and socioeconomic status may also contribute towards hepatic fat accumulation,[16] we were unable to explore these influences because this information was not available for most of the cohort. A major strength of this study is that HFF was measured by MRI, which is the most precise non-invasive method to capture HFF.[55] Furthermore, this study included children across the spectrum of weight classes, in contrast to many of the prior studies of genetics in pediatric NAFLD, which focused exclusively on obese children.[10]

Although there is a substantial body of literature to support that genes play a role in NAFLD,[3] our results suggest that poor cardiometabolic health may be more important to consider when predicting HFF in young populations. In this multiethnic cohort of youth, genetic risk factors for HFF had the highest prevalence among those with Hispanic ethnicity, and these genetic risk factors were only associated with HFF among Hispanics. PNPLA3 rs738409 was not associated with HFF among NHWs, despite a frequency of 0.23 for the risk allele. Further studies are necessary to elucidate whether genetic risk is an important predictor of HFF among populations of NHWs who have more severe levels of HFF, and whether genetic factors are particularly useful for HFF risk profiling among lean individuals.