Mendelian randomization analysis of n-6 polyunsaturated fatty acid levels and pancreatic cancer risk

Abstract

Background:

Whether circulating polyunsaturated fatty acids (PUFA) levels are associated with pancreatic cancer risk is uncertain. Mendelian randomization (MR) represents a study design using genetic instruments to better characterize the relationship between exposure and outcome.

Methods:

We utilized data from genome-wide association studies within the Pancreatic Cancer Cohort Consortium and Pancreatic Cancer Case-Control Consortium, involving approximately 9,269 cases and 12,530 controls of European descent, to evaluate associations between pancreatic cancer risk and genetically predicted plasma n-6 PUFA levels. Conventional MR analyses were performed using individual-level and summary-level data.

Results:

Using genetic instruments, we did not find evidence of associations between genetically predicted plasma n-6 PUFA levels and pancreatic cancer risk (estimates per one standard deviation increase in each PUFA-specific weighted genetic score using summary statistics: Linoleic acid – odds ratio (OR) = 1.00, 95% confidence interval (CI) 0.98-1.02; arachidonic acid – OR = 1.00, 95% CI 0.99-1.01; dihomo-gamma-linolenic acid – OR = 0.95, 95% CI 0.87-1.02). The OR estimates remained virtually unchanged after adjustment for covariates, using of individual level data or summary statistics, or stratification by age and sex.

Conclusions:

Our results suggest that variations of genetically determined plasma n-6 PUFA levels are not associated with pancreatic cancer risk.

Impact:

These results suggest that modifying n-6 PUFA levels through food sources or supplementation may not influence risk of pancreatic cancer.

Introduction

Pancreatic cancer (PC) remains one of the deadliest cancers (1). Polyunsaturated fatty acids (PUFA), linked to the inflammatory process, may influence PC development (2). However, evidence from epidemiological studies is inconsistent (3). For example, associations with n-3 PUFA intake were inverse, positive, or null, and associations with n-6 PUFA intake were positive or null across different studies. Conventional epidemiologic study designs may suffer from methodological limitations, such as reverse causation, selection bias, and uncontrolled confounding (4). We therefore conducted a Mendelian randomization (MR) analysis using genetic variants as instrumental variables. Higher proportions of variance of n-6 PUFA levels were explained by variants compared with n-3 PUFA (5,6). We thus focused on n-6 PUFA in our analysis.

Materials and Methods

Instrumental variables

We identified single nucleotide polymorphisms (SNPs) associated with plasma or red blood cell (RBC) levels of n-6 PUFAs [linoleic acid (LA), arachidonic acid (AA), adrenic acid (AdrA), gamma linolenic acid (GLA), and dihomo-gamma-linolenic acid (DGLA)] from the genome-wide association studies (GWAS) catalog and from published literature (up to November 2018) (6). We selected SNPs associated at P<5×10⁻⁸ that were independent from each other (r² < 0.1). For correlated SNPs, the SNP with a lower p-value was selected unless an independent association was reported, in which case both were selected. We used estimates of association with plasma PUFA levels for our analyses. For SNPs initially reported to be associated with RBC PUFA levels, we checked their associations with plasma levels in the GWAS conducted by the Cohorts for Heart and Aging Research in Genomic Epidemiology consortium (CHARGE) (6); if estimates were significant (P<0.05) we included them in our analyses.

Genetic association datasets for PC risk

For evaluation of associations with PC risk, we used data from four GWAS conducted in the Pancreatic Cancer Cohort Consortium (PanScan) and the Pancreatic Cancer Case-Control Consortium (PanC4) (7). Detailed information on quality control and imputation has been provided elsewhere (7). Data from approximately 9,269 cases and 12,530 controls of European ancestry were used. Only variants with imputation quality of r² ≥ 0.3 were retained.

MR analysis

We performed separate MR analyses for each type of PUFA. Based on power estimation (https://shiny.cnsgenomics.com/mRnd/), the minimal detectable odds ratios per standard deviation of genetically predicted PUFA levels at 80% power and alpha of 0.05 ranged from 1.08 to 1.13 for LA, 1.06 to 1.21 for AA, 1.13 to 1.15 for AdrA, 1.17 to 1.27 for GLA, and 1.08 to 1.11 for DGLA. We created a weighted genetic score to represent the genetically estimated PUFA level using information from published GWAS of PUFA plasma levels and data from PanScan/PanC4 GWAS. For each subject, a weighted genetic score (wGRS) was created as the weighted sum of the number of association alleles at each locus multiplied by the point estimate for the association with plasma PUFA level:

$$wGRS=\sum _{i=1}^n{\beta }_{i}SN{P}_i$$

where β_i is the regression coefficient of the ith SNP for the PUFA, and SNP_i is the dosage of the association alleles (0,1,2) of the ith SNP. All association alleles were converted to correspond to increased PUFA levels. We used logistic regression models to assess the associations between wGRS and PC risk. Besides unadjusted analyses, we performed analyses adjusting for age, sex, and the top principal components. We stratified by age (<50 years, 50-70 years, ≥70 years) and sex. We also used the summary statistics of the PanScan/PanC4 GWAS to estimate the MR associations using the fixed-effects inverse variance-weighted approach (4).

Results

The instruments used for each of the n-6 PUFA are included in Table 1. There was no evidence of association (at P<0.01) between any of the wGRS and common PC risk factors. We did not detect any statistically significant associations between genetically predicted plasma n-6 PUFA levels and PC risk (Table 2). The estimates for AdrA and GLA had wide confidence intervals, consistent with the estimated lower power for these. The associations remained virtually unchanged regardless of covariate adjustment, analyzing individual level versus summary statistics data, or within strata of age or sex.

Table 1.

Genetic instruments for plasma phospholipid levels of n-6 polyunsaturated fatty acids (PUFAs, % of total fatty acids) that were genome-wide significant (P<5 × 10⁻⁸) in previous GWAS

Chr	SNP	GRCh37/hg19 position	Allelea	EAF	β	S.E.	P-value	% VEb per allele	% VE per IVc	F-statistic per IVd
Linoleic acid (LA, 18:2n6)
10	rs10740118	65101207	G/C	0.56	0.2484	0.0431	8.08 × 10−9	0.2–0.7	9.4-25.1	452 – 1461
11	rs174547	61570783	C/T	0.32	1.4737	0.0417	4.98 × 10−274	7.6–18.1
11	rs2727270	61603237	T/C	0.44	0.69	0.07	2.60 × 10−21	0.5–2.4
16	rs16966952	15135943	A/G	0.31	0.3512	0.0439	1.23 × 10−15	0.5–2.5
16	rs2280018	15150833	A/C	0.38	0.38	0.05	3.60 × 10−14	0.6-1.4
Arachidonic acid (AA, 20:4n6)
11	rs174547	61570783	T/C	0.68	1.6909	0.0253	3.30 × 10−971	3.7-37.6	4.1-44	311-5708
11	rs102275	61557803	T/C	0.68	2.49	0.1	6.60 × 10−147	0.3-5.8
16	rs16966952	15135943	G/A	0.69	0.1989	0.0314	2.43 × 10−10	0.1-0.6
Adrenic acid (22:4,n6)
11	rs174547	61570783	T/C	0.67	0.0483	0.0019	6.26 × 10−140	7.8-10.9	7.8-10.9	1844-2667
Gamma linolenic acid (GLA; 18:3,n6)
11	rs174547	61570783	T/C	0.67	0.0156	0.0009	2.29 × 10−72	2.2-4.6	2.5-6.4	186-497
16	rs16966952	15135943	G/A	0.69	0.0061	0.0009	5.05 × 10−11	0.3-1.8
11	rs10899123*	75501207	C/G	0.91	0.0055	0.0014	9.97 × 10−5	NA
Dihomo-gamma-linolenic acid (DGLA; 20:3,n6)
11	rs174547	61570783	C/T	0.33	0.3550	0.0136	2.63 × 10−151	8.7-11.1	13.5-26.3	850-1944
11	rs968567	61595564	T/C	0.16	0.29	0.02	1.30 × 10−42	1.4-7.9
16	rs16966952	15135943	G/A	0.69	0.2204	0.013	7.55 × 10−65	2.0-4.5
16	rs2280018	15150833	C/A	0.61	0.16	0.02	4.50 × 10−25	1.4-2.8

Abbreviations: EAF=effect allele frequency; IV=instrumental variable; s.e.=standard error; SNP=single-nucleotide polymorphism.

^athe first listed allele represents the effect allele associated with an increase level of corresponding PUFA, the second allele represents the alternative allele

^b% variation explained (VE)=(2 × β² × EAF × (1−EAF)/var(PUFA)) × 100 unless indicated in paper such as in Guan et al (6).

^c% VE per IV=sum of the %VE per allele for each SNP included in the IV.

^dF-statistic is a measure of the strength of the genetic instrument and is calculated as follows: (R² × (n-1-k))/((1-R²) × k), where R²=% variation explained, n=sample size, k=total number of instrumental variables.

^*genetic variant rs10899123 showed an association at 5 × 10⁻⁸<P<0.05 in the Cohorts for Heart and Aging Research in Genomic Epidemiology studies (Guan et al, 2014), although it showed a GWAS significant association in the Hu et al (2016). It was included in the genetic instrument in sensitivity analyses while did not significantly change the association. The analyses excluding it in the instrument were reported in Table 2.

Table 2.

Associations between one standard deviation increase in PUFA-specific wGRSs and pancreatic cancer risk in PanScan and PanC4 studies^*

		Linoleic acid (LA)			Arachidonic acid (AA)			Dihomo-gamma-linolenic acid (DGLA)
Subgroup	Cases/Controls	OR	95% CI	P-value	OR	95% CI	P-value	OR	95% CI	P-value
Overalla	9,269/12,530	0.99	0.97, 1.01	0.30	1.01	1.00, 1.02	0.37	0.94	0.87, 1.01	0.10
Overallb	9,206/12,525	1.00	0.98, 1.02	0.89	1.00	0.99, 1.01	0.53	0.94	0.87, 1.02	0.13
Data Sourceb
PanScan1	1,746/1,812	1.03	0.97, 1.08	0.31	0.99	0.97, 1.01	0.35	0.97	0.81, 1.16	0.72
PanScan2	1,768/1,841	0.99	0.93, 1.04	0.59	1.01	0.99, 1.03	0.47	0.97	0.82, 1.15	0.73
PanScan3	1,528/5,080	0.99	0.93, 1.05	0.66	1.01	0.98, 1.03	0.56	0.91	0.76, 1.10	0.34
PanC4	4,164/3,792	1.00	0.96, 1.03	0.91	1.01	0.99, 1.02	0.53	0.94	0.83, 1.06	0.30
Overallc	9,040/12,496	1.00	0.97, 1.03	0.95	1.00	0.99, 1.02	0.73	0.95	0.87, 1.03	0.21
Aged
>70	3,494/3,385	1.02	0.98, 1.06	0.42	1.00	0.98, 1.01	0.68	1.02	0.90, 1.17	0.73
50-70	3,917/6,916	0.99	0.96, 1.03	0.74	1.01	0.99, 1.02	0.55	0.91	0.81, 1.02	0.11
≤50	1,795/2,224	0.98	0.93, 1.03	0.42	1.01	0.99, 1.04	0.32	0.90	0.75, 1.07	0.24
Sexe
Male	4,985/7,801	1.00	0.97, 1.03	0.94	1.00	0.99, 1.02	0.66	0.97	0.87, 1.08	0.55
Female	4,221/4,225	1.00	0.97, 1.04	0.99	1.00	0.99, 1.02	0.71	0.99	0.95, 1.04	0.77

^*results for adrenic acid and gamma linolenic acid not shown; their associations are not significant, with relatively wide confidence intervals

^aORs (odds ratios) and 95% CIs (confidence intervals) estimated using individual level data without adjustment, and represent one s.d. increase in each PUFA-specific wGRS

^bORs and 95% CIs estimated using individual level data with adjustment of age (under 50, 50-60, 60-70, 70-80, above 80), sex, and ten or seven principal components for PanScan and PanC4 data respectively, and represent one s.d. increase in each PUFA-specific wGRS

^cORs and 95% CIs estimated using summary statistics data

^dORs and 95% CIs estimated using individual level data with adjustment of age, sex, and ten or seven principal components for PanScan and PanC4 data respectively, and represent one s.d. increase in each PUFA-specific wGRS

^eORs and 95% CIs estimated using individual level data with adjustment of age (under 50, 50-60, 60-70, 70-80, above 80), and ten or seven principal components for PanScan and PanC4 data respectively, and represent one s.d. increase in each PUFA-specific wGRS

Discussion

We did not observe significant associations between genetically predicted n-6 PUFA levels and PC risk in the Panscan/PanC4 subjects. As the proportion of variance of n-6 PUFA that can be explained by the summed association magnitudes of these GWAS-identified loci is relatively high, there is reasonable statistical power to detect any meaningful associations. PUFA has been reported to be associated with colorectal cancer risk based on MR analysis (8). Most dietary sources of n-6 PUFA are consumed infrequently. A limitation of this study is that there is no information for the genetic variants associated with total n-6 PUFA levels in previous literature, and thus we could not determine the association of total n-6 PUFA and pancreatic cancer risk using genetic instruments. Alternative designs of a direct assessment of dietary sources and measurement of PUFA levels in blood at repeat time points can better characterize the relationship between PUFA and PC risk. Further studies are also needed to investigate the potential relationships in subjects of other populations.

Abbreviations list:

PC

Pancreatic cancer

PUFA

polyunsaturated fatty acids

MR

Mendelian randomization

OR

odds ratio

CI

confidence interval

SNPs

single nucleotide polymorphisms

RBC

red blood cell

LA

linoleic acid

AA

arachidonic acid

AdrA

adrenic acid

GLA

gamma linolenic acid

DGLA

dihomo-gamma-linolenic acid

GWAS

genome-wide association studies

CHARGE

Cohorts for Heart and Aging Research in Genomic Epidemiology consortium

PanScan

Pancreatic Cancer Cohort Consortium

PanC4

the Pancreatic Cancer Case-Control Consortium

wGRS

weighted genetic score