Genetically predicted plasma phospholipid arachidonic acid concentrations and 10 site-specific cancers in UK Biobank and genetic consortia participants: A Mendelian randomization study

Summary

Background & aims

Arachidonic acid (AA) is metabolized by cyclooxygenases and lipoxygenases to pro-inflammatory eicosanoids, which according to experimental research modulate tumor cell proliferation, differentiation, and apoptosis. We employed the Mendelian randomization design to test the hypothesis that higher plasma phospholipid AA concentrations are associated with increased risk of 10 site-specific cancers.

Methods

Two genetic variants associated with plasma phospholipid concentrations of AA (rs174547 in FADS1 [P=3.0×10^-971] and rs16966952 in PDXDC1 [P=2.4×10^-10]) in the Cohorts for Heart and Aging Research in Genomic Epidemiology Consortium were used as genetic instruments. The associations of those variants with cancer were taken from the UK Biobank (n=367,643), FinnGen consortium (n=135,638), International Lung Cancer Consortium (n=27,209), Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome consortium (n=140,254), Breast Cancer Association Consortium (n=228,951), Ovarian Cancer Association Consortium (n=66,450), and BioBank Japan (n=212,453).

Results

Higher genetically predicted plasma phospholipid AA concentrations were associated with increased risk of colorectal and lung cancer. Results were consistent across data sources and variants. The combined odds ratios per standard deviation increase of AA concentrations were 1.08 (95% CI 1.05-1.11; P=6.3×10^-8) for colorectal cancer and 1.07 (95%CI 1.05-1.10; P=3.5×10^-7) for lung cancer. Genetically predicted AA concentrations had a suggestive positive association with esophageal cancer (odds ratio 1.09; 95% CI 1.02-1.17; P=0.016) but were not associated with cancers of the stomach, pancreas, bladder, prostate, breast, uterus, or ovary.

Conclusion

These results indicate that AA may be implicated in the development of colorectal and lung cancer and possibly esophageal cancer. Treatments with plasma AA-lowering properties should be evaluated for clinical benefit.

1. Introduction

Arachidonic acid (AA, 20:4n-6) is a long-chain polyunsaturated fatty acid (PUFA) present in human phospholipid cell membranes. AA is obtained from animal food sources, particularly meat, fish, and eggs, or derived from precursor PUFAs, including linoleic acid (Fig. 1). The metabolism of AA by cyclooxygenase and lipoxygenase pathways generates eicosanoids such as prostaglandins, thromboxanes, and leukotrienes (Fig. 1). Experimental research indicates that AA-derived eicosanoids are primarily pro-inflammatory and promote carcinogenesis by modulating tumor cell proliferation, differentiation, apoptosis, and angiogenesis [1, 2]. Yet, it is unknown whether higher AA concentrations are causally associated with carcinogenesis in humans. Epidemiological data on the association between AA biomarker concentrations and cancer risk are scarce, have not assessed the temporal relationship and used varying sources of blood measurement (such as whole blood, plasma, serum, and erythrocytes) [3]. However, a strong body of observational data, together with the Rothwell trial data are at least strongly suggestive of protective effects of aspirin and other non-steroidal anti-inflammatory drugs for a range of cancers, most notably those in the gastrointestinal tract [4–9]. Mechanistically, these drugs inhibit the cyclooxygenase pathway, implicating AA metabolism in carcinogenesis.

Fig. 1.

Simplified overview of AA, EPA, and DHA biosynthesis from precursor PUFAs and the metabolism of these PUFAs by cyclooxygenases (COXs) and 5-lipoxygenase (5-LOX) into eicosanoids. The main food sources of AA, EPA, and DHA are indicated. PG, prostaglandin; TX, thromboxane; LK, and leukotriene.

Mendelian randomization (MR) is a method for appraising and strengthening causal inference in epidemiological studies by leveraging genetic variants as instrumental variables for the risk factor. As genetic variants are distributed randomly when passed from parents to offspring, a genetic variant that alters the levels of the risk factor (e.g., AA concentrations) is generally unrelated to other risk factors (e.g., lifestyle habits). Therefore, confounding is diminished in an MR study.

Here, we utilized the MR approach and large-scale cohort and genetic consortia data to test the hypothesis that higher plasma phospholipid concentrations of AA are associated with an increased risk of 10 site-specific cancers, including gastrointestinal tract cancers and other major cancers. Delta-5 desaturase, encoded by fatty acid desaturase 1 (FADS1), plays a key role in AA biosynthesis but is also involved in the biosynthesis of eicosapentaenoic acid (EPA, 20:5n-3), which can be converted into docosahexaenoic acid (DHA, 22:6n-3) (Fig. 1). Experimental evidence indicates that these two long-chain n-3 PUFAs might have anticancer activities by suppressing the biosynthesis of AA-derived eicosanoids in favor of EPA-derived eicosanoids [1, 10]. Therefore, in sensitivity analyses, we explored the associations between genetically predicted plasma phospholipid EPA and DHA concentrations and cancer risk.

2. Methods

2.1. Instrumental variables

As instrumental variables we selected single nucleotide polymorphisms (SNPs) associated with plasma phospholipid concentrations of AA [11], EPA [12], and DHA [12] at P<5×10^-8 in meta-analyses of genome-wide association studies in five population-based cohort studies, comprising up to 8866 individuals of European ancestry from the Cohorts for Heart and Aging Research in Genomic Epidemiology Consortium. Characteristics and summary statistics for the five SNPs employed as instrumental variables are presented in Table 1. Two SNPs that were used as instruments for AA and EPA were located in or near FADS1 and were in linkage disequilibrium (r²=0.86 in European populations and r²=1.0 in Japanese populations). One AA-associated SNP was located in PDXDC1, and two EPA- and DHA-associated SNPs were located in or near ELOVL2 and were in linkage disequilibrium (r²=0.97 in European populations and r²=1.0 in Japanese populations). Each SNP explained from 0.4% to 1.7% of the variance in EPA and DHA concentrations and up to 37.6% (for the FADS1 variant) of the variance in AA concentrations [11, 12].

Table 1. Characteristics and summary statistics for the SNPs employed as instrumental variables for plasma phospholipid levels of AA, EPA, and DHA ^a.

PUFA	SNP/rsID	Chr	Nearby gene	EA/OA	% VE	β	SE	P value
AA (20:4n-6)	rs174547b	11	FADS1	T/C	3.7-37.6c	1.69	0.02	3.0×10-971
AA (20:4n-6)	rs16966952	16	PDXDC1	G/A	0.1-0.6	0.20	0.03	2.4×10-10
EPA (20:5n-3)	rs174538b	11	FADS1	G/A	1.7	0.08	0.005	5.4×10-58
EPA (20:5n-3)	rs3798713d	6	ELOVL2	C/G	0.4	0.035	0.005	1.9×10-12
DHA (22:6n-3)	rs2236212d	6	ELOVL2	G/C	0.7	0.11	0.014	1.3×10-15

Chr, chromosome; EA, effect allele; OA, other allele; SE, standard error; VE, variance explained.

^aSNPs were identified in meta-analyses of genome-wide association studies in five population-based cohorts [11, 12].

^bRs174547 and rs174538 are in linkage disequilibrium (r²=0.86 in European populations and r²=1.0 in Japanese populations).

^cPercentage variance explained for Atherosclerosis Risk in Communities, Coronary Artery Risk Development in Young Adults, Cardiovascular Health Study, Invecchiare in Chianti, and Multi-Ethnic Study of Atherosclerosis was 37.6, 26.8, 37.5, 3.7, and 22.9, respectively.

^cRs3798713 and rs2236212 are in linkage disequilibrium (r²=0.97 in European populations and r²=1.0 in Japanese populations).

2.2. Outcome data sources

Association estimates for the PUFA-associated SNPs with cancer were obtained from the UK Biobank (n=367,643) and genome-wide association study consortia, including the FinnGen consortium (n=135,638) [13], International Lung Cancer Consortium (n=27,209) [14], Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome consortium (n=140,254) [15], Breast Cancer Association Consortium (n=228,951) [16], Ovarian Cancer Association Consortium (n=66,450) [17] and BioBank Japan (n=212,453) [18]. Participants were of European ancestry, except in BioBank Japan, which included Japanese participants. For UK Biobank, we estimated the SNP–cancer associations ourselves using logistic regression adjusted for age, sex, and the first ten genetic principal components. Our analyses of UK Biobank data and classification of different cancers have been described previously [19]. From the consortia, we acquired summary statistics data (i.e., beta coefficients and standard errors) for the SNP–cancer associations, which adjusted for age, sex, and genetic principal components in most studies [13–18]. Estimates for prostate cancer were derived from analyses of men only, and estimates for breast, endometrial, and ovarian cancer came from analyses of women only.

The UK Biobank and genome-wide association studies were approved by a relevant ethical review board, and participants had given informed consent. Our analyses based on UK Biobank and summary statistics consortia data were approved by the Swedish Ethical Review Authority.

2.3. Pleiotropy assessment

We searched PhenoScanner, a database of human genotype-phenotype associations [20], for potential pleiotropic associations of the AA-associated SNPs with risk factors for cancer. The SNP in FADS1 (rs174547) was associated with several fatty acids, though most strongly with AA, but not with recognized modifiable risk factors for cancer, including smoking, alcohol intake, physical activity, body mass index or other measures of obesity, or type 2 diabetes at the genome-wide significance threshold. The AA-increasing allele of rs16966952 in PDXDC1 was associated with higher dihomo-γ-linolenic acid and gamma-linolenic acid concentrations, lower linoleic acid concentrations, and higher whole body water mass, fat-free mass, height, weight, and basal metabolic rate at genome-wide significance.

2.4. Statistical analysis

Odds ratios (OR) of cancer were estimated using the Wald ratio and inverse-variance weighted methods. Fixed-effects meta-analysis was applied to combine the OR estimates from different studies and consortia, and heterogeneity between estimates was quantified with the I² statistic [21]. We scaled the OR estimates per standard deviation (SD) increase in plasma phospholipid concentrations (% of total fatty acids) using SD units (1.96%, 0.30%, and 0.89% for AA, EPA, and DHA, respectively) from a population-based cohort study [22]. To remove pleiotropic effects of AA on the association between plasma phospholipid EPA concentrations and cancer, we conducted a sensitivity analysis using only the ELOVL2 variant as instrumental variable for EPA. The Bonferroni method was applied to correct for multiple testing, and associations with P values <0.005 (P=0.05/10 association tests) were regarded statistically significant. All analyses were conducted using Stata/SE 14.2 (StataCorp, College Station, TX, USA).

3. Results

The associations of plasma phospholipid AA concentrations predicted by SNPs in FADS1 and PDXDC1 with 10 site-specific cancers are displayed in Fig. 2. Genetically predicted plasma phospholipid AA concentrations were statistically significantly positively associated with risk of colorectal and lung cancer. The combined ORs per one SD increase of plasma phospholipid AA concentrations were 1.08 (95% confidence interval [CI] 1.05-1.11, P=6.3×10^-8) for colorectal cancer and 1.07 (95% CI 1.05-1.10, P=3.5×10^-7) for lung cancer, with low or no heterogeneity among estimates from different data sources (I²=7.2% and 0%, respectively). Both SNPs used as instrumental variables for AA concentrations were positively associated with colorectal and lung cancer, albeit with low precision of the estimates for the PDXDC1 SNP (Supplementary Fig. 1). Genetically predicted plasma phospholipid AA concentrations were likewise positively associated with esophageal cancer, but the association was not statistically significant at the Bonferroni threshold (OR 1.09, 95% CI 1.02-1.17, P=0.016). There was no association between genetically predicted plasma phospholipid AA concentrations and risk of cancers of the stomach, pancreas, bladder, prostate, breast, uterus or ovary.

Fig. 2.

Associations of genetically predicted plasma phospholipid AA levels with site-specific cancers. SD, standard deviation. Data for bladder cancer were not available in BioBank Japan. BBJ, BioBank Japan; BCAC, Breast Cancer Association Consortium; CI, confidence interval; OR, odds ratio; ILCCO, International Lung Cancer Consortium; OCAC, Ovarian Cancer Association Consortium; PRACTICAL, Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome consortium; SD, standard deviation.

Plasma phospholipid EPA concentrations predicted by SNPs in FADS1 and ELOVL2 showed similar patterns of associations as AA (Supplementary Table 1). However, no association was observed when confining the genetic instrument to the SNP in ELOVL2 (OR 1.00, 95% CI 0.80-1.24, P=0.763 for colorectal cancer; OR 1.07, 95% CI 0.86-1.32, P=0.551 for lung cancer; and OR 0.95, 95% CI 0.55-1.62, P=0.839 for esophageal cancer). There was no association between plasma phospholipid DHA concentrations predicted by an SNP in ELOVL2 and any cancer (Supplementary Table 1).

4. Discussion

This MR study based on large-scale cohort and consortia data showed that higher genetically predicted plasma phospholipid concentrations of AA were associated with an increased risk of colorectal and lung cancer. There was also a suggestive positive association between AA concentrations and esophageal cancer. The associations were coherent across data sources and observed in both European and Asian populations. Genetically predicted plasma phospholipid EPA concentrations, driven by only the FADS1 variant, showed similar patterns of associations as AA, whereas no association was observed for plasma phospholipid DHA concentrations.

Our findings are in agreement with previous MR studies of genetically predicted AA concentrations and colorectal cancer risk in European populations [23–25]. With regard to lung cancer, a previous study showed that a variant in the FADS1 gene (rs174548) was statistically significantly associated with lung cancer risk in 13,821 Europeans and 18,471 Asians [26]. In addition, genetically predicted AA concentrations (per SD increase) were associated with an OR of lung cancer of 1.08 (95% CI 1.00-1.16) in UK Biobank (1863 lung cancer cases) [25]. Our results based on data from UK Biobank and the International Lung Cancer Consortium, FinnGen consortium, and BioBank Japan provide further support for an association between higher AA concentrations and increased risk of lung cancer. As in our study, no association was observed between genetically predicted AA concentrations and risk of breast, prostate, and pancreatic cancer in a previous MR study in UK Biobank participants [25]. We are not aware of any previous study of genetically predicted AA concentrations or FADS1 polymorphisms in relation to esophageal cancer, and MR studies of AA and other cancers are scarce.

Experimental evidence suggests that eicosanoids generated from EPA and DHA are anti-inflammatory and may exert anti-carcinogenic effects by suppressing AA-derived eicosanoid biosynthesis and possibly other direct mechanisms [1]. The observed positive association of genetically predicted plasma phospholipid EPA concentrations with risk of colorectal and lung cancer in this MR study may be driven by the concomitant increase in AA concentrations as delta-5 desaturase (encoded by FADS1) is involved in both EPA and AA biosynthesis. In support for this, no association was found between genetically predicted plasma phospholipid EPA and cancer risk when restricting the instrument to the variant in ELOVL2, which has no pleiotropic effect with plasma phospholipid AA concentrations. Randomized controlled trials of n-3 fatty acid supplementation (EPA and DHA) yielded negative results, but did show some non-significant, modest increases in risk [27, 28]. In the Vitamin D and Omega-3 Trial, the hazard ratio for cancer in the n-3 fatty acid group, compared to the placebo group, was 1.23 (95% CI 0.83-1.83) for colorectal cancer for the entire follow up period, and 1.13 (95% CI 1.00-1.28) for any cancer after excluding the first two years of follow-up [28]. Likewise, a meta-analysis of randomized controlled trials published prior to the Vitamin D and Omega-3 Trial showed a non-significant 10% increased cancer incidence in the n-3 fatty acid supplementation group compared with the placebo group (relative risk 1.10; 95% CI 0.97-1.24) [27].

An advantage of our MR study over previous MR studies on the topic is that we assessed the associations of plasma phospholipid AA concentrations with a broad range of cancers, including cancers that have not been previously studied in relation to AA concentrations, as well as the use of large-scale cohort and consortia data in different populations, comprising both European and Asian participants, thereby increasing statistical power. A strength our MR study compared with randomized trials is that genetically proxied plasma phospholipid AA concentrations represent lifelong higher AA exposure as genotype is fixed at conception.

A limitation of this MR study is that the major determinant of plasma phospholipid AA concentration is a variant in FADS1, which is also associated with other fatty acids in the delta-5 desaturase pathways. Furthermore, genetic associations with downstream mediators of AA metabolism are not available, but would provide mechanistic insight. Nonetheless, both variants used to proxy AA concentrations were positively associated with colorectal and lung cancer. This along with biological plausibility for carcinogenic effects of AA-derived eicosanoids [1, 2] lend support that higher AA levels are causally associated with an increased risk of colorectal and lung cancer. The FADS1 variant was not associated with other major risk factors for cancer, but the variant in PDXDC1 was associated with body composition measures, including higher whole body water mass, fat-free mass, height, and weight, as well as basal metabolic rate. However, results were similar when restricting the analysis to the FADS1 variant.

Another shortcoming is that the number of cases was relatively limited for some cancers, in particular for esophageal, stomach, and pancreatic cancer. The precision of the estimates was therefore low in analyses of those cancers and we cannot rule out any small effects. A further potential limitation is that the genetic variants used to proxy AA concentration were identified in populations of European ancestry and were used to investigate the association between genetically predicted AA concentrations and cancer risk in a Japanese population. Nonetheless, the major genetic variant related to plasma phospholipid AA concentration (rs174547) in European populations is in high linkage disequilibrium (r²>0.9) with another variant (rs174548) that has been shown to be strongly associated with AA concentration in an East Asian population [26]. Another possible limitation is that sex-specific instruments and effect estimates for AA were not available.

This study is also limited by the lack of information on histological type of esophageal cancer (squamous cell and adenocarcinoma). Several genes (e.g., cyclooxygenase-2) in AA metabolism have been shown to be dysregulated in esophageal squamous cell carcinogenesis [29]. The magnitude of the association between genetically predicted AA levels and esophageal cancer was marginally stronger in BioBank Japan (in which most cases are squamous cell carcinoma [30]) than in UK Biobank and FinnGen (larger proportion of adenocarcinoma than in Japanese populations). The incidence of some gastrointestinal tract cancers, particularly esophageal and stomach cancer, is much higher in Japanese than in European populations. Hence, the larger number of esophageal cancer cases in BioBank Japan may also explain the stronger association in this cohort. Likewise, the larger number of colorectal cancer cases in BioBank Japan than in UK Biobank and FinnGen may explain the stronger association between AA concentrations and colorectal cancer in the Japanese cohort.

Our findings suggest that reducing dietary AA intake may play a role in reducing colorectal and lung cancer incidence. Furthermore, inhibition of delta-5 desaturase activity and thus AA biosynthesis may represent a therapeutic strategy for lowering the incidence of these cancers, which is supported by the present MR investigation.

5. Conclusions

This MR study supports the hypothesis that AA may be implicated in the development of colorectal cancer and lung cancer and possibly esophageal cancer. There was no evidence that AA plays a role in other common cancers. Safe treatments with plasma AA-lowering properties should be evaluated for clinical benefit in preventing colorectal and lung cancer.

Abbreviations

AA

arachidonic acid

BBJ

BioBank Japan

BCAC

Breast Cancer Association Consortium

CI

confidence interval

DHA

docosahexaenoic acid

EPA

eicosapentaenoic acid

FADS1

fatty acid desaturase 1

ILCCO

International Lung Cancer Consortium

OCAC

Ovarian Cancer Association Consortium

OR

odds ratio

PRACTICAL

Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome

PUFA

polyunsaturated fatty acid

SNP

single nucleotide polymorphisms

Funding sources

SCL is supported by grants from the Swedish Research Council for Health, Working Life and Welfare, the Swedish Research Council, and the Swedish Heart-Lung Foundation. AMM is supported by EC-Innovative Medicines Initiative (BigData@Heart). SB is supported by a Sir Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal Society (204623/Z/16/Z). This work was supported by funding from the National Institute for Health Research (to S.B.) [Cambridge Biomedical Research Centre at the Cambridge University Hospitals NHS Foundation Trust] [*]. *The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.