Arachidonic acid and cancer risk: a systematic review of observational studies

Background An n-6 essential fatty acid, arachidonic acid (ARA) is converted into prostaglandin E2, which is involved in tumour extension. However, it is unclear whether dietary ARA intake leads to cancer in humans. We thus systematically evaluated available observational studies on the relationship between ARA exposure and the risk of colorectal, skin, breast, prostate, lung, and stomach cancers. Methods We searched the PubMed database for articles published up to May 17, 2010. 126 potentially relevant articles from the initial search and 49,670 bibliographies were scrutinised to identify eligible publications by using predefined inclusion criteria. A comprehensive literature search yielded 52 eligible articles, and their reporting quality and methodological quality was assessed. Information on the strength of the association between ARA exposure and cancer risk, the dose-response relationship, and methodological limitations was collected and evaluated with respect to consistency and study design. Results For colorectal, skin, breast, and prostate cancer, 17, 3, 18, and 16 studies, respectively, were identified. We could not obtain eligible reports for lung and stomach cancer. Studies used cohort (n = 4), nested case-control (n = 12), case-control (n = 26), and cross-sectional (n = 12) designs. The number of subjects (n = 15 - 88,795), ARA exposure assessment method (dietary intake or biomarker), cancer diagnosis and patient recruitment procedure (histological diagnosis, cancer registries, or self-reported information) varied among studies. The relationship between ARA exposure and colorectal cancer was inconsistent based on ARA exposure assessment methodology (dietary intake or biomarker). Conversely, there was no strong positive association or dose-response relationship for breast or prostate cancer. There were limited numbers of studies on skin cancer to draw any conclusions from the results. Conclusions The available epidemiologic evidence is weak because of the limited number of studies and their methodological limitations, but nonetheless, the results suggest that ARA exposure is not associated with increased breast and prostate cancer risk. Further evidence from well-designed observational studies is required to confirm or refute the association between ARA exposure and risk of cancer.


Background
Cancer remains an important health problem worldwide. It is estimated that 58.8 million people died of all causes in 2004 [1]. Deaths from cancer represented around one-eighth of these deaths, although many people who died had cancer even though it was not the direct cause of death. By 2030, it is projected that there will be approximately 26 million new cancer cases and 17 million cancer deaths per year [2]. Given these considerations, the prevention of cancer is a major public health issue around the world.
It is well established that dietary and other lifestyle factors play an important role in cancer control. In terms of dietary factors, earlier studies suggested a relationship between fat intake and the risk of several types of cancer. Prospective cohort studies found no association between fat intake and breast cancer, but a randomised trial organised within the Women's Health Initiative trial suggested a 9% reduction of borderline significance in breast cancer occurrence with decreased fat intake [3][4][5]. Analysis of the information in the Multiethnic Cohort Study found that intake of different types of fat indicated no association with overall prostate cancer risk or with non-localised or high-grade prostate cancer [6]. A prospective cohort study and a clinical trial failed to find evidence for an association between fat intake and colorectal cancer [7,8]. A dietary intervention study demonstrated that a reduction in fat intake reduces the risk of skin cancer [9,10], but the evidence from observational studies [11,12] has been controversial. Japan is a high-risk area for stomach and lung cancer, but no association with fat intake and these types of cancer has been suggested [2].
Essential fatty acids, namely n-3 and n-6 fatty acids, are involved in many important biological functions [13][14][15][16]. They play a structural role in cell membranes, influencing their fluidity and membrane enzyme activities; in addition, some are the precursors of prostaglandins and other lipid mediators. Arachidonic acid (ARA) is an n-6 essential fatty acid and also a major constituent of biomembranes. It is released from membranes by phospholipase A 2 and converted into various lipid mediators that exert many physiological actions [17][18][19]. Many studies have shown that lipid mediators derived from ARA, particularly prostaglandin E 2 (PGE 2 ), are associated with various diseases, which is mainly based on the fact that cyclooxygenase (COX) inhibitors are effective against those conditions [20][21][22][23][24]. PGE 2 is regarded as enhancing tumour extension as well, but it has been suggested that some other ARA mediators inhibit tumour growth [21][22][23][24][25]. In animal models, ARA administration did not affect tumour extension [26,27]. Some observational studies also suggested no relationship between ARA exposure and cancer risk [28,29]. However, there are the inconsistent observational studies that ARA exposure was positively correlated with the risk of colorectal cancer [30,31]. ARA is one of the major polyunsaturated fatty acid, and this inconsistency is not negligible.
No systematic review or meta-analysis has been conducted to evaluate the long-term effects of ARA intake and blood or tissue ARA composition on the risk of colorectal, skin, breast, prostate, lung, and stomach cancers in free-living populations. The objective of this study was to systematically evaluate available observational studies on the relationship between ARA intake and blood or tissue composition of ARA and the risk of these types of cancer.

Search strategy
The PubMed database (http://www.ncbi.nlm.nih.gov/ pubmed/) was searched for observational studies on the relationship between dietary or blood ARA levels with cancer risk that were published up to May 17, 2010. To identify target articles effectively, the strategy for the PubMed search was as follows: keywords for outcome and study types were adopted as commonly used terms representing cancer and study design, whereas terms for exposure were selected from specific words that stand for "arachidonic acid" (see Additional file 1). The initial PubMed search yielded 126 potentially relevant articles.

Study selection
Inclusion criteria were English articles that reported original data on the relationship between ARA exposure (intake or blood level) and target cancer risk in freeliving adults. Eligible study designs were cohort, casecontrol, or cross-sectional studies, and target types of cancer were colorectal, skin, breast, prostate, lung, or stomach cancer. Also included were studies investigating tissue ARA levels and target cancer risk. The study selection process is presented in Figure 1. We omitted reports in which titles or abstracts indicated that: (1) they were not human studies; (2) they were limited to special populations such as people with unusual eating habits; (3) they were intervention studies; or (4) they were not about the target cancers and fatty acids (not fat). We then evaluated the full text of the passed articles. Titles and abstracts of 126 identified publications from the PubMed database were checked and reviewed against the predefined inclusion criteria, and afterward, the full text of 52 articles were similarly assessed for eligibility by three authors (SK, CH, and HT, not independently). The 49,670 bibliographies of these full-text articles were scrutinised to identify additional eligible publications. One article on breast cancer was excluded because an inaccuracy of ARA assessment was clearly reported, although this article met the inclusion criteria described above [32]. Finally, 52 eligible articles were included in this review: 21 and 31 articles were obtained from primary PubMed searches and bibliographies, respectively.

Quality assessment and data extraction
Quality assessment was conducted based on the reporting quality and methodological quality of each study. The reporting quality shows whether the necessary information for observational studies is well indicated. It is the number of fulfilled items from the Strengthening the Reporting of Observational Studies in Epidemiology Statement (STROBE) checklist and varied 0 to 34 [33]. The reporting quality of included observational studies was assessed individually by two reviewers (CH and HT) and then confirmed by another two authors (SK and MS). The methodological quality, a level of suitability of methods used in a study, was assessed by two authors (SK and MS) qualitatively from the following methodological aspects used in the article: subject selection, ARA exposure assessment, diagnosis or recruitment procedure of cancer patients, methods for controlling confounders, and statistical analysis.
For each eligible article, the following information was tabulated: authors and year of publication, study settings and design, subject characteristics (such as age, sex, and number), matching strategy (if applicable), ARA exposure assessment used (as well as information about validity or precision), outcome assessment, adjusted confounders, reporting quality score from the STROBE checklist, and main findings from the fully adjusted model. Case-control studies were classified into two groups based on whether they reported temporal study settings information between exposure and outcome assessment: "case-control study (temporal relationship among exposure and outcome is demonstrated)" was defined as articles in which ARA exposure preceded the occurrence of cancer, whereas "case-control study (temporal relationship among exposure and outcome is unclear)" did not describe sufficient temporal information about exposure and outcome assessment.
Our qualitative definition of the study quality was as below: the reporting quality score under 13 or the insufficient temporal information, low; the other studies were qualitatively divided into high/medium/low according to their strength and weakness. A meta-analysis was not conducted because of the heterogeneity among studies, particularly subject characteristics and exposure/outcome assessment, and the insufficient number of studies with high methodological quality suitable for a metaanalysis. Therefore, qualitative assessment of ARA intake and cancer risk is presented in this review.

Results
For colorectal, skin, breast, and prostate cancer, 52 eligible articles were selected from potentially related reports and were included in the present systematic review ( Figure 1); the number of each was 17, 3, 18, and 16 studies, respectively. In contrast, we could not identify eligible reports for lung and stomach cancer.

Colorectal cancer
Major characteristics are shown in Table 1     and case-control studies were adjusted for well-known potential confounders, such as family history, body weight and smoking, and specific factors for colorectal cancer, such as body mass index (BMI), physical activity, alcohol drinking and total energy. No confounding factors were adjusted for in eight articles.
Dietary ARA intake was estimated in two cohort studies and four case-control studies. Median dietary ARA intake ranged widely from 0.008 to 0.15 g/day, or from 0.04% to 0.07% of energy. Two articles reported a significant increase in colorectal cancer risk. Muff et al. indicated that colorectal cancer risk was significantly increased in the third and fourth quintiles of ARA intake, and that the overall trend was significant (P for trend = 0.03). Nkondjock et al. reported significantly increased colorectal cancer risk in the third and fourth quartiles and significance in the overall trend (P for trend = 0.001).
In seven case-control studies and three cross-sectional studies, the exposure was indicated as blood ARA levels. The precision of blood analysis was mentioned in only four reports, and blinded fatty acid measurement was conducted in only three reports. Five articles showed a significant trend of decreasing colorectal cancer risk or a significant difference of blood ARA levels in cancer subjects. Kuriki et al. found that colorectal cancer risk was significantly decreased in the highest tertile of erythrocyte ARA levels, and that the overall trend was significant (P for trend < 0.05). The remaining four reports, Ghadimi et al., Hietanen et al., Neoptolemos et al. (1988), and Almending et al., were a case-control study with little temporal information between exposure and outcome or a cross-sectional study.
One case-control study with little temporal information between exposure and outcome and two crosssectional studies investigated tissue ARA levels. The precision of tissue analysis was mentioned in only one article, and none reported masking of disease status. Their reporting quality was generally low.

Skin cancer
Only three articles were included in the present systematic review. Major characteristics are shown in Table 2 [48][49][50]. Their exposure assessment and subjects' characteristics were too diverse to be compared to each other.
Dietary ARA intake was estimated in one cohort study and three case-control studies. These four showed no significant change in breast cancer risk except in the second quartile of ARA intake in the report by Nkondjock et al.
Six case-control studies and three cross-sectional studies investigated blood ARA levels. The precision of blood analysis was reported in only five articles, and blinded fatty acid measurement was conducted in only two articles. Three articles indicated significant differences in breast cancer risk; however, they were a casecontrol study with little temporal information between exposure and outcome or a cross-sectional study. Five case-control studies and two cross-sectional studies examined tissue ARA levels. The precision of tissue analysis was mentioned in only three articles, and only in one report fatty acids measurement was performed in a blinded fashion. A significant change in breast cancer risk or a significant difference in tissue ARA level was not found, except for breast tissue triglyceride ARA levels in a report by Zhu et al. and breast tissue phosphatidylcholine ARA levels in a report by Williams et al.
One cohort study and three case-control studies examined dietary ARA intake. They showed no significant change in prostate cancer risk according to increased ARA intake.
Blood ARA levels were estimated in nine case-control studies and three cross-sectional studies. The precision of blood analysis was mentioned in only five articles, and masking of disease status was conducted in only four. Ukori et al. (2010) reported that prostate cancer risk of African-Americans decreased in the fourth quartile of blood ARA level, and that the overall trend was significant (P for trend < 0.05). A significant change in prostate cancer risk or a significant difference in blood ARA levels was not found in the other 11 articles.
Three cross-sectional studies examined tissue ARA levels. All of them reported significant decreases of tissue ARA levels in cancer subjects; however, their reporting quality was generally quite low. None of them mentioned the precision of tissue analysis and masking of groups.

Discussion
In the present review, we systematically reviewed observational studies investigating the association between ARA and cancer of six organs in free-living populations. Fifty-two eligible articles were obtained from our search strategy, and 31 out of the 52 articles were identified from hand searches for references ( Figure 1). Thus, reference searching serves an important role in comprehensive literature searches. This pointed out the characteristics of the reporting style of the observational studies for ARA and cancer risk.
Among the 31 eligible articles from reference searches, 22 were not recognised by our PubMed search formula due to keywords related to "exposure", three were not recognised due to keywords related to "study types", and six were not recognised due to both. For "exposure" terms, 26 articles could be identified by the addition of the search term "fatty". The remaining two articles related to the term "exposure" reported fatty acid compositions of tissues only. In the case of "study type" terms, none of the nine articles used a general study design word (i.e., cohort, case-control, or cross-sectional), although the STROBE statement recommends that authors should indicate the study design with a commonly used term in the title or abstract. These reporting characteristics made it difficult to effectively search for observational studies with a focus on individual fatty acids such as ARA. We therefore adopted the search strategy described above.
The findings from articles for colorectal cancer differ depending on the methodology of ARA exposure assessment. A positive dose-response relationship between dietary ARA intake and colorectal cancer was indicated      in two reports [30,31], whereas four articles [38,40,43,46] indicated a negative association or significant ARA decrease with blood ARA levels, and no article reported a positive relationship between colorectal cancer risk and tissue ARA level. These inconsistent results seem to indicate that there is little firm evidence that ARA correlates with the risk of cancer.
There were limited numbers of studies on skin cancer, and they varied in the assessment method used for ARA exposure and the target cancer. It is therefore impossible to draw any conclusions from the results.
Among studies for breast and prostate cancer, a strong positive association and a clear dose-response relationship between increased cancer risk and ARA exposure were not observed, although the results were replicated in different settings using different methods. This suggests that ARA exposure is not associated with increased breast and prostate cancer risk.
We suppose that the contradictory findings mentioned above were caused by four main factors. First, methodologies for estimating dietary ARA intake have not been developed sufficiently. Most adults with mixed diets consume approximately 50 to 250 mg of ARA per day from foodstuffs [82][83][84], whereas some articles on colorectal and prostate cancer have reported lower values [30,36,39,68,74]. Various validated questionnaires were used in articles which assessed dietary ARA intake, but the validation was not conducted for ARA specifically; total fat, saturated fatty acids, monounsaturated fatty acids, polyunsaturated fatty acids, linoleic acid, or eicosapentaenoic acid intake was assessed, but ARA was not. Since the validity of the estimation of dietary ARA intake is sometimes not enough [82], it should be considered whether exposure assessment is conducted with an appropriate method.
Second, assessment of ARA biomarkers such as blood or tissue ARA levels was diverse; assessed blood fractions included erythrocyte, serum, plasma, or total blood. Tissue sampling was conducted from adipose tissue of buttock or malignant target cancer tissue (i.e., colon, skin, breast, and prostate). Individual biomarkers of fatty acids represent intakes for different time periods [85]. Serum or plasma levels of ARA are considered to reflect dietary intake over a few days, whereas erythrocyte and tissue ARA composition serve as more long-term biomarkers. Overall, habitual dietary ARA intake could not be assessed sufficiently in articles that measured ARA composition of serum or plasma, and this might be one of the causes of inconsistent results among eligible articles.
Third, methodologies of participant selection used in the reports could lead to selection bias. A nested case-control study from a physicians' health study (aspirin and betacarotene intervention study) did not consider trial intervention through the participant selection procedure [73]. Also, reporting of participant selection was important; among the total of 52 articles, 28 did not sufficiently report the eligibility criteria, and the sources and methods of participant selection, and 31 did not fully describe the numbers of individuals at each stage of the study and reasons for nonparticipation. This makes it difficult to estimate the effect of selection bias, and therefore, the relationship between ARA intake and cancer risk could not be determined.
Fourth, publication bias based on findings of a significant association could exist, especially in breast and prostate cancers. We evaluated publication bias qualitatively, not using any statistical tests. Most of the significant results were found in the studies with low reporting quality. There is a possibility of publication bias. The results of the studies with low reporting quality may tend to be significant by chance, due to the lack of appropriate design. This suggests that publication bias may affect our review result on breast and prostate cancers, but the effect should be small, because we did not give importance to these studies with low reporting quality.
The biological plausibility of the relationship between ARA intake and cancer risk is still being debated. Previous clinical studies with aspirin or nonsteroidal antiinflammatory drugs (NSAIDs) have suggested that the cyclooxygenase metabolites of ARA may be associated with risk for colorectal, breast, and prostate cancers [21][22][23][24]. Many observational studies, however, have failed to find any association between ARA intake, or its level in blood or tissue, and cancer risk. These controversial findings may be explained by the following three reasons. First, ARA levels of blood or tissue may not always represent dietary intake. Garland et al. and Kobayashi et al. reported that correlations between dietary estimates and the ARA contents of adipose tissue or serum phospholipids were very low [82,86]. Second, the increment of blood or tissue ARA levels may not be connected with the amount of ARA metabolites. Our previous study, in which the supplementation of 240 or 720 mg of ARA per day in healthy elderly persons for four weeks was conducted, investigated plasma PGE 2 and urinary PGE 2 metabolites [87]. Their concentrations did not differ significantly with regard to ARA supplementation or time points, although plasma ARA compositions increased dose-dependently. Third, ARA metabolites that are produced by pathways other than the cyclooxygenase pathway may decrease cancer risk. LXA 4 is an anti-inflammatory mediator produced by the lipoxygenase pathway and is regarded to be a suppressor of tumour growth based on anti-angiogenic properties [25,88]. Aspirin or NSAIDs may not only inhibit the production of cyclooxygenase metabolites, but also divert ARA into the lipoxygenase pathway. However, it is unclear whether LXA 4 contributes to reduced cancer risk in humans.
In the present study, we reviewed all bibliographies of full-text articles for potential inclusion because reference      searches serve an important role in comprehensive literature searches. 49,670 articles were listed, and 99.9% of them were not eligible. We considered that this large exclusion resulted from the many articles in which ARA was not described at all; therefore, we tried another reference search from the bibliographies of articles after their exclusion. Fifty full-text articles from the PubMed database and 146 from the reference search mentioned ARA. A total of 13,657 articles were listed in their bibliographies and their number was reduced to a third; however, we could select 30 eligible articles out of 31 articles that were selected from all full-text searching. The one article remaining that was not selected was a report on skin cancer [50]. This might have resulted from the smaller number of articles identified on PubMed for skin cancer (10 reports) than for colorectal, breast, or prostate cancers (48, 31, or 41 reports). This suggests that reference searches from bibliographies of articles including ARA are more efficient when enough articles are identified from the PubMed database. This systematic review has limitations. First, studies for inclusion could not be selected independently by two or more reviewers. Our inclusion/exclusion criteria were clear and there were few differences which depended on who was in charge; however that may have introduced a potential selection bias. Second, our search was restricted to English publications and articles from the PubMed database. Furthermore, articles that investigated tissue levels of ARA as an exposure assessment could not be identified comprehensively. We did not set the search terms for ARA levels of tissue before the PubMed search, and identified the articles in the reference search. Third, the search term "fatty" or "fatty acid" was not used in the PubMed search. It led to the efficient search but may cause the possibility that the review may not be completed. Fourth, quality assessment of observational studies is difficult because of the heterogeneity of study designs and methods. The reporting quality was quantitatively expressible using the STROBE checklist; in contrast, the methodological quality could not be quantified and was qualitatively estimated by two independent reviewers. This may have seriously influenced the results and conclusions of the present review.
Note that there are insufficient studies to draw any firm conclusions about the relationship between ARA and cancer risk. Further evidence from well-designed observational studies is required.

Conclusions
In conclusion, we systematically identified articles that investigated the association between dietary ARA intake or its biomarkers and the risk of colorectal, skin, breast, prostate, lung, or stomach cancer, and only a limited number of observational studies were found (17,3,18,16,0, and 0 studies were found on colorectal, skin, breast, prostate, lung, or stomach cancer, respectively). Furthermore, most studies had one or more critical limitations, such as the obscurity of temporal information about exposure and outcome, the methodology of ARA exposure assessment, and inadequate treatment of potential confounding factors. These studies did not sufficiently demonstrate any relationships between ARA exposure and cancer risk; however, they seem to suggest that ARA exposure was not related to increased breast or prostate cancer risk because strong positive associations and clear dose-response relationships were not observed. Findings concerning the association between ARA exposure and colorectal cancer were inconsistent between studies. Thus, further evidence from well-designed observational studies is required to confirm or refute the association between ARA exposure and cancer risk.

Additional file
Additional file 1: PubMed search terms and strategies.