Wood cookstove use is associated with gastric cancer in Central America and mediated by host genetics

Biomass cookstove food preparation is linked to aero-digestive cancers, mediated by ingested and inhaled carcinogens (e.g., heterocyclic amines, and polycyclic aromatic hydrocarbons). We investigated the association between gastric adenocarcinoma, wood cookstove use, H. pylori CagA infection and risk modification by variants in genes that metabolize and affect the internal dose of carcinogens. We conducted a population-based, case–control study (814 incident cases, 1049 controls) in rural Honduras, a high-incidence region with a homogeneous diet and endemic H. pylori infection, primarily with the high-risk CagA genotype. We investigated factors including wood cookstove use, H. pylori CagA serostatus, and 15 variants from 7 metabolizing genes, and the interactions between wood stove use and the genetic variants. Male sex (OR 2.0, 1.6–2.6), age (OR 1.04, 1.03–1.05), wood cookstove use (OR 2.3, 1.6–3.3), and CagA serostatus (OR 3.5, 2.4–5.1) and two SNPs in CYP1B1 (rs1800440 and rs1056836) were independently associated with gastric cancer in multivariate analysis. In the final multivariate model, a highly significant interaction (OR 3.1, 1.2–7.8) was noted between wood cookstove use and the rs1800440 metabolizing genotype, highlighting an important gene-environment interaction. Lifetime wood cookstove use associates with gastric cancer risk in the high-incidence regions of Central America, and the association is dependent on the rs1800440 genotype in CYP1B1. H. pylori CagA infection, wood cookstove use and the rs1800440 genotype, all of which are highly prevalent, informs who is at greatest risk from biomass cookstove use.


Study design
We conducted a population-based, case-control study based in the mountainous regions of Honduras.This region has among the highest incidence rates in the western hemisphere, with a high prevalence of chronic H. pylori infection (80-90%) [25][26][27][28][29] .Incident GC cases were enrolled prospectively from the two district hospitals (Santa Rosa de Copán and Siguatepeque) that serve the mountainous rural areas of west-central Honduras.The diagnosis of GC was based on endoscopic appearance and confirmatory histopathology.Household interviews were conducted for randomly selected healthy controls from a wide distribution of villages in the region and the catchment area, as described previously 25,28 .Once consented, patients and control subjects would undergo the study interview with review of exclusionary health criteria, as well as the demographic, health assessment, and epidemiology questionnaires.Lifetime woodstove use was dichotomized (yes, no) since we observed minimal lifetime variation in usage nor transition to other types of cook stoves in the Honduras rural populations.

H. pylori infection and CagA assessment
A validated multiplex serology was used to determine H. pylori and CagA serostatus, the dominant bacterial risk genotype for GC [30][31][32] .We focused on CagA, the principal oncoprotein, as the region has an extremely high, H. pylori prevalence.The multiplex serology panel was developed by the Germany Cancer Research Center ("DKFZ").In brief, the H. pylori proteins were recombinantly expressed as Glutathione-S-transferase (GST)-tag fusion proteins in Escherichia coli BL21 and affinity-purified on glutathione-coated fluorescently labeled polystyrene beads (Luminex Corp.).A mixture of the differently labeled and antigen-loaded beads was incubated with serum to allow binding of serum antibodies to the H. pylori proteins.Bound serum antibodies were detected by a biotin-labeled anti-human IgM/IgA/IgG secondary antibody and Streptavidin-R-phycoerythrin.The Luminex 200 analyzer (Luminex Corp.) distinguished between the bead type and the bound antigen and quantified the amount of bound serum antibody as median fluorescence intensity (MFI) of 100 beads per type measured.The CagA antigen-specific cutoff was used 30 .

Genotype analysis
Human DNA was isolated from whole blood samples with the Qiagen Puregene® kit and genotyped on the Mas-sARRAY® Sequenom platform at Vanderbilt University Medical Center (VANTAGE Core facility).We tested for Hardy-Weinberg equilibrium and removed SNPs that deviated in cases and controls at p < 1.0 × 10 -6 .We also removed SNPs that had a minor allele frequency (MAF) < 0.05 using PLINK (version 1.9) [33][34][35] .We tested for Linkage Disequilibrium in PLINK and Haploview (version 4.2) regardless of case-control designation to determine the number of independent tests for the False Discovery Rate (FDR) threshold and limit multiple testing 36 .

Statistical analysis
We compared differences between case and control subjects using t-tests for the continuous variable age or the chi-square test for the categorical variables sex, wood stove use, and bacterial CagA serostatus.Age was tested for normality using the Shapiro-Wilk's method with the R "stats" package (R version 4.0.4).Age was negatively skewed as GC is the final step of progressive gastric disease, and the population was predominately older 37 .Therefore, age was log transformed, and the analyses were run with the transformed variable, but the deviation from a normal distribution did not impact the results substantively.The untransformed age variable is presented in all the results.
Univariate logistic regression models were used to estimate risk of GC associated with individual variables, including all SNPs.Logistic regression was used to estimate odds ratios (OR) for adjusted models that included combinations of the variables age, sex, CagA serostatus, wood stove use, and each SNP individually.Daily tobacco use was very low (20 pack-year history, 4.8%), and was not included in the model.For each SNP, we also assessed multivariable models for individuals without missing data that included an interaction term with wood stove use.A likelihood ratio test using R package "lmtest" was used to assess the goodness of fit of competing statistical models, i.e., those without the interaction and those with the interaction (R version 4.0.4).The R command "p.adjust" was used to calculate the q-values using the Benjamini-Hochberg test with a False Discovery Rate (FDR) level of 0.1 for all tests (R version 4.0.4).We confirm that all research was performed in accordance with relevant guidelines/regulations and informed consent was obtained from all participants and/or their legal guardians.Research involving was performed in accordance with the Declaration of Helsinki.

Ethics committee approvals
The study was approved by the institutional review boards of The University of Alabama at Birmingham, Vanderbilt University and the Ministry of Health in western Honduras.

Consent to participate
Informed consent was obtained from all individual participants included in the study.

Results
A total of 814 gastric cancer patients and 1049 population controls were enrolled, of whom genetic and H. pylori data were available for 1,425 participants (Table 1).The median ages of the cases and controls were 65 and 54, respectively.Approximately one-fifth of cases were under the age of 55 (n = 170 of 814).Overall, 87% and 85% of subjects were positive for H. pylori infection and CagA, respectively.Wood stove use was observed in 92% of the gastric cancer cases and in 78% of the population-based controls.Gastric cancer cases were more likely ), wood stove use (OR = 2.57, p = 5.52 × 10 -8 ), and bacterial CagA serostatus (OR = 3.51, p = 4.20 × 10 -13 ) all associated with an increased risk of gastric cancer (Table 2).In a multivariate model including age, sex, wood stove use and bacterial CagA serostatus, these variables all remained significantly associated with gastric cancer.Of the 15 SNPs genotyped, none were out of Hardy-Weinberg equilibrium (p < 1.0 × 10 -6 for both cases and controls), but 2 SNPs had a minor allele frequency of less than 0.05 and were excluded from further analyses (Table S1).Allele frequencies of the SNPs were within the expected ranges, based on the Latin American populations in the 1000 Genomes databases (Version 3).Linkage Disequilibrium (L.D.) was tested for the 6 SNPs in the NAT1 gene that met inclusion criteria.Three pairs showed evidence of significant correlation (r 2 > 0.4), and we randomly removed one SNP from each pair to reduce the multiple testing burden (Fig. S1).The data cleaning resulted in 10 SNPs being tested in the final analysis.Genotype counts are presented in Table S3.
Initial unadjusted association analyses identified two SNPs that associated with GC under a dominant model where the minor allele was used as the referent.The two SNPs, rs1800440 and rs1056836, in cytochrome P450 family 1 subfamily B member 1 (CYP1B1), were both significantly associated with GC after FDR adjustment for multiple testing (adjusted q = 0.07 and q = 0.003, respectively) and were not in L.D. (R 2 = 0.03) (Table 3).In models adjusted for age, sex, bacterial CagA serostatus, wood stove use, and each SNP singly, both the CYP1B1 SNP rs1800440 (adjusted q = 0.023) and the CYP1B1 SNP rs1056836 (adjusted q = 0.023) remained significant (Table 3).Additive and recessive genetic models were also run but no SNPs were significant after FDR correction in the univariate analyses (Table S4).
We used three statistical models of increasing complexity to assess patterns of association (Table S5).The first multivariable model included the core variables age, sex, wood stove use, bacterial CagA serostatus, and the two CYP1B1 SNPs.All variables were significant (Table 4).Multivariable model 2 included the core variables, rs1800440, and the interaction of wood stove use and rs1800440.All variables remained significant except for wood stove use and rs1800440 (Table 4).Multivariable model 3 included the same variables as multivariable model 2 and also included SNP rs1056836 with all variables remaining significant except for wood stove use and www.nature.com/scientificreports/rs1800440 (Table 4).In both models 2 and 3, the interaction term was statistically significant.The likelihood ratio test identified model 3 with the interaction of wood stove use and rs1800440 as the best model (p = 0.025), and it was used for interpretation.Remarkably, the odds ratio of the interaction term in this model was of similar magnitude as the highly associated bacterial CagA serostatus.
Our results showed that genotype did not associate with GC risk for individuals who did not use a wood stove (p = 0.25, OR = 0.59).Nor did wood stove use associate with risk for individuals with GG/GA genotypes (p = 0.80, OR = 0.90).However, wood stove use was associated with GC risk in AA individuals compared to AA individuals who did not use a wood stove (p < 0.0001, OR = 2.73).Finally, for individuals who used a wood stove, AA genotype individuals were at increased risk (p = 0.0007, OR = 1.83) whereas in the absence of wood stove use the same genotype did not associate with GC (Fig. 1; Table 5).Similarly, wood stove use was not associated with GC for people who were GG/GA (Table S6).Multivariable models for rs105836, including the interaction of rs1056836 and wood stove showed no significant interaction between this SNP and wood stove use (Table S7).None of the remaining eight SNPs showed any interactions with woodstove use (Table S8).

Discussion
In the rural Central America Four region, we found a significant association between wood cookstove use and gastric adenocarcinoma.Wood stove use increased the risk of GC by more than 2.3 times after adjusting for demographic and risk factors (age, sex, CagA serostatus).This association was independent of the CagA genotype, the principal H. pylori virulence and GC risk factor.Importantly, the wood stove use association was significantly modified by the rs1800440 variant of CYP1B1, a gene involved in HCA and PAH metabolism, a determinant of the "internal dose".This SNP was only significant in the context of wood cookstove use, and neither wood stove use nor the SNP were significant when the interaction of this SNP and wood stove use was included in the model, implying strong biological mediation by genotype.Specifically, the rs1800440 AA genotype was high risk when wood stoves were used, but not otherwise, and wood stoves were only significant in the presence of this genotype.These two risk factors are highly prevalent in Central American populations.
The gene-environment interaction between wood stove use and genetics greatly informs our understanding with respect to who is most susceptible to adverse effects of cooking with a wood stove.Individuals who use a wood stove are at an increased risk of developing gastric cancer when they are homozygous for the major rs1800440 allele, whereas they do not appear to be at increased risk in the presence of a minor allele.Similarly, those who do not use a wood stove do not have a significant association based on genotype.This finding supports the concept that elevated risk associated with wood stove use may be mediated by HCA and PAH exposure generated in food preparation with wood stove use and is consistent with the recognized link between meat consumption and gastric and colon cancers.Latin American populations have a high prevalence of the rs1800440 risk allele frequency (86%-88%), as do African and Asian populations with 100% and 99% prevalence, respectively 38 .
Studies examining the rs1800440 CYP1B1 polymorphism and cancer risk have been inconclusive.The interaction of exposures, such as wood stove use and the variant, may explain the discordant literature related to CYP1B1 genotypes [39][40][41][42][43][44][45][46][47][48][49][50] .Our study found that the common variant of the phase I CYP1B1, rs1800440, conferred approximately 40% increased risk of gastric cancer even after adjusting for age, sex, CagA status and wood stove use in analyses without a SNP-wood stove interaction term.However, we also noted that the rs1800440 SNP is only significant in the presence of wood stove exposure, a potential explanation of why some prior studies did not find any association between CYP1B1 rs1800440 and gastric cancer 39 .In colorectal cancer studies, a meta-analysis found no association between colorectal cancer and rs1800440; however, there was considerable heterogeneity among the studies 36 .Carriers of the minor allele have been shown to confer either decreased risk of colorectal cancer, increased risk, or no risk in different studies 41,42 .Interestingly, in a Chinese population, the rs1800440 genotype AG associated with decreased hepatocellular carcinoma risk compared to the AA (homozygous major allele) genotype (adjusted OR = 0.33) when adjusting for similar variables to our study as well as smoking 51 .Importantly, the rs1800440 CYP1B1 minor allele (G) encodes an amino acid substitution Asn453Ser with Ser being likely deleterious with respect to protein function according to Polyphen-2 (score = 0.906) (Fig. S2) 52 .This is consistent with a functional allele increasing internal doses of carcinogens, as in the case of the AA genotypes associating with risk in wood stove users.Our analyses showing a significant interaction between genotype and wood stove use may explain the inconsistencies as prior evidence also indicated that this phase I enzyme minor allele polymorphism may degrade more rapidly than the wild type enzyme and have a drastically shorter halflife, and therefore reduced metabolic activation of estrogens, PAHs, and HCAs 48 .
Limited studies have examined the relationship between wood cookstove use and cancer, and specifically gastric cancer.Compounds generated by wood stoves are modified by metabolizing enzymes, and therefore an interaction between genes for these enzymes is not unexpected.The limited literature indicates that exposure to biomass burning and wood stove cooking has been linked to a variety of cancers of the upper aero-digestive tract, including lung, gastric and esophageal cancer 24,[53][54][55][56][57][58][59][60][61][62] .The use of wood stoves, diets low in fruits and vegetables,  5).www.nature.com/scientificreports/and higher elevation households have all been noted as significant risk factors in an epidemiologic profile of GC in Peru 24 and the association between cancers of the upper aero-digestive tract was confirmed in southern Brazil after adjusting for tobacco, alcohol, or dietary factors 53 .In addition, the use of biomass fuels was linked to gastrointestinal cancers, including esophageal cancer and gastric cancer, in a large Iranian cohort of men and women 62 .However, how these factors interact with enzymes that activate and deactivate carcinogens has not to our knowledge been studied previously.Regular wood stove use also contribute to gastric cancer risk through the generation of PAHs.An association has been detected between biomass smoke exposure and gastric cancer, as well as biomass smoke exposure and urinary 1-hydroxypyrene (1-OHP), a PAH metabolite, with a significant trend for dose in a study from Zambia.However, no association was detected between the metabolite and gastric cancer status 61 .Notably, the exposures in sub-Saharan Africa may be less than Latin America as the cook stoves are typically outdoors and often involve the boiling of meats and foods.A recent study of Shanghai women found that women diagnosed with gastric cancer had higher urinary levels of 1-hydroxypyrene glucuronide (1-OHPG), a closely related PAH metabolite 63 .
There is considerable potential for mitigation with improved cookstove design.The Global Alliance for Clean Cook stoves has proposed deployment of 100 million improved stoves 64 .Various designs and interventions have been implemented for the past two decades in order to decrease smoke exposure and deforestation, yet exposure to organic carcinogens in well-done meat and foods cooked at high temperature remain a challenge.Populations with a high prevalence of the risk rs1800440 genotype may benefit most from reduced wood stove use or a changes in food preparation.
The strengths of our study include the regional approach in a rural LMIC population with a high gastric cancer incidence, endemic H. pylori cagA infection, and a homogeneous diet.Not having precise quantification of HCA and PAH exposures was one of the study limitations.We did not examine dietary information and meat intake to further quantify HCA and PAH exposure, but this potential influence is dampened by the generally uniform diet in this rural region over time.Lastly, our asymptomatic population-based controls did not undergo endoscopy, and a limited number of control subjects may have had precancerous conditions or early gastric cancer, but this would only decrease the power to detect associations.

Conclusions
Wood cookstove use contributes to gastric cancer risk in the high-incidence regions of mountainous Central America, where H. pylori cagA infection is endemic.We found that the effect is mediated by host genetics, specifically the CYP1B1 genotype, wherein wood cookstove usage increases risk only in individuals genetically predisposed, to thereby increase the conversion of pro-carcinogenic compounds to carcinogens.Further studies are indicated, as our results underscore the broad health and environmental impact of biomass cookstove use.

Figure 1 .
Figure 1.Interaction between wood stove use and the rs1800440 CYP1B1 SNP.The effect of the rs1800440 AA genotype is dependent on wood stove use.Those that have the rs1800440 AA genotype and use a wood stove have an increased risk of gastric cancer compared to those with the rs1800440 GG or GA genotype and use a wood stove (OR = 1.83; 1.29, 2.60); p = 0.0007).The results derived from Model 3 (Table5).

Table 1 .
The characteristics of the gastric cancer cases and population controls.
a There were 814 gastric cancer patients and 1,049 population controls enrolled; Complete genotyping and H. pylori serology data were available for 1,425 participants.Characteristics

Table 2 .
Univariate and multivariate odds ratios for gastric cancer outcomes.95% CI 95% confidence interval.a Multivariate model includes age, sex, wood stove use, and CagA positive serostatus.

Table 5 .
Associations of multivariable models with gastric cancer stratified by wood stove use.