Non-Hodgkin Lymphoma, Body Mass Index, and Cytokine Polymorphisms: A Pooled Analysis from the InterLymph Consortium

Immune dysregulation plays a pivotal role in lymphomagenesis, and epidemiologic research has tended to concentrate on factors and exposures that interact with the immune system. In this regard, obesity, which can cause a mild chronic inflammatory state, has been suggested to potentially increase the likelihood of lymphoid malignancy development. Earlier, InterLymph pooled analyses reported that obesity was associated with an increased risk of diffuse large B-cell lymphoma (DLBCL; refs. 1, 2), and recent meta-analyses provide further support for this hypothesis (3, 4).

Obesity-related inflammation is thought to result from the proinflammatory cytokines and chemokines that are produced by adipocytes and macrophages in adipose tissue (5). With weight gain, the numbers of adipocytes and macrophages increase as adipose tissue expands, increasing production of cytokines such as TNFα, leptin, IL1β, and IL6, as well as chemokines and acute phase proteins (5). Ideally, to investigate whether increased levels of inflammation-related cytokines modulate the association between obesity and lymphoid neoplasms, serum levels of cytokines would be measured before cancer diagnosis. In the absence of such measurements, single nucleotide polymorphisms (SNP) within genes that express cytokines may act as surrogates that indicate variation in risk of lymphoid neoplasms with obesity. Several putative functional SNPs in candidate cytokine genes were selected a priori by the InterLymph consortium due to their role in lymphoid development, and also in the pro-/anti-inflammatory pathways which may be altered in the obese state. Among these cytokine SNPs, TNF (−308G>A, rs1800629), LTA (−252A>G, rs909253), and IL10 (−3575T>A, rs1800890) have been associated with lymphoid neoplasms and DLBCL in particular (6–8). There has, however, been little exploration of the relationship between obesity and cytokines on the risk of these malignancies (9–11). Here, we investigate gene–environment interactions between body mass index (BMI) and cytokine SNPs using data from case–control studies included in the International Lymphoma Epidemiology Consortium.

Through the InterLymph consortium, nine case–control studies conducted in the United States and five European countries between 1988 and 2008 that had individual level data on BMI and cytokine polymorphisms contributed to this pooled analysis. Data were provided via the InterLymph Data Coordinating Center (DCC) at the Mayo Clinic (Rochester, MN) which was established in 2009 to centrally standardize and harmonize study data so that consistent datasets could be produced to expedite pooling projects. Descriptions of the included studies have been published (12–20); a brief outline is given in Table 1. Cases were ascertained using rapid identification techniques and controls were randomly selected from population registers (6 studies), outpatient clinics (1 study) or hospital in-patients with non-neoplastic conditions (2 studies). Each study had the appropriate ethical committees' approval and participants gave their informed consent.

Diagnoses of lymphoid neoplasms were pathologically confirmed and coded to the World Health Organization International Classification for Oncology Version 3 (ICDO3; 7 studies), REAL (Connecticut), and Working Formulation (UCSF) classifications. Diagnostic codes from the different schemas were bridged by the DCC (Mayo Clinic, Rochester, MN) using the same approach as in previous InterLymph analyses (21). The analysis here reports on specific lymphoid neoplasms: diffuse large B-cell lymphoma (DLBCL: ICDO3 codes 9679, 9680, 9684), follicular lymphoma (FL: 9690, 9691, 9695, 9698), and chronic lymphocytic leukemia/small lymphocytic lymphoma (CLL/SLL: 9670, 9823) and all combined (defined by the above ICDO3 codes and 9671, 9673, 9675, 9687, 9689, 9699, 9700, 9701, 9702, 9705, 9708, 9709, 9714, 9716, 9717, 9718, 9719, 9728, 9729, 9826, 9827, 9832, 9833, 9591, and 9727). As most studies did not recruit cases with HIV-associated lymphoid neoplasms, these diagnoses were excluded.

Findings for the individual effects of BMI and cytokine SNPs on the risk of lymphoid neoplasms have been reported for the InterLymph studies (1, 6–8). In all studies, adult height and weight were self-reported, with information on weight requested for 1 year (NCI-SEER, UCSF, Connecticut), 2 years (Mayo Clinic), or 5 years (United Kingdom) before diagnosis or interview date or usual weight (SCALE, EpiLymph). BMI, calculated from weight in kilograms and height in meters, was classified according to World Health Organization guidelines as: normal weight (18.5–<25 kg/m²); overweight (25–<30 kg/m²); or obese (≥30 kg/m²); the 1% of the study population who were underweight (<18.5 kg/m²) were excluded from the analyses (22). BMI as a continuous variable was defined as per 5 kg/m² increase above 18.5 kg/m². Cytokine SNPs were tested using the TaqMan platform (Applied Biosystems), Pyrosequencing, custom Illumina GoldenGate 1,536 SNP oligonucleotide pool (OPA), or iSelect (6–8, 14). Twelve SNPs in nine candidate genes were investigated: 2q14, IL1A −889C→T (rs1800587; 4 studies, 2,195 cases, and 2,082 controls), IL1B −511C→T (rs16944; 3 studies, 1,843 cases, and 1,695 controls), and IL1B −31C→T (rs1143627; 4 studies, 2,188 cases, and 2,099 controls); in 2q14.2, IL1RN 9589A→T (rs454078; 3 studies, 1,673 cases, and 1,598 controls); in 4q26–27, IL2 −384T→G (rs2069762; 4 studies, 2,185 cases, and 2,085 controls); in 7p21, IL6 −174G→C (rs1800795; 4 studies, 2,203 cases, and 2,095 controls) and IL6 −597G→A (rs1800797; 3 studies, 1,679 cases, and 1,591 controls); in 1q31–32, IL10–3575T→A (rs1800890; 9 studies, 5,015 cases, and 5,061 controls) and IL10–1082A→G (rs1800896; 7 studies, 2,844 cases, and 3,328 controls); in 6p21.3, TNF–308G→A (rs1800629; 9 studies, 4,979 cases, and 4,752 controls) and LTA252A→G (rs909253; 9 studies, 5,067 cases, and 4,879 controls); and in 16q21, CARD15 Ex11–35→C (rs2066847; 7 studies, 4,267 cases, and 4,092 controls). SNPs were modeled as dichotomous variables assuming dominant inheritance (heterozygous/homozygous variant versus homozygous wild-type genotypes) as suggested by InterLymph analyses (6–8), to increase power and reduce the number of statistical tests. Because of potential ethnic differences in body fat and SNP distributions, analyses were restricted to persons who described themselves as of White European descent.

Risk estimates were calculated using unconditional logistic regression adjusted for study, sex, and age. Main and joint associations with BMI and each SNP on the risk of lymphoid neoplasms were estimated. Additive interactions between SNP and BMI were estimated by the relative excess risk due to interaction (RERI). When BMI was a categorical variable, the 95% confidence intervals (CI) for RERI were estimated using likelihood-based 95% CI (23). For BMI as a continuous variable, 10,000 bootstrapping samples (without replacement) of the original sample size were taken from the dataset and the 95% CIs were the 2.5th and 97.5th centile of the bootstrap sampling distribution (24).

Analyses were repeated for DLBCL, FL, and CLL/SLL, and all controls were used irrespective of the individual studies' matching techniques. Heterogeneity between study-specific risk estimates was considered present when a test for interaction between the variable of interest and study was statistically significant (P < 0.05). Potential sources were investigated using sensitivity analyses by: study design (population- or hospital-based); diagnosis classification; participation rates; continent; proportions of cases and controls with SNP data; whether the controls' SNP data were in Hardy–Weinberg equilibrium; or where there was no relationship between obesity and the SNP among controls. All analyses were conducted using Stata 13.1.

Data were received for a total of 5,844 cases and 6,167 controls. The majority of cases were diagnosed with mature B-cell neoplasms (90%), comprising DLBCL (28%), FL (23%), CLL/SLL (19%), and other B-cell subtypes (20%); 6% were T-cell in origin and 4% had no immunophenotype or subtype recorded. A higher proportion of cases were men (53%) and the median age at diagnosis was 60 years. Controls were more likely to be women, of younger age, and higher socioeconomic status than cases (Table 2).

Table 3 shows findings for the twelve cytokine polymorphisms among the subsets of subjects who had genotype data for each SNP as well as BMI data. Positive associations were found with DLBCL for TNF-308G>A (OR = 1.24; 95% CI, 1.07–1.44), IL10-1082A>G (OR = 1.14; 95% CI, 1.00–1.31), and CARD15 Ex11-35>C (OR = 1.25; 95% CI, 1.10–1.56); with FL for the two IL10 SNPs (IL10-3575T>A: OR = 1.15; 95% CI, 1.04–1.28; IL10-1082A>G: OR = 1.10; 95% CI, 1.05–1.15); and with CLL/SLL for IL1RN 9589A>T (OR = 1.50; 95% CI, 1.17–1.91). A few negative associations were also found for FL with IL6 -597G>A (OR = 0.81; 95% CI, 0.78–0.85) and LTA 252A>G (OR = 0.93; 95% CI, 0.87–0.99); and for CLL/SLL with IL1B -511C>T (OR = 0.84; 95% CI, 0.81–0.87). Table 4 shows findings between DLBCL, FL, and CLL/SLL and being overweight or obese for the subsets of subjects with data for the five positively associated polymorphisms; findings for BMI in the subsets for the other seven polymorphisms were similar (data not shown). Risk estimates were increased for DLBCL among obese individuals compared with those of normal weight in subsets with TNF-308G>A (OR = 1.33; 95% CI, 1.02–1.73), IL10-1082A>G (OR = 1.39; 95% CI, 1.12–1.73) and CARD15 Ex11-35>C (OR = 1.41; 95% CI, 1.04–1.91) data. For FL, there was no evidence that being obese increased risk (OR = 1.00; 95% CI, 0.81–1.24; OR = 1.13; 95% CI, 0.91–1.40 in the IL10-3575T>A and IL10-1082A>G subsets for example); while for CLL/SLL, some decreased associations with obesity were found (OR = 0.80; 95% CI, 0.69–0.93 in the IL10-3575T>A subset for example).

For DLBCL, the only subtype associated with obesity, tests for departure from additive interaction showed weak evidence for an additional effect of obesity and TNF-308GA+AA on DLBCL risk (P_interaction = 0.13); there was no evidence for the other two polymorphisms positively associated with this subtype (P_interaction = 0.77 and 0.85 for IL10-1082A>G and CARD15 Ex11-35>C, respectively). Table 5 shows joint associations of BMI and TNF-308G>A on DLBCL risk, with the RERI; for completeness, joint associations and RERIs of BMI and the other polymorphisms for all three subtypes are given in Supplementary Tables S1 and S2 for categorical and continuous BMI, respectively. For DLBCL, risk estimates were increased among those who were overweight or obese irrespective of TNF-308G>A status (overweight and GG: OR = 1.21; 95% CI, 1.02–1.44; overweight and GA+AA: OR = 1.31; 95% CI, 1.00–1.71; obese and GG: OR = 1.25; 95% CI, 1.00–1.56). However, being both obese and having the TNF-308A allele almost doubled the risk estimate compared with persons of normal weight who did not carry the A allele (OR = 1.93; 95% CI, 1.27–2.94); an additional risk from being obese and having TNF-308A variant rather than having either risk factor alone was suggested (RERI = 0.41; 95% CI, −0.05–0.84). Similarly, increased trends with 5 kg/m² increase in BMI above 18.5 kg/m² were seen in homozygous wild types and variant TNF-308A carriers (OR = 1.14; 95% CI, 1.07–1.22; OR = 1.19; 95% CI, 1.12–1.27, respectively). The corresponding RERI of 0.05 (95% CI, −0.005–0.08) suggests that with every 5 kg/m² rise in BMI, the risk of DLBCL is 0.05 more than if there was no interaction.

Joint associations of BMI and TNF-308G>A genotype on DLBCL risk were not consistent across studies (P_{heterogeneity} = 0.02). Associations were similar among North American studies, with evidence of additive interaction (RERI = 1.27; 95% CI, 0.48–2.08); but not among European studies (RERI = −0.25; 95% CI, −0.84–0.28; Table 6). Heterogeneity was present among studies that were population-based, which used the ICDO3 disease classification, where participation rates were 70% or more, where 90% or more of subjects were genotyped, where control distributions of TNF genotypes were in Hardy–Weinberg equilibrium; or where BMI and TNF-308G>A genotypes were not correlated among controls. In these groupings, the risk estimates of being obese and having TNF-308GA+AA genotype tended to be similar, and RERIs were all above zero, although most were not statistically significant.

This InterLymph analysis of the joint associations of BMI and cytokine polymorphisms on the risk of the three most common lymphoid neoplasms found some evidence of interaction between obesity and TNF-308G>A (rs1800629). For DLBCL, the risk was greatest among those who were obese and carried the TNF-308A allele, although risk was also increased among the overweight regardless of TNF status. The associations showed some variation between studies, but these differences were not explained by study design, disease classification, or other factors. On the other hand, being obese and carrying TNF-308A did not increase the risk of either FL or CLL/SLL. Besides TNF-308G>A, other cytokine SNPs in IL1A (rs1800587), IL1B (rs16944, rs1143627), IL1RN (rs454078), IL2 (rs2069762), IL6 (rs1800795, rs1800797), IL10 (rs1800890, rs1800896), LTA (rs909253), and CARD15 (rs2066847) showed little evidence of altering the non-Hodgkin lymphoma (NHL) risk associated with being overweight.

Since publication of the InterLymph pooled analysis of BMI, where we reported that obesity increased DLBCL risk (1), several studies including cohorts have also found this relationship (25–29) while others have not (11, 30–36). In summarizing published data for DLBCL and obesity, two meta-analyses have noted an increased risk (3, 4), the latest including all but the most recent publications (27, 28). Unfortunately, however, several studies which reported no association with total NHL did not stratify their data by subtype (32, 37–39). To further explore the mechanisms underlying the relationship between obesity and lymphoma, chronic inflammation involving cytokine production is one possible pathway that has been investigated mostly by examining SNPs (9–11), although the functional roles for some are not yet conclusive (40). One study measured prediagnosis serum levels of cytokines and found no association between TNF levels and NHL among either normal or overweight persons (9). When examining SNPs in TNF, Wang and colleagues noted an excess risk of DLBCL among obese individuals carrying the TNF-308A allele and, although Chen and colleagues did not report joint associations, a similar finding among overweight women is suggested (crude OR = 1.8; 95% CI, 1.0–3.2; refs. 10, 11). To our knowledge, these are the only studies to have investigated cytokine SNPs in relation to the effect obesity may have on lymphoma risk, and both are included in this pooled analysis.

TNF has been implicated in the relationship between obesity and several other cancers including breast, endometrium, and gastrointestinal (41–43); the promotion of tumor cell proliferation through activation of NFκB being suggested as the most likely explanation (44). In obesity, B cells, T cells, and macrophages infiltrate the expanding adipose tissue, and not only lead to, but also maintain, a chronic inflammatory state (5). The macrophages secrete most of the TNF produced by adipose tissue, which escapes into circulation to bind to and activate its receptor TNFR, which is expressed in all human tissues (5, 41). TNF activates IL6 in adipose tissue and downstream of both cytokines are the NFκB and STAT3 cycles. These pathways have important roles in lymphocyte development, function, and survival, and deregulation of these cycles are seen in lymphoid malignancies including DLBCL, the most common aggressive subtype examined here (26). Obesity-related lymphomagenesis is likely to be complex involving the actions of additional proinflammatory cytokines and immunomodulatory mediators that trigger downstream targets that promote the clonal expansion and transformation of B cells with premalignant lesions. Further studies will be needed to investigate possible disease mechanisms.

When assessing gene–environment interactions, differential misclassification can bias the interaction estimate in either direction (45). Our data may not be free from differential case–control participation and reporting. Among controls, obesity-related health problems may have influenced their participation and for cases, although rapid ascertainment techniques were employed, those with poor survival, which may be related to different degrees of adiposity (46), could have been missed. Our anthropometric data were self-reported and BMI could be biased toward “normal” weight as respondents tend to overestimate their height and underestimate their weight to varying degrees dependent on their gender and age (47). Cases' responses could also have been influenced by weight loss associated with lymphoma, although several studies attempted to compensate for this by requesting weight at a year or more before diagnosis. The effect of participation bias on our finding of an interaction will be limited if obesity and TNF-308G>A are associated in the general population. TNF-308G>A SNP has been suggested to be related, albeit weakly, to obesity, but the mechanism for TNF gene involvement in obesity pathogenesis is unclear (48). If there is an association, persons carrying the variant allele may be under- (or over-) represented in our data if body fatness is related to participation, or the stratum-specific frequencies on gene and BMI category could be inaccurate, biasing the interaction estimate in either direction. Among our controls, TNF genotype and BMI overall were not correlated in all but two of our studies (NCI-SEER, EpiLymph-Spain); the removal of these did not alter our findings.

Strengths of our study include its large sample size, giving the potential to examine interactions and explore differences in interactions among the most common lymphoid neoplasms. Obesity prevalence varies across countries, which could relate differently to subjects' participation and responses in the studies included. Most studies had not published on this topic before, and while data were a subset of studies and subjects included in the main effects analyses, the risk estimates for BMI and SNPs were consistent with those published previously (1, 6–8). A reduced risk of CLL/SLL with obesity was found in some subsets of data, which could relate to disease-related weight loss; but in larger InterLymph datasets, no obesity associations for CLL/SLL have been reported (1). Other limitations are the low power to assess interactions in less common lymphoid neoplasms and some SNPs which were tested in only a few studies. Indeed, statistically significant interactions were found with SNPs genotyped in all nine studies, and others with fewer may have shown an effect had we had more data. Many other candidate cytokine SNPs were not associated with lymphoid neoplasms, and so it is not surprising that no joint association was found between obesity and these SNPs. BMI in this analysis related to weight at older age and there was no evidence that associations were different among those aged under or over 65 years; data on BMI in young adulthood, which has been associated with DLBCL (2, 31), were too sparse. As for generalizability, our findings may not translate to all populations as our subjects were Caucasian and resident in developed nations. Furthermore, the BMI was developed to estimate body fat in Caucasians of working age and so may not be applicable to other groups; no other indicators such as waist-to-hip ratio were available. There is certainly variation in obesity rates worldwide (49) and also the distribution of TNF-308G>A genotypes and those of the other SNPs studied here differ between populations too (50).

In conclusion, we found some evidence of interaction between TNF-308G>A and BMI on the risk of DLBCL. The increased risk among persons carrying the variant TNF-308A allele and being obese was not necessarily consistent across studies, and the possibility that differential biases affected the findings cannot be ruled out. One way to potentially reduce these biases is to examine gene–environment associations in large cohort studies using more specific measures of adiposity such as total body fat, and obtaining data on circulating levels of TNF and other cytokines. Furthermore, within InterLymph, genome-wide association scans are now completed for more studies than included here (51), and findings from these may identify other adipose tissue–related cytokines and adipokines.

P.M. Bracci reports receiving commercial research support from Navidea Biopharmaceuticals, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Kane, C.F. Skibola, E.A. Holly, L.M. Morton, S.S. Wang, T. Zheng

Development of methodology: E.A. Holly, T. Zheng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.M. Bracci, J.R. Cerhan, K.E. Smedby, E.A. Holly, M. Maynadié, A.J. Novak, T.J. Lightfoot, S.M. Ansell, A.G. Smith, M. Liebow, M. Melbye, L.M. Morton, S. de Sanjosé, S.L. Slager, T. Zheng, E. Roman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Kane, P.M. Bracci, K.E. Smedby, L.M. Morton, S. de Sanjosé, S.L. Slager, S.S. Wang, T. Zheng

Writing, review, and/or revision of the manuscript: E. Kane, C.F. Skibola, P.M. Bracci, J.R. Cerhan, L. Costas, K.E. Smedby, E.A. Holly, T.J. Lightfoot, S.M. Ansell, M. Liebow, M. Melbye, L.M. Morton, S. de Sanjosé, S.S. Wang, Y. Zhang, T. Zheng, E. Roman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.M. Bracci, J.R. Cerhan, M. Melbye, Y. Zhang, T. Zheng

Study supervision: C.F. Skibola, E.A. Holly, T. Zheng

Studies that contributed data to this pooled analysis were supported by: NCI contracts N01-PC-65064, N01-PC-67008, N01-PC-67009, N01-PC-67010, N02-PC-71105 and the Intramural Research Program of the NCI, NIH, and Public Health Service (NCI-SEER study); grants R01 CA92153 and P50 CA97274, NIH (Mayo Clinic; to J.R. Cerhan); grants R01 CA45614, R03 CA89745, R01 CA87014, NCI 263-MQ-701711, U01 CA66529, R01 CA104682, and CA154643, NIH and NCI, and collection of cancer incidence data was supported by the California Department of Public Health as part of the statewide cancer reporting program; the NCI's SEER Program under contract HHSN261201000140C awarded to CPIC; and the CDC's National Program of Cancer Registries, under agreement # U58DP003862-01 awarded to the California Department of Public Health (E.A. Holly, UCSF); CA62006, NCI (T. Zheng, Connecticut); Leukaemia & Lymphoma Research, United Kingdom (E. Roman, ELCCS); R19-A2364, Lundbeck Foundation Grant, DP 08-155 Danish Cancer Society Grant, 5R01 CA69669-02 NIH, Danish Cancer Research Foundation Grant and Plan Denmark (M. Melbye, SCALE-Denmark); 2009/659, the Swedish Cancer Society, 20110209, Stockholm County Council, 02 6661, Swedish Cancer Society grant and the Strategic Research Program in Epidemiology at Karolinska Institute (K.E. Smedby, SCALE-Sweden); QLK4-CT-2000-00422 and FOOD-CT-2006-023103 European Commission; CIBERESP, PI11/01810, RCESP C03/09, RTICESP C03/10, and RTIC RD06/0020/0095, the Spanish Ministry of Health (grant references), 051210 the Marató de TV3 Foundation, 2009SGR1465 the Agència de Gestió d'Ajuts Universitaris i de Recerca—Generalitat de Catalunya (who had no role in the data collection, analysis, or interpretation of the results); 2014SGR756-F, PI11/01810, PI14/01219 and CM13/00232; NO1-CO-12400, the NIH, the Compagnia di San Paolo—Programma Oncologia, StSch4261 and StSch4420 grants, the Federal Office for Radiation Protection, DJCLS-R04/08 the José Carreras Leukemia Foundation grant (S. de Sanjosé, M. Maynadié, EpiLymph studies).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.