Genetic Variation in Tumor Necrosis Factor and the Nuclear Factor-κB Canonical Pathway and Risk of Non-Hodgkin's Lymphoma

Non-Hodgkin's lymphoma (NHL) is a heterogeneous malignancy of uncontrolled proliferation of clonal B cells at different stages of maturation, although clonal T-cell and NK-cell malignancies can also occur. Immune dysfunction has been clearly associated with NHL risk (1), and there is accumulating evidence for polymorphic variation in immune genes that control immune function and regulation as susceptibility loci for NHL (2, 3). The most robust finding to date is for the tumor necrosis factor (TNF) G-308A single nucleotide polymorphism (SNP), which has been associated with higher constitutive and inducible expression of TNF-α in several model systems (4, 5) and increased susceptibility to several infectious and inflammatory conditions (6). In the InterLymph pooled analysis of 3,586 NHL patients and 4,018 controls, the variant heterozygotes [odds ratio (OR)_GA, 1.18; 95% confidence interval (95% CI), 1.04-1.33] and homozygotes (OR_AA, 1.25; 95% CI, 0.91-1.70) were associated with increased risk of NHL, and this association was specific to diffuse large B-cell lymphoma (DLBCL; OR_GA, 1.29; 95% CI, 1.10-1.51 and OR_AA, 1.65; 95% CI, 1.16-2.34) but not follicular lymphoma (7). In addition, several haplotypes involving TNF and lymphotoxin A (LTA) have also been implicated in NHL (7-10).

One important biological function of TNF is activation of the canonical (“classic”) pathway of nuclear factor-κB (NF-κB), which leads to the transcription of a large number of target inflammatory and other genes related to cellular growth, differentiation, and apoptosis (Fig. 1). NF-κB is a family of evolutionarily conserved transcription factors that consists of five members in mammals: reticuloendotheliosis (Rel or c-Rel), RelA (p65), RelB, NF-κB1 (p50), and NF-κB2 (p52). Key canonical pathway NF-κB dimers are composed of p50, p65, and c-Rel and inhibitory proteins IκBα, IκBβ, and IκBε. Inappropriate activation of NF-κB has been linked to lymphomagenesis (11) and to inflammatory processes associated with autoimmune disease, asthma, and atopy (12), which themselves are known NHL risk factors (1).

In a prior analysis (3), we focused on gene discovery and reported results for the strongest associations of 1,253 immune genes with risk of NHL. In the current analysis, we have added additional genotyping data and have focused on a candidate pathway analysis; specifically, we evaluated the association of TNF and LTA SNPs and haplotypes with risk of NHL. Furthermore, because binding of TNF to its receptor triggers the NF-κB canonical pathway, leading to transcriptional activation of multiple genes that influence inflammation and immune response, we evaluated whether genetic variation in this pathway was also associated with risk of NHL.

Full details of this ongoing, clinic-based case-control study conducted at the Mayo Clinic have been reported previously (3). This study was reviewed and approved by the Human Subjects Institutional Review Board at the Mayo Clinic, and all participants provided written informed consent. Briefly, eligible patients were within 9 months of their first NHL diagnosis, ages ≥20 years, HIV negative, and were residents of Minnesota, Iowa, or Wisconsin at the time of diagnosis. All cases were reviewed and histologically confirmed by a Mayo Clinic hematopathologist and classified according to the WHO criteria (13). Of the 956 eligible cases, 626 (65%) participated, 109 (11%) refused, 19 (2%) were unable to be contacted, and 202 (21%) had their eligibility expire mainly due to not completing consenting (within 9 months of diagnosis) or data collection (within 12 months of diagnosis) in the required timeframe. The median time from diagnosis to enrollment was 53 days (10th percentile, 4 days; 90th percentile, 138 days), and only 3% of patients had their eligibility expire due to death. Clinic-based controls were randomly selected from Mayo Clinic Rochester patients ages ≥20 years, who were residents of Minnesota, Iowa, or Wisconsin, and were being seen for a prescheduled medical examination in the general medicine divisions of the Department of Medicine. Patients were not eligible if they had a history of lymphoma, leukemia, or HIV infection. Controls were frequency matched to cases by 5-year age group, gender, and county of residence. Of the 818 eligible controls, 572 (70%) participated, 239 (29%) refused, and 7 (1%) had their eligibility expire.

All subjects agreeing to participate were asked to complete a self-administered risk factor questionnaire and provide a blood sample. DNA was extracted from samples using a standard procedure (Gentra). This (phase 1) analysis includes 498 cases and 497 controls enrolled from September 1, 2002 to September 30, 2005 and who had a DNA sample available on October 1, 2005.

This analysis of TNF, LTA, and the NF-κB canonical pathway (Table 1) was part of a larger genotyping project to assess the role of immune and other candidate genes in the etiology and prognosis of NHL (3). Most of the genes and SNPs reported here were from the ParAllele (now Affymetrix) Immune and Inflammation SNP panel that included 1,253 genes and 9,412 SNPs (14). This genotyping was supplemented by a second round of genotyping using a custom Illumina GoldenGate (15) 384 SNP OPA that included SNPs from 100 candidate genes. For both rounds of genotyping, we used tagging SNPs that covered 5 kb upstream and downstream of each gene with minor allele frequency (MAF) of ≥0.05 and pairwise r² threshold of 0.8 supplemented by validated nonsynonymous SNPs. Across the two platforms, the sample success rate was >98% (similar for both platforms), the assay call rate was >93% (98.8% for the ParAllele platform and 93.5% for the Illumina platform), and the concordance rate for blinded duplicates was >98% (similar for both platforms).

After excluding non-Caucasian or Hispanic subjects (n = 16), cases with Hodgkin's lymphoma or other diagnoses (n = 23), subjects with call rates <95% (n = 16), quality control failures (n = 1), and subjects not genotyped on both platforms (n = 23), there were 916 subjects with combined genotype data. Of the 9,796 combined SNPs genotyped, we excluded SNPs with call rates <95% (n = 724), SNPs with two or more nonmissing differences or other quality-control issues (n = 10), and SNPs not mapped to National Center for Biotechnology Information Build 36, dbSNP 126 of the human genome (n = 20), leaving 9,042 SNPs. There were 72 SNPs that were duplicated across platforms, and genotypes from these SNPs were compared. The concordance was 99.70%, although one SNP was found to have a large number of discordant genotypes (160 of 916 subjects) between the platforms and was subsequently dropped from further analysis; the remaining SNPs had only a total of 40 differences (99.94% concordance). Of the 71 remaining duplicated SNPs, the SNP with the highest platform-specific SNP call rate was chosen, leaving a total of 8,969 SNPs. Finally, SNPs that had a MAF of <1% in cases and controls combined were excluded (n = 935), leaving a total of 916 subjects (441 cases and 475 controls) and 8,034 SNPs available for analysis.

Allele frequencies from cases and controls were estimated using observed genotype frequencies. The genotype frequencies in the controls were compared with allele frequencies expected under Hardy-Weinberg equilibrium using a Pearson goodness-of-fit test or Fisher's exact test (MAF < 0.05). In this analysis, there were six SNPs from the candidate genes that had a Hardy-Weinberg P < 0.05 (see Supplementary Table S1); none of these SNPs were excluded from the analysis.

Two methods were used when analyzing the association between each gene and risk of NHL: haplotype analysis and principal components. For the haplotype analysis, all SNPs from a gene were used to determine haplotype frequencies, and a global score test was used, as implemented in the S-Plus program haplo.stats (16). As a global gene test, we used principal components to create uncorrelated components that are linear combinations of the SNPs from a gene. These components were then ranked according to the amount of the total SNP variance explained. The resulting smallest subset of components that accounted for at least 90% of the variability among the SNPs was included in a multivariable logistic regression model. A likelihood ratio test was then used to jointly test the significance of the selected principal components. This method decreases the dimensionality of the correlated SNPs by reducing the number of independent degrees of freedom (17). Gene-level tests with P < 0.05 were declared of interest, which is justified for the TNF and LTA genes given their high prior probability (7). We also used a P < 0.05 threshold for the NF-κB pathway genes, although to address concerns about multiple testing, we assessed the overall significance of the P values for our gene-level tests using the tail strength methodology (18).

Individual SNPs were examined using unconditional logistic regression to estimate ORs and corresponding 95% CIs separately for heterozygotes and minor allele homozygotes using homozygotes for the major allele as the reference. A P_trend was calculated assuming an ordinal (log-additive) genotypic relationship. This scoring scheme has been shown to be robust from departures from an additive model (19) and it makes no assumption about Hardy-Weinberg equilibrium (20). For the analysis of tagSNPs from the NF-κB pathway, a P < 0.05 in the setting of a global gene test of P < 0.05 was declared of interest. To evaluate the effect of multiple testing for the tagSNPs, we calculated the tail strength for the set of 54 P values (18). We also calculated individual q values for each ordinal P_trend tests (21) and declared q values < 0.1 to be of interest.

Analyses were implemented using SAS (SAS Institute; version 8, 1999) and S-Plus (Insightful; version 7.05, 2005) software systems. All analyses were adjusted for the design variables of age, gender, and county of residence. These analyses were restricted to subjects whose self-reported race was Caucasian. Furthermore, we previously tested and found no evidence of population stratification in our data (3) using STRUCTURE (22).

There were 441 cases and 475 controls available for analysis. Cases were slightly younger and somewhat less likely to have attended graduate or professional school compared with controls but were well matched on gender and state of residence (Table 2). The most common NHL subtypes were chronic lymphocytic leukemia (CLL)/small lymphocytic lymphoma (SLL; n = 123), follicular (n = 113), and DLBCL (n = 69).

Five previously published SNPs in LTA and TNF with putative functional significance were first evaluated with risk of NHL and each of the common subtypes. As shown in Table 3, none of the SNPs overall were individually associated with risk of NHL, with the exception of a suggestive positive association for TNF G-308A (rs1800629), where there was evidence increased risk for the AA genotype. Formal evaluation of the recessive model gave suggestive evidence (P = 0.067) for the A allele being association with overall NHL risk (OR, 2.14; 95% CI, 0.95-4.83). In subtype analysis, similar results were seen for both follicular lymphoma (OR, 2.71; 95% CI, 0.93-7.85) and DLBCL (OR, 3.31; 95% CI, 0.98-11.1) but not CLL/SLL (OR, 0.81; 95% CI, 0.17-3.82). Although there was no effect of the LTA C-91A SNP with risk of all NHL, there were inverse associations for the AA genotype for both CLL/SLL (P = 0.03) and follicular NHL (P = 0.05). There was also a positive association for GA heterozygotes (no AA homozygotes observed) in TNF G-238A SNP but for follicular lymphoma only (P = 0.01).

We next conducted a haplotype analysis of the five LTA/TNF SNPs according to a previously published report from a U.S. population (8) but found no global association (P = 0.72). As shown in Table 3, our estimated haplotype frequencies were quite similar to the previous study, and we observed an association for the C-G-C-A-G haplotype (OR, 1.26; 95% CI, 1.07-1.48) compared with the A-A-C-G-G haplotype, which was also seen for CLL/SLL (OR, 1.37; 95% CI, 1.08-1.75) and follicular lymphoma (OR, 1.35; 95% CI, 1.06-1.72) but not DLBCL. In addition to the previously reported haplotype, we also observed suggestive positive associations with the other two remaining common haplotypes. The latter haplotype associations were also observed for follicular lymphoma and CLL/SLL but were not observed for DLBCL.

We also evaluated the restricted haplotype based on the TNF G-308A and the LTA A252G loci, which was reported in the InterLymph pooled analysis (7). Compared with the GA haplotype, there was a significantly elevated risk for the AG haplotype (OR, 1.16; 95% CI, 1.00-1.33) but not the GG haplotype (OR, 1.08; 95% CI, 0.91-1.27); full details are provided in Supplementary Table S2. In subtype analysis, the AG haplotype showed a suggestive association with both CLL/SLL (OR, 1.20; 95% CI, 0.99-1.45) and DLBCL (OR, 1.20; 95% CI, 0.93-1.54) but not follicular lymphoma (OR, 1.05; 95% CI, 0.83-1.31).

We next conducted gene-level tests for TNF, LTA, and nine genes from the NF-κB canonical pathway. The candidate genes in this pathway are shown in Table 4, and coverage ranged 6% to 100%. Only NFKB1 showed evidence for an association with NHL (P = 0.049) by principal components analysis, whereas none of the haplotype tests approached statistical significance at P < 0.05. There were no individual-level SNP tests that achieved statistical significance outside of the gene-level test (Supplementary Table S1). Of the 10 SNPs in NFKB1 (Table 5), only rs4648022 showed a statistically significant association with NHL (P = 0.0037), although rs10489114 showed a suggestive association (P = 0.063) but was uncommon (MAF in controls = 0.01). The linkage disequilibrium structure of NFKB1 is shown in Fig. 2. SNP rs4648022 is an intronic SNP, with a MAF of 0.06 in cases and 0.10 in controls. The ordinal OR was 0.59 for each variant allele (95% CI, 0.41-0.84), and this association was similar for CLL/SLL (OR, 0.58; 95% CI, 0.32-1.02), follicular lymphoma (OR, 0.58; 95% CI, 0.32-1.05), and DLBCL (OR, 0.56; 95% CI, 0.26-1.20).

To assess the effect of multiple testing on our results, we calculated the tail strength of P values from the gene-level tests (n = 10; tail strength = -0.14) and 54 SNP-level tests (n = 54; tail strength = -0.15), and for both, we found that the distribution of P values was no different from chance. We also calculated q values for the 54 SNPs and found that all q values were >0.10, except rs4648022, which had a q = 0.071, suggesting that the association for that SNP was noteworthy.

We provide additional evidence for a role of genetic variation in TNF and LTA SNPs and haplotypes in the etiology of NHL and also provide some of the first data to support an association of genetic variation in NFKB1 with NHL risk. Although our sample size was modest for a genetic study, we were reasonably powered to detect main effects for more common alleles in the setting of evaluating hypotheses with relatively high prior probabilities for NHL associations with TNF (7) and genes in the canonical NF-κB pathway (11). Furthermore, the most significant SNP from NFKB1 remained noteworthy after accounting for multiple testing. Nevertheless, these data should be interpreted with caution and viewed as an initial step in the evaluation of the role of genetic variation in canonical NF-κB pathway in the etiology of NHL.

Our finding of an association with the TNF G-308A SNP (ordinal OR, 1.14; 95% CI, 0.89-1.46) is consistent for all NHL with the InterLymph pooled results (ordinal OR, 1.16; 95% CI, 1.04-1.28), although closer inspection of our data suggests a recessive model and positive associations for both DLBCL and follicular lymphoma as opposed to the association specific to DLBCL observed in the InterLymph results (7). In similar fashion, our LTA A252G SNP (ordinal OR, 1.10; 95% CI, 0.90-1.34) was consistent with the InterLymph results (ordinal OR, 1.05; 95% CI, 0.95-1.15), and the TNF-LTA AG haplotype (compared with the GA haplotype) was associated risk in our study (OR, 1.16; 95% CI, 1.00-1.33) and InterLymph (OR, 1.29; 95% CI, 1.14-1.47). As with InterLymph, this association was seen for DLBCL but not follicular lymphoma, although we also saw an association for CLL/SLL, which was not specifically evaluated in InterLymph.

Our findings for other candidate TNF (C-857T and G-238A) and LTA (C-91A and C-857T) SNPs were notable only for the LTA C-91A SNP, where we saw a weak and not statistically significant inverse association with all NHL (ordinal OR, 0.87; 95% CI, 0.72-1.06), which is consistent with prior reports (8, 9, 23). However, we also observed stronger inverse associations for this SNP with CLL/SLL (OR, 0.71; 95% CI, 0.52-0.96) and follicular lymphoma (OR, 0.73; 95% CI, 0.53-1.00) but not DLBCL, which was not consistent with other studies that found either no subtype difference (8, 23) or an association specific to DLBCL (9). Adding these additional TNF and LTA SNPs into a previously published haplotype (8), our strongest results were seen for the C-G-C-A-G haplotype (compared with the most common A-A-C-G-G haplotype) based on two LTA and three TNF SNPs (OR, 1.26; 95% CI, 1.07-1.48), which was nearly identical to the haplotype estimate from the National Cancer Institute-Surveillance, Epidemiology, and End Results study (OR, 1.31; 95% CI, 1.06-1.63; ref. 8). However, unlike the National Cancer Institute-Surveillance, Epidemiology, and End Results study, our haplotype associations were evident for CLL/SLL and follicular lymphoma but were not seen for DLBCL. Overall, SNP and haplotype analysis suggest that TNF G-308A, LTA C-91A, or LTA A252G, or a SNP in linkage disequilibrium with one or more of these SNPs, plays a role in the etiology of NHL, although the specifics of NHL subtype associations is still evolving. Given that all of these studies were conducted in White populations of mainly northern European ancestry and the MAFs were all nearly identical, the discrepancies in specific associations may be due to statistical variability that planned pooling studies will be able to more robustly address.

Of the NF-κB canonical pathway genes that we evaluated, we observed an association only for NFKB1. NFKB1 encodes two proteins, a non-DNA-binding protein, p105, and a DNA-binding protein, p50; the major form of NF-κB is composed of p50 along with the p65 (RelA). Karban et al. (24), identified a common insertion/deletion promoter polymorphism (-94ins/delATTG) in NFKB1 and provided evidence for functionality in a reporter assay as well as an association with ulcerative colitis, although in a meta-analysis there was no association for either ulcerative colitis or Crohn's disease (25). The same polymorphism has also been associated with bladder cancer (26) and melanoma risk (27) but not CLL, renal cell carcinoma, or colon cancer (26). We did not have data on the -94ins/delATTG polymorphism; the significant SNP in our study was an intronic tagSNP; thus, if an association of NFKB1 is replicated, fine mapping would be required to identify the causal variant(s).

NF-κB signaling has essential roles in lymphocyte development, activation, proliferation, and survival, and aberrant signaling has been linked to lymphomagenesis, including Hodgkin's lymphoma, T- and B-cell NHLs, and multiple myeloma (11). It is well known that TNF signaling through TNFR1 leads to activation of the canonical NF-κB pathway, and p50 (encoded by NFKB1) is a component of the major NF-κB dimer (along with RelA) from the canonical pathway (28). The p50-RelA dimer translocates to the nucleus to activate gene transcription, leading to a broad amplification of the inflammatory response, including an increase in proinflammatory cytokines (e.g., interleukin-6, interleukin-4, interleukin-5, TNF-α and interleukin-1β), as well as chemokines (e.g., interleukin-8 and RANTES), and adhesion molecules (e.g., vascular cell adhesion molecule-1, intercellullar adhesion molecule-1, and E-selectin). Additional effects include activation of positive cell cycle regulators (e.g., cyclin D1, cyclin D2, c-myc, and c-myb) and antiapoptotic factors (e.g., caspases and BCL2 family members; ref. 11). Although we observed an association with NFKB1, we cannot rule fully out other genes in this pathway as tagSNP coverage was <70% for several genes that we studied. In addition, there are other genes within the canonical pathway that we did not evaluate. We also cannot rule out that our findings are false positives. Finally, there is evidence that other cytokines, including CD40L (29) and BLyS (30), are also associated with NHL risk, but these cytokines signal mainly through the NF-κB alternative pathway, which we were not able to evaluate.

There are several strengths to our study, including the careful characterization of cases and controls, central pathology review, extensive genotyping quality controls, and use of the common HapMap SNPs to tag the genes of interest. We have evaluated previously the potential for population stratification in our study population and found that this is not likely (3). Limitations of our study include the relatively small sample size for NHL subtypes, lack of high gene coverage for some of the genes, and an all-White population, limiting generalizability. Although our study was not population-based, by restricting to regional cases and controls (Minnesota, Iowa, and Wisconsin), we decrease referral bias and ensure that our cases and controls are derived from the same underlying population. Of note, the MAF for the TNF and LTA genotypes and haplotypes in our study were concordant with population-based data (8).

In summary, our study provides further evidence of a role for genetic variation in the TNF G-308A SNP with risk of NHL as well as LTA/TNF haplotypes. Additionally, we provide new evidence that genetic variation in NFKB1 is also associated with risk, which will require replication. These findings provide additional clues to the etiology of this cancer and support identifying additional genes and environmental exposures that affect NF-κB signaling, with the ultimate goal of identifying novel prevention approaches.

No potential conflicts of interest were disclosed.

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.

We thank Sondra Buehler for editorial assistance.