Search tips
Search criteria 


Logo of plosonePLoS OneView this ArticleSubmit to PLoSGet E-mail AlertsContact UsPublic Library of Science (PLoS)
PLoS ONE. 2008; 3(7): e2816.
Published online 2008 July 30. doi:  10.1371/journal.pone.0002816
PMCID: PMC2474696

Polymorphisms in the Estrogen Receptor 1 and Vitamin C and Matrix Metalloproteinase Gene Families Are Associated with Susceptibility to Lymphoma

Derya Unutmaz, Editor



Non-Hodgkin lymphoma (NHL) is the fifth most common cancer in the U.S. and few causes have been identified. Genetic association studies may help identify environmental risk factors and enhance our understanding of disease mechanisms.

Methodology/Principal Findings

768 coding and haplotype tagging SNPs in 146 genes were examined using Illumina GoldenGate technology in a large population-based case-control study of NHL in the San Francisco Bay Area (1,292 cases 1,375 controls are included here). Statistical analyses were restricted to HIV- participants of white non-Hispanic origin. Genes involved in steroidogenesis, immune function, cell signaling, sunlight exposure, xenobiotic metabolism/oxidative stress, energy balance, and uptake and metabolism of cholesterol, folate and vitamin C were investigated. Sixteen SNPs in eight pathways and nine haplotypes were associated with NHL after correction for multiple testing at the adjusted q<0.10 level. Eight SNPs were tested in an independent case-control study of lymphoma in Germany (494 NHL cases and 494 matched controls). Novel associations with common variants in estrogen receptor 1 (ESR1) and in the vitamin C receptor and matrix metalloproteinase gene families were observed. Four ESR1 SNPs were associated with follicular lymphoma (FL) in the U.S. study, with rs3020314 remaining associated with reduced risk of FL after multiple testing adjustments [odds ratio (OR) = 0.42, 95% confidence interval (CI) = 0.23–0.77) and replication in the German study (OR = 0.24, 95% CI = 0.06–0.94). Several SNPs and haplotypes in the matrix metalloproteinase-3 (MMP3) and MMP9 genes and in the vitamin C receptor genes, solute carrier family 23 member 1 (SLC23A1) and SLC23A2, showed associations with NHL risk.


Our findings suggest a role for estrogen, vitamin C and matrix metalloproteinases in the pathogenesis of NHL that will require further validation.


Non-Hodgkin lymphoma (NHL) is the most common hematopoietic cancer diagnosed in the U.S. Rates have increased steadily over the last 30 years ( Known risk factors include family history, exposure to infectious agents and immunodeficiency associated with congenital conditions and post-transplant therapy[1], though the causes for most cases are unknown. Genetic and environmental factors that influence the growth, survival and capacity of B-cells with pre-neoplastic changes to undergo transformation[2] or that impact the quality of tumor-specific immune responses may play integral roles in lymphomagenesis. Approximately 80% of lymphomas are of B-cell origin that arise during germinal center (GC) and post-GC reactions as part of otherwise normal B-cell differentiation[3]. Diffuse large B-cell lymphoma (DLBCL), follicular lymphoma (FL) and small lymphocytic lymphoma/chronic lymphocytic leukemia (SLL/CLL) are major lymphoma histologies that collectively comprise ~75% of all NHL in industrialized countries. Whereas DLBCL is a clinically aggressive disease for which incidence increases with age, FL and SLL/CLL present a more indolent clinical course. Molecular and phenotypic differences between NHL histologies suggest they may have distinct etiologies, though overlapping mechanisms are likely. The study of candidate single nucleotide polymorphisms (SNPs) may help to identify environmental risk factors, further elucidate disease mechanisms and discover novel targets for lymphoma treatment regimens. Thus, we undertook a genetic study of 768 coding and haplotype tagging SNPs in 146 genes in a population-based case-control study of NHL in the San Francisco Bay Area (1,292 cases, 1,375 controls). The genes investigated were primarily selected based on findings from our earlier genetic studies[2], [4][12] in pathways involved in steroidogenesis, immune function, cell signaling, sunlight exposure, xenobiotic metabolism/oxidative stress, energy balance, as well as uptake and metabolism of cholesterol, folate and vitamin C (Table 1). The most compelling associations were tested in an independent population-based case-control study of lymphoma based in Germany.

Table 1
Pathways/gene functional groups of analyzed candidate genes in the San Francisco Bay Area NHL study.


Assessment of sample completion rates and SNP call rates

Of the 768 SNPs genotyped in DNA of 2,667 study participants (1,292 cases and 1,375 controls) from a large population-based case-control study with more than 4,100 participants, eight samples were excluded from further analyses due to missing SNP data (>10%) yielding a sample completion rate of 99.7%. After restriction to white non-Hispanics, 959 cases and 1,049 controls were included in the final analysis. A total of 62 SNPs with call rates <90% were dropped from the analysis yielding a SNP call rate of 91.9%. Additional dropped SNPs included those with minor allele frequencies <5% (n = 29), X chromosome SNPs (n = 7) and SNPs with Hardy-Weinberg equilibrium p-values<0.01 (n = 33) leaving a total of 637 SNPs for subsequent analyses. Two pairs of replicate DNAs on each plate revealed genotype concordance rates of >99% within and between plates. Genotype results concurred (>98.5%) with data from the quality control Taqman genotyping of 10 SNPs on the same study population. Single-marker allelic tests for all SNPs were performed using χ2 statistics. SNPs and haplotypes with the strongest statistical associations (q-values≤0.10) adjusted for multiple comparisons using the false discovery rate (FDR) are shown in Tables 2 and and3,3, respectively, while p-values for all SNPs are available online (website). None of the associated SNPs violated HWE at the p<0.05 level (Table 2).

Table 2
SNPs associated with NHL with corrected p-values (q)<0.10* for the San Francisco Bay Area NHL case-control study.
Table 3
Associated haplotypes with corrected p-values (q)<0.10 for the San Francisco Bay Area NHL case-control study.

Gene- and SNP-based analyses

SNPs in vitamin C receptor genes influence risk of NHL, FL and SLL/CLL

SNP-specific analyses revealed that both SNPs genotyped in the vitamin C receptor gene, solute carrier family 23 member 1 (SLC23A1), rs6596473 (IVS13+2515G>C) and rs11950646 (IVS9−110A>G), influenced risk of FL (p = 2.5×10−2, q = 8.4×10−2, Table 2). Those homozygous variant for either SNP had an 80% elevated risk of FL (Table S1). These SNPs captured 75% of the haplotype structure in SLC23A1 (R2 = 0.99) though no haplotype associations were found.

An intronic SLC23A2 SNP rs1776948 (IVS2−17750G>A) influenced NHL (p = 4.7×10−3, q = 4.3×10−2), FL (p = 2.0×10−2, q = 8.4×10−2) and SLL/CLL risk (p = 1.1×10−3, q = 3.0×10−2). In logistic regression analyses, the rs1776948 homozygous variant 17750AA genotype was positively associated with disease risk (Table S1). Notable associations were observed for two additional intronic SLC23A2 SNPs, rs6133175 (IVS4+1767A>G) and rs1715364 (IVS3+14205T>C) (q<0.05), where the homozygous variant genotypes yielded two-fold increased risks for SLL/CLL (Table S1). TFSEARCH ( predicted that the SLC23A2 rs1715364 wild type T allele creates binding sites for the GATA-1 and GATA-2 enhancer proteins that may increase SLC23A2 expression. LD structure for SLC23A2 is shown in Figure S1. Given the size of this gene (~158 Kb), the seven SLC23A2 tag SNPs genotyped only captured 35% of its haplotype structure at the R2 = 0.94 level. Though gene coverage was low, haplotype associations were found with NHL, DLBCL and SLL/CLL risk (Table 3). A two-SNP AA haplotype (rs1776948 and rs6139587) was positively associated with NHL and SLL/CLL and an AAC haplotype derived from minor alleles for rs1715385 and rs1715364 and the wildtype allele for rs6133175 was associated with increased risk of NHL and DLBCL. However, p-values derived from SLC23A2 haplotype associations predicted no greater risks beyond those observed for the individual SNPs.

Matrix metalloproteinase (MMP)3 and MMP9 SNPs influence FL and SLL/CLL risk

A significant association was found between FL and a non-synonymous MMP3 45K>E SNP, (rs679620, A>G) (p = 3.7×10−4; q = 1.2×10−2; Table 2). Logistic regression analyses revealed that the MMP3 45EE allelotype was associated with a 70% increased risk for FL (Table S2). RESCUE ESE[13] predicted that the MMP3 45E variant introduces two exonic splicing enhancer motifs that could augment exon 2 splicing and increase MMP3 expression. This SNP, together with rs615098, located in the MMP3 promoter region, captured 44% of all common HapMap SNPs in the region at a mean R2 of 0.99.

Two non-synonymous SNPs in MMP9 (279Q>R, rs17576; 668R>Q, rs2274756) influenced SLL/CLL status (p = 7.0×10−3, q = 9.5×10−2; p = 1.0×10−2, q = 9.5×10−2, respectively; Tables 2, S2). For the 668Q allele, ESEfinder identified an extra SF2/ASF motif that could enhance exon 2 splicing and increase MMP9 expression. Interestingly, the 668R>Q SNP was in complete LD with a MMP9 promoter SNP, rs3918241 (D' = 1.0), where the variant A allele was positively associated with SLL/CLL (p = 1.0×10−2; q = 9.5×10−2; Tables 2, Table S2). TFSEARCH predicts that the rs3918241 A allele creates extra GATA-1, Oct-1, GATA-2 and GATA-X binding motifs, suggesting increased promoter activity for the variant A allele. MMP9 variation was organized into a single LD block spanning the gene and its promoter region (Figure S2). Thus, the five MMP9 SNPs genotyped captured 78% of all common variation for this region at a mean R2 of 0.98.

Sex hormone SNPs influence risk of lymphoma

The regression analyses (see methods) revealed an association between the cytochrome P450 (CYP)19A1 gene and risk for FL (p = 6.3×10−3). Eight CYP19A1 SNPs influenced risk for FL and SLL/CLL, though considering a q<0.10, only rs1870049 (IVS1+5469T>C) remained associated with disease status (p = 6.0×10−4; q = 4.9×10−2, Table 2). In logistic regression analyses, the rs1870049 homozygous variant CC genotype was associated with a three-fold increased risk for FL (Table S3). Four additional CYP19A1 SNPs, rs10046 (3′UTR, 1678T>C), rs2289105 (IVS7−79T>C), rs2899472 (IVS4−1334C>A), rs9944225 (IVS3+7029C>A) and rs4774584 (IVS1−26860G>A) influenced SLL/CLL risk (Table S2) and rs2008691 (IVS1−13201A>G) was positively associated with the group of “other” lymphoma subtypes (Table S3). CYP19A1 SNP associations did not differ by sex (data not shown). We also found a CYP19A1 five-SNP high-risk FL haplotype (p = 6.5×10−4, q = 2.9×10−2, Table 3) that included the rs1870049 variant C allele. A haplotype with the rs1870049 T allele did not influence FL risk, suggesting that the C allele was driving the haplotype association. Analysis of LD structure throughout the region revealed that the haplotype lies within an 18 kb region of relatively high LD spanning the 5′ UTR of CYP19A1 (Figure S3). The 14 genotyped SNPs, uniformly distributed throughout the gene, captured 66% of CYP19A1 common genetic variation at a mean R2 of 0.96.

An intronic SNP, rs3020314 (IVS4+5029T>C), in the estrogen receptor alpha (ESR1) gene was associated with risk of FL (p = 1.6×10−3, q = 7.0×10−2, Table 2). Logistic regression analyses revealed that the rs3020314 homozygous variant IVS4+5029CC genotype was associated with a 58% reduced risk of FL (Table S4). This SNP was in almost complete LD (D'>0.9,R2>0.90) with three other SNPs that form part of a 14 kb haplotype block of 9 common SNPs located between exons 4 and 5. Homozygous variant genotypes for three additional ESR1 SNPs, rs488133 (promoter−3604C>T), rs532010 (IVS1+1419T>C) and rs827423 (IVS1−7534T>C) also were associated with FL (Table S4). These three SNPs formed part of a major haplotype block that extended 43 kb from the 5′ upstream region to exon 2, which includes another 32 HapMap genotyped SNPs. LD structure analysis (Figure S4) revealed that some of the 11 ESR1 SNPs genotyped in this study are located at the borderlines of blocks close to recombination hotspots. These SNPs, at a density of one every ~28 Kb, captured only 25% of the common variation in ESR1.

Other genetic associations with NHL

In the cholesterol genes, considering q-values<0.10, we found an association between NHL and the nuclear receptor subfamily 1, group H, member 2 (NR1H2) gene (p = 1.8×10−3, q = 3.5×10−2) in the regression analysis. NR1H2 is involved in cholesterol homeostasis and bile acid metabolism. For FL, we found gene and SNP (20T>C) associations with farnesyl-diphosphate farnesyltransferase 1(FDFT1), which plays a key role in cholesterol biosynthesis (gene, p = 4.1×10−3; q = 7.8×10−2; rs1047643, p = 8.4×10−4, q = 5.2×10−2, Table 2). According to ESEfinder, the rs1047643 C allele introduces an extra SRp55 motif that may influence pre-mRNA processing or splicing efficiency. This SNP was included in a FL high-risk FDFT1 haplotype (p = 1.6×10−3, q = 6.4×10−2, Table 3), though the p-value was no more significant for the haplotype than for the rs1047643 SNP association. No other FDFT1 SNP associations were observed. The four FDFT1 SNPs genotyped captured 52% of common variation in FDFT1 (R2 = 0.92).

In cell signaling genes, we found an association with NHL and the BCL6 gene (p = 3.4×10−3, q = 7.8×10−2), though no single SNP significantly influenced risk. We also found that an intronic SNP rs760462 (IVS3+2100G>A) in integrin beta 2 (ITGB2) influenced FL status (p = 9.7×10−4, q = 7.2×10−2, Table 2), where the 2100AA genotype increased FL risk three-fold (Table 3). No other associations were found with ITGB2, however the six SNPs genotyped captured only 21% of the variation in the ITGB2 gene (R2 = 0.99).

For SLL/CLL, the variant allele for rs4149244 (IVS9+1573C>T) in the epoxide hydrolase 2 (EPHX2) gene was positively associated with disease risk (p = 1.2×10−3, q = 4.5×10−2, Table 2). TFSEARCH predicts that the T allele introduces a CdxA binding site, suggesting that this allele may act to enhance EPHX2 gene expression. Gene coverage for the four EPHX2 SNPs genotyped was 58% (R2 = 0.96). The rs414924 T allele also comprised a two-SNP haplotype that was associated with FL risk (p = 1.0×10−3, q = 2.0×10−2, Table 3), though again the haplotype was no more likely to contribute to risk than the rs4149244 SNP. Finally, a five-SNP haplotype in the apoptosis/signaling gene, BCL2-interacting killer (BIK) was associated with an increased risk for SLL/CLL (Table 2), though no SNP associations were detected. Gene coverage for the seven BIK SNPs genotyped was 75% (R2 = 0.95).

Validation Study

We next tested for association of significant findings in an independent case-control study of lymphoma in Germany that comprised 494 NHL cases and 494 age and sex-matched controls. An additional 111 Hodgkin lymphoma cases and 111 matched controls also were analyzed. The criterion used for selection of SNPs in Table 2 was that there was >1 associated signal in a particular pathway and in more than one gene. When SNPs were closely linked (R2>0.8), only one SNP was chosen for validation. Eight SNPs were genotyped in the SLC23A1 (rs6596473, rs11950646) and SLC23A2 (rs1715364), MMP3 (rs679620), MMP9 (rs3918241), CYP19A1 (rs1870049, 2899472) and ESR1 (3020314) genes (Table 4).

Table 4
Odds ratios (OR) and 95% confidence intervals (CI) for associated SNPs in the SLC23A1, SLC23A2, MMP3, MMP9, CYP19A1 and ESR1 genes in the San Francisco Bay Area NHL and German studies.

The SLC23A1 rs6596473 and rs11950646 SNPs associated with increased FL risk in the U.S. study (OR = 1.8, 95% CI = 1.2–2.9, p = 8.8×10−3 and OR = 1.8, 95% CI = 1.2–2.8, p = 8.3×10−3, respectively), did not influence risk of FL in the German study. However, in the German study, risk estimates were elevated for CLL (OR = 3.4, 95% CI = 0.91–13, p = 7.0×10−2; OR = 2.6, 95% CI = 0.85–7.9, p = 9.0×10−2, respectively) and ORs were reduced for DLBCL (OR = 0.39, 95% CI = 0.13–1.1, p = 8.0×10−2 and OR = 0.44, 95% CI = .17–1.1, p = 9.0×10−2, respectively). Consistent with the U.S. study, the SLC23A2 rs1715364 CC genotype was associated with an elevated OR for SLL/CLL (U.S.: OR = 1.9, 95% CI = 1.2–3.1, p = 5.5×10−3; Germany: OR = 1.8, 95% CI = 0.77–4.1, p = 1.8×10−1), though the confidence interval included one.

For MMP3 rs679620 where the US study found an increased risk for FL associated with the homozygous variant genotype (OR = 1.7, 95% CI = 1.1–2.6, p = 1.9×10−2), the German study found no associations with NHL. However, the homozygous variant genotype was associated with an increased risk for HL (OR = 2.3, 95% CI = 0.99–5.4, p = 5.4×10−2). Consistent with findings in the U.S. study, homozygosity for the MMP9 rs3918241 variant resulted in an elevated OR for NHL approaching significance (US: OR = 1.7, 95% CI = 0.95–3.0, p = 7.4×10−2; Germany: OR = 2.5, 95% CI = 0.80–8.1; p = 1.0×10−1), however, there was no increased risk of CLL with the variant allele as found in the U.S. study.

In the German study, CYP19A1 rs1870049 failed Pyrosequencing genotyping and no differences in rs2899472 allele frequencies were observed between cases and controls. The most robust findings were for ESR1 rs3020314 where, as in the US study, heterozygous and homozygous variant genotypes were inversely related with FL risk in the German study (U.S.: heterozygous OR = 0.64, 95% CI = 0.46–0.88, p = 6.2×10−3; homozygous variant OR = 0.42, 95% CI = 0.23–0.77, p = 5.1×10−3; Germany: heterozygous OR = 0.47, 95% CI = 0.22–1.0, p = 5.7×10−2; homozygous variant OR = 0.24, 95% CI = 0.06–.094, p = 4.0×10−2). The ESR1 rs3020314 variant allele also was associated with HL in the German study (OR = 0.61, 95% CI = 0.35–1.1, p = 7.7×10−2).


Here, we report that SNPs in the vitamin C receptor and matrix metalloproteinase gene families and in genes involved in estrogen responsiveness may be risk alleles for NHL. Our main strategy was to use a hypothesis-driven approach to further examine genetic findings based on our earlier studies by genotyping SNPs in related pathways or in genes that share common biological functions to those previously associated with NHL [2], [4][12]. We selected potential functionally relevant SNPs including non-synonymous coding SNPs, and 1–15 haplotype tagging SNPs were genotyped based on known patterns of LD using data available from the HapMap CEU population. These tag SNPs captured an average 39% of all common SNPs (R2>0.8) in the studied genes and their flanking regions. We note that the low coverage in some genes may have precluded us from detecting additional NHL risk alleles. Analyses were restricted to white non-Hispanics with ancestry data to reduce population stratification. In 959 cases and 1,049 controls, we identified 14 unique SNPs in eight pathways and seven haplotypes associated with NHL after correction for multiple testing at the adjusted q<0.10 (estimated FDR) level, and seven SNPs and eight haplotypes remained significantly associated with lymphoma risk at the q<0.05 level. We utilized in silico algorithms to predict likely functional effects for associated SNPs. Next, in an independent case-control study of lymphoma, we performed a small replication analysis of associated SNPs that were chosen based on the criteria that multiple SNP-NHL associations were observed in the same gene or gene family. In doing so, we found a strong inverse association between FL and an intronic SNP in ESR1 in both study populations. Other associations were observed between risk of lymphoma subtypes and SLC23A1, SLC23A2, and MMP9 SNPs. Although several of these associations occurred within different NHL subtypes across the two study populations, the overall consistency of associations with specific genes provides some support that these genes might be associated with NHL risk. Replication in additional populations and investigation of molecular similarities and differences of NHL subtypes such as through tumor microarray studies will be needed to clarify these findings.

Vitamin C and risk of NHL

The associations we identified between NHL and SNPs in the vitamin C receptor genes may provide indirect evidence to support epidemiological data showing inverse associations between NHL risk and intakes of fruits and vegetables, which are primary sources of dietary vitamin C [14], [15]. Here, we observed that homozygous variant genotypes for two closely linked intronic SLC23A1 SNPs, rs6596473 and rs11950646, were positively associated with FL in the U.S. study. However, in the German study, these genotypes were inversely and positively associated with DLBCL and SLL/CLL risk, respectively, suggesting that further study will be needed to determine the relevance, if any, of these findings. In the U.S. analysis of seven SLC23A2 SNPs, we found that variant alleles for rs6133175 and rs1715364 increased SLL/CLL risk and rs1776948 for FL and SLL/CLL. In genotyping SLC23A2 rs1715364 in the German study, the variant CC genotype was associated with an elevated OR for CLL, consistent with findings from the U.S. study. The haplotype analyses supported the genotype results but added nothing further to the individual SNP associations.

SLC23A1 and 2 encode two vitamin C transport proteins, SVCT1 and 2, respectively[16] that exhibit non-redundant functions[17]. The lethality of slc23a1 and slc23a2 mouse knockouts highlights the relevance of these proteins in maintaining vitamin C homeostasis [17], [18]. In humans, SVCT1 is expressed in kidney, intestinal, hepatic and placental tissues and is critical for vitamin C absorption and reabsorption, whereas SVCT2 is expressed in most tissues and is essential for vitamin C bioaccumulation[19]. Thus, the influence of SLC23A1 and SLC23A2 genetic variation on NHL risk suggests that both vitamin C uptake and storage may be involved in the pathogenesis of lymphoma. In silico models suggest that the SLC23A2 rs1715364 T allele may augment SLC23A2 expression and increase vitamin C bioavailability, which will need to be tested in functional studies. By extension, if the variant C allele results in lower SLC23A2 expression, then the increased NHL risk associated with the variant C allele might suggest that reduced vitamin C bioavailability increases lymphoma risk.

As with the majority of SNPs investigated in this study, SLC23A1 and SLC23A2 SNPs were chosen as tag SNPs to gain gene coverage to capture potential causal variants. Though SLC23A1 and SLC23A2 have extensive sequence homology, SLC23A2 is a 10-fold larger gene with a much more complex locus architecture (Figure S1). Whereas, two SLC23A1 tag SNPs captured 75% of its haplotype structure, seven SLC23A2 tag SNPs captured only 35% of its haplotype structure at an R2>0.94, warranting the need for additional fine mapping of the SLC23A2 gene.

In pooled analyses of two NHL case-control studies, we previously reported that SNPs that impair anti-oxidant capacity may influence lymphomagenesis[5]. Vitamin C is an essential enzyme cofactor in reactions catalyzed by several metal-dependent oxygenases[20]. It also functions as a scavenger of reactive oxygen species[21] and, thus, may play an important role in preventing oxidative stress induced DNA damage by quenching free radical formation. Other possible modes of action may involve vitamin C's anti-tumorigenic role in supporting proper collagen formation and matrix stabilization[22]. Further, in an in vitro study of T-cell leukemia, vitamin C abrogated cell proliferation and induced apoptosis through p53, p21, Bcl-2 and Bax modulation[23]. Additional studies will be needed to clarify the role of vitamin C in the pathogenesis of NHL.

MMP SNPs influence NHL, FL and SLL/CLL risk

Extracellular matrix degradation is involved in many normal physiological processes including development, growth and tissue repair. However, high MMP-3 and -9 proteolytic enzyme levels also may support tumor cell proliferation, angiogenesis, invasion and metastasis[24], [25] by degrading type IV collagen, penetrating the basement membrane and invading surrounding tissues[26]. High MMP expression is found in lymphoid cells and malignantly transformed cells[27]. Recent studies highlight the relevance of MMPs and angiogenesis in lympho-hematological malignancies such as B-cell lymphoma, multiple myeloma, CLL and acute lymphocytic and myeloid leukemias[28]. While the MMP family has been related to cancer invasion and metastases, our study suggests that variation in MMP genes also plays a role in lymphomagenesis. Here we found that homozygosity for a non-synonymous MMP3 45K>E SNP, rs679620, was associated with increased risk of FL in the U.S. study, and for HL in the German study. It is biologically feasible that the MMP3 45EE allelotype may influence lymphomagenesis since it is predicted by RESCUE ESE to enhance exon 2 splicing, which may increase MMP3 expression. In the U.S. study, increased FL and SLL/CLL risk also was found in those possessing variant alleles for an MMP9 promoter SNP, rs3918241, and two non-synonymous MMP9 SNPs, rs17576 (279Q>R) and rs2274756 (668R>Q). According to ESEfinder, the rs3918241 variant A allele may enhance MMP9 promoter activity and the rs2274756 variant 668Q allelotype creates an extra binding motif that may enhance exon 2 splicing. Thus, the rs3918241 and rs2274756 risk alleles for SLL/CLL are predicted to increase MMP9 expression. We tested the association of rs3918241 in the German study and found an elevated though non-significant increase in NHL risk, consistent with the U.S. study, though there was no evidence of an association with FL and CLL as in the U.S. study. The 4% frequency of the homozygous variant genotype may have precluded us from observing these associations, which will require the study of this SNP in larger populations. Since rs3918241 and rs2274756 are in complete LD, additional studies also will be needed to discern which, if any, of these functionally relevant SNPs may be causally related to NHL.

In earlier studies, we found associations with NHL and SNPs in energy regulation genes such as leptin (LEP), leptin receptor (LEPR) and ghrelin (GHRL)[9], [11], [12]. We hypothesized that leptin may promote lymphomagenesis through mitogenic and anti-apoptotic effects on B-cell populations through LEPR-mediated upregulation of BCL-2, MMPs, and tissue inhibitors of metalloproteinases. The present study lends further credence to the role of MMPs in lymphomagenesis. Interestingly, our study found multiple genetic associations with LEPR, GHRL and SNPs in other energy regulatory genes, though the adjusted q-values were >0.10. In light of these findings, we plan to further study the relationship between energy regulation and MMP SNPs in lymphomagenesis.

ESR1 and CYP19A1 gene variants in association with lymphoma risk

Previously, we reported a CYP17A1−34T>C high-risk genotype for DLBCL in two independent case-control studies of NHL[7], [10]. CYP17A1 produces key enzymes that catalyze the synthesis of estrone and dehydroepiandrosterone, the precursors for estrogen and testosterone. These findings prompted us to explore SNPs in additional genes involved in steroidogenesis and in estrogen and testosterone uptake and metabolism. Unfortunately, in the present study, we were unable to obtain genotype calls for CYP17A1−34T>C due to poor genotype clustering. Additionally, we had insufficient SNP coverage across ESR1 since some tags were located in the lower LD regions (Figure S4). Nonetheless, in the U.S. study, our results revealed associations with NHL and multiple SNPs in the CYP19A1 and ESR1 genes. CYP19A1 encodes for aromatase, a key enzyme in estrogen production[29] that converts androstenedione and testosterone to estradiol and estrone, respectively. ESR1 encodes estrogen receptor 1 (ERα) that plays an important role in mediating estrogen action in target tissues. Moreover, ERα signaling regulates estrogen production by direct modulation of the CYP19A1 promoter[30]. Here we found seven CYP19A1 SNPs associated with forms of NHL and one, rs1870046, remained associated with FL (q<0.10). In the replication study, rs1870046 was not successfully genotyped and no associations were found between SLL/CLL and rs2899472. Given the extent of associated alleles in the U.S. study and that only one SNP was investigated in the German study, further investigation of associated alleles in the CYP19A1 gene is warranted.

The most significant finding in this study was the inverse association between FL and the C allele in ESR1 rs3020314 that was corroborated with FL and HL in the German study. Since estrogen exerts a biphasic dose-dependent effect on the immune system, this association may be biologically plausible. Notable but somewhat complex hormonal-driven associations have been reported between other ESR1 gene variants and breast cancer, spontaneous abortion, osteoporosis[31], age of menarche[32], genital abnormalities in men[33] and fertility in men and women. ESR1 is also expressed by thymocytes and B- and T-cells[34] and Esr1 knockout mice exhibit compromised thymic[35] and B-cell development[36], insulin resistance, impaired glucose tolerance and obesity[37]. Several epidemiological studies have reported that oral contraceptive use and hormone replacement therapy are associated with lower lymphoma risk [7], [38], [39]. Thus, large pooled gene-environment studies of ESR1 gene variants and measures of hormone levels or history of hormone use may be useful to further explore the role of hormones in the pathogenesis of lymphoma.

Lack of additional genetic associations

Few, if any, additional SNP and haplotype associations were observed in other pathways investigated in this study. For instance, BIK and EPHX2 were the only genes associated with NHL in the immunity and xenobiotic metabolism/oxidative stress pathways, respectively, and no NHL-genetic associations were observed in the sunlight/vitamin D or folate metabolism pathways. Previous studies suggest that SNPs in these pathways may influence lymphoma risk [2]. Our inability to validate these associations here may be that our study was underpowered to detect significant associations after adjusting for multiple comparisons; that our study had incomplete gene coverage; or that other genetic mechanisms such as copy number variation influenced the results. Thus, negative results reported here do not dismiss the potential relevance of these SNPs/pathways in lymphomagenesis, which will require further study in larger populations.

Perspectives and Future Directions

This study reports novel associations with common variants in ESR1 and in the vitamin C receptor and matrix metalloproteinase gene families. Of particular importance is the replication of a significant association with FL and the ESR1 rs3020314 SNP in a second independent study of lymphoma, which suggests that estrogen bioavailability may be relevant in the pathogenesis of lymphoma. We also believe that the genetic associations that did not reach statistical significance in the replication study are noteworthy, but due to the sample size, may have been underpowered to detect subtle allelic associations. Given the fact that we adjusted for multiple comparisons for pathways and did not adjust for all comparisons with all NHL subtypes, there still is an increased chance for type 1 errors and it is possible that some of our results are false positive associations. These findings will need further validation in larger studies. Owing to the small number of SNPs analyzed in the replication study, investigation of associated SNPs identified in the U.S. population will be ongoing. Further replication, fine mapping and functional studies will also be needed to determine the causal variants, and pooled data from consortium studies will be necessary to discern more modest effects of low penetrance SNPs or for rare alleles. The verification of these findings could have important implications by providing mechanistic clues in the etiology of lymphoma and in identifying useful therapeutic targets. Interestingly, most associations were found for FL and SLL/CLL and few were identified for DLBCL. This may be due to the heterogeneity of the disease since GC- and post-GC types of DLBCL exhibit significant molecular diversity at the gene expression and phenotype levels [45][47] that may require further stratification to detect underlying DLBCL risk alleles. It may also be that our hypothesis-driven approach failed to interrogate the appropriate pathways. To overcome this, we are currently performing a whole genome scan that will potentially lead to the identification of novel DLBCL susceptibility alleles.

Materials and Methods

U.S. Study Population

Cases were identified within ~1 month of diagnosis by the Northern California Cancer Center's rapid case ascertainment. For this interim analysis which includes only a subset of the study participants, eligible patients were diagnosed with incident NHL from 10/2001–10/2005. All eligible patients were 20–84 years old and residents of one of six Bay Area counties at the time of diagnosis. There were 1704 eligible cases interviewed between 10-2001 and 10-2005. An additional 452 cases died prior to initial contact, 272 could not complete an interview in English, 70 had physician indicated contraindications to contact, 161 could not be located/moved, 155 were too ill and 360 refused to participate. Pathology reports and diagnostic materials were obtained from the diagnosing pathology department and re-reviewed by an expert pathologist to confirm the diagnosis and to provide a consistent classification of NHL subtypes using the REAL/WHO[40] lymphoma classification. Control participants were identified by random digit dial supplemented by random sampling of Centers for Medicare & Medicaid Services lists for those ≥65 years old. Eligibility criteria for controls were identical to cases with the exception of NHL diagnosis. Control participants were frequency matched to cases by age within five years, sex and county of residence. Among 1939 eligible random-digit-dial controls, 1313 completed in-person interviews, 85 were too ill and 541 refused to participate. Among 1475 eligible Medicare controls, 768 completed in-person interviews, 97 were too ill and 610 refused to participate. Demographic and risk factor data including HIV status, self-reported race and ethnicity, and paternal and maternal ancestry, were collected from all study participants during in-person interviews. No proxy interviews were conducted. The median time from diagnosis to venipuncture was 6.5 months. Blood and/or buccal cell specimens were collected for eligible participants who agreed to biospecimen collection (87% of cases, 88% of controls).

German Study Population

Detailed information regarding the design of this population-based case-control study has been published elsewhere[41]. Briefly, the study was carried out from 1999–2002 in six regions of Germany among 18–80 year-old adults. The study included 710 cases of DLBCL, FL, CLL, multiple myeloma (MM), MALT lymphoma, T-cell lymphoma and Hodgkin lymphoma (HL) and 710 controls individually matched by sex, age (+/−1 year of birth) and study region. Of these, there were 494 NHL cases and 111 HL cases and matched controls. The median time from diagnosis to blood collection was 27 days and the response rate for cases and controls was 87% and 44%, respectively. Analyses for all NHL did not include MM or HL. For subtype-specific analyses, only FL, DLBCL, CLL and HL were considered. Incident lymphoma cases were recruited from hospitals and office-based physicians involved in diagnosis and treatment of lymphoma in the study regions, interviewed by trained interviewers and provided blood samples. Cases were classified according to the WHO system based on pathology reports and 20% of cases were re-reviewed by a team of lymphoma pathologists.

The U.C. San Francisco Committee on Human Research, the U.C. Berkeley Committee for Protection of Human Subjects, and the Ethics Committee, European Union, approved study protocols. All study participants provided written informed consent prior to interview and biospecimen collection.

Pathway, Gene and SNP Selection

Eight pathways or groups of genes with common biological function in a total of 146 genes (Table 1) were chosen to investigate a priori hypotheses (Table S5) in steroidogenesis, immune function, cell signaling, sunlight exposure/vitamin D, xenobiotic metabolism/oxidative stress, energy balance, cholesterol synthesis/metabolism and diet (folate and vitamin C uptake and/or metabolism) pathways. 1–15 SNPs in each gene and a total of 768 SNPs were selected using a minimum minor allele frequency criterion of ≥10%, based upon HapMap data release 12 (~October 2004) for CEU Caucasian samples. Additional coding and regulatory SNPs, especially those with non-synonymous amino acid changes, were identified using Ensembl, NCBI, and SNPper databases [42][47]. For all genes, haplotype tagging SNPs were selected using Haploview[48] after downloading SNP genotyping information available for the CEPH population (Utah residents with ancestry from northern and western Europe) from the HapMap database (Build 34)[43], [49]. SNP validation scores between 0 and 1 were generated using the Illumina Assay Design Tool to estimate the likelihood that SNPs would be successfully genotyped. Through an iterative process, we dropped SNPs with validation scores <0.6 and replacement haplotype tagging SNPs were identified using Haploview until a complete panel of 768 SNPs were chosen. This SNP panel (GS0006402-OPA) was custom designed by Illumina.


For the U.S. study, genotyping was performed using the Illumina GoldenGate assay on a BeadStation 500G Genotyping System. Genotype calls were made using the GenCall version software package (Illumina, Inc., San Diego, CA). Summary files were generated using the GTS Reports version All SNPs were examined for cluster separation using Illumina quality scores that were automatically generated by the software. Poorly performing loci were excluded, designated with a GenTrain score <0.4 or a cluster separation score <0.6. SNPs were further excluded for scattered or overlapping clusters, for controls not in Hardy-Weinberg equilibrium, for low frequency (<0.05), or for DNA samples with low intensity. Individual DNA samples were excluded from the study if there was no signal across SNPs. Genotyping of the German study was performed by Pyrosequencing™ (Biotage, Uppsala, Sweden)[50].

Quality assurance

To validate genotyping results, each 96-well plate contained 90 randomized study samples, two randomized replicate study samples placed side-by-side and three laboratory control DNA samples. DNA also was genotyped from six CEPH families (Coriell Institute, Camden, NJ). Genotyping of 10 SNPs was performed for validation purposes using TaqMan® genotyping assays and the ABI Prism 7700 Sequence Detection System (Applied Biosystems, Foster City, CA).

Coverage estimation

Coverage which indicates the fraction of common HapMap markers successfully tagged by the set of genotyped SNPs was estimated using the pairwise tagging method in Haploview[48], which is based on the Tagger algorithm ( SNPs genotyped were forced into a tag list and neighboring HapMap SNPs were explicitly prohibited from being chosen as tags. We specified that all HapMap markers being captured by the set of tags should be correlated at R2≥0.8 with at least one marker in the set. HapMap markers with MAF<0.05 were excluded.

Statistical Analyses

All statistical analyses were restricted to HIV-negative participants of white non-Hispanic origin (959 cases, 1,049 controls). Because self identified ethnicity and ancestry correlates very well with genetic structure [51], using this information greatly minimizes the potential effect of underlying population stratification on the risk estimates. Hardy-Weinberg equilibria for genotype frequencies among controls were determined by exact tests using a Markov-chain method (Genepop v3.4).

SNP analysis

Exact tests were used to generate p-values to test associations of case-control status and SNP associations. For each pathway, P-values were corrected for multiple comparisons (q-values) using the false discovery rate[52]. For all NHL, and for each subtype tested, the following total number of SNP tests were performed in each pathway: steroidogenesis, 88; immunity, 75; cell signaling, 88; sunlight/vitamin D, 53; xenobiotic/metabolism, 40; energy balance, 155; cholesterol, 62; and diet, 26. Fifty intergenic SNPs were excluded from the pathway analyses. A threshold of q<0.10 was used in reporting the most relevant findings in Tables 2 and and3.3. ORs and 95% CIs as estimates of the relative risk were computed adjusted by age and sex from unconditional logistic regression [53] without taking into account multiple testing. In genotype analyses, the wild-type category (chosen either as the most common homozygous genotype or arbitrarily if the same) was the reference group. The odds ratios from the unconditional logistic models were adjusted for the frequency-matched variables age in 5-year groups and sex. Analyses were performed on all NHL cases and then stratified by DLBCL, FL, SLL/CLL and “others.” We did not adjust for all multiple comparisons for all SNPs in all types of lymphoma.

Gene analysis

We tested for multi-way interactions within sets of SNPs spanning a gene by fitting the genotypes to a linear regression model and applying a t-test to the case-control distributions. P-values were generated by permuting cases and controls, repeating the experiment 10,000 times per gene. Results were adjusted for multiple hypotheses testing using the FDR for each functional group/pathway. For “all NHL” and for each subtype tested, the following total number of gene tests were performed: steroidogenesis, 22; immunity, 22; cell signaling, 19; sunlight/vitamin D, 12; xenobiotic/metabolism, 14; energy balance, 38; cholesterol, 19; and diet, 9.

Haplotype analysis

SNPs were partitioned into blocks of limited diversity using HAP[54], which uses criteria similar to previous methods[55]. For each block of limited diversity, genotypes were phased using PHASE[56]. For each common haplotype (MAF>5%), a χ2-test with one degree of freedom was performed with counts of haplotype occurrences in cases and controls. Results were adjusted for multiple hypotheses using the FDR for each functional group/pathway. For “all NHL” and for each subtype tested, the following total number of haplotype tests were performed: steroidogenesis, 44; immunity, 38; cell signaling, 45; sunlight/vitamin D, 31; xenobiotic/metabolism, 19; energy balance, 91; cholesterol, 36; and diet, 12.

Replication study

Conditional logistic regression analysis was used to estimate ORs and 95% CIs comparing cases (all NHL cases combined –which excluded MM and HL- and individual FL, DLBCL and SLL/CLL subtypes and separate analyses for HL) to respective controls in association with genotypes, adjusting for gender, age and study region.

In silico algorithms

The FASTSNP web server[57] was used to interrogate 11 external web servers including RESCUE-ESE[13], ESEfinder[58] and PolyPhen[59] to predict the functionality of SNPs.

Supporting Information

Figure S1

LD plot characterizing haplotype blocks in SCL23A2. The 7 SLC23A2 SNPs genotyped (rs2298174, rs1629176, rs1715385, rs6133175, rs1715364, rs1776948 and rs6139587 highlighted in green) captured 35% of genetic variation for the SLC23A2 locus. Detailed description of LD plots shown in Figures S1, S2, S3 and S4. R2 values are indicated in percentages within squares in the LD plot, where red indicates regions of high LD and white represents regions of low LD. Blocks without numbers indicate r2 = 1 (100%). At the top, figures display a track with HapMap genotyped SNPs and genes within the genomic region. SNPs that have been genotyped in this study are highlighted in green and marked with lines from the SNP position to the LD chart. LD blocks were generated using the Confidence Intervals method included in Haploview4.0 (, that creates 95% confidence bounds on D' considered to be in strong LD where 95% of the comparisons made are informative. Genotype data was retrieved from HapMap release 22/phaseII (NCBI B36 assembly), for the CEPH (Centre d'Etude du Polymorphisme Humain) from Utah (CEU). Coverages were obtained using the tagging method included in Haploview as described in the manuscript.

(59.29 MB DOC)

Figure S2

LD plot characterizing haplotype blocks in MMP9. The 5 MMP9 SNPs genotyped (rs6094237, rs4810482, rs3918241, rs17576 and rs2274756), highlighted in green, captured 15 out of 19 variations (78%) with r2>0.8. There is, therefore, good coverage of the MMP9 gene.

(2.58 MB DOC)

Figure S3

LD plot characterizing haplotype blocks in CYP19A1. The 14 CYP19A1 SNPs genotyped (rs10046, rs2289105, rs2899472, rs9944225, rs4775936, rs2008691, rs4774584, rs7181886, rs936306, rs1004982, rs2470146, rs2470144, rs1870049 and rs752760 highlighted in green) captured 66% of genetic variation for CYP19A1.

(72.13 MB DOC)

Figure S4

LD plot characterizing haplotype blocks in ESR1. The 11 ESR1 SNPs genotyped (rs2881766, rs488133, rs532010, rs827423, rs1709182, rs1913474, rs3020314, rs722208, rs3020411, rs2813545 and rs910416 highlighted in green) captured 66 out of 262 SNPs at r2>0.8, providing 25% gene coverage.

(70.04 MB DOC)

Table S1

Odds ratios (OR) and 95% confidence intervals (CI) for associated SLC23A1 and SLC23A SNPs in the San Francisco Bay Area NHL study.

(0.11 MB DOC)

Table S2

Odds ratios (OR) and 95% confidence intervals (CI) for associated MMP1, 3 and 9 SNPs in the San Francisco Bay Area NHL study.

(0.14 MB DOC)

Table S3

Odds ratios (OR) and 95% confidence intervals (CI) for associated CYP19A1 SNPs in the San Francisco Bay Area NHL study.

(0.16 MB DOC)

Table S4

Odds ratios (OR) and 95% confidence intervals (CI) for associated ESR1 SNPs in the San Francisco Bay Area NHL study.

(0.12 MB DOC)

Table S5

A priori hypotheses tested.

(0.06 MB DOC)


Competing Interests: The authors have declared that no competing interests exist.

Funding: This work was supported by NIH grant numbers CA122663 (CFS), CA104862 (MTS) and CA45614, CA89745, CA87014 (EAH), and the German José Carreras Leukaemia foundation (DJCLS_R04/08) (AN).


1. Alexander DD, Mink PJ, Adami HO, Chang ET, Cole P, et al. The non-Hodgkin lymphomas: a review of the epidemiologic literature. Int J Cancer. 2007;120(Suppl 12):1–39. [PubMed]
2. Skibola CF, Curry JD, Nieters A. Genetic susceptibility to lymphoma. Haematologica. 2007;92:960–969. [PMC free article] [PubMed]
3. Rui L, Goodnow CC. Lymphoma and the control of B cell growth and differentiation. Curr Mol Med. 2006;6:291–308. [PubMed]
4. Forrest MS, Skibola CF, Lightfoot TJ, Bracci PM, Willett EV, et al. Polymorphisms in innate immunity genes and risk of non-Hodgkin lymphoma. Br J Haematol. 2006;134:180–183. [PubMed]
5. Lightfoot TJ, Skibola CF, Smith AG, Forrest MS, Adamson PJ, et al. Polymorphisms in the oxidative stress genes, superoxide dismutase, glutathione peroxidase and catalase and risk of non-Hodgkin's lymphoma. Haematologica. 2006;91:1222–1227. [PubMed]
6. Lightfoot TJ, Skibola CF, Willett EV, Skibola DR, Allan JM, et al. Risk of non-Hodgkin lymphoma associated with polymorphisms in folate-metabolizing genes. Cancer Epidemiol Biomarkers Prev. 2005;14:2999–3003. [PubMed]
7. Skibola CF, Bracci PM, Paynter RA, Forrest MS, Agana L, et al. Polymorphisms and haplotypes in the cytochrome P450 17A1, prolactin, and catechol-O-methyltransferase genes and non-Hodgkin lymphoma risk. Cancer Epidemiol Biomarkers Prev. 2005;14:2391–2401. [PubMed]
8. Skibola CF, Forrest MS, Coppede F, Agana L, Hubbard A, et al. Polymorphisms and haplotypes in folate-metabolizing genes and risk of non-Hodgkin lymphoma. Blood. 2004;104:2155–2162. [PubMed]
9. Skibola CF, Holly EA, Forrest MS, Hubbard A, Bracci PM, et al. Body mass index, leptin and leptin receptor polymorphisms, and non-hodgkin lymphoma. Cancer Epidemiol Biomarkers Prev. 2004;13:779–786. [PubMed]
10. Skibola CF, Lightfoot T, Agana L, Smith A, Rollinson S, et al. Polymorphisms in cytochrome P450 17A1 and risk of non-Hodgkin lymphoma. Br J Haematol. 2005;129:618–621. [PubMed]
11. Skibola DR, Smith MT, Bracci PM, Hubbard AE, Agana L, et al. Polymorphisms in ghrelin and neuropeptide Y genes are associated with non-Hodgkin lymphoma. Cancer Epidemiol Biomarkers Prev. 2005;14:1251–1256. [PubMed]
12. Willett EV, Skibola CF, Adamson P, Skibola DR, Morgan GJ, et al. Non-Hodgkin's lymphoma, obesity and energy homeostasis polymorphisms. Br J Cancer. 2005;93:811–816. [PMC free article] [PubMed]
13. Fairbrother WG, Yeh RF, Sharp PA, Burge CB. Predictive identification of exonic splicing enhancers in human genes. Science. 2002;297:1007–1013. [PubMed]
14. Chang ET, Smedby KE, Zhang SM, Hjalgrim H, Melbye M, et al. Dietary factors and risk of non-hodgkin lymphoma in men and women. Cancer Epidemiol Biomarkers Prev. 2005;14:512–520. [PubMed]
15. Talamini R, Polesel J, Montella M, Dal Maso L, Crovatto M, et al. Food groups and risk of non-Hodgkin lymphoma: a multicenter, case-control study in Italy. Int J Cancer. 2006;118:2871–2876. [PubMed]
16. Savini I, Rossi A, Pierro C, Avigliano L, Catani MV. SVCT1 and SVCT2: key proteins for vitamin C uptake. Amino Acids 2007 [PubMed]
17. Boyer JC, Campbell CE, Sigurdson WJ, Kuo SM. Polarized localization of vitamin C transporters, SVCT1 and SVCT2, in epithelial cells. Biochem Biophys Res Commun. 2005;334:150–156. [PubMed]
18. Sotiriou S, Gispert S, Cheng J, Wang Y, Chen A, et al. Ascorbic-acid transporter Slc23a1 is essential for vitamin C transport into the brain and for perinatal survival. Nat Med. 2002;8:514–517. [PubMed]
19. Kuo SM, MacLean ME, McCormick K, Wilson JX. Gender and sodium-ascorbate transporter isoforms determine ascorbate concentrations in mice. J Nutr. 2004;134:2216–2221. [PubMed]
20. Linster CL, Van Schaftingen E. Vitamin C. Biosynthesis, recycling and degradation in mammals. Febs J. 2007;274:1–22. [PubMed]
21. Karyotou K, Donaldson RP. Ascorbate peroxidase, a scavenger of hydrogen peroxide in glyoxysomal membranes. Arch Biochem Biophys. 2005;434:248–257. [PubMed]
22. Catani MV, Savini I, Rossi A, Melino G, Avigliano L. Biological role of vitamin C in keratinocytes. Nutr Rev. 2005;63:81–90. [PubMed]
23. Harakeh S, Diab-Assaf M, Khalife JC, Abu-el-Ardat KA, Baydoun E, et al. Ascorbic acid induces apoptosis in adult T-cell leukemia. Anticancer Res. 2007;27:289–298. [PubMed]
24. Curran S, Murray GI. Matrix metalloproteinases in tumour invasion and metastasis. J Pathol. 1999;189:300–308. [PubMed]
25. Stamenkovic I. Matrix metalloproteinases in tumor invasion and metastasis. Semin Cancer Biol. 2000;10:415–433. [PubMed]
26. Sato H, Kida Y, Mai M, Endo Y, Sasaki T, et al. Expression of genes encoding type IV collagen-degrading metalloproteinases and tissue inhibitors of metalloproteinases in various human tumor cells. Oncogene. 1992;7:77–83. [PubMed]
27. Kuittinen O, Apaja-Sarkkinen M, Turpeenniemi-Hujanen T. Gelatinases (MMP-2 and MMP-9), TIMP-1 expression and the extent of neovascularization in aggressive non-Hodgkin's lymphomas. Eur J Haematol. 2003;71:91–99. [PubMed]
28. Ribatti D, Scavelli C, Roccaro AM, Crivellato E, Nico B, et al. Hematopoietic cancer and angiogenesis. Stem Cells Dev. 2004;13:484–495. [PubMed]
29. Brodie AM, Schwarzel WC, Brodie HJ. Studies on the mechanism of estrogen biosynthesis in the rat ovary–I. J Steroid Biochem. 1976;7:787–793. [PubMed]
30. Yang C, Yu B, Zhou D, Chen S. Regulation of aromatase promoter activity in human breast tissue by nuclear receptors. Oncogene. 2002;21:2854–2863. [PubMed]
31. Greendale GA, Chu J, Ferrell R, Randolph JF, Jr, Johnston JM, et al. The association of bone mineral density with estrogen receptor gene polymorphisms. Am J Med. 2006;119:S79–86. [PubMed]
32. Stavrou I, Zois C, Chatzikyriakidou A, Georgiou I, Tsatsoulis A. Combined estrogen receptor alpha and estrogen receptor beta genotypes influence the age of menarche. Hum Reprod. 2006;21:554–557. [PubMed]
33. Watanabe M, Yoshida R, Ueoka K, Aoki K, Sasagawa I, et al. Haplotype analysis of the estrogen receptor 1 gene in male genital and reproductive abnormalities. Hum Reprod. 2007;22:1279–1284. [PubMed]
34. Haruki Y, Seiki K, Enomoto T, Fujii H, Sakabe K. Estrogen receptor in the “non-lymphocytes” in the thymus of the ovariectomized rat. Tokai J Exp Clin Med. 1983;8:31–39. [PubMed]
35. Bolstad J. [Hepatitis B: infective agent, infection transmission, infection recipient]. Sykepleien. 1976;63:1198–1200. [PubMed]
36. Thurmond TS, Murante FG, Staples JE, Silverstone AE, Korach KS, et al. Role of estrogen receptor alpha in hematopoietic stem cell development and B lymphocyte maturation in the male mouse. Endocrinology. 2000;141:2309–2318. [PubMed]
37. Gallagher CJ, Keene KL, Mychaleckyj JC, Langefeld CD, Hirschhorn JN, et al. Investigation of the estrogen receptor-alpha gene with type 2 diabetes and/or nephropathy in African-American and European-American populations. Diabetes. 2007;56:675–684. [PubMed]
38. Lee JS, Bracci PM, Holly EA. Non-Hodgkin Lymphoma in Women: Reproductive Factors and Exogenous Hormone Use. Am J Epidemiol 2008 [PMC free article] [PubMed]
39. Nelson RA, Levine AM, Bernstein L. Reproductive factors and risk of intermediate- or high-grade B-Cell non-Hodgkin's lymphoma in women. J Clin Oncol. 2001;19:1381–1387. [PubMed]
40. Jaffe ES HN, Diebold J, Müller-Hermelink HK. Annals of Oncology. Bethesda, MD: 1998. World Health Organization Classification of lymphomas: a work in progress. In: Hematopathology Section LoP, National Cancer Institute, editor. [PubMed]
41. Becker N, Deeg E, Nieters A. Population-based study of lymphoma in Germany: rationale, study design and first results. Leuk Res. 2004;28:713–724. [PubMed]
42. Clamp M, Andrews D, Barker D, Bevan P, Cameron G, et al. Ensembl 2002: accommodating comparative genomics. Nucleic Acids Res. 2003;31:38–42. [PMC free article] [PubMed]
43. Phillips C. Online resources for SNP analysis: a review and route map. Mol Biotechnol. 2007;35:65–98. [PubMed]
44. Riva A, Kohane IS. SNPper: retrieval and analysis of human SNPs. Bioinformatics. 2002;18:1681–1685. [PubMed]
45. Sherry ST, Ward MH, Kholodov M, Baker J, Phan L, et al. dbSNP: the NCBI database of genetic variation. Nucleic Acids Res. 2001;29:308–311. [PMC free article] [PubMed]
46. Smigielski EM, Sirotkin K, Ward M, Sherry ST. dbSNP: a database of single nucleotide polymorphisms. Nucleic Acids Res. 2000;28:352–355. [PMC free article] [PubMed]
47. Wheeler DL, Church DM, Lash AE, Leipe DD, Madden TL, et al. Database resources of the National Center for Biotechnology Information: 2002 update. Nucleic Acids Res. 2002;30:13–16. [PMC free article] [PubMed]
48. Barrett JC, Fry B, Maller J, Daly MJ. Haploview: analysis and visualization of LD and haplotype maps. Bioinformatics. 2005;21:263–265. [PubMed]
49. The International HapMap Project. Nature. 2003;426:789–796. [PubMed]
50. Fakhrai-Rad H, Pourmand N, Ronaghi M. Pyrosequencing: an accurate detection platform for single nucleotide polymorphisms. Hum Mutat. 2002;19:479–485. [PubMed]
51. Tang H, Quertermous T, Rodriguez B, Kardia SL, Zhu X, et al. Genetic structure, self-identified race/ethnicity, and confounding in case-control association studies. Am J Hum Genet. 2005;76:268–275. [PubMed]
52. Benjamini Y, Hochberg Y. Controling the false discovery rate: a practical and powerful approach to multiple testing. Series B (methodological). Journal of the Royal Statistical Society. 1995;57:289–399.
53. Breslow N, NE. D. Lyon: IARC; 1980. Statistical methods in cancer research. [PubMed]
54. Halperin E, Eskin E. Haplotype reconstruction from genotype data using Imperfect Phylogeny. Bioinformatics. 2004;20:1842–1849. [PubMed]
55. Zhang C, Chen K, Seldin MF, Li H. A hidden Markov modeling approach for admixture mapping based on case-control data. Genet Epidemiol. 2004;27:225–239. [PubMed]
56. Stephens M, Smith NJ, Donnelly P. A new statistical method for haplotype reconstruction from population data. Am J Hum Genet. 2001;68:978–989. [PubMed]
57. Yuan HY, Chiou JJ, Tseng WH, Liu CH, Liu CK, et al. FASTSNP: an always up-to-date and extendable service for SNP function analysis and prioritization. Nucleic Acids Res. 2006;34:W635–641. [PMC free article] [PubMed]
58. Cartegni L, Wang J, Zhu Z, Zhang MQ, Krainer AR. ESEfinder: A web resource to identify exonic splicing enhancers. Nucleic Acids Res. 2003;31:3568–3571. [PMC free article] [PubMed]
59. Ramensky V, Bork P, Sunyaev S. Human non-synonymous SNPs: server and survey. Nucleic Acids Res. 2002;30:3894–3900. [PMC free article] [PubMed]
60. Rothman N, Skibola CF, Wang SS, Morgan G, Lan Q, et al. Genetic variation in TNF and IL10 and risk of non-Hodgkin lymphoma: a report from the InterLymph Consortium. Lancet Oncol. 2006;7:27–38. [PubMed]
61. Hartge P, Lim U, Freedman DM, Colt JS, Cerhan JR, et al. Ultraviolet radiation, dietary vitamin D, and risk of non-Hodgkin lymphoma (United States). Cancer Causes Control. 2006;17:1045–1052. [PubMed]
62. Purdue MP, Hartge P, Davis S, Cerhan JR, Colt JS, et al. Sun exposure, vitamin D receptor gene polymorphisms and risk of non-Hodgkin lymphoma. Cancer Causes Control. 2007;18:989–999. [PubMed]
63. Skibola CF. Obesity, diet and risk of non-Hodgkin lymphoma. Cancer Epidemiol Biomarkers Prev. 2007;16:392–395. [PMC free article] [PubMed]
64. Holly EA, Eberle CA, Bracci PM. Prior history of allergies and pancreatic cancer in the San Francisco Bay area. Am J Epidemiol. 2003;158:432–441. [PubMed]
65. Holly EA, Lele C, Bracci PM, McGrath MS. Case-control study of non-Hodgkin's lymphoma among women and heterosexual men in the San Francisco Bay Area, California. Am J Epidemiol. 1999;150:375–389. [PubMed]
66. Fortuny J, de Sanjose S, Becker N, Maynadie M, Cocco PL, et al. Statin use and risk of lymphoid neoplasms: results from the European Case-Control Study EPILYMPH. Cancer Epidemiol Biomarkers Prev. 2006;15:921–925. [PubMed]
67. Cohen JI. HMG CoA reductase inhibitors (statins) to treat Epstein-Barr virus-driven lymphoma. Br J Cancer. 2005;92:1593–1598. [PMC free article] [PubMed]
68. Li YC, Park MJ, Ye SK, Kim CW, Kim YN. Elevated levels of cholesterol-rich lipid rafts in cancer cells are correlated with apoptosis sensitivity induced by cholesterol-depleting agents. Am J Pathol. 2006;168:1107–1118. [PubMed]

Articles from PLoS ONE are provided here courtesy of Public Library of Science