|Home | About | Journals | Submit | Contact Us | Français|
Sarcoidosis is a multi-system inflammatory disease with organ involvement that varies by race and sex. Family studies indicate that genes play a role in the etiology and extent of organ involvement in sarcoidosis. In this study, we evaluated whether 25 variants distributed in 19 genes with a known role in inflammation were associated with erythema nodosum status in 659 sarcoidosis patients and 658 controls from A Case–Control Etiologic Study of Sarcoidosis (ACCESS). We found no association with affectation status; however, a variant in the promoter of tumor necrosis factor (TNF) at position −308 was found to be associated with erythema nodosum in Caucasian sarcoidosis patients (study-wide P = 0.027). When separated by sex, a variant in intron 1 of lymphotoxin-α (LTA), a gene adjacent to TNF, was associated with erythema nodosum in female Caucasian sarcoidosis patients (study-wide P = 0.027). These DNA variants frequently occur together in Caucasians, and each variant has individually been associated with erythema nodosum in sarcoidosis patients. These results confirm that variation in the LTA/TNF gene cluster modifies a major skin manifestation of sarcoidosis and may explain the higher rate of erythema nodosum in females with sarcoidosis.
Sarcoidosis is a multi-systemic inflammatory disorder characterized by noncaseating epithelial cell granulomas. Sarcoidosis has a variable age of onset, usually 20–40 years. Inflammation is most commonly found in lungs and lymph nodes. Sarcoidosis occurs 10 times more frequently in African Americans than in Caucasians (James, 1992). However, the familial relative risk estimate for siblings and parents is considerably lower in African Americans (3.1) compared to Caucasians (16.6) (Rybicki et al., 2001a). Familial clustering suggests that shared genetic and/or environmental components, especially in Caucasians, contribute to sarcoidosis (Rybicki et al., 2001b). Indeed, the major histocompatibility locus on chromosome 6 has been implicated as an etiologic factor in this disorder (Schurmann et al., 2000). In addition, genome-wide linkage and fine mapping demonstrated that butyrophilin-like 2, a member of the immunoglobin super-family in the major histocompatibility region, was associated with sarcoidosis (Valentonyte et al., 2005). Recent work suggests butyrophilin-like 2 regulates T-cell activation by binding to a receptor expressed on activated T-cells and inhibiting T-cell proliferation (Nguyen et al., 2006).
Erythema nodosum (EN) is a major skin manifestation of sarcoidosis. It is characterized by hard, painful nodules occurring in 3–44% of patients and occurs more frequently in women and, in some studies, Caucasians (Siltzbach et al., 1974; Baughman et al., 2001; Grunewald and Eklund, 2007; 1999). Löfgren's syndrome, originally defined as sarcoidosis with presence of EN and bilateral hilar lymphadenopathy (Lofgren and Lundback, 1952a), has been associated with single nucleotide polymorphisms (SNPs) in the HLA-DRB1/LTA/TNF locus (Seitzer et al., 1997; Swider et al., 1999; Labunski et al., 2001; Mrazek et al., 2005; Grunewald and Eklund, 2007). However, the studies have not been convincingly replicated. Some were limited to small groups of patients and some did not distinguish between variants analyzed as causal for sarcoidosis and variants modifying sarcoidosis phenotype. In addition, diversity of study design and selection of controls add to the challenge of evaluating these studies.
To examine the association between the LTA/TNF locus and EN, we analyzed sarcoidosis patients and non-disease controls recruited by A Case–Control Etiologic Study of Sarcoidosis (ACCESS) on the basis of sarcoidosis affectation status alone. SNPs in 19 genes involved in inflammatory processes were genotyped and the significance for association was derived at a study-wide level. Finally, patients were analyzed in groups to determine if SNP variants accounted for higher rates of EN in women and in Caucasians.
Twenty-five SNPs in 19 genes were typed in 659 sarcoidosis patients and 658 controls recruited by ACCESS (Table 1). The genotype distribution of 23 SNPs met the Hardy–Weinberg Equilibrium (HWE) predictions in African Americans and Caucasians. SNP rs1799987 met HWE predictions in African Americans only and SNP rs333 met HWE predictions in Caucasians only. Owing to racial differences in allele frequencies, Caucasian patients were analyzed separately (24 tests) from African-American patients (24 tests). No SNPs reached significance in the Caucasian or the African-American group, even when separated by sex. Study-wide P-values and Q-values for all analyses are available in Tables S1-S3.
EN phenotype data were available for 362 Caucasian sarcoidosis patients (100% of Caucasian cases) and 294 African-American patients (98.9% of African-American cases). In these populations, 8.1% of Caucasian sarcoidosis patients had EN (5.5% of male patients; 10% of female patients). In addition, 9.5% of ACCESS African-American sarcoidosis patients were affected with EN (2.6% of male patients; 12.2% of female patients). The higher prevalence of EN in female patients in ACCESS has been reported by others, but the similar prevalence by race contrasts with other studies (Baughman et al., 2001). As no SNPs were significant factors in previous analysis of sarcoidosis disease status, no corrections for potential confounding were needed. When both sexes were analyzed together, genotypes at tumor necrosis factor (TNF) SNP 308 showed an association with EN in Caucasian patients, with an uncorrected P-value of 0.002 and study-wide P-value of 0.027 (Table 2; data for all SNPs in Tables S4 and S6). Similarly, the Q-value derived from the false discovery rate (FDR) for the association with the TNF-308 genotype is 0.048, indicating that if a P-value of 0.002 was set as a study-wide threshold, less than 5% of SNPs reaching that level would be expected to be false positives (Table 2). Odds ratio analysis revealed that Caucasian sarcoidosis patients with the TNF-308 AA genotype had an 8-fold higher risk of developing EN compared to patients with the GG genotype (homozygous odds ratio 8.182 (2.449, 27.339); Table 2).
When all SNPs were tested for association in patients with EN according to sex, the TNF-308 genotype showed an association with EN in Caucasian women (P = 0.002). This association was just below the threshold for study-wide significance using the permutation method. The Q-value derived by FDR was 0.024, indicating a low probability that the observed association was a false positive (Table 3). In addition, the homozygous odds ratio for TNF-308 was 10.667 (2.495, 45.603), indicating an over 10-fold higher risk of EN in females with the AA genotype compared to the GG genotype. This is consistent with the homozygous odds ratio of 8.182 in Caucasians as a whole.
Interestingly, Caucasian women with EN also showed a strong association with genotypes at a SNP in lymphotoxin-α (LTA), a gene immediately adjacent to TNF (study-wide P = 0.027; Q < 0.001; Table 3). Female patients with the LTA GG genotype had an 11-fold higher risk compared to female patients with the AA genotype (homozygous odds ratio 11.333 (3.182, 40.370); Table 3). Although no genotypes achieved study-wide significance in African-Americans (Table S4 and S5), there was a trend toward an association between EN and LTA genotypes in African-American women (Table 3). The Q-value for an association between LTA genotypes and EN in African-American women was 0.114, indicating a relatively low (11.4%) chance that the observed association was a false positive if P = 0.047 was defined as the threshold for a significant P-value (Table 3).
Haplotype analysis using the Exhaustive Allelic Case–Control method (Lin et al., 2004) indicated that risk alleles for EN are frequently found together (that is, high linkage disequilibrium) at the LTA/TNF locus. Upon investigation of this region using Caucasian data publicly available from HapMap (http://www.hapmap.org), the D′ measure of linkage disequilibrium was shown to be 1.0, indicating an extremely high degree of correlation between the two SNPs. These SNPs are present in the same haplotype block and the TNF-308 A allele is found only with the LTA G allele (Barrett et al., 2005).
Large, well-characterized patient collections such as ACCESS provide an excellent resource to evaluate the association between candidate genes and disease manifestations. Several prior studies implicated genetic variation at the LTA/TNF locus as a risk factor for development of EN in Caucasian sarcoidosis patients (Seitzer et al., 1997; Swider et al., 1999; Labunski et al., 2001; Mrazek et al., 2005; Grunewald and Eklund, 2007). However, when attempting to evaluate the genetic factors for a modifying role, it is important to differentiate whether the genetic variants contribute to affectation status (for example, causing sarcoidosis) or whether they only associate with disease complication (for example, EN). Thus, we first evaluated whether any SNPs were associated with sarcoidosis disease status in Caucasians (362 cases/364 controls) using an unconditional logistic regression adjusted for each matching factor. As none of the SNPs were associated with affectation status, we assessed whether Caucasian sarcoidosis patients with and without EN (27 with EN/334 without) were associated with any of 24 variants in inflammatory genes. By structuring the second phase of analysis in this manner, we evaluated candidate variants only for a role in the presence or absence of EN within sarcoidosis. The study design had two additional strengths. First, analysis of patients recruited on the basis of sarcoidosis rather than EN minimized ascertainment bias. Second, typing of all patients for 24 variants in 19 biologically implicated candidate genes and use of a study-wide threshold for significance provided a stringent test for association.
Löfgren's syndrome, defined as sarcoidosis with the presence of EN and bilateral hilar lymphadenopathy (Lofgren and Lundback, 1952a, b), has been associated with SNPs in the HLA-DRB1/LTA/TNF locus in Caucasians and in patients of unspecified race (Seitzer et al., 1997; Swider et al., 1999; Labunski et al., 2001; Mrazek et al., 2005; Grunewald and Eklund, 2007). Seitzer et al. (1997) found TNF-308 A overrepresented in sarcoidosis patients classified as Löfgren's syndrome (16 patients) compared to non-Löfgren's sarcoidosis patients (85 patients) and healthy controls (216 controls) (Seitzer et al., 1997). They also noted a trend toward an association of an SNP in the first intron of LTA (the same SNP in this study) in the Löfgren's syndrome patients (Seitzer et al., 1997). The same group reported an association between TNF-308 A and Löfgren's syndrome in a second study involving 16 Löfgren's patients, 62 non-Löfgren's sarcoidosis patients, and 50 controls (Swider et al., 1999). Labunski et al. (2001) showed an association between TNF-308 A and 10 sarcoidosis patients with EN compared to 10 patients with EN without underlying sarcoidosis and 17 control patients with granulomas annulare or acute drug eruption. Mrazek et al. (2005) also found TNF-308 A overrepresented in 16 Löfgren's syndrome sarcoidosis patients compared to 98 non-Löfgren's sarcoidosis patients and 232 controls. In addition, LTA + 252 G (the same SNP in this study and the one by Seitzer et al., 1997) was also associated with the 16 Löfgren's syndrome sarcoidosis patients compared to 98 non-Löfgren's sarcoidosis patients and 232 controls, as well as an additional control group of 193 individuals (Mrazek et al., 2005). These studies provided congruent evidence of a role for the LTA/TNF locus in development of skin complications in sarcoidosis.
A prior study using a subset of the ACCESS population, which detected an increase in a genetic marker of immunoglobulin κ chains in African-American patients without EN compared to non-disease matched controls, did not detect an association of TNF-308 A in 278 Caucasian sarcoidosis patients with EN (proportion unreported) compared with matched non-sarcoidosis controls, or when sarcoidosis patients without EN (proportion unreported) were compared with matched non-sarcoidosis controls (Pandey and Frederick, 2002). Using the entire ACCESS collection of Caucasian patients (362 with genotype and racial data), we demonstrated a significant association between LTA/TNF variants and EN in Caucasian sarcoidosis patients. The different conclusions drawn by the two ACCESS studies could be due to the different numbers of patients studied. However, of potentially greater importance is that the first ACCESS study did not compare Caucasian sarcoidosis patients with and without EN to each other as was done in this study.
It is interesting to note that no previous studies evaluated male and female patients separately for genetic association with TNF or LTA even though the prevalence of EN is increased in female sarcoidosis patients. Grouping of ACCESS Caucasian patients with EN by sex reveals that female patients have a robust association with LTA and a near-study-wide significant association with TNF. These two genes are next to each other and variants in each gene that associate with EN are almost always found together (Barrett et al., 2005; Mrazek et al., 2005). Thus, either variant, both variants, or an as yet unidentified variant residing nearby appears responsible for the major skin complication of sarcoidosis. These variants may contribute to EN by altering the levels of TNF, LTA, or both or by some other mechanism. There is evidence that TNF-308 A may increase TNF levels by increasing the transcription of TNF (Kroeger et al., 1997; Wilson et al., 1997). It is noteworthy that when Caucasian sarcoidosis patients with “high risk” LTA genotypes are removed, prevalence of EN is unchanged in male patients (5.5%) but decreases from 10 to 6.2% in female patients, suggesting that the higher prevalence of EN in female patients is due to variation in the LTA/TNF locus. This sex-dependent effect may not be limited to Caucasians, as female African-American sarcoidosis patients in this study also displayed evidence of association between LTA/TNF variants and EN.
TNF is a compelling biological candidate for the sex-dependent difference in EN as it has been shown to be upregulated by estradiol, which in turn regulates the fate of T-cells (Cutolo et al., 2003; Hirano et al., 2006). The presence of high levels of estradiol in female patients may enhance the physiological effect of the genetic variation in TNF, leading to substantial elevation of this potent cytokine. In support of this conjecture is the observation that the prevalence of EN in male and female patients is similar before puberty (Gordon, 1961; Soderstrom and Krull, 1978). LTA could also be involved, as it has 50% homology with TNF and both TNF and LTA share receptors and act as modulators of the immune response (Aggarwal et al., 1985; Porter, 1990; Bradley, 2008). An association between LTA/TNF locus variation and presence of EN increases confidence that these protein products play an etiologic role in skin sarcoidosis. Thus, therapies directed against these proteins may be effective, especially for female patients (Sheskin, 1980; Alexis and Strober, 2005; Baughman et al., 2006; Faber et al., 2006).
Sarcoidosis patients and controls were recruited by ACCESS: a multi-center study designed to examine the etiology of sarcoidosis and the socioeconomic status and clinical course of patients with sarcoidosis (ACCESS Research Group, 1999). Center study protocols were approved by their local Institutional Review Board, and Declaration of Helsinki protocols were followed. The experiments in this work were approved by the Johns Hopkins University Institutional Review Board. Written informed consent was received from all participating individuals. Seven hundred six matched pairs (1412 individuals) were recruited between November 1996 and June 1999 (Rybicki et al., 2001a). All patients had active sarcoidosis confirmed by biopsy within 6 months prior to recruitment. The study design and sarcoidosis assessment instruments of ACCESS have been described in detail (ACCESS Research Group, 1999; Judson et al., 1999). Briefly, organ involvement was assessed in each subject by a physician participating in ACCESS using a system based on the subject's history, physical examination, and laboratory tests (Judson et al., 1999). For each patient, biopsy data and laboratory results were compared with organ involvement data to confirm the diagnosis. Patients were matched with controls based on age, sex, self-reported ethnicity, and geographic area. Poor quality or lack of DNA prevented genotyping of 60 individuals in ACCESS. Thirty-five genotyped individuals missing racial data were excluded. A total of 726 Caucasian samples (362 cases/364 controls) and 591 African-American samples (297 cases/294 controls) were analyzed.
Twenty-five SNPs in genes encoding proteins involved in T-cell-mediated inflammatory responses were chosen based upon biological relevance or previously published association with sarcoidosis (Table S7). Genotyping was performed by hybridization of PCR-amplified DNA fragments encompassing each SNP to allele-specific oligonucleotides in linear arrays (Baxter et al., 2001; Wang et al., 2002; Barcellos et al., 2004; Buranawuti et al., 2006). Multiplex PCR with biotinylated primers was performed as previously reported (Baxter et al., 2001; Wang et al., 2002; Barcellos et al., 2004; Buranawuti et al., 2006). Arrays were hybridized and washed using a BeeBlot machine (Product Number 1003, manufactured by Bee Robotics Ltd., UK) with BeeBlot Programmable Software V2.1b. Assay conditions were as follows: hybridization 53°C for 15 minutes; wash 25°C for 1 minute; conjugation 54°C for 5 minutes; wash 25°C for 1 minute; wash 54°C for 12 minutes; citrate 25°C for 5 minutes; substrate 25°C for 25 minutes; deionized water 25°C for 5 minutes (3×). Genotypes were determined using image analysis software (StripScan 5.3.5) provided by Roche Molecular Systems.
Allele and genotype frequencies were estimated using the genassoc package available for STATA (http://www.stata.com) and confirmed in SAS version 8.2 (http://www.sas.com/SAS). For allele frequencies please see Table S8. Tests of fit to Hardy–Weinberg proportions were also performed in controls using genassoc (http://www-gene.cimr.cam.ac.uk/clayton/software/stata/genassoc/). SNP rs1799987 (http://www.ncbi.nlm.nih.gov/SNP) in Caucasians and SNP rs333 in African Americans did not conform to HWE predictions and were removed from further analysis in these groups.
Logistic regression was used to estimate genotype odds ratios for each SNP, and Wald tests were performed to assess statistical significance by computing study-wide P-values in SAS version 8.2. Permutation testing, through random shuffling of case-control status, was performed for the 24 SNPs conforming to HWE predictions in African Americans or Caucasians. We obtained an empirical P-value estimated by the proportion of permuted data sets with a Wald test for an SNP greater than the observed test value for that SNP. To correct for false positives due to multiple testing, we extended the permutation procedure to obtain a study-wide threshold for a given type 1 error rate that accounts for all SNP tests performed. This was constructed using distribution of maximum Wald values (maxT) across 24 tests of single SNPs at each permutation. The maxT distribution accounts for 24 tests while allowing for correlations between SNPs when computing a multiple-test adjusted empirical P-value for each SNP (Westfall et al., 2002). We calculated study-wide empirical P-values for each SNP as proportions of maximum Wald test values among permuted data sets greater than the observed value for each SNP. One thousand permutations were performed and we considered SNPs with a study-wide P ≤ 0.05 statistically significant associations. Analyses using 10,000 permutations did not produce P-values that were different from those with 1,000 permutations. Testing and permutation methods did not directly incorporate the matched design of ACCESS. We adjusted for matching factors using unconditional logistic regression that adjusted for each factor while estimating genotype odds ratios, rather than using a conditional logistic regression model without these covariates. This approach causes minimal bias of odds ratio estimates when adjusting by matching factors, as long as they are not very strong confounders (Breslow et al., 1978). Given different allele frequencies and potential for different risk effect sizes between races, we analyzed Caucasian and African-American samples separately. Haplotype analysis was performed using the Exhaustive Allelic Case–Control method as described earlier (Lin et al., 2004).
FDR was also estimated by computing a Q-value from an uncorrected P-value for each test, using the distribution of observed test statistics (Storey and Tibshirani, 2003). A Q-value represents the proportion of false positives controlled for by accepting significance at a particular observed P-value. In this study, we considered Q-values of 0.05 or less as an acceptable proportion of false positives in combination with a study-wide P0.05. Q-values were estimated from single-test empirical P-values by the bootstrap estimation method using the Q-value package in R (http://cran.r-project.org/) (Benjamini and Hochberg, 1995; Storey and Tibshirani, 2003).
Table S1. Study-wide P-values in analyses of sarcoidosis status in sarcoidosis patients and controls.
Table S2. FDR Q-values and uncorrected P-values in analyses of sarcoidosis status in African-American sarcoidosis patients and controls.
Table S3. FDR Q-values and uncorrected P-values in analyses of sarcoidosis status in Caucasian sarcoidosis patients and controls.
Table S4. Study-wide P-values in analyses of erythema nodosum status in sarcoidosis patients.
Table S5. FDR Q-values and uncorrected P-values in analyses of erythema nodosum status in African-American sarcoidosis patients.
Table S6. FDR Q-values and uncorrected P-values in analyses of sarcoidosis status in Caucasian sarcoidosis patients and controls.
Table S7. Previous evidence for gene involvement with sarcoidosis.
Table S8. SNP number and allele frequencies by race.
Appendix S1. ACCESS members.
We thank Nulang Wang and Catherine Yurk for genotyping assistance. ACCESS Research Group details are in Appendix S1. Funded by National Institutes of Health grants HL-68927, HL-68019, and HL 99-024; contracts (N01-HR-56065, 56066, 56067, 56068, 56069, 56070, 56071, 56072, 56073, 56074, and 56075) with the National Heart, Lung, and Blood Institute; the Life and Breath Foundation, and the Hospital for Consumptives of Maryland (Eudowood) Foundation. This work was performed in Baltimore, MD, USA.
Conflict of Interest: Lori L. Steiner is employed by Roche Molecular Systems, which provided research reagents at no cost to Johns Hopkins. Garry R. Cutting is a consultant for Roche Molecular Systems and receives less than $1,000 USD per year for consulting fees. The remaining authors state no conflict of interest.