Associations between cytokine gene variations and severe persistent breast pain in women following breast cancer surgery

Patients and Settings

This study is part of a larger, longitudinal study that evaluated for neuropathic pain and lymphedema in a sample of women who underwent breast cancer surgery.[44; 46] Patients were recruited from Breast Care Centers located in a Comprehensive Cancer Center, two public hospitals, and four community practices. Patients were eligible to participate if they: were an adult woman (≥18 years) who would undergo breast cancer surgery on one breast; were able to read, write, and understand English; agreed to participate; and gave written informed consent. Patients were excluded if they were having breast cancer surgery on both breasts and/or had distant metastasis at the time of diagnosis. A total of 516 patients were approached to participate and 410 were enrolled in the study (response rate 79.5%). The major reasons for refusal were: too busy, overwhelmed with the cancer diagnosis, or insufficient time available to do baseline assessment prior to surgery.

Subjective Measures

The demographic questionnaire obtained information on age, education, ethnicity, marital status, employment status, living situation, and financial status. The Karnofsky Performance Status (KPS) scale is widely used to evaluate functional status in patients with cancer and has well established validity and reliability.[35; 36] Patients rated their functional status using the KPS scale that ranged from 30 (I feel severely disabled and need to be hospitalized) to 100 (I feel normal; I have no complaints or symptoms). Patients were asked to indicate if they exercised on a regular basis (yes/no format).

The Self-Administered Comorbidity Questionnaire (SCQ) is a short and easily understood instrument that was developed to measure comorbidity in clinical and health service research settings.[60] The questionnaire consists of 13 common medical conditions that were simplified into language that could be understood without any prior medical knowledge. Patients were asked to indicate if they had the condition using a “yes/no” format. If they indicated that they had a condition, they were asked if they received treatment for it (yes/no; proxy for disease severity) and did it limit their activities (yes/no; indication of functional limitations). Patients were given the option to add three additional conditions not listed on the instrument. For each condition, a patient can receive a maximum of 3 points. Because there are 13 defined medical conditions and 2 optional conditions, the maximum score totals 45 points if the open-ended items are used and 39 points if only the closed-ended items are used. The SCQ has well-established validity and reliability and has been used in studies of patients with a variety of chronic conditions.[4; 8; 41; 60; 61]

Preoperative and persistent, as well as acute postoperative pain ratings were evaluated using the Breast Symptoms Questionnaire (BSQ) and Postsurgical Pain Questionnaire, respectively. The BSQ consists of two parts. Part 1 obtained information on the occurrence of pain and the occurrence of other symptoms in the breast scar area (i.e., swelling, numbness, strange sensations, hardness). The additional symptoms that were assessed were identified in studies by Tasmuth and colleagues.[65; 66] If the patient had pain in the breast scar area, they completed Part 2 of the BSQ. Patients were asked to rate the intensity of their average and worst pain using a 0 (no pain) to 10 (worst imaginable pain) NRS. A NRS is a valid and reliable measure of pain intensity.[31]

The Postsurgical Pain Questionnaire evaluated pain intensity in the first 24 to 48 hours after surgery. Average and worst pain were rated using a 0 (no pain) to 10 (worst imaginable pain) NRS. This questionnaire was completed during the month 1 study visit.

Study Procedures

The study was approved by the Committee on Human Research at the University of California, San Francisco and by the Institutional Review Boards at each of the study sites. During the patient’s preoperative visit, a clinician explained the study to the patient and determined her willingness to participate. For those women who were willing to participate, the clinician introduced the patient to the research nurse. The research nurse met with the women, determined eligibility, and obtained written informed consent prior to surgery. After obtaining written informed consent, patients completed the enrollment questionnaires (Assessment 0).

Patients were contacted two weeks after surgery to schedule the first postsurgical appointment. The research nurse met with the patients either in their home or in the Clinical Research Center at 1, 2, 3, 4, 5, and 6 months after surgery. During each of the study visits, the women completed the study questionnaires, and provided information on new and ongoing treatments. Over the course of the study, patients’ medical records were reviewed for disease and treatment information.

Characterization of the persistent breast pain phenotype

Characterization of the persistent breast pain phenotype used in this study was described previously.[46] At each assessment, patients were asked, “Are you experiencing pain in your affected breast?” If the patient reported pain, she was asked to rate her “current pain at its worst” using a 0 (no pain) to 10 (worst pain) NRS. Prior to conducting GMM analyses, patients who reported no pain in their affected breast for all 6 assessments (i.e., enrollment and 2, 3, 4, 5, and 6 months) were identified (N=126; 31.7%) and were not included in the GMM analysis. For the remaining 272 women, the six ratings of worst breast pain were used in the GMM analysis to assign each individual into a latent class. Acute pain ratings obtained at the 1-month follow-up assessment were excluded from the model. The high prevalence of pain at the month 1 assessment reduced the variability in pain ratings among the patients. This reduced variability prohibited the determination of latent classes when month 1 ratings were included in the GMM.

A single, unadjusted growth curve that represented the “average” change trajectory was estimated for the sample. Then, the number of latent growth classes that best fit the data was identified using guidelines recommended in the literature.[34; 48; 68] Model fit was assessed statistically by identifying the model with the lowest Bayesian Information Criterion (BIC). The parametric bootstrapped likelihood ratio test (BLRT) was used to evaluate whether a model with K classes fit the data better than a model with K-1 classes. In addition to using the BLRT to compare models, the Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR) for the “K” versus “K-1” class models were examined. When the VLMR test is non-significant, it provides evidence that the K-class model is not better than the K-1 class model. The fourth index used to evaluate model fit was entropy, with >.80 being preferred.[7; 47] Finally, the best fitting model was visually inspected by plotting observed against model-predicted values to determine whether the predicted trajectories followed the empiric trajectories for the classes, and to evaluate whether the predicted plots “made sense” theoretically and clinically.[34; 48; 68] The GMM analyses were done using MPlus 6.1.[47]

Descriptive statistics and frequency distributions for the no breast pain and severe breast pain classes were generated for demographic and clinical characteristics using Stata version 12.1 (StataCorp, College Station, TX). Independent sample t-tests, Mann-Whitney U tests, Chi square tests, and Fisher’s Exact tests were used to evaluate for differences in demographic and clinical characteristics between the two breast pain classes. Logistic regression analysis was performed to evaluate the association between phenotypic characteristics and pain group membership. All phenotypic characteristics that were identified in the bivariate analyses as being different between the pain classes were evaluated for inclusion in the multivariate analysis. A backwards stepwise approach was used to create a parsimonious model. Only predictors with a p-value of <.05 were retained in the final model. These predictors were used in the logistic regression analyses to evaluate the associations between genotype and pain group membership.

Genotype determination

Genomic deoxyribonucleic acid (DNA) was extracted from peripheral blood mononuclear cells using the PUREGene DNA Isolation System (Invitrogen, Carlsbad, CA). DNA was available from 310 of the 398 patients. DNA samples were quantitated with a Nanodrop Spectrophotometer (ND-1000; Nanodrop Products, Wilmington, DE) and normalized to a concentration of 50 ng/µL (diluted in 10 mM Tris/1 mM EDTA). Genotyping was performed blinded to clinical status and positive and negative controls were included. Samples were genotyped using the Golden Gate genotyping platform (Illumina, San Diego, CA) and processed according to the standard protocol using GenomeStudio (Illumina, San Diego, CA). Two blinded reviewers visually inspected signal intensity profiles and resulting genotype calls for each single nucleotide polymorphism (SNP). Disagreements were adjudicated by a third reviewer.

A combination of tagging SNPs and literature driven SNPs (i.e., reported as being associated with altered function and/or symptoms) were selected for analysis. Tagging SNPs were required to be common (i.e., defined as having a minor allele frequency (MAF) of ≥.05) in public databases (e.g., HapMap, release 24). Tag SNP selection was performed across all four available HapMap samples (i.e., CEU, YRI, JPT, CHB). In order to ensure robust genetic association analyses, quality control filtering of SNPs was performed. SNPs with call rates <95% (n=0), sample call rates <95% (n=0), Hardy-Weinberg p < .001 (n=11), and/or a MAF of <5% (n=11) were excluded. A total of 82 SNPs from 15 inflammatory cytokine genes (i.e., interferon gamma (IFNG): 5 SNPs; IFNG receptor 1 (IFNGR1): 1 SNP; IL1B: 12 SNPs; IL1R1: 4 SNPs; IL1 receptor 2 (IL1R2): 3 SNPs; IL2: 3 SNPs; IL4: 2 SNPs; IL6: 9 SNPs; IL8: 3 SNPs; IL10: 7 SNPs; IL13: 4 SNPs; IL17A: 5 SNPs; nuclear factor kappa beta-1 (NFKB1): 11 SNPs; NFKB2: 4 SNPs; TNFA: 9 SNPs) passed all quality control filters and are included in subsequent analyses. Potential functional roles of SNPs associated with persistent breast pain were examined using PUPASuite 2.0,[11] a comprehensive search engine that examines for a series of putative functional effects (i.e., non-synonymous changes, altered transcription factor binding sites, exonic splicing enhancing or silencing, splice site alterations, microRNA target alterations).

Statistical analysis

Allele and genotype frequencies were determined by gene counting. Hardy-Weinberg equilibrium was assessed by the Chi-square test. Measures of linkage disequilibrium (i.e., D’ and r²) were computed from the patients’ genotypes with Haploview 4.2. Linkage disequilibrium (LD)-based haplotype block definition was based on the D’ confidence interval method.[21]

For SNPs that were members of the same haploblock, haplotype analyses were conducted in order to localize the association signal within each gene and to determine if haplotypes improved the strength of the association with the phenotype. Haplotypes were constructed using the program PHASE version 2.1.[63] In order to improve the stability of haplotype inference, the haplotype construction procedure was repeated five times using different seed numbers with each cycle. Only haplotypes that were inferred with probability estimates of ≥.85, across the five iterations, were retained for downstream analyses. Haplotypes were evaluated assuming a dosage model (i.e., analogous to the additive model).

Ancestry informative markers (AIMs) were used to minimize confounding due to population stratification.[24; 27; 67] Homogeneity in ancestry among patients was verified by principal component analysis[51] using Helix Tree (Golden Helix, Bozeman, MT). Briefly, the number of principal components (PCs) was sought which distinguished the major racial/ethnic groups in the sample by visual inspection of scatter plots of orthogonal PCs (i.e., PC 1 versus PC2, PC2 versus PC3). This procedure was repeated until no discernible clustering of patients by their self-reported race/ethnicity was possible (data not shown). The first three PCs were selected to adjust for potential confounding due to population substructure (i.e., race/ethnicity) by including the three covariates in all regression models. One hundred and six AIMs were included in the analysis.

For association tests, three genetic models were assessed for each SNP: additive, dominant, and recessive. Barring trivial improvements (i.e., delta <10%), the genetic model that best fit the data, by maximizing the significance of the p-value, was selected for each SNP. Logistic regression analysis, that controlled for significant covariates, as well as genomic estimates of and self-reported race/ethnicity, was used to evaluate the associations between genotype and pain group membership. A backwards stepwise approach was used to create a parsimonious model. Except for genomic estimates of and self-reported race/ethnicity, only predictors with a p-value of <.05 were retained in the final model. Genetic model fit and both unadjusted and covariate-adjusted odds ratios were estimated using Stata version 12.1.

As was done in our previous studies,[28; 44] based on recommendations in the literature,[26; 58] the implementation of rigorous quality controls for genomic data, the nonindependence of SNPs/haplotypes in LD, and the exploratory nature of the analyses, adjustments were not made for multiple testing. Significant SNPs identified in the bivariate analyses (Table 1) were evaluated further using regression analyses that controlled for differences in phenotypic characteristics, potential confounding due to population stratification, and variation in other SNPs/haplotypes within the same gene. Only those SNPs that remained significant are included in the final presentation of the results. Therefore, the significant independent associations reported are unlikely to be due solely to chance. Unadjusted associations are reported for all SNPs passing quality control criteria in Supplemental Table 1 to allow for subsequent comparisons and meta-analyses.

Differences in demographic and clinical characteristics between pain classes

Of the 398 women who completed the presurgical assessment, 126 (31.7%) were classified into the no breast pain class and 46 (11.6%) were classified into the severe breast pain class. Differences in demographic and clinical characteristics among the four breast pain classes at the time of enrollment are described in detail elsewhere.[46] Differences in demographic and clinical characteristics between the no breast pain and severe breast pain classes are provided in Table 2.

Women who were classified into the severe breast pain class were significantly younger, had fewer years of education, and were more likely to have an annual household income below $20,000 than women in the no breast pain class. In terms of ethnicity, post-hoc analyses revealed that the representation of Whites was greater in the no breast pain class (73%) than in the severe breast pain class (41%) (Bonferroni-corrected p-value = 0.006).

In terms of preoperative clinical characteristics, women in the severe breast pain class reported a higher number of comorbidities (i.e., SCQ score), lower functional status (i.e., KPS score), and were more likely to have a history of depression, back pain, and rheumatoid arthritis than women in the no breast pain class. Forty-three percent of women in the severe breast pain class, compared to 2.4% in the no breast pain class, reported pain in the affected breast prior to surgery. Women in the severe breast pain class were more likely to report swelling, numbness, strange sensations, and hardness in their affected breast prior to surgery compared to women in the no breast pain class.

Differences between the no breast pain and severe breast pain classes were found in a number of surgical and postoperative characteristics. Compared to the no breast pain class, women in the severe pain class had a greater number of lymph modes removed, reported higher average and worst postoperative pain scores, were more likely to have undergone an axillary lymph node dissection, and were more likely to have re-excision or mastectomy within 6 months after surgery.

Regression analysis for phenotypic characteristics

As shown in Table 3, the only predictors that remained significant in the final regression model of clinical, demographic, and surgical characteristics were the occurrence of pain in the breast prior to surgery and the severity of average postoperative pain. For patients who reported pain in their affected breast prior to surgery, the odds of being in the severe pain class increased 8.71-fold (95% confidence interval (CI): 1.14, 66.48; p=0.037). For each one-unit increase in the severity of average postoperative pain, the odds of being in the severe pain class increased 2.02-fold (95% confidence interval (CI): 1.43, 2.84; p<0.001).

Regression analyses for candidate genes

As summarized in Table 1, no associations were found between pain group membership and SNPs in IFNGR1, IL1B, IL2, IL6, IL8, IL17A, NFKB2, and TNFA. However, the genotype frequency was significantly different between the no breast pain and severe breast pain classes for 13 SNPs and 3 haplotypes among 7 genes (IFNG: 2 SNPs, 1 haplotype; IL1R1: 1 SNP; IL1R2: 1 SNP; IL4: 1 SNP; IL10: 3 SNPs, 1 haplotype; IL13: 4 SNPs, 1 haplotype; NFKB1: 1 SNP).

In order to better estimate the magnitude (i.e., odds ratio, OR) and precision (i.e., CI) of genotype on pain group membership, multivariate logistic regression models were fit. In addition to genotype, the phenotypic variables included in the regression models were genomic estimates of and self-reported race/ethnicity (i.e., White, Black, Asian, Hispanic/Mixed ethnic background/other), occurrence of pain in the affected breast prior to surgery, and severity of average postoperative pain. As shown in Table 4, the genetic associations that remained significant were for IL1R2 rs11674595 and IL10 haplotype A8 (see Table 4 and Figures 1 and 2).

In the regression analysis for IL1R2 rs11674595 (Figure 1), individuals who were homozygous for the rare “C” allele (i.e., TT+TC versus CC) had a 36.1-fold increase in the odds of belonging to the severe breast pain class (95% CI: 2.02, 643.37, p=0.015).

In the regression analysis for IL10 haplotype A8, each dose of this haplotype decreased the odds of belonging to the severe breast pain class by 79% (95% CI: 9%, 95%, p=0.037). The IL10 haplotype A8 is composed of seven SNPs (i.e., rs3024505 “C“ allele, rs3024498 “G“ allele, rs3024496 “C“ allele, rs1878672 “G“ allele, rs1518111 “A“ allele, rs1518110 “T“ allele, rs3024491 “T“ allele) (Figure 2).