Vachon, Celine M.

Abstract Motivation: Batch effects are due to probe-specific systematic variation between groups of samples (batches) resulting from experimental features that are not of biological interest. Principal component analysis (PCA) is commonly used as a visual tool to determine whether batch effects exist after applying a global normalization method. However, PCA yields linear combinations of the variables that contribute maximum variance and thus will not necessarily detect batch effects if they are not the largest source of variability in the data. Results: We present an extension of PCA to quantify the existence of batch effects, called guided PCA (gPCA). We describe a test statistic that uses gPCA to test whether a batch effect exists. We apply our proposed test statistic derived using gPCA to simulated data and to two copy number variation case studies: the first study consisted of 614 samples from a breast cancer family study using Illumina Human 660 bead-chip arrays, whereas the second case study consisted of 703 samples from a family blood pressure study that used Affymetrix SNP Array 6.0. We demonstrate that our statistic has good statistical properties and is able to identify significant batch effects in two copy number variation case studies. Conclusion: We developed a new statistic that uses gPCA to identify whether batch effects exist in high-throughput genomic data. Although our examples pertain to copy number data, gPCA is general and can be used on other data types as well. Availability and implementation: The gPCA R package (Available via CRAN) provides functionality and data to perform the methods in this article. Contact: reesese@vcu.edu or eckel@mayo.edu Supplementary information: Supplementary data are available at Bioinformatics online.

AbstractWe previously reported an increased risk of monoclonal gammopathy of undetermined significance (MGUS) in first‐degree relatives of MGUS and multiple myeloma patients. Here, we examine whether primary cytogenetic categories of myeloma differ between patients with and without a family history of MGUS or myeloma. We studied 201 myeloma patients with available data on family history and molecular cytogenetic classification. Myeloma with trisomies was more common in probands who had an affected first‐degree relative with MGUS or myeloma compared with those without a family history (46.9% vs. 33.5%, P = 0.125); however, the difference was not statistically significant. Additional studies on the cytogenetic types of myeloma associated with familial tendency are needed.

Abstract Triple-negative breast cancers (TNBC), defined by the absence of estrogen receptor, progesterone receptor, and HER-2 expression, account for 12% to 24% of all breast cancers. TNBC is associated with early recurrence of disease and poor outcome. Germline mutations in the BRCA1 and BRCA2 breast cancer susceptibility genes have been associated with up to 15% of TNBC, and TNBC accounts for 70% of breast tumors arising in BRCA1 mutation carriers and 16% to 23% of breast tumors in BRCA2 carriers. Whether germline mutations in other breast cancer susceptibility genes also predispose to TNBC remains to be determined. Common variation in a subset of the 72 known breast cancer susceptibility loci identified through genome-wide association studies and other large-scale genotyping efforts have also been associated with risk of TNBC (TOX3, ESR1, RAD51L1, TERT, 19p13.1, 20q11, MDM4, 2p24.1, and FTO). Furthermore, variation in the 19p13.1 locus and the MDM4 locus has been associated with TNBC, but not other forms of breast cancer, suggesting that these are TNBC-specific loci. Thus, TNBC can be distinguished from other breast cancer subtypes by a unique pattern of common and rare germline predisposition alleles. Additional efforts to combine genetic and epidemiologic data are needed to better understand the etiology of this aggressive form of breast cancer, to identify prevention and therapeutic targets, and to impact clinical practice through the development of risk prediction models. Cancer Res; 73(7); 2025–30. ©2012 AACR.

Abstract Purpose: Mammographic breast density (MBD) is decreased by tamoxifen, but the effect of aromatase inhibitors is less clear. Experimental Design: We enrolled early-stage postmenopausal patients with breast cancer initiating adjuvant aromatase inhibitor therapy and ascertained mammograms before and at an average 10 months of aromatase inhibitor therapy. We matched cases to healthy postmenopausal women (controls) from a large mammography screening cohort on age, baseline body mass index, baseline MBD, and interval between mammograms. We estimated change in MBD using a computer-assisted thresholding program (Cumulus) and compared differences between cases and matched controls. Results: In predominantly White women (96%), we found 14% of the 387 eligible cases had a MBD reduction of at least 5% after an average of 10 months of aromatase inhibitor therapy. MBD reductions were associated with higher baseline MBD, aromatase inhibitor use for more than 12 months, and prior postmenopausal hormone use. Comparing each case with her matched control, there was no evidence of an association of change in MBD with aromatase inhibitor therapy [median case–control difference among 369 pairs was −0.1% (10th and 90th percentile: −5.9%, 5.2%) P = 0.51]. Case–control differences were similar by type of aromatase inhibitor (P's 0.41 and 0.56); prior use of postmenopausal hormones (P = 0.85); baseline MBD (P = 0.55); and length of aromatase inhibitor therapy (P = 0.08). Conclusions: In postmenopausal women treated with aromatase inhibitors, 14% of cases had a MBD reduction of more than 5%, but these decreases did not differ from matched controls. These data suggest that MBD is not a clinically useful biomarker for predicting the value of aromatase inhibitor therapy in White postmenopausal women. Clin Cancer Res; 19(8); 2144–53. ©2013 AACR.

Abstract Background: Our genome-wide association study (GWAS) of chronic lymphocytic leukemia (CLL) identified 4 highly correlated intronic variants within the IRF8 gene that were associated with CLL. These results were further supported by a recent meta-analysis of our GWAS with two other GWAS of CLL, supporting the IRF8 gene as a strong candidate for CLL risk. Methods: To refine the genetic association of CLL risk, we conducted Sanger sequencing of IRF8 in 94 CLL cases and 96 controls. We then conducted fine mapping by genotyping 39 variants (of which 10 were identified from sequencing) in 745 CLL cases and 1,521 controls. We also assessed these associations with risk of other non-Hodgkin lymphoma (NHL) subtypes. Results: The strongest association with CLL risk was observed with a common single-nucleotide polymorphism (SNP) located within the 3′ untranslated region (UTR) of IRF8 (rs1044873, log additive OR = 0.7, P = 1.81 × 10−6). This SNP was not associated with the other NHL subtypes (all P > 0.05). Conclusions: We provide evidence that rs1044873 in the IRF8 gene accounts for the initial GWAS signal for CLL risk. This association appears to be unique to CLL with little support for association with other common NHL subtypes. Future work is needed to assess functional role of IRF8 in CLL etiology. Impact: These data provide support that a functional variant within the 3′UTR of IRF8 may be driving the GWAS signal seen on 16q24.1 for CLL risk. Cancer Epidemiol Biomarkers Prev; 22(3); 461–6. ©2013 AACR.

Summary A recent meta‐analysis of three genome‐wide association studies of chronic lymphocytic leukaemia ( CLL ) identified two common variants at the 6p21.31 locus that are associated with CLL risk. To verify and further explore the association of these variants with other non‐ H odgkin lymphoma ( NHL ) subtypes, we genotyped 1196 CLL cases, 1699 NHL cases, and 2410 controls. We found significant associations between the 6p21.31 variants and CLL risk (rs210134: P = 0·01; rs210142: P = 6·8 × 10 −3 ). These variants also showed a trend towards association with some of the other NHL subtypes. Our results validate the prior work and support specific genetic pathways for risk among NHL subtypes.

Abstract Monoclonal gammopathy of undetermined significance (MGUS) is a precursor to multiple myeloma and related diseases. The majority of MGUS cases are characterized by the expression of immunoglobulin heavy chain (IgH); however, approximately one-fifth of MGUS cases do not present with IgH expression. These individuals have light chain MGUS (LC-MGUS), a precursor to light-chain MM (LC-MM). LC-MGUS is characterized by an absence of IgH and an abnormal free light chain ratio (FLC-R), indicating excessive expression of either κ or ≤ free light chains. LC-MGUS has a prevalence of 0.8% in the general population over the age of 50, as compared with a prevalence of 3.2% for heavy-chain MGUS (HC-MGUS) (Dispenzieri, 2010). There is currently little understanding of the association between family history of HC-MGUS or multiple myeloma (MM) and LC-MGUS. We investigated whether the prevalence of LC-MGUS is elevated in first-degree relatives of patients with either HC-MGUS or MM. HC-MGUS probands expressed IgH and were recruited through a population-based study of Olmsted County residents; MM probands consisted of both HC-MM and LC-MM and were recruited through the Mayo Clinic Hematology practice. Serum samples were collected from first-degree relatives of probands older than 40 years of age and subjected to electrophoresis, immunofixation, and free light chain assay. Prevalence of LC-MGUS was calculated in all first-degree relatives combined, and in relatives from both HC-MGUS and MM families separately. We compared the prevalence in family members to that in a well-defined Olmsted County, MN cohort (Dispenzieri, 2010) of patients aged 50 and over. A total 911 relatives of 232 MM and 97 MGUS patients were studied. LC-MGUS was detected in 17/911 (1.9%) relatives, for an age- and sex-adjusted prevalence of 1.7% (95% CI, 0.9% to 2.6%). Prevalence among relatives of MM probands (age- and sex- adjusted rate 2.1%; 95% CI, 1.0% to 3.2%) was higher than among relatives of MGUS probands (age- and sex- adjusted rate 0.6%; 95% CI, 0.0%-1.6%). There was an increased risk of LC-MGUS in relatives compared to the reference population which was predominantly due to the increased risk among relatives of MM probands (risk ratio=3.4, 95% CI 2.0, 5.51) as opposed to relatives of MGUS probands (risk ratio=1.1, 95% CI, 0.3 to 4.5). Among MM probands, there was no significant difference between the prevalence increase in relatives of the 32 probands with LC-MM and that of the relatives of the 200 probands with HC-MM (3.8 vs 3.3, ratio=1.2, 95% CI: 0.3-5.1, p=0.85). We conclude that the prevalence of LC-MGUS is significantly higher among first degree family members of all MM probands compared to the general population. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 5485. doi:1538-7445.AM2012-5485

Abstract INTRODUCTION: Mammographic breast density is a strong risk factor for breast cancer (BC). We established the Mayo Mammography Health Study (MMHS) cohort study at the Mayo Clinic in Rochester, Minnesota (MN) to examine the association of breast density with BC. We evaluated the influence of the image acquisition technique on the density and BC association. METHODS: From October 2003 to September 2006, all women scheduled for screening mammography at the Mayo Clinic were invited to participate. Eligible women were residents of MN, Iowa (IA) or Wisconsin (WI); age 35+; and had no personal history of BC. A risk factor questionnaire, consent form and permission to link to tumor registries were obtained. Incident BC was identified through 2009 by linkage to the Mayo and state cancer registries. A case-cohort of all incident BCs and 2300 randomly selected women (the subcohort) were used to examine the association of breast density and BC using digitized film mammograms at the time of enrollment while controlling for the influence of the acquisition parameters (peak kilovoltage, milliampere-second, and compressed breast thickness). Two density measures were considered: a quantitative percent density (PD) measure estimated using the computer-assisted thresholding program, Cumulus (University of Toronto), and the 4-category clinical BI-RADS measure. Proportional hazards regression was used to calculate hazards ratios (HR) and 95% confidence intervals (CI) for BC associated with quartiles of PD (0-6.2%,6.3-14.9%, 15.0-25.7% and 25.8%+) and BI-RADS categories 1-4 (almost entirely fatty to extremely dense). All models included age, postmenopausal hormone use (PMH), BMI, and menopausal status. The influence of acquisition parameters was evaluated by examining models with and without their inclusion. RESULTS: A total of 20,982 (50%) women participated in MMHS; 1058 (5%) with a prior history of BC were excluded, for a total cohort of 19,924. Compared to nonresponders, responders were younger (57.5 vs. 58.4 yrs), more likely to have ever used PMH (45% vs. 33%), and more likely to have a BC family history (19% vs 16%). The case-cohort consisted of 317 incident cases and 2300 women in the subcohort. Women were excluded from the current analysis if PD could not be estimated or if acquisition parameters were not available, leaving 249 cases and 1937 in the subcohort. As expected, PD was associated with BC [HR (95% CI): 1.0 (REF), 2.1 (1.4-3.1), 3.0 (2.0-4.5), and 4.6 (3.0-7.0) for quartiles; p-trend<0.001]. Controlling for acquisition parameters attenuated the association [HR (95% CI): 1.0 (REF), 2.3 (1.5-3.4), 2.4 (1.6-3.7), and 3.0 (1.8-5.0) for quartiles; p-trend<0.001]. Results for BI-RADS density were similar to those for PD. CONCLUSION: This study confirms that breast density is a significant risk factor for BC and demonstrates that the acquisition technique confounds the density and BC risk association. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 3716. doi:10.1158/1538-7445.AM2011-3716

Abstract We performed a meta-analysis of 3 genome-wide association studies to identify additional common variants influencing chronic lymphocytic leukemia (CLL) risk. The discovery phase was composed of genome-wide association study data from 1121 cases and 3745 controls. Replication analysis was performed in 861 cases and 2033 controls. We identified a novel CLL risk locus at 6p21.33 (rs210142; intronic to the BAK1 gene, BCL2 antagonist killer 1; P = 9.47 × 10−16). A strong relationship between risk genotype and reduced BAK1 expression was shown in lymphoblastoid cell lines. This finding provides additional support for polygenic inheritance to CLL and provides further insight into the biologic basis of disease development.