DNA Methylation Markers for Breast Cancer Detection in the Developing World

Abstract

Purpose:

An unmet need in low resource countries is an automated breast cancer-detection assay to prioritize women who should undergo core breast biopsy and pathological review. Therefore, we sought to identify and validate a panel of methylated DNA markers to discriminate between cancer and benign breast lesions using cells obtained by fine needle aspiration (FNA).

Experimental Design:

Two case-control studies were conducted comparing cancer and benign breast tissue identified from clinical repositories in the U.S., China, and South Africa for marker selection/training (N=226) and testing (N=246). Twenty-five methylated markers were assayed by quantitative multiplex-methylation-specific PCR (QM-MSP) to select and test a cancer-specific panel. Next, a pilot study was conducted on archival FNAs (49 benign, 24 invasive) from women with mammographically suspicious lesions using a newly developed, 5-hour, quantitative, automated cartridge system. We calculated sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) compared to histopathology for the marker panel.

Results:

In the discovery cohort, 10/25 markers were selected that were highly methylated in breast cancer compared to benign tissues by QM-MSP. In the independent test cohort, this panel yielded an AUC of 0.937 (95% CI = 0.900–0.970). In the FNA pilot, we achieved an AUC of 0.960 (95% CI = 0.883–1.0) using the automated cartridge system.

Conclusions:

We developed and piloted a fast, and accurate methylation marker-based automated cartridge system to detect breast cancer in FNA samples. This quick ancillary test has the potential to prioritize cancer over benign tissues for expedited pathologic evaluation in poorly resourced countries.

INTRODUCTION

In developing countries of the world, breast cancer is the most frequently diagnosed cancer in women and the leading cause of cancer death (1, 2). Breast cancer incidence is rising globally because of longer life expectancies, decreased burden of infectious diseases, and changes in reproductive risk factors (3–5). Decreases in overall breast cancer mortality in the U.S. and Canada are associated with advances in screening and in adjuvant therapy (6, 7). Low resource countries have poor breast cancer survival rates partly because of lack of organized screening programs, late stage at diagnosis and limited access to timely, standard treatment (1, 5). To reduce the global burden of breast cancer, we require novel approaches to breast cancer screening, early detection programs and access to treatment in low and middle-income countries of the world (8, 9).

In underdeveloped countries, there is a critical shortage of clinical pathologists and resources to perform the complex histopathologic evaluation of breast biopsies (10, 11). Currently, even after the breast lesion is biopsied, there may be delays of up to 10 months for a definitive diagnosis in many South American and African countries (10, 11). Therefore, developing a molecular test for quick and accurate determination of a suspicious breast lesion will enable more efficient use of limited resources and earlier detection of breast cancer.

Hypermethylation of CpG islands located in the promoter regions of tumor suppressor genes is recognized as one of the earliest, most frequent, and robust alterations in cancer development (12, 13). However, low-level methylation occurs in certain CpG regions in benign conditions (14–19). Euhus et al. performed a genome-wide search for differentially methylated markers in cancer and benign breast epithelial cell lines treated with 5-azacytidine. A select number of markers were evaluated in fine needle aspirations (FNAs) of 97 cancer and 327 benign disease cases. Significant differential methylation was reported for CPNE8, PSAT1, CXCL14, and CLDN1 and GNE (14). However, no further validation of these markers was performed. Other researchers have reported modest differences for methylation markers in serum that failed to provide sufficient ability to distinguish between cancer and benign/normal controls (20, 21). Thus, better markers and further studies are warranted.

In this paper, we report the selection of a panel of DNA methylation markers that distinguishes between benign and cancer breast tissues with high sensitivity and specificity. We incorporated the panel into an automated breast-cancer detection cartridge system that quantitated gene methylation with a high level of accuracy within 5 hours. We tested this assay in a blinded pilot study of archival breast FNA samples from cancer and benign lesions. Our results suggest that the cartridge assay has potential to detect breast cancer in FNA samples of suspicious breast lesions identified by mammography or ultrasound-based screening, and thereby could have utility to prioritize breast lesions that require biopsy especially in low resource countries.

MATERIALS AND METHODS

Study Design.

Fifty cases (invasive ductal carcinoma and ductal carcinoma in situ) and 50 controls (benign breast disease) were sought from each of the three participating sites, U.S., China, and S. Africa, a design that would result in a total of 300 samples, or 75 cases and 75 controls in each of the training and test phase of development. With 75 samples per group, the precision of a one sided, 95% confidence bound is less than 0.1, and precision is even better for sensitivities and specificities near 1.0. However, collection exceeded expectation and we were able to increase the sample size by more than 50%. ductal cancer and benign breast tissues from each of three geographical regions, U.S., China, and S. Africa. Four hundred and forty nine formalin-fixed paraffin-embedded (FFPE) breast tissues from the three regions were randomly assigned to training (N=226) and test (N=223) case-control sets (Figure 1) using the Excel RANDBETWEEN function. The randomization was carried out separately for each geographical region, and type of lesion to ensure balance (Table 1). Later, 23 invasive lobular carcinomas (ILC) were added to the test set to ensure inclusion of this significant pathological subset of breast carcinoma. Our study design (Figure 1) was as follows: First, 25 hypermethylated candidate gene markers were screened by QM-MSP in the training cohort to identify a minimal marker panel of 10-genes using the predetermined criteria described below. This marker panel was evaluated to establish the methylation threshold best able to distinguish between benign and cancer samples. Due to limited available DNA, it was not possible to evaluate all candidate markers in all samples of the training cohort, so the final set of markers was measured in 206 of the 226 samples. Similarly, the marker panel was examined by QM-MSP in a test cohort of 220 samples, consisting of 197 samples with sufficient DNA plus an additional new sample set of 23 ILCs. We evaluated the utility of the assay methylation threshold (defined in the training cohort) for the novel gene panel (Figure 1).

Figure 1. Overview of the study workflow and sample selection.

FFPE, formalin fixed paraffin embedded; IDC, invasive ductal carcinoma; ILC, invasive lobular carcinoma; DCIS, ductal carcinoma in situ; QM-MSP, quantitative multiplex methylation-specific PCR; AUC, area under the ROC curve; FNA, fine needle aspirate. Total FFPE samples: n=472 (449 randomized to test and training Sets+23 ILC added to test set)

Table 1.

Patient Characteristics

Sample Sets	Training (FFPE)				Test (FFPE)				Pilot(FNA
Region; Told	US	China	S. Africa	Total	US	China	S Africa	Total	Total
Sample N; Total	87	89	50	226	109	88	49	246	76
Sample N: win 10 marker CMI	68	88	50	206	85	87	48	220	73
Invasive Ductal Carcinoma, N	29	29	13	71	29	30	13	72	24
Receptor status
ER/PR +, HER2−	11	19	8	38	3	18	5	26	16
ER/PR +, HER2 +	2	2	3	7	1	3	3	7	2
ER/PR −, HER2 +	2	4	0	6	1	1	1	3	1
ER/PR −, HER2 −	4	4	2	10	11	7	3	21	3
Unknown	10	0	0	10	13	1	1	15	2
Age (in years)
Median	56	53	50	54	54	50	55	54	59
Range	25–87	23–66	32–74	23–87	25–70	35–75	34–88	25–88	30–87
Invasive Lobular Carcinoma, N	-	-	-	-	23	-	-	23	0
Receptor status
ER/PR +	-	-	-	-	19	-	-	19	-
ER/PR −	-	-	-	-	4	-	-	4	-
Unknown	-	-	-	-	-	-	-	-	-
Age (in years)
Median	-	-	-	-	58	-	-	58	-
Range	-	-	-	-	38–81	-	-	38–81	-
Ductal Carcinoma In Situ, N	18	8	13	39	18	7	13	38	0
Age (in years)
Median	58	54	59	56	56	53	46	53	-
Range	42–78	25–71	37–76	25–78	42–79	45–72	26–82	26–82	-
Benign breast disease, N	40	52	12	104	39	51	12	102	52
Age (in years)
Median	51	46	39	47	52	48	37	47	41
Range	26–78	26–68	17–62	17–78	36–75	23–73	19–60	19–75	19–74
Normal breast; N	-	-	12	12	-	-	11	11	0
Age (in years)
Median	-	-	45	45	-	-	45	45	-
Range	-	-	37–96	37–96	-	-	23–69	23–69	-

Lastly, these findings were translated to a newly developed, automated breast cancer detection DNA methylation cartridge. We performed a pilot study of archival FNAs (N=76) collected from Portugal and Hong Kong (Figure 1).

The study, performed on archival, formalin fixed cancer, benign and normal breast samples was approved by Institutional Review Boards of Johns Hopkins, USA; Emory University School of Medicine, USA; Renmin Hospital, China; University of Witwaterstrand, S. Africa; Instituto de Patologia e Imunologia Molecular da Universidade do Porto, Portugal; and The Chinese University of Hong Kong. The study was conducted in accordance with Declaration of Helsinki and the U. S. Common Rule. Patients provided informed consent to use excess tissue and cells for research. Consent requirement was waived at The Chinese University of Hong Kong, and excess slides of FNA biopsy were used.

Patient Materials.

DNA was extracted from FFPE tissues (N=472; 449 randomized to training and test + 23 ILC in test) obtained from Johns Hopkins Surgical Pathology, Renmin Hospital of Wuhan University, China, and National Health Laboratory Service, S. Africa, following review by a breast pathologist to confirm correct classification. Breast cancer samples included invasive ductal carcinoma (IDC), invasive lobular carcinoma (ILC), and ductal carcinoma in situ (DCIS). Two non-cancer groups were studied: normal breast and benign breast disease. The cancer FFPE blocks ranged in age from 2–28 years (median 4 years), and the benign/normal blocks ranged in age from 2–19 years (median 2 years). A large majority of our FFPE samples (cancer: 71%; benign: 89%) were 2–10 years old. Archival diagnostic stained slides of FNA smears from 24 cases of IDC and 52 benign lesions (26 fibroadenoma, 19 benign ductal epithelium, 5 fibrocystic, 2 cyst) were obtained from Medical Faculty of University of Porto (Portugal, N = 33) and The Chinese University of Hong Kong (China, N = 43). Of these, 3 benign lesions were discarded for lack of sufficient DNA. In the FNA pilot, the median age of the FNA slides from IDC was 10 years (range of 1–11 years) and for benign disease was 3 years (range of 1–11 years). Patient characteristics are provided in Table 1. The immunohistochemical subtype information for estrogen/progesterone receptor (ER/PR) and HER2 was available for a subgroup of the invasive cancer samples in the training and test sets (N=108).

Assay Development, Marker Selection and Training.

Quantitative multiplex methylation-specific PCR (QM-MSP)(22, 23) was used in the training and test sets for marker selection and evaluation. Twenty-five individual markers were assayed using DNA from FFPE tissues (N = 226) in the training cohort. Marker selection criteria required first, that markers show considerably higher median methylation in cancer than in benign/normal tissue (P<0.001, based on the Mann-Whitney test). Secondly, to further minimize the risk of false positives, benign and normal samples were required to be methylated at lower levels and less frequently compared to cancer samples. Specifically, we calculated the 75^th percentile of methylation separately in tumor and benign samples, discarding any markers in which the 75^th percentile of normal methylation was high, or where the difference between normal and tumor was small (Supplementary Table S1 and Figure S1). The selected markers were then evaluated as a panel in the training set of N =206 samples with sufficient DNA. QM-MSP values for the panel were expressed as Cumulative Methylation Index (CMI), the sum of percent methylation for each gene in the panel (22, 23). Using Receiver Operating Characteristic (ROC) curve analysis we identified the laboratory CMI threshold for the study that maximized specificity while maintaining sensitivity at 90%.

Marker Testing.

The marker panel was assayed in the test FFPE samples by QM-MSP using locked parameters defined in the training set. The independent test set (N=220) consisted of 66 IDC, 23 ILC, 30 DCIS, 91 benign and 10 normal breast tissues. For the marker panel, sensitivity, specificity and Area Under Curve (AUC) were evaluated.

Differences in DNA Methylation Based on Immunohistochemical (IHC) Subtype and Geographic Origin.

We evaluated differences in DNA methylation among four IHC subtypes of breast cancer, using clinical data for expression of estrogen receptor/progesterone receptors (ER/PR) and HER2: ER/PR+, HER2-; ER/PR+, HER2+; ER/PR-,HER2+; ER/PR-,HER2- (triple-negative breast cancer, TNBC) in samples from training and test sets (total N = 108). CMI of the 10-gene panel for the four IHC subtypes was compared using the Kruskal-Wallis test. Relative differences in percent methylation of each gene was evaluated in two-way comparisons by the Mann-Whitney test. The Fishers Exact test was performed to evaluate the difference in frequency of samples that test positive (higher than the ROC CMI threshold) among the IHC subtypes. For the marker panel, sensitivity, specificity, and AUC were evaluated for tumors from each subtype.

We evaluated tumors from three geographical regions, U.S., China, and S. Africa, for differences in DNA methylation of the 10-gene panel. Cumulative methylation and individual gene methylation was assessed for the cancer and benign samples by the Kruskal-Wallis test. For the marker panel, sensitivity, specificity, and AUC were evaluated for tumors from each region.

Assay Development on the GeneXpert® Platform.

Cartridge Design.

The GeneXpert® System (Cepheid) consists of a computer preloaded with software for running GeneXpert® cartridges tailored for PCR-based diagnostics. The automated methylation-specific PCR system described in this study is a prototype in development, not for use in diagnostic procedures, and has not been reviewed by any regulatory body. The assay has three single-use cartridges, one of which holds reagents required for bisulfite treatment of DNA (Cartridge A) and the two remaining cartridges hold reagents for quantitative PCR of Marker Set 1 or Marker Set 2 (Cartridge B). Each marker set consists of 5 target genes and ACTB as the internal reference, and utilizes 6 fluorophores.

Cartridge Performance.

To determine the sensitivity of detection of methylated DNA in the cartridge, CAMA-1 breast cancer cells (200, 100, and 50 cells; ATCC, authenticated by the Johns Hopkins Genetic Resources Core Facility), that are fully methylated by QM-MSP for each of the genes in our marker panel, were tested in six replicates. To generate standard curves for quantitation of methylation in the cartridge and for testing interassay reproducibility, CAMA-1 cells were mixed with fully unmethylated human sperm DNA (HSD) prior to lysis.

Sample Preparation.

Archival stained FNA slides were soaked in xylene to remove the coverslip, and scraped into proteinase K/FFPE lysis buffer solution (20 μL of proteinase K and 1.2 mL lysis buffer; FFPE Lysis Kit, 900–0697, Cepheid), digested at 80°C for 30 min, mixed with 1.2 mL of ethanol and loaded into the bisulfite conversion Cartridge A. Methylated CAMA-1 cells, or CAMA-1 cells mixed with unmethylated HSD DNA were similarly digested and loaded into Cartridge A. Upon completion of the reaction, bisulfite-treated DNA was transferred to the Set 1-Cartridge B or Set 2-Cartridge B to quantitate methylation.

Quantitation of DNA Methylation.

Ct values were obtained using the GeneXpert® software for methylated targets and ACTB reference (Ct = the cycle threshold at which signal fluorescence exceeds background). For calculating % methylation, the Δ Ct (Ct Gene – Ct ACTB) value of each target gene was extrapolated from historical standard curves of mixtures of methylated and unmethylated DNA ranging from 100% to 3% methylation. This enabled quantitation of cumulative methylation (CM), which is the sum of % methylation for all genes in the marker panel.

Interassay Reproducibility was analyzed in the cartridge by performing 6 replicate assays using mixtures of fully methylated CAMA-1 cells and fully unmethylated HSD to achieve methylation values ranging from 100% to 3%. Reproducibility was reported as Coefficient of Variation (CV) as calculated by GraphPad Prism version 7.04.

Interoperator Reproducibility was tested by two individuals who performed the assay using separate aliquots of the same lysate from 23 archival FNA samples. The R package psych (https://CRAN.R-project.org/package=psych) was used to evaluate interoperator reproducibility in cumulative methylation values using intraclass correlation coefficient (ICC). Cohen’s kappa coefficient (κ), calculated using the R package fmsb and function Kappa.test, was used to measure interrater agreement for presence or absence of cancer cells in the FNA based on quantitative methylation values.

Clinical Pilot Testing of FNA Samples.

A pilot study for breast cancer detection was performed on 76 FNAs for the selected 10-gene panel of methylated markers in the automated GeneXpert® cartridge. Percent methylation for each marker analyzed in the cartridge was calculated by interpolation, using standard curves as described above, and evaluated by nonlinear fit regression in GraphPad Prism and expressed as cumulative methylation.

RESULTS

Selection of Methylated Gene Markers.

We evaluated 25 candidate gene markers in the formalin-fixed paraffin-embedded (FFPE) archival training set samples by QM-MSP to identify those with the highest sensitivity and specificity for cancer based on histopathology of the core biopsy (Figure 1). Candidate gene markers were among those previously identified by us as having frequent measurable gene hypermethylation in ER/PR+, HER2-; ER/PR+, HER2+; ER/PR-, HER2+ and ER/PR-, HER2- breast cancer (23, 24). Nine of the 25 markers were discarded based on their inability to significantly distinguish between cancers (IDC/DCIS) versus benign/normal breast tissues in the training set. Four additional candidate markers were discarded either because normal/benign samples had high methylation, or because the absolute difference in methylation levels was small. Among the 12 markers thus selected, TWIST1 and MAL were discarded (Supplementary Figure S1). The final panel consisted of 10 markers: AKR1B1, APC, CCND2, COL6A2, HIST1H3C, HOXB4, RASGRF2, RASSF1, TMEFF2, and ZNF671. Each of the gene markers showed significantly higher methylation in cancer than benign/normal samples (Figure 2).

Figure 2. Performance of the 10 individual gene markers in the Training set.

Box- Whiskers plots (Tukey method) depict the percent methylation (Y-axis) for each of the 10-gene markers in FFPE tissues obtained from patients with IDC/DCIS versus Benign/Normal. Percent methylation in the benign/normal tissues, sub-classified according to histologic types is shown. Mann-Whitney P-values are indicated. Sample number (N) is shown below the X-axis. IDC, invasive ductal carcinoma; DCIS, ductal carcinoma in situ; Fibro, fibroadenoma; Pap, papilloma; UDH, usual ductal hyperplasia.

Evaluation of Performance of the 10-gene Marker Panel.

Using training set samples, Cumulative methylation for the 10-gene panel was higher in cancer compared to benign/normal samples as shown in the histogram (Figure 3A) and in the box-plot (Figure 3A, inset). ROC analyses of the training set established the laboratory threshold that provided the optimal specificity while retaining ≥ 90% sensitivity. A sensitivity of 90% (95% CI = 82–95%), a specificity of 88% (95% CI = 80–93%), and an AUC = 0.948 (95% CI = 0.914–0.976) was achieved in the training data at a threshold of 14.5 CMI units (Figure 3A, inset).

Figure 3. Performance of the 10-gene marker panel in distinguishing between benign and cancer tissue.

FFPE samples of IDC/DCIS and Benign/Normal tissues were assayed by QM-MSP. Histogram plots indicate the percent methylation (colored segment) and cumulative methylation (bar height, Y-axis) for each of the 10-gene markers in the panel. Insets of 1) Box plots show the median cumulative methylation of cancer (IDC/DCIS) versus benign/normal tissues. 2) ROC analyses indicate the discriminatory power of the 10-gene marker panel. A. Training set. The laboratory threshold that provided highest sensitivity and specificity for detection of cancer in the training cohort was 14.5 CMI, based on ROC analysis. B. Test set. Conditions established in the training set were locked and performance of the test set was evaluated as in A. Insets of 1) Box plots show the median cumulative methylation of cancer (IDC/DCIS) versus benign/normal tissues. 2) ROC analyses indicate the discriminatory power of the 10-gene marker panel. The dashed line indicates our target sensitivity at 90%, and dot indicates the cutoff at 14.5 CMI units on the curve. CMI, cumulative methylation index; IDC, invasive ductal carcinoma; ILC, invasive lobular carcinoma, DCIS, ductal carcinoma in situ; FA, fibroadenoma; Pap, papilloma; UDH, usual ductal hyperplasia, and cysts. ROC, receiver operator characteristic; AUC, area under the ROC curve. Mann-Whitney P-values are indicated.

Verification of the Performance of the 10-gene Marker Panel.

Parameters defined in the training set were locked for analysis in the independent test set of 220 cancer, benign and normal breast tissues using QM-MSP. We calculated the cumulative methylation in the test set for all 10 genes in the panel, shown as a histogram (Figure 3B). The marker panel in the test set was significantly more methylated in IDC/ILC/DCIS compared to benign/normal tissues as shown in the box plot (P<0.0001; Figure 3B, inset; Mann-Whitney). Using ROC analyses, the sensitivity achieved was 87% (95% CI = 80–93%) and specificity was 88% (95% CI = 80–94%) at a CMI threshold of 14.5 units (AUC = 0.937, 95% CI = 0.900–0.970; P<0.0001) (Figure 3B, inset). We also analyzed the performance of the assay for detecting IDC, ILC and DCIS individually (Supplementary Figure S2). For DCIS alone the sensitivity was 77% (95% CI = 58–90%) and specificity was 88% (95% CI = 80–94%), and the AUC was 0.881 (95% CI = 0.777–0.962) (Supplementary Figure S2). For detection of IDC and ILC alone, sensitivity was higher at 91% (95% CI = 81–97%) and 91% (95% CI = 72–99%) respectively, specificity was 88% and AUC was (95%CI = 80–94%) (Supplementary Figure S2).

Influence of Age on DNA Methylation in Test and Training sets.

Age of blocks.

Formalin and other fixation methods used in preparation of FFPE tissues can result in degradation of DNA, which may increase with the age of the blocks (25, 26). Since our FFPE tissues ranged in age from 2–28 years, we tested the influence of block age on CMI by QM-MSP assay. Our results showed that linear regression model of CMI as a function of FFPE block age, fit on benign/normal samples, demonstrates a modest but not significant decrease in methylation with greater block age (P=0.064; R² = 0.017) (Supplementary Figure S4) as assessed by QM-MSP.

Age of patients.

Age-related DNA hypermethylation in normal tissues has been observed to be associated with the subsequent development of cancer (27–29). This raises a concern for cancer detection tests since the presence of hypermethylation in normal tissue can be a source of false positives. Therefore, we investigated whether CMI of our marker panel was associated with patient age. A linear regression model of CMI fit on benign/normal samples demonstrated a modest but not significant increase in methylation with greater age (P = 0.063, R² = 0.017; Supplementary Figure S3A) with the QM-MSP assay. A coefficient = 0.011 corresponds with a doubling of CMI from 5 to 10 units over 91 years. We also performed logistic regression models of misclassification rates as a function of age, When fit on benign/normal samples, these data show a modest increase in the log-odds of false positives with increased age (P = 0.054; Supplementary Figure S3B). Fit on tumor samples, these data show a decrease in the risk of false negatives for older patients, although the effect is very small and not statistically significant (P = 0.478).

Methylation Frequency by Immunohistochemical (IHC) Subtype and Region.

IHC subtype.

Using a laboratory threshold value of 14.5 CMI units, each of the subtypes had a similar percentage of tumors that scored positive by QM-MSP (86–100%; P = 0.50 by Fisher’s Exact, Supplementary Table S2A) and no significant difference (P = 0.077) in CMI of the 10-gene panel (Supplementary Figure S5A). However, between the four subtypes the individual markers showed a varying extent of methylation (Supplementary Table S2B and Figure S5A). For example, ZNF671 was significantly more methylated in ER/PR-, HER2- (TNBC) compared to ER/PR+, HER2- tumors (P<0.0001). For ER/PR-, HER2+ tumors the methylation of APC, RASGRF2, RASSF1, and TMEFF2 was higher than for other IHC subtypes (P values not adjusted) (Supplementary Table S5 and Table S2B).

Region.

Cumulative methylation of the 10-gene panel did not differ significantly between the regions of U.S., China, and S. Africa (P = 0.265 for benign/normal and P = 0.474 for cancer, Kruskal-Wallis; Supplementary Figure S5B; Supplementary Table S3). However, for individual markers, regional differences were detected for CCND2, COL6A2, HIST1H3C and ZNF671 (Kruskal-Wallis P values = 0.010, 0.003, 0.003, 0.043 respectively) (Supplementary Figure S5B). AUC values showed modest differences between U.S. (AUC = 0.941; 95% CI = 0.890–0.983), China (AUC = 0.901; 95% CI = 0.813–0.970), and S. Africa (AUC = 0.958; 95% CI 0.872–1.00) (Supplementary Figure S5C).

Collectively, the 10-gene marker panel performed reliably as a “pan breast cancer” detection tool across the main immunohistochemical subtypes of breast cancer and within the three distinct regional populations evaluated.

Development of an Automated Cartridge for Quantitative Analysis of the 10-gene Marker Panel.

QM-MSP requires substantial technical expertise and a relatively long 10-day turn-around time from DNA extraction to quantitation of cumulative methylation of a 10-gene marker panel. Therefore, we endeavored to develop the assay on the automated GeneXpert® assay platform that requires minimal technical training, could be completed within a few hours of sample collection, and could be widely applied in breast screening clinics in low resource regions of the world. The distribution of the gene markers in the two cartridges was determined by systematically optimizing combinations of the six gene probes and their fluorophore-tags, PCR reagents, PCR cycle numbers and external amplicon dilutions. The optimized conditions were tested using mixtures of fully methylated CAMA-1 breast cancer cells and unmethylated human sperm DNA.

Interassay Reproducibility.

To test interassay reproducibility we analyzed Set 1 and Set 2 markers in the cartridge in 6 replicate assays. We examined a range of dilutions of fully methylated CAMA-1 cell DNA diluted with human sperm DNA, which is unmethylated for all 10 markers, keeping the amount of input DNA constant (1 ng total DNA). The assays were highly reproducible over a wide range of dilutions with the coefficient of variation (CV) ranging from 6% to 23% (Figure 4A). The 10-marker set maintained log2 linearity for each marker (r² =0.78–0.98) in the dilution range of 3.12% to 100% of methylated DNA, indicating sensitivity to detect as low as 3% methylated target DNA.

Figure 4. Assay validation- Interassay and interoperator reproducibility of the automated cartridge assay.

A. Box-whisker plots show the median and full range of cumulative methylation (CM) (Y-axis) for the 10-gene marker panel using fully methylated CAMA-1 DNA diluted with unmethylated human sperm DNA. The figure shows interassay reproducibility using five-six replicates for each sample (X-axis). Statistical significance is indicated by P values. B. Interoperator reproducibility was evaluated for 10 genes in a set of 23 patient FNA lysates analyzed independently by two investigators. CM measurements for the 10 markers between operators showed an intraclass coefficient (ICC) of 0.99. Kappa statistic for interrater reliability, using the threshold of 14.5 CMI from the training cohort, was 0.91 (95% CI: 0.75–1.08, p value: 6.76e-6).

Interoperator Reproducibility.

Interoperator reproducibility of the cartridge assay was tested by two individuals blinded to the identity of the samples using two aliquots of the same FNA lysate (12 IDC and 11 benign). Cumulative methylation measurements of the 10 markers was highly reproducible between operators, with an intraclass correlation coefficient (ICC) of 0.99 (Figure 4B). Kappa statistic for interrater reliability, using the threshold of 14.5 CMI from the test cohort, was 0.91 (95% CI: 0.75–1.08, p value: 6.76e-6).

Clinical Pilot Testing of FNA Samples.

Archival, stained FNA cells from mammography-detected lesions (24 IDC and 52 benign breast lesions) were obtained from Hong Kong and Portugal, and analyzed in a blinded manner. Three benign samples with <50 cells and ACTB Cts >35 were excluded from data analysis. Robust marker methylation, quantitated using dilution curves ranging from 100% - 3.12% methylation for assessing percent methylation in the sample (Supplementary Figure S6), was detectable in FNA samples of all except one cancer, as shown in the histogram in Figure 5. Difference in cumulative methylation between cancer and benign samples was highly significant (P<0.0001, Mann-Whitney) as shown in the box plot in Figure 5, inset. The GeneXpert® system achieved a detection sensitivity of 96% (95% CI = 79–100%) and a specificity of 90% (95% CI = 78–97%) using ROC threshold of 14.5 CM (P<0.0001, AUC = 0.960, 95% CI = 0.883 – 1.00) (Figure 5, inset, Supplementary Table S4). The assay, from sample preparation to results, was completed within 5 hours.

Figure 5. Pilot test of the automated breast cancer detection cartridge on FNA samples.

Histogram shows the 10-gene marker panel distinguished between IDC (N=24) and benign (N=49) FNA samples from Portugal and Hong Kong, using a cutoff of CM 14.5 that was derived from training cohort of FFPE samples (denoted by horizontal line). Insets of 1) Box plot shows that cumulative methylation is significantly higher in IDC compared to benign breast FNA. 2) ROC analyses indicate the discriminatory power of the 10-gene marker panel. The dashed line indicates our target sensitivity at 90%, and dot indicates the cutoff at 14.5 CMI units on the curve. CM, cumulative methylation, IDC: Invasive ductal carcinoma, FNA: fine needle aspiration, N= number of samples.

DISCUSSION

In this paper, we report the systematic development of a methylated 10-gene marker panel that distinguishes between benign and cancer tissues with a high level of sensitivity and specificity. We also incorporated this panel into an automated breast cancer detection cartridge on the GeneXpert® system to reduce time and increase throughput. The cartridge assay performed with good accuracy, with a sensitivity of 96% (95% CI = 79–100%) and specificity of 90% (95% CI = 78–97%); AUC = 0.960, (95% CI = 0.883 – 1.00) in a pilot study of archival, clinical FNA samples from Portugal and Hong Kong. The assay is simple to perform and has a time-to-result of less than five hours. Thus, the assay could be easily deployed to underserved regions of the world for the discrimination of breast cancer from benign lesions. Our results support the utility of assessing methylated markers in FNA for breast cancer screening in women with suspicious lesions.

Our data and outcomes attest to the robustness of the 10-gene panel of markers (Figures 3 and 5) to distinguish between benign and cancer lesions. We did not observe significant differences in cumulative methylation of the 10-gene marker panel by geography or IHC subtype, indicating that our marker panel captures the diversity across different geographical regions and ER/PR/HER2 and triple negative IHC-subtypes of breast cancer. Some genes in the panel, such as APC, CCND2, RASGRF2 and TMEFF2 were frequently methylated in ER+/PR+ tumors, APC, AKR1B1 and RASSF1 were frequently methylated in HER2 over-expressed tumors, while ZNF671 was more frequently hypermethylated in TNBC, also consistent with reports in the literature (30–32).

Our study has several strengths as well as a few limitations. We chose to develop a DNA methylation-based assay because aberrant DNA methylation is detectable very early in nearly all breast cancers (23, 33, 34). As has been well established, gene promoter-associated CpG island methylation can silence tumor suppressor genes, much like gene deletions or mutations. Moreover, methylation alterations, in contrast to mutations, are extremely common in breast cancer (35–37). This allowed us to develop a marker set capable of detecting breast cancer in more than 90% of the samples in ethnically varied populations of the world irrespective of IHC subtype. Another important feature of the test is the high specificity in differentiating between normal and benign tissues displayed in large sample sets for discovery, training and testing that included a wide variety of benign lesions encompassing cysts, usual ductal hyperplasia, fibroadenoma and papilloma. A key limitation is that the FNA pilot was small (n=73) and used archival samples of convenience. Further validation of the GeneXpert® assay in prospective blinded clinical trials of FNA samples of benign and malignant lesions in various clinical settings is necessary to assess the full potential of this assay to reduce the time to breast cancer detection.

Our findings also demonstrated several advantages of automated GeneXpert® breast cancer detection using the GeneXpert® cartridge. Careful selection of a 10-gene marker panel using large cohorts of cancer and benign tissue was performed using a highly sensitive laboratory assay, QM-MSP. But, QM-MSP requires technical expertise and is too time consuming for easy operation in the field. The cartridge advantages are as follows: 1) Automated, quick, bisulfite conversion of the sample. 2) In each of two cartridges, six fluorophores in multiplex allowed simultaneous amplification of 5 test markers and the actin control, reducing assay time from ten days for QM-MSP for the ten-gene marker panel to 5 hours or less. 3) Both interassay and interoperator tests validated its precision. 4) The cartridge assay performed with the predicted high accuracy in a pilot study of 73 FNAs from mammographically suspicious breast lesions. 5) The simplified assay, data analysis and interpretation are automated, leaving little margin for operator-errors and; finally, 6) this newly developed methylation cartridge could be easily adapted for detection of other cancers.

In conclusion, our study describes the development of a methylated DNA-based breast cancer detection assay from marker discovery to clinical pilot testing, for use on the automated GeneXpert® system. We have prototyped the first automated breast cancer detection cartridge and demonstrated its analytical and clinical validity. We have defined and validated cutoffs for the discrimination of IDC/ILC/DCIS from benign breast disease. Further, we have demonstrated the feasibility of applying this assay to FNA samples collected from mammographically detected cancer and benign breast lesions. Our future steps are to prospectively validate the FNA-based molecular detection test in major hospitals in the US, China, and African Nations. This automated 5-hour assay could enable our ultimate goal to expedite diagnosis and thereby reduce mortality from breast cancer in the underdeveloped nations of the world.

STATEMENT OF TRANSLATIONAL RELEVANCE.

Breast cancer screening in developed countries has saved millions of lives. In developing countries, delayed diagnosis of breast cancer is a major contributor to high breast cancer mortality rates. Timely diagnosis is dependent on rapid pathologic review of patient samples, a service that is often not available. We propose that a quick ancillary test to prioritize malignant versus benign cases could lead to earlier detection and better survival from cancer. DNA methylation is a molecular hallmark of breast cancer. We developed a panel of cancer-specific DNA methylation markers for use in our automated breast cancer detection cartridge. In a pilot study of fine needle aspirates of lesions detected during breast imaging, the cartridge system was 96% sensitive and 88% specific compared to cytopathology. This automated 5-hour assay has the potential to expedite the diagnosis of breast cancer in low resource settings and thereby help reduce the global mortality rates.