WO2017015125A1 - Stratifying breast cancer risk in women with sclerosing adenosis - Google Patents

Abstract

Materials and methods for individualized prediction of breast cancer risk in women with sclerosing adenosis are provided herein.

Description

STRATIFYING BREAST CANCER RISK IN WOMEN

WITH SCLEROSING ADENOSIS

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority to U.S. Application Serial No. 62/196,169, filed on July 23, 2015. The disclosure of the prior application is considered part of the disclosure of this application, and is incorporated in its entirety into this application.

STATEMENT AS TO FEDERALLY SPONSORED RESEARCH

This invention was made with government support under CA116201 awarded by the National Institutes of Health. The government has certain rights in the invention.

TECHNICAL FIELD

This document relates to materials and methods involved in assessing the risk of breast cancer in subjects having benign breast disease, and particularly in subjects with sclerosing adenosis. This document also relates to methods and materials for treating subjects having sclerosing adenosis and determined to be at increased risk of developing breast cancer.

BACKGROUND

Sclerosing adenosis (SA) is a benign proliferative epithelial lesion of the breast that is found in -22% of all benign breast biopsies. As a group, women with S A have an increased future risk of breast cancer (BC), with a relative risk of more than double as compared to the general population risk. In a research cohort including 2672 women with SA, 311 breast cancers occurred with a median follow-up of 15 years, for an annual breast cancer risk of 0.8% per year. Considering that -1,000,000 benign breast biopsies are performed in the US every year, about 220,000 US women per year are diagnosed with SA, or 2.2 million over 10 years. With an annual breast cancer risk of 0.8% per person per year, 17,600 of the 220,000 women diagnosed annually would be expected to develop breast cancer over the following 10 years. Adding in other women with SA diagnosed every successive year for a total of 10 years, an estimated cumulative 96,760 breast cancers would be expected to develop in all women diagnosed with SA over a 10 year period.

SUMMARY

This document is based, at least in part, on the discovery that benign breast biopsies contain transcriptional alterations that can be used to predict future breast cancer incidence within ten years, demonstrating that essential elements of malignancy are present many years prior to cancer development. This document also is based, at least in part, on the identification of a gene panel that can dramatically improve risk stratification in women with SA. As described herein, the TTC10 gene signature model and a ten gene signature model (also referred to herein as the TEN gene signature model) was developed and used to identify a high risk group for cancer development. The TTC10 gene signature model and the TEN gene signature model can be assessed on formalin-fixed, paraffin-embedded (FFPE) biopsies, and constitutes a prognostic biomarker for SA.

The materials and methods described herein provide an RNA-based assay that can be performed on whole tissue sections from FFPE tissue blocks of the benign breast biopsy tissue. Compared to the standard risk assessment tool (BCRAT; online at cancer.gov/bcrisktool/), which has an AUC of 0.64, adding the results of the gene signature panel improves the AUC to 0.91, greatly enhancing the ability to identify the high risk subset of women with SA who will likely develop breast cancer in the absence of intervention. This assay can allow the identification of high and lower risk women with SA for whom clinical management can be tailored; high risk women can be advised to undergo increased surveillance and prevention therapies, and lower risk women may reduce mammographic screening to a biennial interval. If the high risk women can be identified with the use of the gene signature panel at 90% accuracy, prevention therapies can be recommended to these women. For example, tamoxifen, raloxifene, anastrozole, and exemestane can reduce the risk of breast cancer. Women would be more likely to utilize these therapies knowing that they are at significantly increased risk. Therefore, the use of the breast tissue gene signature to identify high risk women with SA could prevent -40,000 breast cancers over a 10 year period, along with their associated treatment costs (e.g., surgery, radiation, and/or chemotherapy). Conversely, the remaining about 2 million women with SA who are found to be at low risk of developing breast cancer could be reassured that their cancer risk is low, and could lengthen mammographic screening intervals from annual to biannual, also resulting in reduction in public health care costs.

In general, one aspect of this document features a method for identifying a mammal having sclerosing adenosis as being likely to develop breast cancer. The method comprises, or consists essentially of, (a) determining that the mammal has a TTC IO gene signature or a TEN gene signature, and (b) classifying the mammal as being likely to develop breast cancer based at least in part on the presence of the TTC IO gene signature or TEN gene signature. The mammal can be a human. The determining can comprise analyzing a biological sample obtained from the mammal, wherein the biological sample comprises breast cells or breast tissue. The method can comprise classifying the mammal as being likely to develop breast cancer within ten years.

In another aspect, this document features a method for identifying a mammal having sclerosing adenosis as not being likely to develop breast cancer. The method comprises, or consists essentially of, (a) determining that the mammal does not have as a TTCI O gene signature or a TEN gene signature, and (b) classifying the mammal as not being likely to develop breast cancer based at least in part on the absence of the TTC IO gene signature or the TEN gene signature. The mammal can be a human. The determining can comprise analyzing a biological sample obtained from the mammal, wherein the biological sample comprises breast cells or breast tissue. The method can comprise classifying the mammal as not being likely to develop breast cancer within ten years.

In another aspect, this document features a method for treating a mammal having sclerosing adenosis. The method comprises, or consists essentially of, administering to the mammal a composition comprising an agent capable of reducing the risk of developing breast cancer. The mammal can be a human. The agent can be tamoxifen, raloxifene, anastrozole, or exemestane. The mammal can be identified as having a TTCI O gene signature or a TEN gene signature prior to the administering step.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Although methods and materials similar or equivalent to those described herein can be used to practice the invention, suitable methods and materials are described below. All publications, patent applications, patents, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control. In addition, the materials, methods, and examples are illustrative only and not intended to be limiting.

The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the invention will be apparent from the description and drawings, and from the claims.

DESCRIPTION OF DRAWINGS

FIG. 1 is a hematoxylin and eosin image showing the histology of SA (arrow) in a field containing two normal lobules (arrowheads). Scale bar = 1 mm.

FIGS. 2A-2E demonstrate the development and validation of the SA TTCIO model. FIG. 2A is a graph plotting the mean area above the receiver operating characteristic (ROC) curve against the number of top genes included in the classifiers. FIG. 2B is a graph plotting average gene expression values, indicating probes that were used for model building (>45% positive expression, p<0.1 for the difference between cases and controls) and locations of which probes were included in the model. Probes passing the filtering threshold are shown in red in the right portion of the graph, those filtered out are shown in blue in the left portion of the graph, and probes selected as final-model-features are shown as large black dots. FIGS. 2C and 2D are graphs plotting the ROC for the SA TTCIO model applied to a training set (FIG. 2C; N=86) and a validation set (FIG. 2D; N=65). FIG. 2E is a graph plotting SA TTC10 predictions for the training and validation dataset cases and controls. The vertical dashed line separates the samples into those predicted to be a TTClO-control (prediction metric less than or equal to 0) or TTClO-case (prediction metric greater than 0).

FIGS. 3A and 3B are graphs plotting the time-to-cancer distributions within the SA TTCIO prediction groups. These Kaplan Meier plots visualize the distribution of actual time-to-cancer within predicted case/control groups in training (FIG. 3A) and validation (FIG. 3B) cohorts. FIGS. 4A-4D indicate that the combination of the SA TTC10 model with BCRAT or BBD-BC models can improves the performance of each. FIGS. 4A and 4B are graphs plotting the ROC for BCRAT (FIG. 4A) and BBD-BC (FIG. 4B) models with a S A training set. FIGS. 4C and 4D are graphs plotting the ROC for SA TTC10 combined with BCRAT (FIG. 4C) and BBD-BC (FIG. 4D) models for a SA validation set.

FIGS. 5A and 5B are graphs plotting the combination of SA TTC10 model metric for replicated samples. FIG. 5A shows the plate 1 vs. plate 2, while FIG. 5B shows the replicates included in the Model Validation dataset.

FIGS. 6A-6C are graphs indicating the development of a gene signature-based model of breast cancer risk for patients with atypia. FIG. 6A is a graph plotting the classification of BC cases (patients who subsequently developed BC) and controls (patients who did not develop BC) in the training dataset according to the model risk prediction. FIG. 6B is a graph plotting survival analysis of predicted risk in the training dataset (HR 6.6; p<0.0001). FIG. 6C is a graph plotting the estimated 10-year risk of developing BC as a continuous function (dashed curved lines, 95% CI; dashed straight lines, estimated risk for scores of 10, 0, and -10). More precise estimates are seen for lower values and lower risk levels because of the greater number of patients, as indicated by the rug plot along the x-axis.

DETAILED DESCRIPTION

Patients with benign breast disease (BBD) are at substantially increased risk for subsequent development of breast cancer. Existing models for risk assessment perform poorly at the individual level, however.

SA is a common BBD lesion, with SA patients representing about 25% of all BBD patients. SA is characterized by epithelial proliferation, disordered acinar architecture, and stromal fibrosis (FIG. 1 ; Jensen et al, Cancer 64(10): 1977-1983, 1989; Visscher et al, Breast Cancer Res Treat 144(1):205-212, 2014, and Hartmann et al, NEngl J Med 353(3):229-237, 2005). Investigation of a BBD cohort revealed that S A was present in 28% of the cohort, and was associated with an approximate doubling of risk of subsequent BC (Visscher et al, supra). The increased risk indicates that premalignant changes are likely to be present in some patients with SA. This document provides RNA-microarray-based transcriptional models for breast cancer risk prediction for patients with SA. As described in the Examples herein, a training set was developed from 86 patients diagnosed with SA, of which 27 subsequently developed cancer within 10 years (cases) and 59 remained cancer-free at 10 years (controls). A diagonal linear discriminate analysis (DLDA)-prediction model for prediction of cancer within 10 years (referred to herein as SATTC10, or a TTC10 gene signature) was generated from transcriptional profiles of FFPE biopsy-derived RNA, and the model was tested on a separate validation case-control set composed of 65 SA patients. These studies revealed that the SA TTC10 gene signature model, composed of 35 gene features, achieved a clear and significant separation between case and control, with receiver operating characteristic (ROC) area under the curve of 0.913 in the training set and 0.836 in the validation set. These studies provided the first demonstration that benign breast tissue contains transcriptional alterations that indicate risk of breast cancer development, and demonstrated that essential precursor biomarkers of malignancy are present many years prior to cancer development. The SATTC10 gene signature model also constitutes a novel prognostic biomarker for patients with S A; S A patients with the signature can be classified as being at increased risk of breast cancer over the next ten years, while SA patients not having the signature can be classified as not being at increased risk of developing breast cancer over the next ten years.

These studies also revealed that the TEN gene signature model, composed of ten gene features, achieved a clear and significant separation between case and control (see Example 3). The presence of, for example, the TEN gene signature can be confirmed when a sample is identified as having altered expression levels (e.g., altered mRNA expression levels) of the AK5, DLK2, EXOC6, ITGA6, ITPRIPLl, KIT, LRRC4B, PSMB1, RGS12, and SORBS 12 genes as described herein. For example, the presence of the TEN gene signature can be confirmed when a sample is identified as having altered expression levels (e.g., altered mRNA expression levels) of the AK5, DLK2, EXOC6, ITGA6, ITPRIPLl, KIT, LRRC4B, PSMB1, RGS12, and SORBS 12 genes as compared to the levels observed from samples obtained from a population of patients with SA who did not develop BC within ten years of diagnosis of a breast condition. The methods described herein provide microarray-based gene signatures to predict risk of later BC from benign breast tissue, and provide a template for a clinical assay with considerable translational potential.

Thus, this document provides methods for determining the risk of breast cancer in a subject (e.g., a mammal such as a human, dog, cat, horse, cow, pig, sheep, goat, monkey, ape, hamster, rat, or mouse) having a BBD such as SA. The methods can include providing a biological sample containing breast cells or breast tissue (e.g., a FFPE biopsy sample) from the subj ect, analyzing the sample according to the methods described in the Examples below to determine whether the subject has a TTC IO gene signature or a TEN gene signature, which indicates that the subject has an increased risk of developing breast cancer. In addition, when it is determined that a subject is at increased risk of breast cancer, the methods provided herein can further include treating the subject in an attempt to reduce the risk of developing breast cancer.

As described herein, subjects having SA can be assessed to determine whether or not they contain breast tissue or breast cells having a TTCI O gene signature or a TEN gene signature. Subjects determined to have a TTC IO gene signature or a TEN gene signature can be classified as being more likely to develop (or having an increased risk of developing) breast cancer, while subjects determined not to have a TTC IO gene signature or a TEN gene signature can be classified as not being more likely to develop (or not having an increased risk of developing) breast cancer, at least within the ten years immediately following the assessment.

Any appropriate method can be used to determine whether or not a biological sample from a subject (e.g., a breast biopsy) contains cells having a TTCI O gene signature or a TEN gene signature. For example, RNA microarray-based analyses such as those described in the Examples below can be used. The genes included in the signature can include those listed in Table 2 or Table 4 herein, for example.

In addition, this document provides methods and materials for treating a subject having SA in order to reduce their risk of developing breast cancer. For example, a subject with SA and identified as having a TTCI O gene signature or a TEN gene signature can be treated with an agent that is capable of reducing the risk of breast cancer. For example, an agent such as tamoxifen, raloxifene, anastrozole, or exemestane, or a combination of such agents, can be administered to reduce the risk of breast cancer.

In some cases, the agent(s) can be formulated into one or more

pharmaceutically acceptable compositions for administration to an SA patient. For example, a therapeutically effective amount of an agent that reduces the risk of breast cancer can be formulated with one or more pharmaceutically acceptable carriers (additives) and/or diluents. A pharmaceutical composition can be formulated for administration in solid or liquid form including, without limitation, sterile solutions, suspensions, sustained-release formulations, tablets, capsules, pills, powders, and granules.

Pharmaceutically acceptable carriers, fillers, and vehicles that may be used in a pharmaceutical composition described herein include, without limitation, ion exchangers, alumina, aluminum stearate, lecithin, serum proteins, such as human serum albumin, buffer substances such as phosphates, glycine, sorbic acid, potassium sorbate, partial glyceride mixtures of saturated vegetable fatty acids, water, salts or electrolytes, such as protamine sulfate, disodium hydrogen phosphate, potassium hydrogen phosphate, sodium chloride, zinc salts, colloidal silica, magnesium trisilicate, polyvinyl pyrrolidone, cellulose-based substances, polyethylene glycol, sodium carboxymethylcellulose, polyacrylates, waxes, polyethylene- poly oxypropylene-block polymers, polyethylene glycol, and wool fat.

A pharmaceutical composition containing one or more agents that reduce the risk of breast cancer can be designed for oral or parenteral (including subcutaneous, intramuscular, intravenous, and intradermal) administration. When being administered orally, a pharmaceutical composition can be in the form of a pill, tablet, or capsule. Compositions suitable for parenteral administration include aqueous and non-aqueous sterile injection solutions that can contain anti-oxidants, buffers, bacteriostats, and solutes which render the formulation isotonic with the blood of the intended recipient; and aqueous and non-aqueous sterile suspensions which may include suspending agents and thickening agents. The formulations can be presented in unit-dose or multi-dose containers, for example, sealed ampules and vials, and may be stored in a freeze dried (lyophilized) condition requiring only the addition of the sterile liquid carrier, for example water for injections, immediately prior to use. Extemporaneous injection solutions and suspensions may be prepared from sterile powders, granules, and tablets.

Such injection solutions can be in the form, for example, of a sterile injectable aqueous or oleaginous suspension. This suspension may be formulated using, for example, suitable dispersing or wetting agents (such as, for example, Tween 80) and suspending agents. The sterile injectable preparation can be a sterile injectable solution or suspension in a non-toxic parenterally-acceptable diluent or solvent, for example, as a solution in 1, 3-butanediol. Examples of acceptable vehicles and solvents that can be used include, without limitation, mannitol, water, Ringer's solution, and isotonic sodium chloride solution. In addition, sterile, fixed oils can be used as a solvent or suspending medium. In some cases, a bland fixed oil can be used such as synthetic mono- or di-glycerides. Fatty acids, such as oleic acid and its glyceride derivatives can be used in the preparation of injectables, as can natural pharmaceutically-acceptable oils, such as olive oil or castor oil, including those in their polyoxyethylated versions. In some cases, these oil solutions or suspensions can contain a long-chain alcohol diluent or dispersant.

In some cases, a pharmaceutically acceptable composition including one or more agents that capable of reducing breast cancer risk can be administered locally or systemically. For example, a composition can be administered locally by injection into lesions at surgery or by subcutaneous administration of a sustained release formulation. In some cases, a composition can be administered systemically, orally or by injection to a mammal (e.g., a human).

Effective doses can vary depending on the severity of the SA, the route of administration, the age and general health condition of the subject, excipient usage, the possibility of co-usage with other therapeutic treatments, and the judgment of the treating physician. In some cases, one or more chemotherapeutic agents can be used in combination with one or more agents that can reduce breast cancer risk. Such chemotherapeutic agents may include, without limitation, taxane therapies, anthracycline therapies, gemcitabine therapies, and other chemotherapies. Examples of taxane therapies include, without limitation, cancer treatments that involve administering taxane agents such as paclitaxel, nanoparticle albumin bound paclitaxel (nab-paclitaxel), docetaxel, or other microtubule disrupting agents such as vinblastine, vincristine, or vinorelbine. In some cases, drugs used to treat gout or colchicine can be used as a mitotic inhibitor to treat a mammal at increased risk of breast cancer. Examples of anthracycline agents such as doxorubicin, liposomal doxorubicin, and epirubicin. Other chemotherapeutics that may be useful include, without limitation, cyclophosphamide, 5-fluorouracil, capecitabine, ixabepilone, erubilin, palbociclib, and methotrexate.

The frequency of administration can be any frequency that reduces the risk of breast cancer over time. For example, the frequency of administration can be from about once a week to about three times a day, or from about twice a month to about six times a day, or from about twice a week to about once a day. The frequency of administration can remain constant or can be variable during the duration of treatment. A course of treatment with a composition containing one or more agents that reduce breast cancer risk can include rest periods. For example, a composition containing one or more such agents can be administered daily over a two week period followed by a two week rest period, and such a regimen can be repeated multiple times. As with the effective amount, various factors can influence the actual frequency of administration used for a particular application. For example, the effective amount, duration of treatment, use of multiple treatment agents, route of administration, and severity of the condition may require an increase or decrease in administration frequency.

An effective duration for administering a composition containing one or more agents capable of reducing breast cancer risk can be any duration that reduces the risk. The effective duration can vary from several days to several weeks, months, or years. In some cases, the effective duration may be for the rest of the patient's lifetime. Multiple factors can influence the actual effective duration used for a particular treatment. For example, an effective duration can vary with the frequency of administration, effective amount, use of multiple treatment agents, route of administration, and severity of the condition being treated.

The invention will be further described in the following examples, which do not limit the scope of the invention described in the claims. EXAMPLES

Example 1 - Development of a gene signature model for BC risk prediction with SA Materials and Methods

Patients and Samples: The study sample included patients selected from the Mayo Clinic BBD Cohort, which has been described elsewhere (Hartmann et al, supra; and Milanese et al, J Natl Cancer Inst 98(22): 1600-1607, 2006). The Mayo BBD Cohort included 9,854 women ages 18 to 85 who had excisional breast biopsy with benign findings between 1967 and 1991. Demographic descriptors and potential breast cancer risk factors were identified via medical record review and from self- response questionnaires. Over a median of 18.9 years of follow-up, 924 of these women have been diagnosed with BC.

Biopsy findings were classified into the following categories: non-proliferative fibrocystic changes (NP), proliferative fibrocystic disease without atypia (PDWA), and proliferative fibrocystic disease with atypia (i.e., atypical hyperplasia; AH) (Dupont and Page, Ή Engl J Med 312(3): 146-151, 1985). SA is a proliferative lesion, without atypia, that consists of enlarged and distorted lobules with prominent myoepithelium and stromal fibrosis. Because of concerns with tissue quality from the older biopsy specimens, eligibility for this study was restricted to the 1,486 women diagnosed with PDWA and S A between the years of 1977 and 1991.

Two sets of breast cancer cases and controls from the SA group were formed for study purposes: one for model development (N=86) and one for model validation (N=65). For each, an age-stratified random sample of women with breast cancer at any time was selected, with selection probabilities proportional to the size of the age strata. Women from the last 10 years of the cohort recruitment period (1982-1991) were preferentially sampled, again under the assumption that tissue quality would be higher for these women than for those from the earlier years of the cohort. An equal number of controls were then frequency -matched to these cases based on 5-year age and year of biopsy categories. The presence of SA lesions in the case/control samples was confirmed and reviewed by a second breast pathologist. Study cohort demographics and clinical characteristics are shown in Table 1.

RNA Extraction and Gene Expression Profiling: RNA was extracted from FFPE samples using the High Pure RNA Paraffin Kit (Roche Diagnostics, Mannheim, Germany). The amount and quality of RNA were assessed with an ND-1000 Spectrophotometer (Nanodrop, Wilmington, DE), and they were considered adequate for further analysis if the optical density 260/280 ratio was > 1.8 and the total RNA yield was > 500 ng. Extracted RNA was labeled and hybridized according to the manufacturer's instruction for the Whole Genome DASL assay (Illumina, San Diego, CA). All samples in the model development set had technical replicates. The first replicate was randomized to one 96-well plate for assay preparation and the second replicate for each sample had a different randomization to a second 96-well plate. For the validation set, only 17 samples were profiled in replicate. Briefly, 200 ng of total RNA was reverse transcribed with biotinylated oligo(dT) and random nonamer primers. The resulting cDNA was annealed to chimeric query oligonucleotides, which contained a gene-specific region and a universal primer sequence for PCR

amplification, and then bound to streptavidin-conjugated paramagnetic particles. The gene-specific oligonucleotides were extended by second-strand cDNA synthesis and then ligated. Subsequently, the products were sequestered by magnetic separation, washed to remove unbound molecules, and then amplified by PCR with fluorophore- labeled universal primers. The resulting PCR products were purified, applied to Human HT-12 v.4 beadchips (Illumina), and then hybridized for 16 hours at 58°C. The beadchips were washed and then scanned in a BeadArray Reader using BeadScan v3 software (Illumina). Quality control parameters were determined to be within normal ranges before proceeding to the final data analysis. Sample probe gene expression values for the 29,377 probes were exported from Illumina Genomestudio and imported into the software R (Team, 2013; online at www.R-project.org/) for normalization, additional quality control, analysis, and prediction model development. The agreement between technical replicates was analyzed for correlation between model metrics in the first and second replicates for the model development set and between the first and second replicates for the validation set, respectively (FIGS. 5A and 5B, respectively).

Gene expression intensities from only the model development set were quantile-normalized in an iterative fashion using a normalization stress metric as described elsewhere (Mahoney et al., BMC Res Notes 6:33, 2013) to exclude failed samples. Samples and their respective replicates were kept for further analysis if the normalization stress metric was less than 0.585=log2(1.5). The model validation set was quantile-normalized separately to the final normalized distribution from the model development set. Normalized gene expression values were transformed to the log2 scale for analysis.

The complete description of the methods involved in development and analysis of the SA "TTC10" (cancer within 10 years) model, including extensive internal model validation, follows.

Model Development: The primary outcome measure used for prediction modeling was the dichotomous time to breast cancer diagnosis within 10 years (TTClO-cases) or no breast cancer within 10 years (TTClO-controls). The model to predict TTClO-cases vs. TTClO-controls was built and tested using the following steps: probe filtering, final-model-feature selection, and final-model building.

Two methods were used for filtering probes for analysis. The first strategy included using no-filtering (i.e., using all probes). The second strategy was to select probes for which more than 45% of the samples had a Genomestudio detection-p- value of less than 0.01 for that probe.

Final-model feature selection was conducted using Monte Carlo cross- validation (MCCV) within the model-development set to determine the optimum number of probes to include in the final-model. The eight main steps in the final- model feature selection were:

(1) Randomly split the model development set into a 2/3 MCCV-training- set and a 1/3 MCCV-test-set while preserving the ratio of TTClO-cases to TTClO-controls.

(2) Using just the MCCV-training set, conduct unequal variance t-tests for each probe to test for differences between TTClO-cases and TTClO-controls.

(3) Rank-order the probes in the MCCV-training set from smallest to largest p-value from the t-test results.

(4) Build a diagonal linear discriminate analysis (DLDA)15 prediction model using gene expression values from the top 2 probes (smallest p-values) and calculate the predicted TTClO-case/control status values of the MCCV- training-set samples along with MCCV-training-set prediction performance and accuracy metrics.

(5) Determine the predicted TTClO-case/control status values for the MCCV-test-set, using the DLDA model and the gene expression values of the top two probes found in step 3. (6) Calculate MCCV-test-set prediction performance and accuracy metrics using the predicted and actual TTClO-case/control status for the MCCV-test- set samples.

(7) Repeat steps 4-6 for the top 3, 4, 5, 10, 15, 20, 25, 30, 35, 40, etc. probes.

(8) Repeat steps 1-7 for a total of 100 MCCV data sets.

MCCV prediction performance and accuracy metrics included area above the ROC curve (AAC) for the MCCV-test-set, MCCV-training-set prediction error, and the MCCV-test-set prediction error. Each of these metrics was summarized by calculating the average metric for filtering strategy and each value of top probes considered across the 100 MCCV data sets. These averages were then plotted versus the number of top probes considered for each of the filtering strategies. The optimal filtering strategy (no-filtering or probe-filtering) and the optimum number of features to include in the final model were chosen based on the local minimum MCCV-test-set average prediction error from the 100 MCCV-test-sets.

All of the plate- 1 replicates were used as the final-model development data set to build the final-model. The filtering strategy selected via the MCCV steps was applied and unequal variance t-tests were used to rank the remaining probes based on the p-values for testing differences between TTClO-cases and TTClO-controls. The list of top ranked probes (probes with the smallest p-values) was cut at the optimum number of probes determined by the MCCV steps above. This resulted in a specific list of probes referred to as the final-model -features. The gene expression values from these final-model-features were then used to build a final DLDA model using the known TTClO-case/control status from the final-model development set. The DLDA prediction model metrics were used to predict the TTC10 case/control status.

Samples with DLDA prediction model metric greater than 0 were predicted to be cases and less than or equal to 0 were predicted to be controls.

Prediction Model Performance and Analysis: The final DLDA model was then used to create predicted values for the model validation set. The gene expression values for the final-model-features from the first replicate of the validation set were used with the DLDA model coefficients from the development set to determine the DLDA model prediction metrics and the predicted TTClO-case/control status values for the validation set. Performance metrics used to evaluate the prediction model included comparing the DLDA model metric and actual TTClO-case/control status using the area under the ROC curve (ROC-AUC), and comparing the actual to predicted TTClO-case/control status using accuracy, sensitivity, specificity, positive predictive value, and negative predictive values. Based on the sampling frame of 1487 women, a 10-year period prevalence of breast cancer of 6% was estimated for calculation of the positive and negative predictive values.

In order to further evaluate the performance of the final model, an assessment was made for how well it predicted the TTCIO status of the samples that were used in the model development, realizing that these metrics are optimistic. The same data that were used to build the final model were then used to create DLDA model prediction metrics and predicted TTClO-case/control values. These metrics were then used to calculate the prediction performance and accuracy metrics for how well the model development set predicted itself. Kaplan-Meier survival curves were built to visualize differences in time-to-cancer between the predicted TTClO-case/control status for both the model development set and model validation set. Since model development was performed for a dichotomous endpoint, HRs and associated p- values will be biased and so are not reported (Dupuy and Simon, J Natl Cancer Inst 99(2): 147-157, 2007).

Since gene expression data from FFPE samples is in general more variable than that from fresh frozen samples, the gene expression values for the final-model- features from the technical replicates from plate-2 in the model development set and the second replicates from the validation set were also used with the final DLDA model to produce DLDA model prediction metrics and TTClO-case/control status. The correlation between the DLDA model metrics was evaluated for the paired- replicates from the model development data set and the validation data set.

Comparisons between the BCRAT, BBD-BC, and the DLDA-models (SA- TTC10) were analyzed in several ways. First, the predicted TTClO-case/control status was compared via the area under the ROC curves (ROC-AUC) for each model individually. Second, the BCRAT and BBD-BC prediction model metrics were separately added in the DLDA final-model build step along with the gene expression values from the final-model-features to produce combined DLDA models for both BCRAT+SA-TTC10 and BBD-BC+SA-TTC10. Only samples that had both the BCRAT and BBD-BC metrics available were included in this analysis. Ten-year risk percentage also was evaluated compared to the DLDA model prediction metric. The ten-year-risk percentage was evaluated at values of the DLDA model prediction metric of -10, 0, and 10.

Results

The development of BC at 10 years was used as the primary end point for model development. The 86 patients in the model development set included 27 TTClO-cases and 59 TTClO-controls, and the 65 patients in the validation set included 10 TTClO-cases and 55 TTClO-controls, with 3 and 12 replicates respectively. The optimum filtering-strategy of 45% detected-p-value< 0.01 and 35 probes was chosen based on the local-minimum (to prevent over-fitting) of the average prediction error from the 100 Monte Carlo cross-validation-test sets (FIG. 2A), which filtered out 8873 probes from further analysis (FIG. 2B). The 35 probes selected as final-model-features are shown in Table 2.

Prediction performance for the model development set was ROC AUC = 0.91 (95% CI, 0.87, 0.95) with a prediction accuracy of 80% (95% CI, 70, 88%; FIG. 2C). The independent validation set had an ROC AUC = 0.84 (95% CI, 0.75, 0.92) with a prediction accuracy of 58% (95% CI, 46, 71%; FIG. 2D). Actual TTCIO case/control status and the continuous version of the DLDA prediction model metric for both the model development and validation sets are displayed (FIG. 2E). Full prediction performance and accuracy metrics are shown in Table 3. Actual time-to-cancer based on the predicted group for each sample is displayed in Kaplan-Meier plots (FIGS. 3A and 3B).

Combined BCRAT+TTC10 and BBD-BC+TTC10 models were built for the samples where both the BCRAT and BBD-BC model metrics were available (FIGS. 4A-4D). All patient samples in the model development set could be used, and 10 TTCIO cases and 41 TTCIO controls were available from the validation set. For the training set, the TTCIO model improved discrimination of the BCRAT model, with AUC=0.64 (95% CI 0.57 - 0.71) for BCRAT alone, 0.91 (95% CI 0.87 - 0.95) for BCRAT combined with TTCIO, and 0.91 (95% CI 0.87 - 0.95) for TTCIO alone (FIG. 4A). The TTCIO model also improved discrimination of the BBD-BC model, with AUC of 0.63 (95% CI 0.56 - 0.70) for BBD-BC alone, 0.93 (95% CI 0.89 - 0.97) for BBD-BC combined with TTCIO, and 0.91 (95% CI 0.87 - 0.95) for TTCIO alone (FIG. 4B). Similarly, for the validation set, the TTCIO model also improved discrimination: AUC = 0.55 for BCRAT alone, 0.79 (95% CI 0.70 - 0.87) for BCRAT combined with TTCIO, and 0.80 (95% CI 0.71 - 0.89) when TTCIO is used alone (FIG. 4C). Improvements of TTCIO with the BBD-BC model are as follows:

AUC=0.75 (95% CI 0.65 - 0.85) for BBD-BC alone, 0.82 (95% CI 0.73 - 0.91) for BBD-BC combined with TTCIO, and 0.80 (95% CI 0.71 - 0.89) for TTCIO alone (FIG. 4D). Thus, in women with SA, the SATTCIO model provides improved and independent risk assessment compared to those of the BCRAT and the BBD-BC models.

Table 1. Characteristics of study set and comparison to overall SA patient cohort.

Family History of Breast Cancer 0.0249

Missing 3 0

None 814 (62.0%) 87 (51.2%)

Weak 337 (25.7%) 56 (32.9%)

Strong 162 (12.3%) 27 (15.9%)

Table 2. TTC10 model features.

Model Gene Entrez

GenelD coefficient Symbol description

ILMN_1653102 -0.948 CCDC64 coiled-coil domain containing 64

ILMN_1656761 -1.098 TGIF1 TGFB-induced factor homeobox 1 hypoxia inducible lipid droplet-

ILMN l 659990 -0.737 HILPDA associated

methylenetetrahydrofolate

ILMN_1674706 0.418 MTHFD2 dehydrogenase2

ILMN_1679252 0.475 ZNF546 zinc finger protein 546

ILMN_1679687 0.420 EXOC6 exocyst complex component 6

ILMN 1681846 0.370 ZNF540 zinc finger protein 540

neuropeptide FF-amide peptide

ILMN_1682503 0.500 NPFF precursor

ribosomal RNA processing 15

ILMN 1685661 0.371 RRP15 homolog

RNA binding protein SI, serine-rich

ILMN_1691843 -2.429 RNPS1 domain

potassium voltage-gated channel,

ILMN_1704063 0.966 KCNH3 subfamily H, member 3

ILMN_1716407 0.407 SORBS2 sorbin and SH3 domain containing 2

ILMN_1717393 0.377 PTCHD1 patched domain containing 1

ILMN_1726873 -3.494 TPCN2 two pore segment channel 2

transcription elongation factor A

ILMN_1726928 0.381 TCEA3 (SII), 3

ILMN_1735762 -0.670 NPNT nephronectin

ILMN_1738229 -3.556 NDRG3 NDRG family member 3

ILMN_1751141 0.406 RGS12 regulator of G-protein signaling 12

La ribonucleoprotein domain family,

ILMN_1752810 0.404 LARP6 member 6

mannosidase, alpha, class 2B,

ILMN_1768510 -4.436 MAN2B2 member 2

pellino E3 ubiquitin protein ligase

ILMN_1780132 0.368 PELI2 family member 2

N-ethylmaleimide-sensitive factor

ILMN_1788268 -3.978 NAPG attachment protein, gamma

proteasome (prosome, macropain)

ILMN_1789176 -3.926 PSMB1 subunit, beta type, 1

ILMN_1793537 0.478 MUC15 mucin 15, cell surface associated inositol 1,4,5-trisphosphate receptor

ILMN_1798373 0.602 ITPRIPL1 interacting protein-like 1

ILMN_1804673 0.498 SLC16A4 solute carrier family 16, member 4 potassium channel tetramerization

ILMN_1809708 0.475 KCTD21 domain containing 21

testis-specific transcript, Y-linked

ILMN_2055391 -0.728 TTTY17A 17A (non-protein coding)

ILMN_2058975 0.619 UFL1 UFM1 -specific ligase 1

ILMN_2297196 -0.613 LRRC4B leucine rich repeat containing 4B gem (nuclear organelle) associated

ILMN_2344002 0.373 GEMIN2 protein 2

ATPase, H+ transporting, lysosomal

ILMN_2353642 -1.721 ATP6V0B 21kDa, VO subunit b

ILMN_2387712 0.454 AK5 adenylate kinase 5

ILMN_3251497 0.377 GSTA1 glutathione S-transferase alpha 1

ILMN_3310790 0.594 MIR626 microRNA 626

Table 3. Model metrics for the TTCIO model, unless otherwise specified.

Example 2 - Atypia model

Within the BBD cohort, 45 women were identified with atypia who subsequently developed BC (cases). Those samples were matched on patient age and era of biopsy with 45 women with atypia that did not develop cancer (controls). RNA was isolated from whole tissue sections from the benign biopsies for these women, and a 40 gene signature classifier was generated based on the Illumina DASL microarray platform. A list of the genes and coefficients is shown in Table 4, with links to relevant known functions. A key feature of this model is that it works very well to determine 10-year breast cancer risk (FIGS. 6A-6C).

Table 4. Atypia model features.

ILMN_3240231 4.9992 SNORA34 Small Nucleolar RNA, H/ACA Box 34

ILMN_2290118 1.0093 MEGF9 Multiple EGF-Like-Domains 9

ILMN_1728298 0.8558 SBK1 SH3 Domain Binding Kinase 1

ILMN_1678531 0.5347 N4BP3 NEDD4 Binding Protein 1

ILMN_1757732 0.8820 OSGIN2 Oxidative Stress Induced Growth Inhibitor

Family Member 2

ILMN_1756784 0.5608 FREQ Neuronal Calcium Sensor 1

ILMN_2378100 2.2459 FBXL5 F-Box And Leucine-Rich Repeat Protein 5

ILMN_1729272 0.9266 DIP2A DIP2 Disco-Interacting Protein 2 Homolog

A (Drosophila)

ILMN_1808500 0.5946 CEP68 Centrosomal Protein Of 68 KDa

ILMN_1769694 -1.7684 ACCN2 Acid-Sensing (Proton-Gated) Ion Channel

1

ILMN_1676555 0.5829 TTC26 Tetratricopeptide Repeat Domain 261

ILMN_2112256 -2.1959 TNFRSF4 Tumor Necrosis Factor Receptor

Superfamily, Member 4

ILMN_1705774 0.7941 TIGD5 Tigger Transposable Element Derived 5

ILMN_2148360 0.5403 ADAM 10 ADAM Metallopeptidase Domain 10

ILMN_3240069 4.2160 SCARNA4 Small Cajal Body-Specific RNA 4

ILMN_3244216 5.1715 SCARNA3 Small Cajal Body-Specific RNA 3

Example 3 - Gene signature model for BC risk prediction with SA

The Nanostring nCounter platform was used to further assess gene signature models for BC risk predictions in those patients with SA. The Nanostring analytical method has improved reproducibility, sensitivity, and reduced background signal relative to microarray analysis, and is optimal for assessing mRNA derived from formalin-fixed, paraffin-embedded (FFPE) samples (Norton et al., PLoS One,

8:e81925 (2013) and Reis et al, BMC Biotechnol, 11 :46 (2011)).

A Nanostring codeset was generated. It contained the original 35 genes from the TTCIO model (see, Table 2) along with 26 additional genes selected for biological relevance. The Nanostring assay was performed on 150 patient samples to produce an SA Nanostring dataset. 15 of the genes exhibited a significant predictive value

(p<0.05) between cases and controls, validating the results provided herein (Table 5).

Table 5. Nanostring assessment of TTCIO model.

probe 10 year tumor hazard ratio wilcoxon p-value

MUC15 1.7136 0.0005

KIT 1.8591 0.001

EXOC6 1.6711 0.0016

SORBS2 2.8659 0.0016 DLK2 1.8729 0.0019

ST6GALNAC5 1.6627 0.0063

AK5 1.6588 0.0069

LRRC4B 0.673 0.0072

BRCA1 1.5649 0.01

RGS12 2.0695 0.0109

DDR1 2.5295 0.0132

ITGA6 2.9241 0.0134

PTCHD1 1.4252 0.0217

PSMB1 1.6954 0.0226

TGIF1 1.8219 0.0454

TNK1 1.6868 0.0606

RNPS1 2.1834 0.07

NPNT 1.3512 0.0924

ITPRIPLl 1.4125 0.0947

DIAPH3 1.0478 0.1062

UIMC1 0.7928 0.1122

TCEA3 1.5051 0.1131

KCTD21 1.5301 0.1171

NAPG 1.4124 0.1222

BTBD11 1.536 0.1285

SLC16A4 1.4663 0.1489

LARP6 1.4148 0.1516

NPFF 0.7305 0.1578

ZNF546 1.2982 0.1693

TTTY17A 0.3133 0.1724

C6orfl50 1.4708 0.1761

RBBP4 1.5609 0.1861

MMP14 1.3094 0.231

STX2 0.9347 0.2327

RRP15 1.5408 0.2343

KCNH3 1.2453 0.2452

MMP17 1.0112 0.2793

TRIM2 0.7623 0.2793

FBX044 1.4968 0.2968

ZRANB3 1.4365 0.3493

SENP7 1.3981 0.3537

MTHFD2 1.295 0.3835

MAN2B2 1.3793 0.4499

GEMIN2 1.1396 0.4707

HMGA1 0.9521 0.4921

HILPDA 1.1194 0.5252

ZNF540 1.2402 0.5336

ATP6V0B 0.8634 0.5393

HOXB6 1.016 0.5406

TPCN2 1.2807 0.5564

TNFSF11 0.9285 0.5691 PELI2 1.1939 0.6154

UFL1 0.8258 0.6521

HSDL1 1.3232 0.6581

GSTA1 0.9353 0.7253

LM07 1.1753 0.7442

CCDC64 1.044 0.7474

NDRG3 1.2433 0.8133

RAC1 1.2222 0.82

USP6NL 1.108 0.8468

EGR2 1.067 0.8907

Analysis of the Nanostring SA dataset are presented as associations with the ten-year endpoint, expressed as a Wilcoxon p-value and hazard ratio. Genes selected for the TEN gene signature model are highlighted in bold.

A gene signature model with ten genes (TEN gene signature model) was selected using the S A Nanostring dataset and the DLDA methodology as described herein. The ten genes were: AK5, DLK2, EXOC6, ITGA6, ITPRIPL1, KIT,

LRRC4B, PSMBl, RGS12, and SORBS 12. This model was assessed using five-fold cross-validation and found to have an average AUC in the training set of 0.78

(expr. train) and an average AUC in the validation set of 0.75 (expr.val). When the clinical variables of patient age and year of biopsy (age.yb) were added to the model, these values improved to 0.80 and 0.75, respectively (Tables 6 and 7).

Table 6. Demographics of model.

These results demonstrate that a model with at least these 10 genes can be a Nanostring based diagnostic for 10-year breast cancer risk in women with sclerosing adenosis. Women who demonstrate significantly elevated risk from this test should consider risk-reduction strategies such as tamoxifen or other antiestrogen cancer prevention therapies, which can result in an up to 75% reduction in breast cancer incidence for high risk populations.

OTHER EMBODIMENTS

It is to be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.

Claims

WHAT IS CLAIMED IS:

1. A method for identifying a mammal having sclerosing adenosis as being likely to develop breast cancer, wherein said method comprises:

(a) determining that the mammal has a TTC 10 gene signature, and

(b) classifying the mammal as being likely to develop breast cancer based at least in part on the presence of said signature.

2. The method of claim 1, wherein the mammal is a human.

3. The method of claim 1, wherein the determining comprises analyzing a biological sample obtained from the mammal, wherein the biological sample comprises breast cells or breast tissue.

4. The method of claim 1, comprising classifying the mammal as being likely to develop breast cancer within ten years.

5. A method for identifying a mammal having sclerosing adenosis as not being likely to develop breast cancer, wherein said method comprises:

(a) determining that the mammal does not have as a TTC10 gene signature, and

(b) classifying the mammal as not being likely to develop breast cancer based at least in part on the absence of the signature.

6. The method of claim 5, wherein the mammal is a human.

7. The method of claim 5, wherein the determining comprises analyzing a biological sample obtained from the mammal, wherein the biological sample comprises breast cells or breast tissue.

8. The method of claim 5, comprising classifying the mammal as not being likely to develop breast cancer within ten years.

9. A method for treating a mammal having sclerosing adenosis, comprising administering to the mammal a composition comprising an agent capable of reducing the risk of developing breast cancer.

10. The method of claim 9, wherein said mammal is a human.

1 1. The method of claim 9, wherein the agent is tamoxifen, raloxifene, anastrozole, or exemestane.

12. The method of claim 9, wherein the mammal is identified as having a TTCI O gene signature prior to said administering step.

13. A method for identifying a mammal having sclerosing adenosis as being likely to develop breast cancer, wherein said method comprises:

(a) determining that the mammal has a TEN gene signature, and

(b) classifying the mammal as being likely to develop breast cancer based at least in part on the presence of said signature.

14. The method of claim 13, wherein the mammal is a human.

15. The method of claim 13, wherein the determining comprises analyzing a biological sample obtained from the mammal, wherein the biological sample comprises breast cells or breast tissue.

16. The method of claim 13, comprising classifying the mammal as being likely to develop breast cancer within ten years.

17. A method for identifying a mammal having sclerosing adenosis as not being likely to develop breast cancer, wherein said method comprises:

(a) determining that the mammal does not have as a TEN gene signature, and

(b) classifying the mammal as not being likely to develop breast cancer based at least in part on the absence of the signature.

18. The method of claim 17, wherein the mammal is a human.

19. The method of claim 17, wherein the determining comprises analyzing a biological sample obtained from the mammal, wherein the biological sample comprises breast cells or breast tissue.

20. The method of claim 17, comprising classifying the mammal as not being likely to develop breast cancer within ten years.

21. A method for treating a mammal having sclerosing adenosis, comprising administering to the mammal a composition comprising an agent capable of reducing the risk of developing breast cancer.

22. The method of claim 21 , wherein said mammal is a human.

23. The method of claim 21 , wherein the agent is tamoxifen, raloxifene, anastrozole, or exemestane.

24. The method of claim 21 , wherein the mammal is identified as having a TEN gene signature prior to said administering step.