Classification performance bias between training and test sets in a limited mammography dataset

OBJECTIVES

To assess the performance bias caused by sampling data into training and test sets in a mammography radiomics study.

METHODS

Mammograms from 700 women were used to study upstaging of ductal carcinoma in situ. The dataset was repeatedly shuffled and split into training (n = 400) and test cases (n = 300) forty times. For each split, cross-validation was used for training, followed by an assessment of the test set. Logistic regression with regularization and support vector machine were used as the machine learning classifiers. For each split and classifier type, multiple models were created based on radiomics and/or clinical features.

RESULTS

Area under the curve (AUC) performances varied considerably across the different data splits (e.g., radiomics regression model: train 0.58-0.70, test 0.59-0.73). Performances for regression models showed a tradeoff where better training led to worse testing and vice versa. Cross-validation over all cases reduced this variability, but required samples of 500+ cases to yield representative estimates of performance.

CONCLUSIONS

In medical imaging, clinical datasets are often limited to relatively small size. Models built from different training sets may not be representative of the whole dataset. Depending on the selected data split and model, performance bias could lead to inappropriate conclusions that might influence the clinical significance of the findings.

ADVANCES IN KNOWLEDGE

Performance bias can result from model testing when using limited datasets. Optimal strategies for test set selection should be developed to ensure study conclusions are appropriate.

Molecular subtypes of high-grade serous ovarian cancer across racial groups and gene expression platforms

Introduction

High-grade serous carcinoma (HGSC) gene expression subtypes are associated with differential survival. We characterized HGSC gene expression in Black individuals and considered whether gene expression differences by race may contribute to poorer HGSC survival among Black versus non-Hispanic White individuals.

Methods

We included newly generated RNA-Seq data from Black and White individuals, and array-based genotyping data from four existing studies of White and Japanese individuals. We assigned subtypes using K-means clustering. Cluster- and dataset-specific gene expression patterns were summarized by moderated t-scores. We compared cluster-specific gene expression patterns across datasets by calculating the correlation between the summarized vectors of moderated t-scores. Following mapping to The Cancer Genome Atlas (TCGA)-derived HGSC subtypes, we used Cox proportional hazards models to estimate subtype-specific survival by dataset.

Results

Cluster-specific gene expression was similar across gene expression platforms. Comparing the Black study population to the White and Japanese study populations, the immunoreactive subtype was more common (39% versus 23%-28%) and the differentiated subtype less common (7% versus 22%-31%). Patterns of subtype-specific survival were similar between the Black and White populations with RNA-Seq data; compared to mesenchymal cases, the risk of death was similar for proliferative and differentiated cases and suggestively lower for immunoreactive cases (Black population HR=0.79 [0.55, 1.13], White population HR=0.86 [0.62, 1.19]).

Conclusions

A single, platform-agnostic pipeline can be used to assign HGSC gene expression subtypes. While the observed prevalence of HGSC subtypes varied by race, subtype-specific survival was similar.

Growth Dynamics of Ductal Carcinoma in Situ Recapitulate Normal Breast Development

Ductal carcinoma in situ (DCIS) and invasive breast cancer share many morphologic, proteomic, and genomic alterations. Yet in contrast to invasive cancer, many DCIS tumors do not progress and may remain indolent over decades. To better understand the heterogenous nature of this disease, we reconstructed the growth dynamics of 18 DCIS tumors based on the geo-spatial distribution of their somatic mutations. The somatic mutation topographies revealed that DCIS is multiclonal and consists of spatially discontinuous subclonal lesions. Here we show that this pattern of spread is consistent with a new 'Comet' model of DCIS tumorigenesis, whereby multiple subclones arise early and nucleate the buds of the growing tumor. The discontinuous, multiclonal growth of the Comet model is analogous to the branching morphogenesis of normal breast development that governs the rapid expansion of the mammary epithelium during puberty. The branching morphogenesis-like dynamics of the proposed Comet model diverges from the canonical model of clonal evolution, and better explains observed genomic spatial data. Importantly, the Comet model allows for the clinically relevant scenario of extensive DCIS spread, without being subjected to the selective pressures of subclone competition that promote the emergence of increasingly invasive phenotypes. As such, the normal cell movement inferred during DCIS growth provides a new explanation for the limited risk of progression in DCIS and adds biologic rationale for ongoing clinical efforts to reduce DCIS overtreatment.

Association of inflammation-related exposures and ovarian cancer survival in a multi-site cohort study of Black women

BACKGROUND

An association was observed between an inflammation-related risk score (IRRS) and worse overall survival (OS) among a cohort of mostly White women with invasive epithelial ovarian cancer (EOC). Herein, we evaluated the association between the IRRS and OS among Black women with EOC, a population with higher frequencies of pro-inflammatory exposures and worse survival.

METHODS

The analysis included 592 Black women diagnosed with EOC from the African American Cancer Epidemiology Study (AACES). Cox proportional hazards models were used to compute hazard ratios (HRs) and 95% confidence intervals (CIs) for the association of the IRRS and OS, adjusting for relevant covariates. Additional inflammation-related exposures, including the energy-adjusted Dietary Inflammatory Index (E-DIITM), were evaluated.

RESULTS

A dose-response trend was observed showing higher IRRS was associated with worse OS (per quartile HR: 1.11, 95% CI: 1.01-1.22). Adding the E-DII to the model attenuated the association of IRRS with OS, and increasing E-DII, indicating a more pro-inflammatory diet, was associated with shorter OS (per quartile HR: 1.12, 95% CI: 1.02-1.24). Scoring high on both indices was associated with shorter OS (HR: 1.54, 95% CI: 1.16-2.06).

CONCLUSION

Higher levels of inflammation-related exposures were associated with decreased EOC OS among Black women.

Ipsilateral Lesion Detection Refinement for Tomosynthesis

Computer-aided detection (CAD) frameworks for breast cancer screening have been researched for several decades. Early adoption of deep-learning models in CAD frameworks has shown greatly improved detection performance compared to traditional CAD on single-view images. Recently, studies have improved performance by merging information from multiple views within each screening exam. Clinically, the integration of lesion correspondence during screening is a complicated decision process that depends on the correct execution of several referencing steps. However, most multi-view CAD frameworks are deep-learning-based black-box techniques. Fully end-to-end designs make it very difficult to analyze model behaviors and fine-tune performance. More importantly, the black-box nature of the techniques discourages clinical adoption due to the lack of explicit reasoning for each multi-view referencing step. Therefore, there is a need for a multi-view detection framework that can not only detect cancers accurately but also provide step-by-step, multi-view reasoning. In this work, we present Ipsilateral-Matching-Refinement Networks (IMR-Net) for digital breast tomosynthesis (DBT) lesion detection across multiple views. Our proposed framework adaptively refines the single-view detection scores based on explicit ipsilateral lesion matching. IMR-Net is built on a robust, single-view detection CAD pipeline with a commercial development DBT dataset of 24675 DBT volumetric views from 8034 exams. Performance is measured using location-based, case-level receiver operating characteristic (ROC) and case-level free-response ROC (FROC) analysis.

Archival single-cell genomics reveals persistent subclones during DCIS progression

Ductal carcinoma in situ (DCIS) is a common precursor of invasive breast cancer. Our understanding of its genomic progression to recurrent disease remains poor, partly due to challenges associated with the genomic profiling of formalin-fixed paraffin-embedded (FFPE) materials. Here, we developed Arc-well, a high-throughput single-cell DNA-sequencing method that is compatible with FFPE materials. We validated our method by profiling 40,330 single cells from cell lines, a frozen tissue, and 27 FFPE samples from breast, lung, and prostate tumors stored for 3-31 years. Analysis of 10 patients with matched DCIS and cancers that recurred 2-16 years later show that many primary DCIS had already undergone whole-genome doubling and clonal diversification and that they shared genomic lineages with persistent subclones in the recurrences. Evolutionary analysis suggests that most DCIS cases in our cohort underwent an evolutionary bottleneck, and further identified chromosome aberrations in the persistent subclones that were associated with recurrence.

Classification performance bias between training and test sets in a limited mammography dataset

Objectives

To assess the performance bias caused by sampling data into training and test sets in a mammography radiomics study.

Methods

Mammograms from 700 women were used to study upstaging of ductal carcinoma in situ. The dataset was repeatedly shuffled and split into training (n=400) and test cases (n=300) forty times. For each split, cross-validation was used for training, followed by an assessment of the test set. Logistic regression with regularization and support vector machine were used as the machine learning classifiers. For each split and classifier type, multiple models were created based on radiomics and/or clinical features.

Results

Area under the curve (AUC) performances varied considerably across the different data splits (e.g., radiomics regression model: train 0.58-0.70, test 0.59-0.73). Performances for regression models showed a tradeoff where better training led to worse testing and vice versa. Cross-validation over all cases reduced this variability, but required samples of 500+ cases to yield representative estimates of performance.

Conclusions

In medical imaging, clinical datasets are often limited to relatively small size. Models built from different training sets may not be representative of the whole dataset. Depending on the selected data split and model, performance bias could lead to inappropriate conclusions that might influence the clinical significance of the findings. Optimal strategies for test set selection should be developed to ensure study conclusions are appropriate.

Tumor-associated nonmyelinating Schwann cell-expressed PVT1 promotes pancreatic cancer kynurenine pathway and tumor immune exclusion

One of the major obstacles to treating pancreatic ductal adenocarcinoma (PDAC) is its immunoresistant microenvironment. The functional importance and molecular mechanisms of Schwann cells in PDAC remains largely elusive. We characterized the gene signature of tumor-associated nonmyelinating Schwann cells (TASc) in PDAC and indicated that the abundance of TASc was correlated with immune suppressive tumor microenvironment and the unfavorable outcome of patients with PDAC. Depletion of pancreatic-specific TASc promoted the tumorigenesis of PDAC tumors. TASc-expressed long noncoding RNA (lncRNA) plasmacytoma variant translocation 1 (PVT1) was triggered by the tumor cell-produced interleukin-6. Mechanistically, PVT1 modulated RAF proto-oncogene serine/threonine protein kinase-mediated phosphorylation of tryptophan 2,3-dioxygenase in TASc, facilitating its enzymatic activities in catalysis of tryptophan to kynurenine. Depletion of TASc-expressed PVT1 suppressed PDAC tumor growth. Furthermore, depletion of TASc using a small-molecule inhibitor effectively sensitized PDAC to immunotherapy, signifying the important roles of TASc in PDAC immune resistance.

Molecular classification and biomarkers of clinical outcome in breast ductal carcinoma in situ: Analysis of TBCRC 038 and RAHBT cohorts

Ductal carcinoma in situ (DCIS) is the most common precursor of invasive breast cancer (IBC), with variable propensity for progression. We perform multiscale, integrated molecular profiling of DCIS with clinical outcomes by analyzing 774 DCIS samples from 542 patients with 7.3 years median follow-up from the Translational Breast Cancer Research Consortium 038 study and the Resource of Archival Breast Tissue cohorts. We identify 812 genes associated with ipsilateral recurrence within 5 years from treatment and develop a classifier that predicts DCIS or IBC recurrence in both cohorts. Pathways associated with recurrence include proliferation, immune response, and metabolism. Distinct stromal expression patterns and immune cell compositions are identified. Our multiscale approach employed in situ methods to generate a spatially resolved atlas of breast precancers, where complementary modalities can be directly compared and correlated with conventional pathology findings, disease states, and clinical outcome.

Abstract PR002: Genomic predictor can discriminate between high- and low-risk DCIS

Introduction: Ductal carcinoma in situ (DCIS) is considered a non-obligate precursor of invasive ductal carcinoma. With the aim of preventing a subsequent invasive cancer, all DCIS lesions are currently treated with surgical excision often supplemented with radiotherapy (RT). To prevent DCIS over- or undertreatment, a reliable marker of DCIS invasiveness risk is urgently needed. Methods: We studied two large DCIS cohorts: the Sloane cohort, a prospective breast screening cohort from the UK (median follow-up of 11 years), and a Dutch population-based cohort (NKI, median follow-up of 13 years). FFPE tissue specimens from patients with pure primary DCIS after breast-conserving surgery (BCS) +/- RT that did develop a subsequent ipsilateral event (DCIS or invasive) were considered as cases, whereas patients that did not develop any form of recurrence up to the last follow-up or death were considered as controls. We performed copy number analysis (CNA) and RNAseq analysis on 229 cases (80 DCIS only recurrences) and 344 controls. Results: DCIS was classified into the PAM50 subtypes using RNAseq data which revealed an enrichment of luminal A phenotype in DCIS that did not recur (P = 0.01, Fisher Exact test). No single copy number aberration was more common in cases compared to controls. RNAseq data did not reveal any genes significantly over/under-expressed in cases versus controls after FDR correction. However, by limiting the analysis to samples that had not had RT and excluding pure DCIS recurrences, we could develop a penalized Cox model from RNAseq data. The model was trained on weighted samples (to correct for the biased sampling of the case-control dataset) from the NKI series with double loop cross-validation. The genes were selected using the Elastic net framework of penalization. Using this predicted hazard ratio, the samples were split into high, medium, and low-risk quantiles, with a recurrence risk of 23%, 7% and 2%, respectively at 5 years (p = 10-10, Wald test). The NKI-trained predictor was independently validated in the Sloane No RT no DCIS recurrence cohort (p = 0.02, Wald test). GSEA analysis revealed proliferation hallmarks enriched in the recurrence predictor (FDR = 0.058). The RNAseq predictor was more predictive of recurrence than PAM50, clinical features (Grade, Her2 and ER) and the 12-gene Oncotype DCIS score (p < 0.001, permutation test using the Wald statistic) in both the NKI and Sloane series. Conclusion: Genomic profiling of two independent series of DCIS with outcome data did not reveal any clear associations with recurrence until analysis was limited to a set of samples who had not had radiotherapy and DCIS recurrences were excluded. We then identified an RNAseq-based classifier that could differentiate primary DCIS in low-, medium-, and high-risk groups, and validated it in an independent cohort. This classifier, if validated in other datasets, will allow us to identify women who do not need intensive treatment for their DCIS.

Citation Format: Maria Roman Escorza, Michael Sheinman, Tycho Bismeijer, Ahmed A. Ahmed, Vandna Shah, Jeffrey R. Marks, Lorraine M. King, Anargyros Megalios, Lindy L. Visser, Marlous Hoogstrat, Helen R. Davies, Tapsi Kumar, Deborah Collyar, Hilary Stobart, Sarah Pinder, Nicholas N. Navin, Andrew Futreal, Serena Nik-Zainal, E. Shelley Hwang, Esther H. Lips, Alastair Thompson, Lodewyk F.A. Wessels, Jelle Wesseling, Elinor J. Sawyer. Genomic predictor can discriminate between high- and low-risk DCIS [abstract]. In: Proceedings of the AACR Special Conference on Rethinking DCIS: An Opportunity for Prevention?; 2022 Sep 8-11; Philadelphia, PA. Philadelphia (PA): AACR; Can Prev Res 2022;15(12 Suppl_1): Abstract nr PR002.

Abstract PR010: Characterizing N-glycan profiles of DCIS progression using tissue imaging MALDI mass spectrometry

Identifying distinct biomarkers of DCIS and determining the invasive potential of DCIS are of great clinical importance. N-linked glycans present on the cell surface of DCIS lesions and surrounding stroma are potential biomarkers of disease and tissue, but remain largely unexplored. An N-glycan targeted imaging mass spectrometry (IMS) approach was initiated to identify N-glycans associated with DCIS and progression in clinical FFPE tissues. A cohort of pathologist annotated DCIS and IDC tissues with a range of disease severity as well as the RAHBT TMA DCIS cohort were evaluated. The RAHBT cohort contains primary DCIS lesions with known long-term outcomes and serves as a good sample set to identify early markers indicating progressive potential, while the DCIS and IDC large tissue biopsies provide information on distinct glycan profiles when DCIS has progressed to IDC. Initially, tissue samples containing DCIS alone or both DCIS and IDC were processed for N-glycan MALDI-IMS analysis, the goal being to establish a consensus panel of each analyte and determine distinct profiles of DCIS when alone vs. when IDC was present. For N-glycans, a peak list of 54 N-glycans was selected for evaluation of each DCIS only tissue, as well as the mixed DCIS/IDC tissues. By highest intensity, the major DCIS-associated glycans were in the high-mannose and paucimannose structure categories. Additionally, a series of tri- and tetra-antennary multi-fucosylated glycans were also detected specifically in the DCIS lesions. In tissues containing both IDC and DCIS lesions, the high mannose glycans were also detected in the IDC regions, and GlcNAc-bisected glycans were seen elevated in these as compared to DCIS-only tissues. For samples in the RAHBT TMA cohort (progressors N=43; non-progressors N=78), the bisected N-glycans seen elevated with IDC were also seen to be significantly increased in the RAHBT samples that would eventually progress to IDC. Additional analyses to evaluate N-glycan isomer distributions for fucosylated and sialylated glycan species are also ongoing. The cumulative N-glycan data will be assessed with the extensive spatial genomic, transcriptomic and immunohistological data already generated for the same samples in the RAHBT cohort. This novel combination of multi-enzymatic digests with histopathology annotations represents an extensive multi-dimensional profile of DCIS and a novel tissue biomarker discovery approach.

Citation Format: Elizabeth N. Wallace, Grace Grimsley, Siri H. Strand, Robert Michael Angelo, Graham Colditz, E. Shelley Hwang, Robert West, Jeffrey R. Marks, Peggi M. Angel, Richard R. Drake. Characterizing N-glycan profiles of DCIS progression using tissue imaging MALDI mass spectrometry [abstract]. In: Proceedings of the AACR Special Conference on Rethinking DCIS: An Opportunity for Prevention?; 2022 Sep 8-11; Philadelphia, PA. Philadelphia (PA): AACR; Can Prev Res 2022;15(12 Suppl_1): Abstract nr PR010.

Abstract B019: Discrete regulation of the collagen proteome among pathological features in DCIS and invasive breast cancer by mass spectrometry tissue imaging

Ductal carcinoma in situ (DCIS) is characterized by inter-tumor heterogeneity that poses a therapeutic challenge due to its unpredictable recurrence and progression to invasive breast cancer (IBC). In recent publications, collagen stromal differences have been reported between patients that progressed to IBC (progressors) and those that did not (non-progressors). However, details on the role spatial regulation of the collagen proteome might play in this progression have yet to be studied. Here, we investigated the pathological distribution of collagen post-translational modifications in a cohort of patients classified as progressors with ipsilateral IBC recurrence compared to non-progressors. Previously published methods for collagen proteomics by targeted tissue mass spectrometry imaging were used. The method reports collagen types and post-translational modifications within the collagen triple-helical region as well as approximately 40 other extracellular matrix (ECM) proteins involved in the regulation of collagen fibers. Initial studies investigated collagen variation in lumpectomies (n=7) with DCIS, DCIS plus invasive ductal carcinoma (IDC) or IDC only. Over 590 peptides were found to be linked to annotated pathologies. A preliminary comparison of DCIS (n=2392 spectra) to IBC (n=4696 spectra) using area under the receiver operating curve (AUROC) ≥0.85 demonstrated that 47 peptides could individually discriminate between DCIS and IBC in this limited cohort. Image segmentation of the 405,652 pixels demonstrated 11 high-level hierarchical groups designating unique spatially localized ECM proteomic groups; these groups overlaid with histopathological features and pathological annotations. A total of 87 samples from the Resource of Archival Breast Tissue (RAHBT) matched with clinical characteristics were also investigated. Cores were histologically diverse within the tissue microarrays, with cribriform, micropapillary, papillary, solid, and comedo necrosis architectural patterns. Initial results suggest certain peptides may differentiate between non-progressors and progressors with ipsilateral IBC recurrence. Our current work focuses on correlating collagen signatures to mixed pathologies and the cellular content of cores. Further investigation of the collagen proteome is warranted. Overall, the data suggest unique collagen signatures in DCIS that could be useful for understanding recurrence and progression to IBC.

Citation Format: Taylor S. Hulahan, Elizabeth N. Wallace, Siri H. Strand, Robert Michael Angelo, Graham Colditz, Eun-Sil Shelley Hwang, Robert West, Laura Spruill, Jeffrey R. Marks, Richard R. Drake, Peggi M. Angel. Discrete regulation of the collagen proteome among pathological features in DCIS and invasive breast cancer by mass spectrometry tissue imaging [abstract]. In: Proceedings of the AACR Special Conference on Rethinking DCIS: An Opportunity for Prevention?; 2022 Sep 8-11; Philadelphia, PA. Philadelphia (PA): AACR; Can Prev Res 2022;15(12 Suppl_1): Abstract nr B019.

Spatial interplay of tissue hypoxia and T-cell regulation in ductal carcinoma in situ

Hypoxia promotes aggressive tumor phenotypes and mediates the recruitment of suppressive T cells in invasive breast carcinomas. We investigated the role of hypoxia in relation to T-cell regulation in ductal carcinoma in situ (DCIS). We designed a deep learning system tailored for the tissue architecture complexity of DCIS, and compared pure DCIS cases with the synchronous DCIS and invasive components within invasive ductal carcinoma cases. Single-cell classification was applied in tandem with a new method for DCIS ductal segmentation in dual-stained CA9 and FOXP3, whole-tumor section digital pathology images. Pure DCIS typically has an intermediate level of colocalization of FOXP3+ and CA9+ cells, but in invasive carcinoma cases, the FOXP3+ (T-regulatory) cells may have relocated from the DCIS and into the invasive parts of the tumor, leading to high levels of colocalization in the invasive parts but low levels in the synchronous DCIS component. This may be due to invasive, hypoxic tumors evolving to recruit T-regulatory cells in order to evade immune predation. Our data support the notion that hypoxia promotes immune tolerance through recruitment of T-regulatory cells, and furthermore indicate a spatial pattern of relocalization of T-regulatory cells from DCIS to hypoxic tumor cells. Spatial colocalization of hypoxic and T-regulatory cells may be a key event and useful marker of DCIS progression.

Abstract 5108: Copy number analysis of pure DCIS and association with recurrence

With the widespread adoption of breast cancer screening the incidence of pure ductal carcinoma in situ (DCIS) has increased. As DCIS is considered a non-obligate precursor of invasive ductal carcinoma most women with pure DCIS are treated with breast conserving surgery (BCS) +/- radiotherapy. However, for many this is likely to be overtreatment as only a minority will develop a subsequent ipsilateral recurrence. Studies also show that only ~60% of these ipsilateral recurrences are invasive disease with the remainder being pure DCIS. To predict which women are most likely to benefit from interventions, there is a need to identify biomarkers that are associated with invasive recurrence. Our aim was to assess whether copy number aberrations (CNAs) could be used to identify DCIS that was likely to recur as invasive disease or remain recurrence-free during long-time follow up.

We performed somatic copy number profiling on 309 pure DCIS samples that had not developed an ipsilateral event (controls), 198 that had developed subsequent ipsilateral invasive disease (INV-cases) and 58 that had developed subsequent ipsilateral pure DCIS (DCIS-cases). The samples were obtained from two large nation-wide cohorts: the Sloane cohort, a prospective breast screening cohort from the UK with a median follow up of 12.5 years and a Dutch population based cohort, with a median follow up of 13 years. CNAs were assessed using the CytoSNP array or low pass whole genome sequencing and analyzed using GISTIC.

Integrative cluster (IntClust) subtyping revealed that only 5 subtypes were well represented in DCIS compared to 10 in invasive disease and the distribution of clusters between INV-cases and controls was similar with the exception of IntClust 4, which was significantly more common in controls (P= 0.025, Fishers exact test). IntClust 4 is characterized to have low levels of genomic instability and a CNA-devoid. INV-cases were globally more aberrant than controls (P = 0.006, Wilcoxon test) as assessed by the chromosomal instability index (CIN) score. GISTIC identified 17 recurrent amplifications, 21 recurrent gains and 22 recurrent losses in the whole cohort. Six of these regions were more common in INV-cases compared to controls: amplifications at 17q24.1 and 8p11.23, losses at 1p36.13 and 11q23.2 ​and gains at 17q21.33​ and 16p (Nominal P < 0.05 and FDR < 0.1, Fishers exact test). Subgroup analysis of ER+, Her2- INV-cases versus controls revealed an additional differential CNA, amplification at 11q13.3 more common in cases.

DCIS-cases had similar CNAs to INV-cases and were more aberrant than controls in terms of CIN score (P < 0.037, Wilcoxon test) but not as aberrant as INV-cases.

In conclusion, we have identified potential CNAs that are associated with invasive recurrence. Further analysis will integrate gene expression with copy number data to identify which genes are being targeted by these CNAs in order to identify pathways important in progression of DCIS.

Citation Format: Ahmed A. Ahmed, Maria Roman-Escorza, Tycho Bismeijer, Michael Sheinman, Vandna Shah, Rana Shami, Jeffrey R. Marks, Lorraine M. King, Anargyros Megalios, Lindy L. Visser, Marlous Hoogstraat, Helen R. Davies, Tapsi Kumar, Deborah Collyar, Hilary Stobart, Sarah Pinder, Nicholas N. Navin, Andrew Futreal, Serena Nik-Zainal, E. Shelley Hwang, Lodewyk F. Wessels, Esther H. Lips, Alastair Thompson, Jelle Wesseling, Elinor J. Sawyer. Copy number analysis of pure DCIS and association with recurrence [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5108.

Genomic analysis defines clonal relationships of ductal carcinoma in situ and recurrent invasive breast cancer

Ductal carcinoma in situ (DCIS) is the most common form of preinvasive breast cancer and, despite treatment, a small fraction (5-10%) of DCIS patients develop subsequent invasive disease. A fundamental biologic question is whether the invasive disease arises from tumor cells in the initial DCIS or represents new unrelated disease. To address this question, we performed genomic analyses on the initial DCIS lesion and paired invasive recurrent tumors in 95 patients together with single-cell DNA sequencing in a subset of cases. Our data show that in 75% of cases the invasive recurrence was clonally related to the initial DCIS, suggesting that tumor cells were not eliminated during the initial treatment. Surprisingly, however, 18% were clonally unrelated to the DCIS, representing new independent lineages and 7% of cases were ambiguous. This knowledge is essential for accurate risk evaluation of DCIS, treatment de-escalation strategies and the identification of predictive biomarkers.

Racial Differences in the Tumor Immune Landscape and Survival of Women with High-Grade Serous Ovarian Carcinoma

BACKGROUND

Tumor-infiltrating lymphocytes (TIL) confer a survival benefit among patients with ovarian cancer; however, little work has been conducted in racially diverse cohorts.

METHODS

The current study investigated racial differences in the tumor immune landscape and survival of age- and stage-matched non-Hispanic Black and non-Hispanic White women with high-grade serous ovarian carcinoma (HGSOC) enrolled in two population-based studies (n = 121 in each racial group). We measured TILs (CD3+), cytotoxic T cells (CD3+CD8+), regulatory T cells (CD3+FoxP3+), myeloid cells (CD11b+), and neutrophils (CD11b+CD15+) via multiplex immunofluorescence. Multivariable Cox proportional hazard regression was used to estimate the association between immune cell abundance and survival overall and by race.

RESULTS

Overall, higher levels of TILs, cytotoxic T cells, myeloid cells, and neutrophils were associated with better survival in the intratumoral and peritumoral region, irrespective of tissue compartment (tumor, stroma). Improved survival was noted for T-regulatory cells in the peritumoral region and in the stroma of the intratumoral region, but no association for intratumoral T-regulatory cells. Despite similar abundance of immune cells across racial groups, associations with survival among non-Hispanic White women were consistent with the overall findings, but among non-Hispanic Black women, most associations were attenuated and not statistically significant.

CONCLUSIONS

Our results add to the existing evidence that a robust immune infiltrate confers a survival advantage among women with HGSOC; however, non-Hispanic Black women may not experience the same survival benefit as non-Hispanic White women with HGSOC.

IMPACT

This study contributes to our understanding of the immunoepidemiology of HGSOC in diverse populations.

Anomaly Detection of Calcifications in Mammography Based on 11,000 Negative Cases

In mammography, calcifications are one of the most common signs of breast cancer. Detection of such lesions is an active area of research for computer-aided diagnosis and machine learning algorithms. Due to limited numbers of positive cases, many supervised detection models suffer from overfitting and fail to generalize. We present a one-class, semi-supervised framework using a deep convolutional autoencoder trained with over 50,000 images from 11,000 negative-only cases. Since the model learned from only normal breast parenchymal features, calcifications produced large signals when comparing the residuals between input and reconstruction output images. As a key advancement, a structural dissimilarity index was used to suppress non-structural noises. Our selected model achieved pixel-based AUROC of 0.959 and AUPRC of 0.676 during validation, where calcification masks were defined in a semi-automated process. Although not trained directly on any cancers, detection performance of calcification lesions on 1,883 testing images (645 malignant and 1238 negative) achieved 75% sensitivity at 2.5 false positives per image. Performance plateaued early when trained with only a fraction of the cases, and greater model complexity or a larger dataset did not improve performance. This study demonstrates the potential of this anomaly detection approach to detect mammographic calcifications in a semi-supervised manner with efficient use of a small number of labeled images, and may facilitate new clinical applications such as computer-aided triage and quality improvement.

Polygenic risk modeling for prediction of epithelial ovarian cancer risk

Polygenic risk scores (PRS) for epithelial ovarian cancer (EOC) have the potential to improve risk stratification. Joint estimation of Single Nucleotide Polymorphism (SNP) effects in models could improve predictive performance over standard approaches of PRS construction. Here, we implemented computationally efficient, penalized, logistic regression models (lasso, elastic net, stepwise) to individual level genotype data and a Bayesian framework with continuous shrinkage, "select and shrink for summary statistics" (S4), to summary level data for epithelial non-mucinous ovarian cancer risk prediction. We developed the models in a dataset consisting of 23,564 non-mucinous EOC cases and 40,138 controls participating in the Ovarian Cancer Association Consortium (OCAC) and validated the best models in three populations of different ancestries: prospective data from 198,101 women of European ancestries; 7,669 women of East Asian ancestries; 1,072 women of African ancestries, and in 18,915 BRCA1 and 12,337 BRCA2 pathogenic variant carriers of European ancestries. In the external validation data, the model with the strongest association for non-mucinous EOC risk derived from the OCAC model development data was the S4 model (27,240 SNPs) with odds ratios (OR) of 1.38 (95% CI: 1.28-1.48, AUC: 0.588) per unit standard deviation, in women of European ancestries; 1.14 (95% CI: 1.08-1.19, AUC: 0.538) in women of East Asian ancestries; 1.38 (95% CI: 1.21-1.58, AUC: 0.593) in women of African ancestries; hazard ratios of 1.36 (95% CI: 1.29-1.43, AUC: 0.592) in BRCA1 pathogenic variant carriers and 1.49 (95% CI: 1.35-1.64, AUC: 0.624) in BRCA2 pathogenic variant carriers. Incorporation of the S4 PRS in risk prediction models for ovarian cancer may have clinical utility in ovarian cancer prevention programs.

Abstract GS4-07: The Breast PreCancer Atlas DCIS genomic signatures define biology and correlate with clinical outcomes: An analysis of TBCRC 038 and RAHBT cohorts

Background. DCIS consists of a molecularly heterogeneous group of premalignant lesions, with variable risk of invasive progression. Understanding biomarkers for invasive progression could help individualize treatment recommendations based upon tumor biology. As part of the NCI Human Tumor Atlas Network (HTAN), we conducted comprehensive genomic analyses on two large DCIS case-control cohorts. Methods. We performed smart3-seq and low-pass whole genome sequencing on two independent, retrospective, longitudinally sampled DCIS case-control cohorts. TBCRC 038 was a multicenter cohort diagnosed with DCIS between 1998 and 2016 at one of the Translational Breast Cancer Research sites; the RAHBT (Resource of Archival Human Breast Tissue) cohort included women identified through the St. Louis Breast Tissue Repository, and the Women’s Health Repository diagnosed between 1997 and 2001. We studied the spectrum of molecular changes present and sought genomic predictors of subsequent ipsilateral breast events (iBEs: DCIS recurrence or invasive progression) in both DCIS epithelium and stroma in formalin fixed paraffin embedded tissue. We generated de novo tumor and stroma-centric subtypes for DCIS that represents fundamental transcriptomic organization. Copy number analysis was performed using low-pass DNA sequencing. Non-negative matrix factorization (NMF) was applied to the RNA expression of all coding genes to identify clusters. A negative-binomial regression model was used to identify differentially expressed genes. Results. We analyzed 677 DCIS samples from 481 patients with 7.1 years median follow-up. In TBCRC samples, we identified three clusters via NMF in TBCRC referred to as ER low, quiescent, and ER high. The ER-low cluster had significantly higher levels of ERBB2 and lower levels of ESR1 compared to quiescent and ER-high clusters. Quiescent cluster lesions were less proliferative and less metabolically active than ER high and ER low subtypes. These findings were replicated in the RAHBT cohort. Focusing on the stromal component of DCIS from laser capture microdissection in RAHBT samples, we identified four distinct DCIS-associated stromal clusters. A “normal-like” stromal cluster with ECM organization and PI3K-AKT signaling; a “collagen-rich” stromal cluster; a “desmoplastic” stromal cluster with high fibroblast and total myeloid abundance, mostly associated with macrophages and myeloid dendritic cells (mDC); and an “immune-dense” stromal cluster. Further, we compared differentially expressed genes in patients with or without subsequent iBEs within 5 years of diagnosis. Hypothesizing that the resulting 812 DE genes (DESeq2) represent multiple routes to subsequent iBEs, we leveraged NMF to identify paths to progression. In both TBCRC and RAHBT cohorts, poor outcome groups exhibited increased ER, MYC signaling, and oxidative phosphorylation, supporting that these pathways are important for DCIS recurrence and progression. Conclusion. Comprehensive genomic profiling in two independent DCIS cohorts with longitudinal outcomes shows distinct DCIS stromal expression patterns and immune cell composition. RNA expression profiles reveal underlying tumor biology that is associated with later iBEs in both cohorts. These studies provide new insight into DCIS biology and will guide the design of diagnostic strategies to prevent invasive progression.

Citation Format: Siri H Strand, Belén Rivero-Gutiérrez, Kathleen E Houlahan, Jose A Seoane, Lorraine M King, Tyler Risom, Lunden Simpson, Sujay Vennam, Aziz Khan, Timothy Hardman, Bryan E Harmon, Fergus J Couch, Kristalyn Gallagher, Mark Kilgore, Shi Wei, Angela DeMichele, Tari King, Priscilla F McAuliffe, Julie Nangia, Joanna Lee, Jennifer Tseng, Anna Maria Storniolo, Alastair Thompson, Gaorav Gupta, Robyn Burns, Deborah J Veis, Katherine DeSchryver, Chunfang Zhu, Magdalena Matusiak, Jason Wang, Shirley X Zhu, Jen Tappenden, Daisy Yi Ding, Dadong Zhang, Jingqin Luo, Shu Jiang, Sushama Varma, Cody Straub, Sucheta Srivastava, Christina Curtis, Rob Tibshirani, Robert Michael Angelo, Allison Hall, Kouros Owzar, Kornelia Polyak, Carlo Maley, Jeffrey R Marks, Graham A Colditz, E Shelley Hwang, Robert B West. The Breast PreCancer Atlas DCIS genomic signatures define biology and correlate with clinical outcomes: An analysis of TBCRC 038 and RAHBT cohorts [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr GS4-07.

Abstract P1-22-05: Identifying predictors of invasive recurrence based on molecular profiles of DCIS lesions

Introduction Ductal carcinoma in situ (DCIS) is a non-obligate precursor of invasive breast cancer. Patients with DCIS are routinely treated by breast-conserving surgery often supplemented by radiotherapy, although many will never develop invasive disease. To date, no robust predictors of invasive breast cancer recurrence following DCIS have been identified. In our efforts to find such predictors, we performed gene expression, copy number and mutation analysis on two large DCIS cohorts with long-term follow-up. Methods Two nested case control series were analyzed, where cases are defined as DCIS with a subsequent invasive breast cancer and controls remained disease free during follow up. Cases and controls were matched on age and on follow up duration and were derived from two nation-wide cohort studies. The Sloane cohort is a prospective breast screening cohort from the UK, median follow up is 6 years (range 1-10). The Dutch cohort is population-based and had a median follow up of 13 years (range 2-23). We performed copy number analysis using CytoSNP array or low pass whole genome sequencing (lpWGS) on 310 controls and 196 cases, and RNA-seq on 295 controls and 206 cases. Results First analyses on the copy number data suggest that cases are genetically more aberrant with multiple regions of amplification compared to controls (p < 0.05). RNA-seq was used to classify DCIS into the PAM50 subtypes which did not appear to be predictive of recurrence. Initial RNA-seq analysis did not show consistent gene expression differences between cases and controls in the Sloane or Dutch cohorts, possibly explained by differences in clinical characteristics of the cohorts. A new computational method has been developed accounting for the differences in follow-up times, results will be presented at SABCS. Targeted sequencing revealed that the most common mutations were in PIK3CA and TP53, but there was no association with recurrence. Conclusion Only small molecular differences were identified between DCIS that recurs as invasive breast cancer and DCIS that remains disease-free. Currently, we are seeking to identify reproducible differences by a combined analysis of two population-based cohorts in a time dependent fashion. These will be presented at the SABCS. This work was supported by Cancer Research UK and by KWF Dutch Cancer Society (ref.C38317/A24043)

Citation Format: Tycho Bismeijer, Ahmed A Ahmed, Michael Sheinman, Maria Roman-Escorza, Vandna Shah, Jeffrey R Marks, Lorraine M King, Anargyros Megalios, Lindy L Visser, Marlous Hoogstraat, Helen R Davies, Tapsi Kumar, Deborah Collyar, Hilary Stobart, Nicholas N Navin, Andrew Futreal, Serena Nik-Zainal, Shelley Hwang, Esther H Lips, Alastair Thompson, Lodewyk FA Wessels, Elinor J Sawyer, Jelle Wesseling, Grand Challenge PRECISION Consortium. Identifying predictors of invasive recurrence based on molecular profiles of DCIS lesions [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P1-22-05.

Transition to invasive breast cancer is associated with progressive changes in the structure and composition of tumor stroma

Ductal carcinoma in situ (DCIS) is a pre-invasive lesion that is thought to be a precursor to invasive breast cancer (IBC). To understand the changes in the tumor microenvironment (TME) accompanying transition to IBC, we used multiplexed ion beam imaging by time of flight (MIBI-TOF) and a 37-plex antibody staining panel to interrogate 79 clinically annotated surgical resections using machine learning tools for cell segmentation, pixel-based clustering, and object morphometrics. Comparison of normal breast with patient-matched DCIS and IBC revealed coordinated transitions between four TME states that were delineated based on the location and function of myoepithelium, fibroblasts, and immune cells. Surprisingly, myoepithelial disruption was more advanced in DCIS patients that did not develop IBC, suggesting this process could be protective against recurrence. Taken together, this HTAN Breast PreCancer Atlas study offers insight into drivers of IBC relapse and emphasizes the importance of the TME in regulating these processes.

Prediction of Upstaging in Ductal Carcinoma in Situ Based on Mammographic Radiomic Features

Background Improving diagnosis of ductal carcinoma in situ (DCIS) before surgery is important in choosing optimal patient management strategies. However, patients may harbor occult invasive disease not detected until definitive surgery. Purpose To assess the performance and clinical utility of mammographic radiomic features in the prediction of occult invasive cancer among women diagnosed with DCIS on the basis of core biopsy findings. Materials and Methods In this Health Insurance Portability and Accountability Act-compliant retrospective study, digital magnification mammographic images were collected from women who underwent breast core-needle biopsy for calcifications that was performed at a single institution between September 2008 and April 2017 and yielded a diagnosis of DCIS. The database query was directed at asymptomatic women with calcifications without a mass, architectural distortion, asymmetric density, or palpable disease. Logistic regression with regularization was used. Differences across training and internal test set by upstaging rate, age, lesion size, and estrogen and progesterone receptor status were assessed by using the Kruskal-Wallis or χ² test. Results The study consisted of 700 women with DCIS (age range, 40-89 years; mean age, 59 years ± 10 [standard deviation]), including 114 with lesions (16.3%) upstaged to invasive cancer at subsequent surgery. The sample was split randomly into 400 women for the training set and 300 for the testing set (mean ages: training set, 59 years ± 10; test set, 59 years ± 10; P = .85). A total of 109 radiomic and four clinical features were extracted. The best model on the test set by using all radiomic and clinical features helped predict upstaging with an area under the receiver operating characteristic curve of 0.71 (95% CI: 0.62, 0.79). For a fixed high sensitivity (90%), the model yielded a specificity of 22%, a negative predictive value of 92%, and an odds ratio of 2.4 (95% CI: 1.8, 3.2). High specificity (90%) corresponded to a sensitivity of 37%, positive predictive value of 41%, and odds ratio of 5.0 (95% CI: 2.8, 9.0). Conclusion Machine learning models that use radiomic features applied to mammographic calcifications may help predict upstaging of ductal carcinoma in situ, which can refine clinical decision making and treatment planning. © RSNA, 2022.

A new method to accurately identify single nucleotide variants using small FFPE breast samples

Most tissue collections of neoplasms are composed of formalin-fixed and paraffin-embedded (FFPE) excised tumor samples used for routine diagnostics. DNA sequencing is becoming increasingly important in cancer research and clinical management; however it is difficult to accurately sequence DNA from FFPE samples. We developed and validated a new bioinformatic pipeline to use existing variant-calling strategies to robustly identify somatic single nucleotide variants (SNVs) from whole exome sequencing using small amounts of DNA extracted from archival FFPE samples of breast cancers. We optimized this strategy using 28 pairs of technical replicates. After optimization, the mean similarity between replicates increased 5-fold, reaching 88% (range 0-100%), with a mean of 21.4 SNVs (range 1-68) per sample, representing a markedly superior performance to existing tools. We found that the SNV-identification accuracy declined when there was less than 40 ng of DNA available and that insertion-deletion variant calls are less reliable than single base substitutions. As the first application of the new algorithm, we compared samples of ductal carcinoma in situ of the breast to their adjacent invasive ductal carcinoma samples. We observed an increased number of mutations (paired-samples sign test, P < 0.05), and a higher genetic divergence in the invasive samples (paired-samples sign test, P < 0.01). Our method provides a significant improvement in detecting SNVs in FFPE samples over previous approaches.

Unmasking the immune microecology of ductal carcinoma in situ with deep learning

Despite increasing evidence supporting the clinical relevance of tumour infiltrating lymphocytes (TILs) in invasive breast cancer, TIL spatial variability within ductal carcinoma in situ (DCIS) samples and its association with progression are not well understood. To characterise tissue spatial architecture and the microenvironment of DCIS, we designed and validated a new deep learning pipeline, UNMaSk. Following automated detection of individual DCIS ducts using a new method IM-Net, we applied spatial tessellation to create virtual boundaries for each duct. To study local TIL infiltration for each duct, DRDIN was developed for mapping the distribution of TILs. In a dataset comprising grade 2-3 pure DCIS and DCIS adjacent to invasive cancer (adjacent DCIS), we found that pure DCIS cases had more TILs compared to adjacent DCIS. However, the colocalisation of TILs with DCIS ducts was significantly lower in pure DCIS compared to adjacent DCIS, which may suggest a more inflamed tissue ecology local to DCIS ducts in adjacent DCIS cases. Our study demonstrates that technological developments in deep convolutional neural networks and digital pathology can enable an automated morphological and microenvironmental analysis of DCIS, providing a new way to study differential immune ecology for individual ducts and identify new markers of progression.

Abstract PR02: Inferring the evolutionary dynamics of ductal carcinoma in situ through multi-regional sequencing and mathematical modeling

Introduction. The natural history of preinvasive breast cancer, or ductal carcinoma in situ (DCIS) remains poorly understood. Overcoming this gap would allow risk-appropriate treatment for patients diagnosed with DCIS. We used a multiregional sequencing approach in combination with mathematical modeling to characterize the evolutionary dynamics of DCIS initiation and progression. Methods. We analyzed a cohort of 18 patients diagnosed with DCIS, either with (n=9) or without (n=9) synchronous invasive cancer. Based on whole exome sequencing, tumor-specific mutation panels were constructed, each targeting 29-75 mutations (median: 60). From each tumor, and using selective ultraviolet radiation fractionation (SURF), we microdissected small spots (encompassing 1-3 duct cross-sections) from 3-4 spatially separated microscope sections (mean slide separation: 1.25cm, range: 0.34-6.0cm). Spots were spatially registered and genotyped based on targeted sequencing of the tumor-specific mutation panels. For each tumor, we performed unsupervised clonal deconvolution of the spot genotypes (CloneFinder) and constructed phylogenetic subclone trees. To quantify the spatial patterns of subclonal mutations, we introduced a dispersion index (DI), ranging from low (DI=0%) to high (DI=100%). To provide a spatio-temporal context for the heterogeneity patterns we developed a family of stochastic mathematical models of DCIS initiation and progression. Thereby, we embedded the evolutionary dynamics of tumor cell expansion in the branching topology of mammary ductal trees. Results. A total of 485 microdissected spots (median per tumor: 23, range: 10-50) were spatially registered and sequenced (median depth: 9,000x). All tumors were multiclonal, containing a median of 5 subclones (range: 2-14). Surprisingly, the correlation between spatial and genomic distances of spots was low. Individual subclones were diffusely dispersed across tumors. DCIS with synchronous DCIS and invasive cancer (mixed DCIS) had a higher mutation dispersion (DI=84.7%) than those without (pure DCIS, DI=70.5%; p=0.03, Wilcoxon rank-sum test). Mixed DCIS also had a higher fraction of spots containing more than one subclone than pure DCIS (median: 30.4% vs 0%, p=0.03). Among 7 mixed DCIS with invasive spots, 5 showed evidence of multiclonal invasion, that is more than one invading subclones were found in both in situ and invasive regions of the tumor. Mathematical modeling analyses show that the observed spatial patterns of genetic heterogeneity are consistent with a single expansion of mixing subclones across the ductal tree architecture. Conclusions. Our findings provide novel insights into the early growth and invasion dynamics of DCIS lesions. Further, we identified potential evolutionary markers for the delineation between indolent (pure) and aggressive (mixed) DCIS. This constitutes an important step towards identification of patients with low-risk DCIS who could be appropriately managed with less aggressive treatment.

Citation Format: Marc D. Ryser, Inmaculada C. Sorribes, Matthew Greenwald, Ethan Wu, Allison Hall, Diego Mallo, Lorraine M. King, Timothy Hardman, Lunden Simpson, Carlo C. Maley, Jeffrey R. Marks, Darryl Shibata, E. Shelley Hwang. Inferring the evolutionary dynamics of ductal carcinoma in situ through multi-regional sequencing and mathematical modeling [abstract]. In: Proceedings of the AACR Virtual Special Conference on Tumor Heterogeneity: From Single Cells to Clinical Impact; 2020 Sep 17-18. Philadelphia (PA): AACR; Cancer Res 2020;80(21 Suppl):Abstract nr PR02.

Abstract 2502: Genetic and functional heterogeneity of DCIS as predictors of invasive cancer

Genetic diversity both between and within individual tumors constitutes a challenge to personalized cancer medicine. Intra-tumor heterogeneity provides the genetic fuel for natural selection in clonal evolution and cancer progression. Tumors with high levels of genetic heterogeneity are hypothesized to be more likely to demonstrate aggressive behavior and progress to invasion and metastasis.

We analyzed the mutational loads from separate areas of pure DCIS and compared this to genetic heterogeneity in DCIS lesions found adjacent to invasive and metastatic cancer. Two spatially distinct areas of DCIS from each case were macro-dissected and the DNA extracted from FFPE samples. To analyze the data, we developed new bioinformatics methods that allowed analysis of small amounts of degraded DNA extracted from FFPE samples across multiple regions. Our bioinformatics pipeline was optimized on a series of 28 independent technical replicates of the same DNA sample sequenced twice, as training tools to find the best filtering parameters.

Whole exome sequencing was performed on each of the two geospatially separated samples for each case. Minimum coverage for inclusion in this study was 40X over at least 50% of the exome. We used the ratio of private mutations (only in 1 area) to public (found in both areas) mutations as a measure of intra-tumor heterogeneity.

We present an approach to measure clonal heterogeneity using a bulk sequencing strategy applied to geospatially distinct foci of DCIS. We found statistically significant difference between DCIS adjacent to invasive disease and metastatic patients' genetic divergence (t-test, p=0.013). Our findings suggest that genetic and functional heterogeneity may play an important evolutionary role as a driver for invasive progression.

Citation Format: Angelo Fortunato, Diego Mallo, Lorraine King, Timothy Hardman, Allison Hall, Jeffrey R. Marks, Shelley Shelley Hwang, Carlo C. Maley. Genetic and functional heterogeneity of DCIS as predictors of invasive cancer [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2502.

Circulating stromal cells as a potential blood-based biomarker for screening invasive solid tumors

3535

Background: Peripheral blood allows for a simple non-invasive method for isolating various cancer associated circulating stromal cells (CStCs) which may predict for cancer presence. Cancer Associated Macrophage-Like cells (CAMLs), a specific CStC, are phagocytic myeloid cells that derive from an immunological response to cancer and emanate from primary tumors. Using a filtration platform we screened the peripheral blood of untreated newly diagnosed cancer patients (n = 308) for CAMLs. In parallel, we screened patients with newly diagnosed non-malignant diseases, i.e. lupus, benign cysts, etc. (n = 39), and healthy control samples (n = 76). We found that CAMLs are highly prevalent (87%) in the blood of cancer patients, but uncommon in non-malignant conditions (20%) & absent in healthy individuals (0%). Methods: Anonymized peripheral blood were taken from 308 cancer patients after confirmation of invasive malignancy [stage I (n = 76), stage II (n = 73), stage III (n = 72), stage IV (n = 65) and unstaged non-metastatic (n = 22)] with pathologically confirmed lung (n = 65), pancreas (n = 53), breast (n = 52), prostate (n = 40), esophageal (n = 30), renal cell (n = 18), hepatocellular (n = 15), neuroblastoma (n = 10), melanoma (n = 8), and other (n = 17). Further, anonymized blood was taken from patients with untreated non-malignant conditions including benign breast masses (n = 19), lupus (n = 11), liver cirrhosis (n = 5), benign prostatic hyperplasia (BPH) (n = 3), and viral infection (n = 1); or from healthy control volunteers (n = 76). CAMLs were isolated from whole peripheral blood by the CellSieve™ microfiltration technique and defined as enlarged, multinuclear cells with cytokeratin and/or CD45/CD14 positive. Results: CAMLs were found in 87% of all cancer patients regardless of stage, ~5.4 CAMLs/7.5mL blood. Specifically, CAMLs were found in 80% of Stage I, 90% Stage II, 89% Stage III, and 97% Stage IV patients. No CAMLs were found in any healthy controls, but were found in 26% of benign breast masses, 18% of lupus, 0% of BPH and 0% of cirrhosis. In total, CAML sensitivity in cancer vs healthy was 87% (CI95% 82-90%), specificity = 100% (CI95% 95-100%), PPV = 100% (CI95% 100%), NPV = 67% (CI95% 58-71%). CAML sensitivity in cancer vs benign was 87% (CI95% 82-90%), specificity = 80% (CI95% 64-91%), PPV = 97% (CI95% 95-98%), NPV = 43% (CI95% 35-51%). Conclusions: CAMLs, a Circulating Stromal Cell subtype, is a sensitive blood based biomarker found in all stages of cancer that is rare in non-malignant conditions and absent in healthy individuals.

Identification of pathogenic ROS1 alterations in cell-free DNA (cfDNA) from patients with breast cancer

1031

Background: ROS1 is an important proto-oncogene involved in the development of various cancers for which we have FDA approved therapies. While activation of the ROS1 tyrosine kinase receptor has been reported in 1-2% of lung cancers, the frequency and type of ROS1 alterations in breast cancer have not been fully explored. We previously described the incidence of ROS1 alterations from breast cancer tissue. The purpose of this study was to identify the incidence of ROS1 genomic alterations occurring in cfDNA from patients with breast cancer. Methods: We queried 16,053 breast cancer samples from the Guardant Health breast cancer database between June 2015 - October 2019 to identify the incidence of ROS1 alterations detected in cfDNA in breast cancer. We identified fusion partner genes and classified each alteration type into the following categories: fusion, single nucleotide variants (SNVs), and indels. Radical amino acid changes occurring at conserved regions across the ROS1 gene were identified. In vitro analyses were used to investigate the effect of ROS1 nonsynonymous mutations on the ROS1 protein. We made associations with ROS1 alterations and co-occurring mutated genes. Results: Nonsynonymous ROS1 alterations from the Guardant Health breast cancer database were found in 162 samples from 142 patients in the 16,053-patient cohort (1%). Alterations found included: 1 (0.6%) ROS1-SLC35F1 fusion, 155 (95.7%) SNVs, and 6 (3.7%) indels. Of the 155 SNVs, we identified 23 (14.8%) mutations occurring in the ROS1 kinase, of which, 20 (12.9%) occurred at highly conserved regions and 15 (9.6%) harbored radical amino acid changes. The top 5 co-occurring mutations in samples with ROS1 alterations were TP53 (50%), PIK3CA (44%), ESR1 (27%), EGFR (21%), and FGFR1 (18%). Conclusions: A modest incidence of ROS1 genomic alterations occurs in cfDNA from patients with breast cancer. New somatic alterations in the ROS1 gene were identified from Guardant Health that were not detected in publicly available databases. A portion of mutations occurred at highly conserved regions across the ROS1 gene suggesting these may be more actionable than currently recognized. In vitro analyses of ROS1 gene activation from these newly discovered somatic alterations are being investigated with results to be reported. Co-occurring mutations reveal a unique genotype associated with ROS1 alterations that may play a biologic role in ROS1-mediated pathogenesis.

Minimal barriers to invasion during human colorectal tumor growth

Intra-tumoral heterogeneity (ITH) could represent clonal evolution where subclones with greater fitness confer more malignant phenotypes and invasion constitutes an evolutionary bottleneck. Alternatively, ITH could represent branching evolution with invasion of multiple subclones. The two models respectively predict a hierarchy of subclones arranged by phenotype, or multiple subclones with shared phenotypes. We delineate these modes of invasion by merging ancestral, topographic, and phenotypic information from 12 human colorectal tumors (11 carcinomas, 1 adenoma) obtained through saturation microdissection of 325 small tumor regions. The majority of subclones (29/46, 60%) share superficial and invasive phenotypes. Of 11 carcinomas, 9 show evidence of multiclonal invasion, and invasive and metastatic subclones arise early along the ancestral trees. Early multiclonal invasion in the majority of these tumors indicates the expansion of co-evolving subclones with similar malignant potential in absence of late bottlenecks and suggests that barriers to invasion are minimal during colorectal cancer growth.

Abstract P2-10-18: Deciphering racial disparities in breast cancer collagen reorganization by targeted extracellular matrix proteomics

Although European American women have a higher incidence of breast cancer, African American women have higher mortality rates with increased occurrence of lethal cancers at a younger age. Breast density, characterized by increases in stroma, is a risk factor for breast cancer and African American women have significantly greater odds of high breast density compared to European American women. Genomic profiling of breast cancer by ancestry shows that stromal collagen type genes vary based on African and European ancestry and correlate to differences in breast cancer progression and survival. Contemporary scientific literature describes breast cancer progression as being linked to a “collagen switch” whereby collagen fibers surrounding the tumor are re-organized during later tumor growth to facilitate metastatic spread of cancer activated fibroblasts. However, mechanisms facilitating the “collagen switch” remain mostly unknown, especially at a translational level. Additionally, collagen re-organization has not yet been evaluated for contributions to racial disparities in breast cancer. Our novel extracellular matrix proteomic studies on 17 patient lumpectomies depict that very specific sites on collagen protein types in breast tumor are post-translationally regulated by hydroxylation of proline (HYP) dependent on race (7 African American, 10 European American; no significant age or receptor status difference). By proteomic sequencing, distinct peptide sequences of collagen 1A1, collagen 1A2, and collagen 3A1 are modified by HYP and are significantly regulated when measured in African American versus European American breast tumors. From 16 collagen type proteins, 143 collagen peptides from all breast tumors were identified with HYP modifications (site probability ≥0.95). A total of 16% of the HYP peptides were unique to African American breast tumors. Imaging mass spectrometry that targets collagens reported that specific peptides sequences showed increased hydroxylation of proline in tumors compared to normal adjacent tissue. Interestingly, but not unexpected, translational and post-translational regulation did not always correlate with transcriptional regulation. Investigation on over 300 tissue microarray cores of breast tumor by second harmonic generation microscopy demonstrated significant ancestry-defined regulation of collagen fibers. For example, compared to breast tumor tissue cores of European American ancestry, African American breast tumor tissue cancer cores showed significant overall increases in collagen fibril length with decreased collagen fibril straightness (two-tailed student’s t-test p-value &lt;0.05). Imaging mass spectrometry that targets collagens done on the identical set of 300 ancestry-defined tissue microarray cores supports that ancestry of breast tumor correlates with site specific collagen modification by hydroxylation of prolines. For instance, COL3A1 peptide GPAGIPx GFPx shows a nearly 3-fold increase in African American breast tumors compared to European breast tumors (Mann-Whitney p-value ≤0.02; area under the receiver operating curve 0.81, p-value ≤0.0001; HYP site probability of 1.00 for each proline designated Px). Since HYP stabilizes collagen triple helices, we hypothesize that HYP alteration at specific sites in collagen protein types from African American breast tumors may reflect a change in collagen flexibility that contributes to an earlier metastatic collagen switch than in European American breast tumors. Current work determines ancestry-related changes in collagen fibers per breast cancer type and per normal adjacent tissue. To summarize, we provide the first evidence that post-translational regulation of collagen types may contribute to racial disparities in breast cancer.

Citation Format: Peggi M Angel, Janet Saunders, Heather Jensen-Smith, Evelyn Bruner, Marvella E. Ford, Savanna Berkhiser, Baylye Boxall, Jennifer Bethard, Lauren E. Ball, Elizabeth S. Yeh, Michael A. Hollingsworth, Anand S. Mehta, Jeffrey R. Marks, Harikrishna Nakshatri, Richard R. Drake. Deciphering racial disparities in breast cancer collagen reorganization by targeted extracellular matrix proteomics [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P2-10-18.

Abstract P1-21-07: The Patient-reported Outcomes after Routine Treatment of Atypical Lesions (PORTAL) study: Pain, psychosocial wellbeing, and quality of life among women undergoing guideline concordant care for DCIS vs. active surveillance for in situ and atypical lesions

Background: Guideline-concordant care (GCC) for DCIS includes surgery, radiation, and endocrine treatment. Active surveillance (AS) is a strategy under study for management of low risk DCIS. The PORTAL Study was designed to evaluate patient reported outcomes (PROs) after GCC for DCIS compared to women who received AS for DCIS combined with women with a history of other atypical lesions (atypical ductal hyperplasia (ADH), or atypical lobular hyperplasia/lobular carcinoma in situ (LN), as proxies for AS-managed DCIS. Methods: The PORTAL Study invited women age≥ 18, diagnosed with DCIS, ADH, or LN between 2012-2017 from 4 academic centers to complete a one-time, cross-sectional survey. Clinical, pathological, and treatment information was obtained from medical record review. The primary outcome was breast/chest wall pain assessed with the Breast Cancer Pain Questionnaire (BCPQ) including severity (10-point scale, ≥3=clinically relevant), a Pain Burden Index (PBI), which is a composite of severity, frequency, and location (breast, arm, side, axilla) and assessments of sensory disturbances, and impact of pain on emotional and physical functioning. Additional PROs included measures of generalized pain (Brief Pain Inventory), anxiety (STAI-Short Form), depression (CES-D), and QOL (Quality of Life in Adult Cancer Survivors). Pain, psychosocial, and QOL outcomes were compared between the GCC vs. AS groups using Wilcoxon Rank Sum and Chi-Square tests. Results: Of 1565 patients invited and sent a survey, 927 (59%) responded to the survey with evaluable pain outcome data. Median time from diagnosis was 3.8 years. Median age at survey completion was 58 (range: 26-94) years; 13% identified as non-White; 4% Hispanic. Among those with DCIS (n=554), 97% had GCC (62%, lumpectomy, 38%, mastectomy; 48%, radiation), representing 58% of participants vs 42% representing AS. The prevalence of clinically relevant pain was higher in the GCC vs. AS group (16.5% vs 9%, p=.0009). Median BCPQ-PBI, sensory disturbance, physical, and emotional impact scores were all higher (p&lt;.0001) in the GCC vs. AS group (Table); BPI scores for pain severity and interference were similar between groups. QOL, anxiety and depressive symptoms were similar among women who had GCC compared to the AS group. Conclusion: Women with DCIS who have undergone GCC experience more breast/chest wall pain and report greater impact of pain on physical and emotional functioning in long term follow-up, compared to women who have undergone AS for DCIS or are managed for other atypical lesions. Given that many women with low risk DCIS may be unlikely to develop invasive cancer, improved understanding of the potential trade-offs of GCC vs AS can help support informed decision making in women with DCIS who are considering their treatment options. Ongoing prospective trials will provide further information regarding risks and benefits of AS vs GCC for women with low risk DCIS.

BCPQ Scores, GCC vs. ASGCCASMean (range)Median (IQR)Mean (range)Median (IQR)p*PBI6.4 (0-80)0 (0-9)2.9 (0-64)0 (0-0)&lt;.0001Sensory disturbance1.4 (0-9)0 (0-2)0.6 (0-9)0 (0-0)&lt;.0001Physical impact9.6 (0-67)0 (0-19)4.4 (0-56)0 (0-0)&lt;.0001Emotional impact1.4 (0-33)0 (0-1)0.6 (0-38)0 (0-0)&lt;.0001*Wilcoxon rank sum test comparing median scores

Citation Format: Shoshana M Rosenberg, Laura H Hendrix, Kristin L Schreiber, Alastair M Thompson, Isabelle Bedrosian, Kevin S Hughes, Thomas Lynch, Desiree Basila, Deborah E Collyar, Elizabeth S Frank, Sonja Darai, Conor Lanahan, Jeffrey R Marks, Jennifer K Plichta, Terry Hyslop, Ann H Partridge, E. Shelley Hwang. The Patient-reported Outcomes after Routine Treatment of Atypical Lesions (PORTAL) study: Pain, psychosocial wellbeing, and quality of life among women undergoing guideline concordant care for DCIS vs. active surveillance for in situ and atypical lesions [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P1-21-07.

The Lineage Determining Factor GRHL2 Collaborates with FOXA1 to Establish a Targetable Pathway in Endocrine Therapy-Resistant Breast Cancer

Notwithstanding the positive clinical impact of endocrine therapies in estrogen receptor-alpha (ERα)-positive breast cancer, de novo and acquired resistance limits the therapeutic lifespan of existing drugs. Taking the position that resistance is nearly inevitable, we undertook a study to identify and exploit targetable vulnerabilities that were manifest in endocrine therapy-resistant disease. Using cellular and mouse models of endocrine therapy-sensitive and endocrine therapy-resistant breast cancer, together with contemporary discovery platforms, we identified a targetable pathway that is composed of the transcription factors FOXA1 and GRHL2, a coregulated target gene, the membrane receptor LYPD3, and the LYPD3 ligand, AGR2. Inhibition of the activity of this pathway using blocking antibodies directed against LYPD3 or AGR2 inhibits the growth of endocrine therapy-resistant tumors in mice, providing the rationale for near-term clinical development of humanized antibodies directed against these proteins.

Prediction of Upstaged Ductal Carcinoma In Situ Using Forced Labeling and Domain Adaptation

OBJECTIVE

The goal of this study is to use adjunctive classes to improve a predictive model whose performance is limited by the common problems of small numbers of primary cases, high feature dimensionality, and poor class separability. Specifically, our clinical task is to use mammographic features to predict whether ductal carcinoma in situ (DCIS) identified at needle core biopsy will be later upstaged or shown to contain invasive breast cancer.

METHODS

To improve the prediction of pure DCIS (negative) versus upstaged DCIS (positive) cases, this study considers the adjunctive roles of two related classes: atypical ductal hyperplasia (ADH), a non-cancer type of breast abnormity, and invasive ductal carcinoma (IDC), with 113 computer vision based mammographic features extracted from each case. To improve the baseline Model A's classification of pure vs. upstaged DCIS, we designed three different strategies (Models B, C, D) with different ways of embedding features or inputs.

RESULTS

Based on ROC analysis, the baseline Model A performed with AUC of 0.614 (95% CI, 0.496-0.733). All three new models performed better than the baseline, with domain adaptation (Model D) performing the best with an AUC of 0.697 (95% CI, 0.595-0.797).

CONCLUSION

We improved the prediction performance of DCIS upstaging by embedding two related pathology classes in different training phases.

SIGNIFICANCE

The three new strategies of embedding related class data all outperformed the baseline model, thus demonstrating not only feature similarities among these different classes, but also the potential for improving classification by using other related classes.

LncRNAs-directed PTEN enzymatic switch governs epithelial-mesenchymal transition

Despite the structural conservation of PTEN with dual-specificity phosphatases, there have been no reports regarding the regulatory mechanisms that underlie this potential dual-phosphatase activity. Here, we report that K27-linked polyubiquitination of PTEN at lysines 66 and 80 switches its phosphoinositide/protein tyrosine phosphatase activity to protein serine/threonine phosphatase activity. Mechanistically, high glucose, TGF-β, CTGF, SHH, and IL-6 induce the expression of a long non-coding RNA, GAEA (Glucose Aroused for EMT Activation), which associates with an RNA-binding E3 ligase, MEX3C, and enhances its enzymatic activity, leading to the K27-linked polyubiquitination of PTEN. The MEX3C-catalyzed PTENK27-polyUb activates its protein serine/threonine phosphatase activity and inhibits its phosphatidylinositol/protein tyrosine phosphatase activity. With this altered enzymatic activity, PTENK27-polyUb dephosphorylates the phosphoserine/threonine residues of TWIST1, SNAI1, and YAP1, leading to accumulation of these master regulators of EMT. Animals with genetic inhibition of PTENK27-polyUb, by a single nucleotide mutation generated using CRISPR/Cas9 (PtenK80R/K80R), exhibit inhibition of EMT markers during mammary gland morphogenesis in pregnancy/lactation and during cutaneous wound healing processes. Our findings illustrate an unexpected paradigm in which the lncRNA-dependent switch in PTEN protein serine/threonine phosphatase activity is important for physiological homeostasis and disease development.

Abstract P4-08-09: DNA methylation markers predict recurrence-free interval in triple-negative breast cancer

BACKGROUND. Chemotherapy remains the treatment mainstay for triple-negative breast cancer (TNBC). Nevertheless, randomized trials have shown that not all TNBC require it, nor does it benefit all patients that receive it. Molecular tools to risk-stratify TNBC are currently lacking. In light of the importance of epigenetic processes modulating gene expression, we performed an array-based genome-wide DNA methylation search in well-documented institutional and clinical trial cohorts of TNBC for markers that can distinguish breast cancers with a favorable natural history from those with a high risk of recurrence.

METHODS. We performed an array-based genome-wide DNA methylation survey of well-documented institutional and clinical trial cohorts of TNBC and conducted molecular marker discovery on institutional TNBCs (115 patient samples; 53 recurrences) treated by locoregional therapy (LRT) alone. The identified hypermethylated gene signatures were then tested in a TNBC cohort (50 patient samples; 16 recurrences) from the no chemotherapy arms of IBCSG trials VIII and IX, and in a separate combined cohort of TNBCs (131 patient samples; 33 recurrences) treated with chemotherapy from an institutional repository and from IBCSG trials VIII and IX. Cross platform validation was conducted using quantitative multiplexed methylation specific PCR (QM-MSP) on hypermethylated markers in samples from both the Discovery Set and IBCSG LRT Test Set.

RESULTS. We identified methylation signatures in the discovery cohort consisting of 100 or 30 CpG probes that discriminated patients who remained recurrence-free from those with recurrent disease. These signatures were then tested in the IBCSG no chemotherapy cohort, and we found that hypermethylation was associated with shorter recurrence-free interval (RFI). A significant association of both 100 CpG (P&lt;0.0001) and 30 CpG (P=0.0021) signatures with shorter RFI was found in the combined institutional and IBCSG chemotherapy cohort. We observed an enrichment of methylation probes residing on chromosome 19, particularly within 19q13.41-43, that significantly correlated with RFI following chemotherapy. QM-MSP results reflected that of the methylation array [Spearman correlation coefficient of r = 0.495 (P = 0.0009)] indicating that the relationship between high methylation and short RFI is detectable independent of analytical platform. We also observed enrichment for Chromosome 19-specific probes within the 100 and 30 probe sets. While only 5% of all CpG markers are located within Chr19, 15% of the 100 CpG set, 37% of the 30 CpG set, and 47% of the 17 CpGs that are statistically significantly correlated with RFI in the chemotherapy group reside on the Chr19, mostly within 19q13.41-43.

CONCLUSIONS. Methylation markers may be of prognostic importance in TNBC and our findings should be validated in additional clinical trial cohorts.

Citation Format: Fackler MJ, Cho SS, Cope LM, Gabrielson E, Wilsbach K, Lynch C, Marks JR, Geradts J, Regan MM, Viale G, Wolff AC, Umbricht CB, Sukumar S. DNA methylation markers predict recurrence-free interval in triple-negative breast cancer [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P4-08-09.

Association of distant recurrence-free survival with algorithmically extracted MRI characteristics in breast cancer

BACKGROUND

While important in diagnosis of breast cancer, the scientific assessment of the role of imaging in prognosis of outcomes and treatment planning is limited.

PURPOSE

To evaluate the potential of using quantitative imaging variables for stratifying risk of distant recurrence in breast cancer patients.

STUDY TYPE

Retrospective.

POPULATION

In all, 892 female invasive breast cancer patients.

SEQUENCE

Dynamic contrast-enhanced MRI with field strength 1.5 T and 3 T.

ASSESSMENT

Computer vision algorithms were applied to extract a comprehensive set of 529 imaging features quantifying size, shape, enhancement patterns, and heterogeneity of the tumors and the surrounding tissue. Using a development set with 446 cases, we selected 20 imaging features with high prognostic value.

STATISTICAL TESTS

We evaluated the imaging features using an independent test set with 446 cases. The principal statistical measure was a concordance index between individual imaging features and patient distant recurrence-free survival (DRFS).

RESULTS

The strongest association with DRFS that persisted after controlling for known prognostic clinical and pathology variables was found for signal enhancement ratio (SER) partial tumor volume (concordance index [C] = 0.768, 95% confidence interval [CI]: 0.679-0.856), tumor major axis length (C = 0.742, 95% CI: 0.650-0.834), kurtosis of the SER map within tumor (C = 0.640, 95% CI: 0.521-0.760), tumor cluster shade (C = 0.313, 95% CI: 0.216-0.410), and washin rate information measure of correlation (C = 0.702, 95% CI: 0.601-0.803).

DATA CONCLUSION

Quantitative assessment of breast cancer features seen in a routine breast MRI might be able to be used for assessment of risk of distant recurrence.

LEVEL OF EVIDENCE

4 Technical Efficacy: Stage 6 J. Magn. Reson. Imaging 2019.

Intra-tumor molecular heterogeneity in breast cancer: definitions of measures and association with distant recurrence-free survival

PURPOSE

The purpose of the study was to define quantitative measures of intra-tumor heterogeneity in breast cancer based on histopathology data gathered from multiple samples on individual patients and determine their association with distant recurrence-free survival (DRFS).

METHODS

We collected data from 971 invasive breast cancers, from 1st January 2000 to 23rd March 2014, that underwent repeat tumor sampling at our institution. We defined and calculated 31 measures of intra-tumor heterogeneity including ER, PR, and HER2 immunohistochemistry (IHC), proliferation, EGFR IHC, grade, and histology. For each heterogeneity measure, Cox proportional hazards models were used to determine whether patients with heterogeneous disease had different distant recurrence-free survival (DRFS) than those with homogeneous disease.

RESULTS

The presence of heterogeneity in ER percentage staining was prognostic of reduced DRFS with a hazard ratio of 4.26 (95% CI 2.22-8.18, p < 0.00002). It remained significant after controlling for the ER status itself (p < 0.00062) and for patients that had chemotherapy (p < 0.00032). Most of the heterogeneity measures did not show any association with DRFS despite the considerable sample size.

CONCLUSIONS

Intra-tumor heterogeneity of ER receptor status may be a predictor of patient DRFS. Histopathologic data from multiple tissue samples may offer a view of tumor heterogeneity and assess recurrence risk.

Multivariate machine learning models for prediction of pathologic response to neoadjuvant therapy in breast cancer using MRI features: a study using an independent validation set

PURPOSE

To determine whether a multivariate machine learning-based model using computer-extracted features of pre-treatment dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) can predict pathologic complete response (pCR) to neoadjuvant therapy (NAT) in breast cancer patients.

METHODS

Institutional review board approval was obtained for this retrospective study of 288 breast cancer patients at our institution who received NAT and had a pre-treatment breast MRI. A comprehensive set of 529 radiomic features was extracted from each patient's pre-treatment MRI. The patients were divided into equal groups to form a training set and an independent test set. Two multivariate machine learning models (logistic regression and a support vector machine) based on imaging features were trained to predict pCR in (a) all patients with NAT, (b) patients with neoadjuvant chemotherapy (NACT), and (c) triple-negative or human epidermal growth factor receptor 2-positive (TN/HER2+) patients who had NAT. The multivariate models were tested using the independent test set, and the area under the receiver operating characteristics (ROC) curve (AUC) was calculated.

RESULTS

Out of the 288 patients, 64 achieved pCR. The AUC values for predicting pCR in TN/HER+ patients who received NAT were significant (0.707, 95% CI 0.582-0.833, p < 0.002).

CONCLUSIONS

The multivariate models based on pre-treatment MRI features were able to predict pCR in TN/HER2+ patients.

LncRNA CamK-A Regulates Ca2+-Signaling-Mediated Tumor Microenvironment Remodeling

Cancer cells entail metabolic adaptation and microenvironmental remodeling to survive and progress. Both calcium (Ca²⁺) flux and Ca²⁺-dependent signaling play a crucial role in this process, although the underlying mechanism has yet to be elucidated. Through RNA screening, we identified one long noncoding RNA (lncRNA) named CamK-A (lncRNA for calcium-dependent kinase activation) in tumorigenesis. CamK-A is highly expressed in multiple human cancers and involved in cancer microenvironment remodeling via activation of Ca²⁺-triggered signaling. Mechanistically, CamK-A activates Ca²⁺/calmodulin-dependent kinase PNCK, which in turn phosphorylates IκBα and triggers calcium-dependent nuclear factor κB (NF-κB) activation. This regulation results in the tumor microenvironment remodeling, including macrophage recruitment, angiogenesis, and tumor progression. Notably, our human-patient-derived xenograft (PDX) model studies demonstrate that targeting CamK-A robustly impaired cancer development. Clinically, CamK-A expression coordinates with the activation of CaMK-NF-κB axis, and its high expression indicates poor patient survival rate, suggesting its role as a potential biomarker and therapeutic target.

Expression of Long Noncoding RNA YIYA Promotes Glycolysis in Breast Cancer

Long noncoding RNA (lncRNA) is yet to be linked to cancer metabolism. Here, we report that upregulation of the lncRNA LINC00538 (YIYA) promotes glycolysis, cell proliferation, and tumor growth in breast cancer. YIYA is associated with the cytosolic cyclin-dependent kinase CDK6 and regulated CDK6-dependent phosphorylation of the fructose bisphosphatase PFK2 (PFKFB3) in a cell-cycle-independent manner. In breast cancer cells, these events promoted catalysis of glucose 6-phosphate to fructose-2,6-bisphosphate/fructose-1,6-bisphosphate. CRISPR/Cas9-mediated deletion of YIYA or CDK6 silencing impaired glycolysis and tumor growth in vivo In clinical specimens of breast cancer, YIYA was expressed in approximately 40% of cases where it correlated with CDK6 expression and unfavorable survival outcomes. Our results define a functional role for lncRNA in metabolic reprogramming in cancer, with potential clinical implications for its therapeutic targeting.Significance: These findings offer a first glimpse into how a long-coding RNA influences cancer metabolism to drive tumor growth. Cancer Res; 78(16); 4524-32. ©2018 AACR.

Abstract 5318: High-grade serous ovarian cancer DNA methylation and survival in African-American women

Ovarian cancer is the most lethal gynecologic cancer in the United States, and African-American (AA) women have the poorest outcomes compared to other racial/ethnic groups. Although several biomarkers have been proposed to establish prognosis in European ancestry (EA) patients, including some DNA methylation markers (FGF4, FGF21, MYLK2, MYLK3, MYL7, and ITGAE) (Phelps et al., 2017), it is unknown if these or other markers are applicable to AA patients. Using data from the African-American Cancer Epidemiology Study (AACES), we evaluated 1) if DNA methylation is associated with residual disease and survival, and 2) if previously reported CpG biomarkers for EA women are related to survival in AA women. 121 AA women with high-grade serous ovarian cancer (HGSOC) were randomly selected from 600 women enrolled in AACES. Clinical records and formalin-fixed, paraffin-embedded (FFPE) tumor tissue were retrieved, and a pathologist reviewed histopathologic slides to confirm diagnosis. 92 HGSOC had complete clinical information. DNA methylation was measured using Illlumina HumanMethylationEPIC. Beta-values were preprocessed (RELIC, Tost+BMIQ and ComBat). Low quality probes were filtered. Tumor purity was estimated using InfiniumPurify; &lt;30% was considered low purity. Four cell estimates (RefFreecellmix) were used as a proxy of cell composition. To evaluate DNA methylation alterations, we used the top 100K most variable CpG sites and a semisupervised recursively partitioned mixture model (ssRPMM) approach to delineate the patients into RPMM classes. We also evaluated the six candidate CpGs from Phelps et al. The dataset was divided into training and validation subsets (50% each); if findings were consistent, a pooled statistic is reported. We used logistic regression to evaluate the association between DNA methylation and residual disease; Cox proportional hazard models were used for survival. Models were adjusted for age at diagnosis, low purity, cell types, neoadjuvant therapy, tissue source (adnexa vs. peritoneum), histology (serous vs. mixed), and residual disease. The RPMM classes were not associated with residual disease. For survival, four RPMM classes were delineated, which we collapsed into two classes. A lower risk of mortality was observed for one of the RPMM classes, HR: 0. 03[95% CI: 0.01-0.12]. This “low risk” RPMM class grouped five CpGs in genes PLEC1, AP5B1, DNAH7 and MAPK15. These genes have been associated with cell motility and ovarian cancer ascites. Among the candidate CpGs, we only observed a trend to better survival per every 10% increase in MYLK3 CpG methylation, HR: 0.51 [95%CI: 0.24-1.10]. These preliminary results suggest that some DNA methylation modifications may identify subgroups of AA women with better survival. Previously reported biomarkers in EA may not be as useful in AA women. Future studies with increased sample size may help to disentangle these associations.

Citation Format: Lucas A. Salas, Lauren C. Peres, Sarah E. Abbott, Casey S. Greene, Jeffrey R. Marks, Anthony J. Alberg, Elisa V. Bandera, Jill S. Barnholtz-Sloan, Ann G. Schwartz, Michele L. Cote, Patricia G. Moorman, Ellen M. Funkhouser, Edward S. Peters, Melissa L. Bondy, Paul D. Terry, Jennifer A. Doherty, Brock C. Christensen, Joellen M. Schildkraut. High-grade serous ovarian cancer DNA methylation and survival in African-American women [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 5318.

CircIRAK3 sponges miR-3607 to facilitate breast cancer metastasis

As a class of endogenous noncoding RNAs, circular RNAs (circRNAs) have been recently identified to regulate tumourigenesis and progression in multiple malignancies. However, the expression profiles and function of circRNAs in breast cancer metastasis are largely unknown. Here, we determined that the expression of a novel circRNA, which we named circIRAK3, was increased in metastatic breast cancer (BC) cells and predictive of BC recurrence. Gain-of-function and loss-of-function studies in BC cells demonstrated that circIRAK3 promoted cell migration, invasion and metastasis in vitro and in vivo but did not affect cell proliferation, colony formation or cell cycle progression. Using circIRAK3 in vivo precipitation and luciferase reporter assays, we identified miR-3607 as a circIRAK3-associated miRNA. Furthermore, RNA sequencing and bioinformatics analysis showed that forkhead box C1 (FOXC1), the target of miR-3607, was downregulated in circIRAK3-silenced cells and mediated circIRAK3-induced BC cell migration. Intriguingly, FOXC1 could, in turn, bind to the IRAK3 promoter, triggering a positive-feedback loop that perpetuated the circIRAK3/miR-3607/FOXC1 signaling axis. Collectively, our findings indicated that circIRAK3 may exert regulatory roles in BC metastasis and may be a potential target for metastatic BC therapy.

Prediction of Occult Invasive Disease in Ductal Carcinoma in Situ Using Deep Learning Features

PURPOSE

The aim of this study was to determine whether deep features extracted from digital mammograms using a pretrained deep convolutional neural network are prognostic of occult invasive disease for patients with ductal carcinoma in situ (DCIS) on core needle biopsy.

METHODS

In this retrospective study, digital mammographic magnification views were collected for 99 subjects with DCIS at biopsy, 25 of which were subsequently upstaged to invasive cancer. A deep convolutional neural network model that was pretrained on nonmedical images (eg, animals, plants, instruments) was used as the feature extractor. Through a statistical pooling strategy, deep features were extracted at different levels of convolutional layers from the lesion areas, without sacrificing the original resolution or distorting the underlying topology. A multivariate classifier was then trained to predict which tumors contain occult invasive disease. This was compared with the performance of traditional "handcrafted" computer vision (CV) features previously developed specifically to assess mammographic calcifications. The generalization performance was assessed using Monte Carlo cross-validation and receiver operating characteristic curve analysis.

RESULTS

Deep features were able to distinguish DCIS with occult invasion from pure DCIS, with an area under the receiver operating characteristic curve of 0.70 (95% confidence interval, 0.68-0.73). This performance was comparable with the handcrafted CV features (area under the curve = 0.68; 95% confidence interval, 0.66-0.71) that were designed with prior domain knowledge.

CONCLUSIONS

Despite being pretrained on only nonmedical images, the deep features extracted from digital mammograms demonstrated comparable performance with handcrafted CV features for the challenging task of predicting DCIS upstaging.

Abstract GS5-04: Prediction of occult invasive disease in ductal carcinoma in situ using deep learning features

Background: Deep learning, especially deep convolutional neural network (CNN), has emerged as a promising approach for many image recognition or classification tasks, demonstrating human or even superhuman performance. Used as feature extractor, some pre-trained CNN models can match or surpass the performance of domain-specific, “handcrafted” features. In this study, we aim to determine whether deep features extracted from digital mammograms using a pre-trained deep CNN are prognostic of occult invasive disease for patients with ductal carcinoma in situ (DCIS) on core needle biopsy.

Materials and Methods: In this retrospective study, we collected digital mammography magnification views for 99 subjects with DCIS at biopsy, 25 of which were subsequently upstaged to invasive cancer. We utilized a deep CNN model that was pre-trained on non-medical images (e.g., animals, plants, instruments) as the feature extractor. Through a statistical pooling strategy, we extracted deep features at different levels of convolutional layers from the lesion areas, without sacrificing the original resolution or distorting the underlying topology. A multivariate classifier was then trained to predict which tumors contain occult invasive disease. This was compared to the performance of traditional “handcrafted” computer vision (CV) features previously developed specifically to assess mammographic calcifications. The generalization performance was assessed using Monte Carlo cross validation and receiver operating characteristic (ROC) curve analysis.

Results: Deep features were able to distinguish DCIS with occult invasion from pure DCIS, with an area under the ROC curve (AUC-ROC) equal to 0.70 (95% CI: 0.68-0.73). This performance was comparable to the "handcrafted" CV features (AUC-ROC = 0.68, 95% CI: 0.66-0.71) that were designed with prior domain knowledge.

Conclusion: In spite of being pre-trained on only non-medical images, the deep features extracted from digital mammograms demonstrated comparable performance to "handcrafted" CV features for the challenging task of predicting DCIS upstaging.

Acknowledgments: This work was supported in part by NIH/NCI R01-CQA185138 and DOD Breast Cancer Research Program W81XWH-14-1-0473.

Citation Format: Lo JY, Grimm LJ, Mazurowski MA, Baker JA, Marks JR, King LM, Maley CC, Hwang E-SS, Shi B. Prediction of occult invasive disease in ductal carcinoma in situ using deep learning features [abstract]. In: Proceedings of the 2017 San Antonio Breast Cancer Symposium; 2017 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2018;78(4 Suppl):Abstract nr GS5-04.

Abstract P2-05-05: Not presented

This abstract was not presented at the symposium.

JAK2-binding long noncoding RNA promotes breast cancer brain metastasis

Conventional therapies for breast cancer brain metastases (BCBMs) have been largely ineffective because of chemoresistance and impermeability of the blood-brain barrier. A comprehensive understanding of the underlying mechanism that allows breast cancer cells to infiltrate the brain is necessary to circumvent treatment resistance of BCBMs. Here, we determined that expression of a long noncoding RNA (lncRNA) that we have named lncRNA associated with BCBM (Lnc-BM) is prognostic of the progression of brain metastasis in breast cancer patients. In preclinical murine models, elevated Lnc-BM expression drove BCBM, while depletion of Lnc-BM with nanoparticle-encapsulated siRNAs effectively treated BCBM. Lnc-BM increased JAK2 kinase activity to mediate oncostatin M- and IL-6-triggered STAT3 phosphorylation. In breast cancer cells, Lnc-BM promoted STAT3-dependent expression of ICAM1 and CCL2, which mediated vascular co-option and recruitment of macrophages in the brain, respectively. Recruited macrophages in turn produced oncostatin M and IL-6, thereby further activating the Lnc-BM/JAK2/STAT3 pathway and enhancing BCBM. Collectively, our results show that Lnc-BM and JAK2 promote BCBMs by mediating communication between breast cancer cells and the brain microenvironment. Moreover, these results suggest targeting Lnc-BM as a potential strategy for fighting this difficult disease.

Can algorithmically assessed MRI features predict which patients with a preoperative diagnosis of ductal carcinoma in situ are upstaged to invasive breast cancer?

PURPOSE

To assess the ability of algorithmically assessed magnetic resonance imaging (MRI) features to predict the likelihood of upstaging to invasive cancer in newly diagnosed ductal carcinoma in situ (DCIS).

MATERIALS AND METHODS

We identified 131 patients at our institution from 2000-2014 with a core needle biopsy-confirmed diagnosis of pure DCIS, a 1.5 or 3T preoperative bilateral breast MRI with nonfat-saturated T1 -weighted MRI sequences, no preoperative therapy before breast MRI, and no prior history of breast cancer. A fellowship-trained radiologist identified the lesion on each breast MRI using a bounding box. Twenty-nine imaging features were then computed automatically using computer algorithms based on the radiologist's annotation.

RESULTS

The rate of upstaging of DCIS to invasive cancer in our study was 26.7% (35/131). Out of all imaging variables tested, the information measure of correlation 1, which quantifies spatial dependency in neighboring voxels of the tumor, showed the highest predictive value of upstaging with an area under the curve (AUC) = 0.719 (95% confidence interval [CI]: 0.609-0.829). This feature was statistically significant after adjusting for tumor size (P < 0.001).

CONCLUSION

Automatically assessed MRI features may have a role in triaging which patients with a preoperative diagnosis of DCIS are at highest risk for occult invasive disease.

LEVEL OF EVIDENCE

4 Technical Efficacy: Stage 3 J. Magn. Reson. Imaging 2017;46:1332-1340.