Generative Adversarial Networks Accurately Reconstruct Pan-Cancer Histology from Pathologic, Genomic, and Radiographic Latent Features.

Artificial intelligence models have been increasingly used in the analysis of tumor histology to perform tasks ranging from routine classification to identification of novel molecular features. These approaches distill cancer histologic images into high-level features which are used in predictions, but understanding the biologic meaning of such features remains challenging. We present and validate a custom generative adversarial network - HistoXGAN - capable of reconstructing representative histology using feature vectors produced by common feature extractors. We evaluate HistoXGAN across 29 cancer subtypes and demonstrate that reconstructed images retain information regarding tumor grade, histologic subtype, and gene expression patterns. We leverage HistoXGAN to illustrate the underlying histologic features for deep learning models for actionable mutations, identify model reliance on histologic batch effect in predictions, and demonstrate accurate reconstruction of tumor histology from radiographic imaging for a 'virtual biopsy'.

Deep learning generates synthetic cancer histology for explainability and education.

Artificial intelligence methods including deep neural networks (DNN) can provide rapid molecular classification of tumors from routine histology with accuracy that matches or exceeds human pathologists. Discerning how neural networks make their predictions remains a significant challenge, but explainability tools help provide insights into what models have learned when corresponding histologic features are poorly defined. Here, we present a method for improving explainability of DNN models using synthetic histology generated by a conditional generative adversarial network (cGAN). We show that cGANs generate high-quality synthetic histology images that can be leveraged for explaining DNN models trained to classify molecularly-subtyped tumors, exposing histologic features associated with molecular state. Fine-tuning synthetic histology through class and layer blending illustrates nuanced morphologic differences between tumor subtypes. Finally, we demonstrate the use of synthetic histology for augmenting pathologist-in-training education, showing that these intuitive visualizations can reinforce and improve understanding of histologic manifestations of tumor biology.

Tumor immune microenvironment of self-identified African American and non-African American triple negative breast cancer.

Differences in the tumor immune microenvironment may result in differences in prognosis and response to treatment in cancer patients. We hypothesized that differences in the tumor immune microenvironment may exist between African American (AA) and NonAA patients, due to ancestry-related or socioeconomic factors, that may partially explain differences in clinical outcomes. We analyzed clinically matched triple-negative breast cancer (TNBC) tissues from self-identified AA and NonAA patients and found that stromal TILs, PD-L1 IHC-positivity, mRNA expression of immune-related pathways, and immunotherapy response predictive signatures were significantly higher in AA samples (p < 0.05; Fisher's Exact Test, Mann-Whitney Test, Permutation Test). Cancer biology and metabolism pathways, TAM-M2, and Immune Exclusion were significantly higher in NonAA samples (p < 0.05; Permutation Test, Mann-Whitney Test). There were no differences in somatic tumor mutation burden. Overall, there is greater immune infiltration and inflammation in AA TNBC and these differences may impact response to immune checkpoint inhibitors and other therapeutic agents that modulate the immune microenvironment.

Residual cancer burden after neoadjuvant chemotherapy and long-term survival outcomes in breast cancer: a multicentre pooled analysis of 5161 patients.

BACKGROUND: Previous studies have independently validated the prognostic relevance of residual cancer burden (RCB) after neoadjuvant chemotherapy. We used results from several independent cohorts in a pooled patient-level analysis to evaluate the relationship of RCB with long-term prognosis across different phenotypic subtypes of breast cancer, to assess generalisability in a broad range of practice settings. METHODS: In this pooled analysis, 12 institutes and trials in Europe and the USA were identified by personal communications with site investigators. We obtained participant-level RCB results, and data on clinical and pathological stage, tumour subtype and grade, and treatment and follow-up in November, 2019, from patients (aged ≥18 years) with primary stage I-III breast cancer treated with neoadjuvant chemotherapy followed by surgery. We assessed the association between the continuous RCB score and the primary study outcome, event-free survival, using mixed-effects Cox models with the incorporation of random RCB and cohort effects to account for between-study heterogeneity, and stratification to account for differences in baseline hazard across cancer subtypes defined by hormone receptor status and HER2 status. The association was further evaluated within each breast cancer subtype in multivariable analyses incorporating random RCB and cohort effects and adjustments for age and pretreatment clinical T category, nodal status, and tumour grade. Kaplan-Meier estimates of event-free survival at 3, 5, and 10 years were computed for each RCB class within each subtype. FINDINGS: We analysed participant-level data from 5161 patients treated with neoadjuvant chemotherapy between Sept 12, 1994, and Feb 11, 2019. Median age was 49 years (IQR 20-80). 1164 event-free survival events occurred during follow-up (median follow-up 56 months [IQR 0-186]). RCB score was prognostic within each breast cancer subtype, with higher RCB score significantly associated with worse event-free survival. The univariable hazard ratio (HR) associated with one unit increase in RCB ranged from 1·55 (95% CI 1·41-1·71) for hormone receptor-positive, HER2-negative patients to 2·16 (1·79-2·61) for the hormone receptor-negative, HER2-positive group (with or without HER2-targeted therapy; p<0·0001 for all subtypes). RCB score remained prognostic for event-free survival in multivariable models adjusted for age, grade, T category, and nodal status at baseline: the adjusted HR ranged from 1·52 (1·36-1·69) in the hormone receptor-positive, HER2-negative group to 2·09 (1·73-2·53) in the hormone receptor-negative, HER2-positive group (p<0·0001 for all subtypes). INTERPRETATION: RCB score and class were independently prognostic in all subtypes of breast cancer, and generalisable to multiple practice settings. Although variability in hormone receptor subtype definitions and treatment across patients are likely to affect prognostic performance, the association we observed between RCB and a patient's residual risk suggests that prospective evaluation of RCB could be considered to become part of standard pathology reporting after neoadjuvant therapy. FUNDING: National Cancer Institute at the US National Institutes of Health.

An Open-Source, Automated Tumor-Infiltrating Lymphocyte Algorithm for Prognosis in Triple-Negative Breast Cancer.

PURPOSE: Although tumor-infiltrating lymphocytes (TIL) assessment has been acknowledged to have both prognostic and predictive importance in triple-negative breast cancer (TNBC), it is subject to inter and intraobserver variability that has prevented widespread adoption. Here we constructed a machine-learning based breast cancer TIL scoring approach and validated its prognostic potential in multiple TNBC cohorts. EXPERIMENTAL DESIGN: Using the QuPath open-source software, we built a neural-network classifier for tumor cells, lymphocytes, fibroblasts, and "other" cells on hematoxylin-eosin (H&E)-stained sections. We analyzed the classifier-derived TIL measurements with five unique constructed TIL variables. A retrospective collection of 171 TNBC cases was used as the discovery set to identify the optimal association of machine-read TIL variables with patient outcome. For validation, we evaluated a retrospective collection of 749 TNBC patients comprised of four independent validation subsets. RESULTS: We found that all five machine TIL variables had significant prognostic association with outcomes (P ≤ 0.01 for all comparisons) but showed cell-specific variation in validation sets. Cox regression analysis demonstrated that all five TIL variables were independently associated with improved overall survival after adjusting for clinicopathologic factors including stage, age, and histologic grade (P ≤ 0.0003 for all analyses). CONCLUSIONS: Neural net-driven cell classifier-defined TIL variables were robust and independent prognostic factors in several independent validation cohorts of TNBC patients. These objective, open-source TIL variables are freely available to download and can now be considered for testing in a prospective setting to assess clinical utility.See related commentary by Symmans, p. 5446.

Immunological Differences Between Immune-Rich Estrogen Receptor-Positive and Immune-Rich Triple-Negative Breast Cancers.

PURPOSE: A subset of estrogen receptor-positive (ER-positive) breast cancer (BC) contains high levels of tumor-infiltrating lymphocytes (TILs), similar to triple-negative BC (TNBC). The majority of immuno-oncology trials target TNBCs because of the greater proportion of TIL-rich TNBCs. The extent to which the immune microenvironments of immune-rich ER-positive BC and TNBC differ is unknown. PATIENTS AND METHODS: RNA sequencing data from The Cancer Genome Atlas (TCGA; n = 697 ER-positive BCs; n = 191 TNBCs) were used for discovery; microarray expression data from Molecular Taxonomy of Breast Cancer International Consortium (METABRIC; n = 1,186 ER-positive BCs; n = 297 TNBCs) was used for validation. Patients in the top 25th percentile of a previously published total TIL metagene score distribution were considered immune rich. We compared expression of immune cell markers, immune function metagenes, and immuno-oncology therapeutic targets among immune-rich subtypes. RESULTS: Relative fractions of resting mast cells (TCGA P adj = .009; METABRIC P adj = 4.09E-15), CD8+ T cells (TCGA P adj = .015; METABRIC P adj = 0.390), and M2-like macrophages (TCGA P adj= 4.68E-05; METABRIC P adj = .435) were higher in immune-rich ER-positive BCs, but M0-like macrophages (TCGA P adj = 0.015; METABRIC P adj = .004) and M1-like macrophages (TCGA P adj = 9.39E-08; METABRIC P adj = 6.24E-11) were higher in immune-rich TNBCs. Ninety-one immune-related genes (eg, CXCL14, CSF3R, TGF-B3, LRRC32/GARP, TGFB-R2) and a transforming growth factor ß (TGF-ß) response metagene were significantly overexpressed in immune-rich ER-positive BCs, whereas 41 immune-related genes (eg, IFNG, PD-L1, CTLA4, MAGEA4) were overexpressed in immune-rich TNBCs in both discovery and validation data sets. TGF-ß pathway member genes correlated negatively with expression of immune activation markers (IFNG, granzyme-B, perforin) and positively with M2-like macrophages (IL4, IL10, and MMP9) and regulatory T-cell (FOXP3) markers in both subtypes. CONCLUSION: Different immunotherapy strategies may be optimal in immune-rich ER-positive BC and TNBC. Drugs targeting the TGF-ß pathway and M2-like macrophages are promising strategies in immune-rich ER-positive BCs to augment antitumor immunity.

High frequency of p16 and SOX10 coexpression but not androgen receptor expression in triple-negative breast cancers.

Triple-negative breast cancers (TNBCs) represent approximately 12-17% of all breast cancers and have distinctively aggressive clinical courses. Because routine biomarkers for breast cancer do not apply for TNBCs, it is essential to find novel prognostic markers and potential targets for therapeutic agents. p16 and SOX10 are emerging biomarkers with relatively unexplored expressions in TNBCs. We present an analysis of the expression of p16 and SOX10 in combination with that of androgen receptor (AR) and cytokeratin (CK) 5/6 in TNBCs. In addition, we used tissue microarrays (TMAs) to compare frequencies of p16 and SOX10 between TNBCs and non-TNBCs. Fifty-six TNBC samples with clinical data were stained immunohistochemically with p16, SOX10, AR, and CK5/6. Fifty-four cases (96.4%) were invasive ductal carcinoma, not otherwise specified, and 46 cases (82.1%) were Nottingham histologic grade 3. The majority of TNBC cases were positive for p16 (n = 44; 78.6%) and SOX10 (n = 48; 85.7%). AR was positive in 15 cases (26.8%). CK5/6 was positive in 24 cases (42.9%), which were classified as basal-like breast cancer (BLBC) subtype. The frequencies of p16 and SOX10 expression in BLBC and non-BLBC subtypes did not reveal significant statistical difference in a separate analysis. Using archived TNBC and non-TNBC TMAs, we observed that 56% of TNBC cases were positive for p16 compared with 16% of non-TNBC cases (p-value <0.0001). SOX10 was positive in 80% of TNBC cases compared with 35% of non-TNBC cases (p-value <0.0001). A significant correlation was observed between p16 and SOX10 coexpression in TNBC cases (n = 56/80, p = 0.02) but not in non-TNBC cases (n = 23/348; p = 0.626). In conclusion, p16 and SOX10 are frequently expressed in TNBC, regardless of CK5/6 expression. Furthermore, p16 and SOX10 are often coexpressed in TNBCs compared with non-TNBCs.

Prospective multi-institutional evaluation of pathologist assessment of PD-L1 assays for patient selection in triple negative breast cancer.

The US Food and Drug Administration (FDA) approved the PD-L1 immunohistochemical assay, SP142, as a companion test to determine eligibility for atezolizumab therapy in patients with advanced triple negative breast cancer (TNBC) but data in lung cancer studies suggest the assay suffers from poor reproducibility. We sought to evaluate reproducibility and concordance in PD-L1 scoring across multiple pathologists. Full TNBC sections were stained with SP142 and SP263 assays and interpreted for percentage (%) immune cell (IC) staining by 19 pathologists from 14 academic institutions. Proportion of PD-L1 positive cases (defined as ≥1% IC) was determined for each assay as well as concordance across observers. We utilized a new method we call Observers Needed to Evaluate Subjective Tests (ONEST) to determine the minimum number of evaluators needed to estimate concordance between large numbers of readers, as occurs in the real-world setting. PD-L1 was interpreted as positive with the SP142 assay in an average 58% of cases compared with 78% with SP263 (p < 0.0001). IC positive continuous scores ranged from 1 to 95% (mean = 20%) and 1 to 90% (mean = 10%) for SP263 and SP142, respectively. With SP142, 26 cases (38%) showed complete two category (<1% vs. ≥1%) concordance; with SP263, 38 cases (50%) showed complete agreement. The intraclass correlation coefficient (ICC) for two category scoring of SP263 and SP142 was 0.513 and 0.560. ONEST plots showed decreasing overall percent agreement (OPA) as observer number increased, reaching a low plateau of 0.46 at ten observers for SP263 and 0.41 at eight observers for SP142. IC scoring with both assays showed poor reproducibility across multiple pathologists with ONEST analysis suggesting more than half of pathologists will disagree about IC scores. This could lead to many patients either receiving atezolizumab when they are unlikely to benefit, or not receiving atezolizumab when they may benefit.

Long-term implementation of a prehospital severity scale for EMS triage of acute stroke: a real-world experience.

BACKGROUND: Data on the implementation of prehospital large vessel occlusion (LVO) scales to identify and triage patients with acute ischemic stroke (AIS) in the field are limited, with the majority of studies occurring outside the USA. OBJECTIVE: To report our long-term experience of a US countywide emergency medical services (EMS) acute stroke triage protocol using the Rapid Arterial oCclusion Evaluation (RACE) score. METHODS: Our prospective database was used to identify all consecutive patients triaged within Lucas County, Ohio by the EMS with (1) a RACE score ≥5, taken directly to an endovascular capable center (ECC) as RACE-alerts (RA) and (2) a RACE score <5, taken to the nearest hospital as stroke-alerts (SA). Baseline demographics, RACE score, time metrics, final diagnosis, treatments, and clinical and angiographic outcomes were captured. The sensitivity and specificity for patients with a RACE score ≥5 with LVO, eligible for mechanical thrombectomy (MT), were calculated. RESULTS: Between July 2015 and June 2018, 492 RA and 1147 SA were triaged within our five-hospital network. Of the RA, 37% had AIS secondary to LVOs. Of the 492 RA and 1147 SA, 125 (25.4%) and 38 (3.3%), respectively, underwent MT (OR=9.9; 95% CI 6.8 to 14.6; p<0.0001). Median times from onset-to-ECC arrival (74 vs 167 min, p=0.03) and dispatch-to-ECC arrival (31 vs 46 min, p=0.0002) were shorter in the RA-MT than in the SA-MT cohort. A RACE cut-off point ≥5 showed a sensitivity and specificity of 0.77 and 0.75 for detection of patients with LVO eligible for MT, respectively. CONCLUSIONS: We have demonstrated the long-term feasibility of a countywide EMS-based prehospital triage protocol using the RACE Scale within our hospital network.

Re-Evaluating E-Cadherin and ß-Catenin: A Pan-Cancer Proteomic Approach with an Emphasis on Breast Cancer.

E-cadherin is conventionally considered to be a good prognostic marker in cancer. The loss of E-cadherin is one of the key hallmarks of epithelial-to-mesenchymal transition, a biological process that promotes cancer cell invasiveness and metastasis. Recent evidence has cast doubt on the importance of epithelial-to-mesenchymal transition in metastasis. The availability of protein-level data in the Cancer Genome Atlas allows for the quantitative analysis of protein and prognosis. The prognostic values of E-cadherin and ß-catenin were revisited across 19 cancer types, and high E-cadherin was found to correlate with good prognosis in most cancers. Conversely, higher E-cadherin and ß-catenin correlated with shorter survival in invasive breast carcinoma. Stratifying breast cancers by histologic subtype revealed that the poor prognosis of E-cadherin and ß-catenin proteins was characteristic of infiltrating ductal, but not lobular, carcinomas. To further corroborate the protein findings and examine cellular localization, immunohistochemistry was used for E-cadherin and ß-catenin in 163 breast patient samples from the Iowa cohort. Most previous studies showing that reduced or absent E-cadherin and ß-catenin was inversely associated with tumor stages in ductal carcinomas were confirmed. Taken together, these results lead us to question the prognostic values of E-cadherin and ß-catenin in ductal carcinomas and indicate a complicated role of E-cadherin and ß-catenin in breast cancer progression.

IDH2 Mutations Define a Unique Subtype of Breast Cancer with Altered Nuclear Polarity.

Solid papillary carcinoma with reverse polarity (SPCRP) is a rare breast cancer subtype with an obscure etiology. In this study, we sought to describe its unique histopathologic features and to identify the genetic alterations that underpin SPCRP using massively parallel whole-exome and targeted sequencing. The morphologic and immunohistochemical features of SPCRP support the invasive nature of this subtype. Ten of 13 (77%) SPCRPs harbored hotspot mutations at R172 of the isocitrate dehydrogenase IDH2, of which 8 of 10 displayed concurrent pathogenic mutations affecting PIK3CA or PIK3R1 One of the IDH2 wild-type SPCRPs harbored a TET2 Q548* truncating mutation coupled with a PIK3CA H1047R hotspot mutation. Functional studies demonstrated that IDH2 and PIK3CA hotspot mutations are likely drivers of SPCRP, resulting in its reversed nuclear polarization phenotype. Our results offer a molecular definition of SPCRP as a distinct breast cancer subtype. Concurrent IDH2 and PIK3CA mutations may help diagnose SPCRP and possibly direct effective treatment. Cancer Res; 76(24); 7118-29. ©2016 AACR.

Androgen receptor expression in breast cancer in relation to molecular phenotype: results from the Nurses' Health Study.

Previous studies have demonstrated that androgen receptor is expressed in many breast cancers, but its expression in relation to the various breast cancer subtypes as defined by molecular profiling has not been studied in detail. We constructed tissue microarrays from 3093 breast cancers that developed in women enrolled in the Nurses' Health Study. Tissue microarray sections were immunostained for estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), cytokeratin 5/6, epidermal growth factor receptor (EGFR) and androgen receptor (ER). Immunostain results were used to categorize each cancer as luminal A or B, HER2 and basal like. The relationships between androgen receptor expression and molecular subtype were analyzed. Overall, 77% of the invasive breast carcinomas were androgen receptor positive. Among 2171 invasive cancers, 64% were luminal A, 15% luminal B, 6% HER2 and 11% basal like. The frequency of androgen receptor expression varied significantly across the molecular phenotypes (P<0.0001). In particular, androgen receptor expression was commonly observed in luminal A (91%) and B (68%) cancers, but was less frequently seen in HER2 cancers (59%). Despite being defined by the absence of ER and PR expression and being considered hormonally unresponsive, 32% of basal-like cancers expressed androgen receptor. Among 246 cases of ductal carcinoma in situ, 86% were androgen receptor positive, but the frequency of androgen receptor expression differed significantly across the molecular phenotypes (P=0.001), and high nuclear grade lesions were less likely to be androgen receptor positive compared with lower-grade lesions. Androgen receptor expression is most commonly seen in luminal A and B invasive breast cancers. However, expression of androgen receptor is also seen in approximately one-third of basal-like cancers, providing further evidence that basal-like cancers represent a heterogeneous group. Our findings raise the possibility that targeting the androgen receptor pathway may represent a novel therapeutic approach to the management of patients with basal-like cancers.

Biologic characteristics of premalignant breast disease.

Breast cancer is the second leading cause of cancer death in women in the United States. While mammography and breast magnetic resonance imaging (MRI) improve detection of early disease, there remains an unmet need for biomarkers for risk stratification, early detection, prediction, and disease prognosis. A number of early breast lesions, from atypical hyperplasias to carcinomas in situ, are associated with an increased risk of developing subsequent invasive breast carcinoma. The recent development of genomic, epigenomic, and proteomic tools for tissue biomarker detection, including array CGH, RNA expression microarrays, and proteomic arrays have identified a number of potential biomarkers that both identify patients at increased risk, as well as provided insights into the pathology of early breast cancer development. This chapter focuses on the detection and application of tissue and serum biomarkers for the identification and risk stratification of early breast cancer lesions.

Anaerobic nitrate-dependent iron(II) bio-oxidation by a novel lithoautotrophic betaproteobacterium, strain 2002.

Microbial nitrate-dependent Fe(II) oxidation is known to contribute to iron biogeochemical cycling; however, the microorganisms responsible are virtually unknown. In an effort to elucidate this microbial metabolic process in the context of an environmental system, a 14-cm sediment core was collected from a freshwater lake and geochemically characterized concurrently with the enumeration of the nitrate-dependent Fe(II)-oxidizing microbial community and subsequent isolation of a nitrate-dependent Fe(II)-oxidizing microorganism. Throughout the sediment core, ambient concentrations of Fe(II) and nitrate were observed to coexist. Concomitant most probable number enumeration revealed the presence of an abundant nitrate-dependent Fe(II)-oxidizing microbial community (2.4 x 10(3) to 1.5 x 10(4) cells g(-1) wet sediment) from which a novel anaerobic, lithoautotrophic, Fe(II)-oxidizing bacterium, strain 2002, was isolated. Analysis of the complete 16S rRNA gene sequence revealed that strain 2002 was a member of the beta subclass of the proteobacteria with 94.8% similarity to Chromobacterium violaceum, a bacterium not previously recognized for the ability to oxidize nitrate-dependent Fe(II). Under nongrowth conditions, both strain 2002 and C. violaceum incompletely reduced nitrate to nitrite with Fe(II) as the electron donor, while under growth conditions nitrate was reduced to gaseous end products (N2 and N2O). Lithoautotrophic metabolism under nitrate-dependent Fe(II)-oxidizing conditions was verified by the requirement of CO2 for growth as well as the assimilation of 14C-labeled CO2 into biomass. The isolation of strain 2002 represents the first example of an anaerobic, mesophilic, neutrophilic Fe(II)-oxidizing lithoautotroph isolated from freshwater samples. Our studies further demonstrate the abundance of nitrate-dependent Fe(II) oxidizers in freshwater lake sediments and provide further evidence for the potential of microbially mediated Fe(II) oxidation in anoxic environments.