A whole slide image-based machine learning approach to predict ductal carcinoma in situ (DCIS) recurrence risk.

BACKGROUND: Breast ductal carcinoma in situ (DCIS) represent approximately 20% of screen-detected breast cancers. The overall risk for DCIS patients treated with breast-conserving surgery stems almost exclusively from local recurrence. Although a mastectomy or adjuvant radiation can reduce recurrence risk, there are significant concerns regarding patient over-/under-treatment. Current clinicopathological markers are insufficient to accurately assess the recurrence risk. To address this issue, we developed a novel machine learning (ML) pipeline to predict risk of ipsilateral recurrence using digitized whole slide images (WSI) and clinicopathologic long-term outcome data from a retrospectively collected cohort of DCIS patients (n = 344) treated with lumpectomy at Nottingham University Hospital, UK. METHODS: The cohort was split case-wise into training (n = 159, 31 with 10-year recurrence) and validation (n = 185, 26 with 10-year recurrence) sets. The sections from primary tumors were stained with H&E, then digitized and analyzed by the pipeline. In the first step, a classifier trained manually by pathologists was applied to digital slides to annotate the areas of stroma, normal/benign ducts, cancer ducts, dense lymphocyte region, and blood vessels. In the second step, a recurrence risk classifier was trained on eight select architectural and spatial organization tissue features from the annotated areas to predict recurrence risk. RESULTS: The recurrence classifier significantly predicted the 10-year recurrence risk in the training [hazard ratio (HR) = 11.6; 95% confidence interval (CI) 5.3-25.3, accuracy (Acc) = 0.87, sensitivity (Sn) = 0.71, and specificity (Sp) = 0.91] and independent validation [HR = 6.39 (95% CI 3.0-13.8), p < 0.0001;Acc = 0.85, Sn = 0.5, Sp = 0.91] cohorts. Despite the limitations of our cohorts, and in some cases inferior sensitivity performance, our tool showed superior accuracy, specificity, positive predictive value, concordance, and hazard ratios relative to tested clinicopathological variables in predicting recurrences (p < 0.0001). Furthermore, it significantly identified patients that might benefit from additional therapy (validation cohort p = 0.0006). CONCLUSIONS: Our machine learning-based model fills an unmet clinical need for accurately predicting the recurrence risk for lumpectomy-treated DCIS patients.

Machine learning-based prediction of breast cancer growth rate in vivo.

BACKGROUND: Determining the rate of breast cancer (BC) growth in vivo, which can predict prognosis, has remained elusive despite its relevance for treatment, screening recommendations and medicolegal practice. We developed a model that predicts the rate of in vivo tumour growth using a unique study cohort of BC patients who had two serial mammograms wherein the tumour, visible in the diagnostic mammogram, was missed in the first screen. METHODS: A serial mammography-derived in vivo growth rate (SM-INVIGOR) index was developed using tumour volumes from two serial mammograms and time interval between measurements. We then developed a machine learning-based surrogate model called Surr-INVIGOR using routinely assessed biomarkers to predict in vivo rate of tumour growth and extend the utility of this approach to a larger patient population. Surr-INVIGOR was validated using an independent cohort. RESULTS: SM-INVIGOR stratified discovery cohort patients into fast-growing versus slow-growing tumour subgroups, wherein patients with fast-growing tumours experienced poorer BC-specific survival. Our clinically relevant Surr-INVIGOR stratified tumours in the discovery cohort and was concordant with SM-INVIGOR. In the validation cohort, Surr-INVIGOR uncovered significant survival differences between patients with fast-growing and slow-growing tumours. CONCLUSION: Our Surr-INVIGOR model predicts in vivo BC growth rate during the pre-diagnostic stage and offers several useful applications.

Novel immunohistochemistry-based signatures to predict metastatic site of triple-negative breast cancers.

BACKGROUND: Although distant metastasis (DM) in breast cancer (BC) is the most lethal form of recurrence and the most common underlying cause of cancer related deaths, the outcome following the development of DM is related to the site of metastasis. Triple negative BC (TNBC) is an aggressive form of BC characterised by early recurrences and high mortality. Athough multiple variables can be used to predict the risk of metastasis, few markers can predict the specific site of metastasis. This study aimed at identifying a biomarker signature to predict particular sites of DM in TNBC. METHODS: A clinically annotated series of 322 TNBC were immunohistochemically stained with 133 biomarkers relevant to BC, to develop multibiomarker models for predicting metastasis to the bone, liver, lung and brain. Patients who experienced metastasis to each site were compared with those who did not, by gradually filtering the biomarker set via a two-tailed t-test and Cox univariate analyses. Biomarker combinations were finally ranked based on statistical significance, and evaluated in multivariable analyses. RESULTS: Our final models were able to stratify TNBC patients into high risk groups that showed over 5, 6, 7 and 8 times higher risk of developing metastasis to the bone, liver, lung and brain, respectively, than low-risk subgroups. These models for predicting site-specific metastasis retained significance following adjustment for tumour size, patient age and chemotherapy status. CONCLUSIONS: Our novel IHC-based biomarkers signatures, when assessed in primary TNBC tumours, enable prediction of specific sites of metastasis, and potentially unravel biomarkers previously unknown in site tropism.

Predicting Metastasis Risk in Pancreatic Neuroendocrine Tumors Using Deep Learning Image Analysis.

BACKGROUND: The prognosis of patients with pancreatic neuroendocrine tumors (PanNET), the second most common type of pancreatic cancer, varies significantly, and up to 15% of patients develop metastasis. Although certain morphological characteristics of PanNETs have been associated with patient outcome, there are no available morphology-based prognostic markers. Given that current clinical histopathology markers are unable to identify high-risk PanNET patients, the development of accurate prognostic biomarkers is needed. Here, we describe a novel machine learning, multiclassification pipeline to predict the risk of metastasis using morphological information from whole tissue slides. METHODS: Digital images from surgically resected tissues from 89 PanNET patients were used. Pathologist-annotated regions were extracted to train a convolutional neural network (CNN) to identify tiles consisting of PanNET, stroma, normal pancreas parenchyma, and fat. Computationally annotated cancer or stroma tiles and patient metastasis status were used to train CNN to calculate a region based metastatic risk score. Aggregation of the metastatic probability scores across the slide was performed to predict the risk of metastasis. RESULTS: The ability of CNN to discriminate different tissues was high (per-tile accuracy >95%; whole slide cancer regions Jaccard index = 79%). Cancer and stromal tiles with high evaluated probability provided F1 scores of 0.82 and 0.69, respectively, when we compared tissues from patients who developed metastasis and those who did not. The final model identified low-risk (n = 76) and high-risk (n = 13) patients, as well as predicted metastasis-free survival (hazard ratio: 4.71) after adjusting for common clinicopathological variables, especially in grade I/II patients. CONCLUSION: Using slides from surgically resected PanNETs, our novel, multiclassification, deep learning pipeline was able to predict the risk of metastasis in PanNET patients. Our results suggest the presence of prognostic morphological patterns in PanNET tissues, and that these patterns may help guide clinical decision making.

Hypoxia-Induced Centrosome Amplification Underlies Aggressive Disease Course in HPV-Negative Oropharyngeal Squamous Cell Carcinomas.

Human papillomavirus-negative (HPV-neg) oropharyngeal squamous cell carcinomas (OPSCCs) are associated with poorer overall survival (OS) compared with HPV-positive (HPV-pos) OPSCCs. The major obstacle in improving outcomes of HPV-neg patients is the lack of robust biomarkers and therapeutic targets. Herein, we investigated the role of centrosome amplification (CA) as a prognostic biomarker in HPV-neg OPSCCs. A quantitative evaluation of CA in clinical specimens of OPSCC revealed that (a) HPV-neg OPSCCs exhibit higher CA compared with HPV-pos OPSCCs, and (b) CA was associated with poor OS, even after adjusting for potentially confounding clinicopathologic variables. Contrastingly, CA was higher in HPV-pos cultured cell lines compared to HPV-neg ones. This divergence in CA phenotypes between clinical specimens and cultured cells can therefore be attributed to an inaccurate recapitulation of the in vivo tumor microenvironment in the cultured cell lines, namely a hypoxic environment. The exposure of HPV-neg OPSCC cultured cells to hypoxia or stabilizing HIF-1α genetically increased CA. Both the 26-gene hypoxia signature as well as the overexpression of HIF-1α positively correlated with increased CA in HPV-neg OPSCCs. In addition, we showed that HIF-1α upregulation is associated with the downregulation of miR-34a, increase in CA and expression of cyclin- D1. Our findings demonstrate that the evaluation of CA may aid in therapeutic decision-making, and CA can serve as a promising therapeutic target for HPV-neg OPSCC patients.

Prognostic Role of Androgen Receptor in Triple Negative Breast Cancer: A Multi-Institutional Study.

BACKGROUND: The androgen receptor (AR) has emerged as a potential therapeutic target for AR-positive triple-negative breast cancer (TNBC). However, conflicting reports regarding AR's prognostic role in TNBC are putting its usefulness in question. Some studies conclude that AR positivity indicates a good prognosis in TNBC, whereas others suggest the opposite, and some show that AR status has no significant bearing on the patients' prognosis. METHODS: We evaluated the prognostic value of AR in resected primary tumors from TNBC patients from six international cohorts {US (n = 420), UK (n = 239), Norway (n = 104), Ireland (n = 222), Nigeria (n = 180), and India (n = 242); total n = 1407}. All TNBC samples were stained with the same anti-AR antibody using the same immunohistochemistry protocol, and samples with ≥1% of AR-positive nuclei were deemed AR-positive TNBCs. RESULTS: AR status shows population-specific patterns of association with patients' overall survival after controlling for age, grade, population, and chemotherapy. We found AR-positive status to be a marker of good prognosis in US and Nigerian cohorts, a marker of poor prognosis in Norway, Ireland and Indian cohorts, and neutral in UK cohort. CONCLUSION: AR status, on its own, is not a reliable prognostic marker. More research to investigate molecular subtype composition among the different cohorts is warranted.

A novel machine learning approach reveals latent vascular phenotypes predictive of renal cancer outcome.

Gene expression signatures are commonly used as predictive biomarkers, but do not capture structural features within the tissue architecture. Here we apply a 2-step machine learning framework for quantitative imaging of tumor vasculature to derive a spatially informed, prognostic gene signature. The trained algorithms classify endothelial cells and generate a vascular area mask (VAM) in H&E micrographs of clear cell renal cell carcinoma (ccRCC) cases from The Cancer Genome Atlas (TCGA). Quantification of VAMs led to the discovery of 9 vascular features (9VF) that predicted disease-free-survival in a discovery cohort (n = 64, HR = 2.3). Correlation analysis and information gain identified a 14 gene expression signature related to the 9VF's. Two generalized linear models with elastic net regularization (14VF and 14GT), based on the 14 genes, separated independent cohorts of up to 301 cases into good and poor disease-free survival groups (14VF HR = 2.4, 14GT HR = 3.33). For the first time, we successfully applied digital image analysis and targeted machine learning to develop prognostic, morphology-based, gene expression signatures from the vascular architecture. This novel morphogenomic approach has the potential to improve previous methods for biomarker development.

Distinctions in Breast Tumor Recurrence Patterns Post-Therapy among Racially Distinct Populations.

BACKGROUND: Clinical studies have revealed a higher risk of breast tumor recurrence in African-American (AA) patients compared to European-American (EA) patients, contributing to the alarming inequality in clinical outcomes among the ethnic groups. However, distinctions in recurrence patterns upon receiving hormone, radiation, and/or chemotherapy between the races remain poorly characterized. METHODS: We compared patterns and rates (per 1000 cancer patients per 1 year) of recurrence following each form of treatment between AA (n = 1850) and EA breast cancer patients (n = 7931) from a cohort of patients (n = 10504) treated between 2005-2015 at Northside Hospital in Atlanta, GA. RESULTS: Among patients who received any combination of adjuvant therapy, AA displayed higher overall rates of recurrence than EA (p = 0.015; HR: 1.699; CI: 1.108-2.606). Furthermore, recurrence rates were higher in AA than EA among stage I (p = 0.031; HR: 1.736; CI: 1.052-2.864) and T1 classified patients (p = 0.003; HR: 2.009; CI: 1.263-3.197). Interestingly, among patients who received neoadjuvant chemotherapy, AA displayed higher rates of local recurrence than EA (p = 0.024; HR: 7.134; CI: 1.295-39.313). CONCLUSION: Our analysis revealed higher incidence rates of recurrence in AA compared to EA among patients that received any combination of adjuvant therapy. Moreover, our data demonstrates an increased risk of tumor recurrence in AA than EA among patients diagnosed with minimally invasive disease. This is the first clinical study to suggest that neoadjuvant chemotherapy improves breast cancer recurrence rates and patterns in AA.

Multi-institutional study of nuclear KIFC1 as a biomarker of poor prognosis in African American women with triple-negative breast cancer.

Nuclear KIFC1 (nKIFC1) predicts worse outcomes in breast cancer, but its prognostic value within racially distinct triple-negative breast cancer (TNBC) patients is unknown. Thus, nKIFC1 expression was assessed by immunohistochemistry in 163 African American (AA) and 144 White TNBC tissue microarrays (TMAs) pooled from four hospitals. nKIFC1 correlated significantly with Ki67 in White TNBCs but not in AA TNBCs, suggesting that nKIFC1 is not merely a surrogate for proliferation in AA TNBCs. High nKIFC1 weighted index (WI) was associated with significantly worse overall survival (OS), progression-free survival (PFS), and distant metastasis-free survival (DMFS) (Hazard Ratios [HRs] = 3.5, 3.1, and 3.8, respectively; P = 0.01, 0.009, and 0.007, respectively) in multivariable Cox models in AA TNBCs but not White TNBCs. Furthermore, KIFC1 knockdown more severely impaired migration in AA TNBC cells than White TNBC cells. Collectively, these data suggest that nKIFC1 WI an independent biomarker of poor prognosis in AA TNBC patients, potentially due to the necessity of KIFC1 for migration in AA TNBC cells.

Biomarkers Predicting Pathologic Complete Response to Neoadjuvant Chemotherapy in Breast Cancer.

OBJECTIVES: Recent studies have shown strong correlation of pathologic complete response (pCR) to neoadjuvant chemotherapy with survival and prognosis in breast cancers. METHODS: Clinical data from 237 breast cancer patients who received neoadjuvant chemotherapy between 2012 and 2014 were reviewed. Correlations were sought between pCR and estrogen receptor (ER), progesterone receptor (PR), and HER2 status; Nottingham and nuclear grades; tumor tubule formation; mitotic score; Ki67 index; and tumoral and stromal lymphocytic infiltration (TLI and SLI, respectively). RESULTS: Of the 237 cases, 104 (43.9%) achieved pCR. The HER2+ and triple negative breast cancer (TNBC) subtypes had higher pCR rates compared with the luminal subtype (ER+ or PR+ and HER2-). ER and PR negativity, HER2 positivity, Nottingham grade 3, increased TLI and SLI, high mitotic count and Ki67 score correlated significantly with pCR in the overall cohort. TLI and SLI correlated significantly with pCR in the HER2+ and TNBC subtypes in multivariate analysis, whereas no biomarkers correlated with pCR in the luminal subtype. CONCLUSIONS: In addition to the pathologic parameters and biomarkers already routinely assessed, evaluation of TLI and SLI may help to better select patients with HER2+ and TNBC for neoadjuvant chemotherapy.

A centrosome clustering protein, KIFC1, predicts aggressive disease course in serous ovarian adenocarcinomas.

BACKGROUND: Amplified centrosomes are widely recognized as a hallmark of cancer. Although supernumerary centrosomes would be expected to compromise cell viability by yielding multipolar spindles that results in death-inducing aneuploidy, cancer cells suppress multipolarity by clustering their extra centrosomes. Thus, cancer cells, with the aid of clustering mechanisms, maintain pseudobipolar spindle phenotypes that are associated with low-grade aneuploidy, an edge to their survival. KIFC1, a nonessential minus end-directed motor of the kinesin-14 family, is a centrosome clustering molecule, essential for viability of extra centrosome-bearing cancer cells. Given that ovarian cancers robustly display amplified centrosomes, we examined the overexpression of KIFC1 in human ovarian tumors. RESULTS: We found that in clinical epithelial ovarian cancer (EOC) samples, an expression level of KIFC1 was significantly higher when compared to normal tissues. KIFC1 expression also increased with tumor grade. Our In silico analyses showed that higher KIFC1 expression was associated with poor overall survival (OS) in serous ovarian adenocarcinoma (SOC) patients suggesting that an aggressive disease course in ovarian adenocarcinoma patients can be attributed to high KIFC1 levels. Also, gene expression levels of KIFC1 in high-grade serous ovarian carcinoma (HGSOC) highly correlated with expression of genes driving centrosome amplification (CA), as examined in publically-available databases. The pathway analysis results indicated that the genes overexpressed in KIFC1 high group were associated with processes like regulation of the cell cycle and cell proliferation. In addition, when we performed gene set enrichment analysis (GSEA) for identifying the gene ontologies associated to KIFC1 high group, we found that the first 100 genes enriched in KIFC1 high group were from centrosome components, mitotic cell cycle, and microtubule-based processes. Results from in vitro experiments on well-established in vitro models of HGSOC (OVSAHO, KURAMOCHI), OVCAR3 and SKOV3) revealed that they display robust centrosome amplification and expression levels of KIFC1 was directly associated (inversely correlated) to the status of multipolar mitosis. This association of KIFC1 and centrosome amplification with HGSOC might be able to explain the increased aggressiveness in this disease. CONCLUSION: These findings compellingly underscore that KIFC1 can be a biomarker that predicts an aggressive disease course in ovarian adenocarcinomas.

Rampant centrosome amplification underlies more aggressive disease course of triple negative breast cancers.

Centrosome amplification (CA), a cell-biological trait, characterizes pre-neoplastic and pre-invasive lesions and is associated with tumor aggressiveness. Recent studies suggest that CA leads to malignant transformation and promotes invasion in mammary epithelial cells. Triple negative breast cancer (TNBC), a histologically-aggressive subtype shows high recurrence, metastases, and mortality rates. Since TNBC and non-TNBC follow variable kinetics of metastatic progression, they constitute a novel test bed to explore if severity and nature of CA can distinguish them apart. We quantitatively assessed structural and numerical centrosomal aberrations for each patient sample in a large-cohort of grade-matched TNBC (n = 30) and non-TNBC (n = 98) cases employing multi-color confocal imaging. Our data establish differences in incidence and severity of CA between TNBC and non-TNBC cell lines and clinical specimens. We found strong correlation between CA and aggressiveness markers associated with metastasis in 20 pairs of grade-matched TNBC and non-TNBC specimens (p < 0.02). Time-lapse imaging of MDA-MB-231 cells harboring amplified centrosomes demonstrated enhanced migratory ability. Our study bridges a vital knowledge gap by pinpointing that CA underlies breast cancer aggressiveness. This previously unrecognized organellar inequality at the centrosome level may allow early-risk prediction and explain higher tumor aggressiveness and mortality rates in TNBC patients.