Search

Gynecologic carcinosarcomas (CS) are biphasic neoplasms composed of carcinomatous (C) and sarcomatous (S) malignant components. Because of their rarity and histologic complexity, genetic and functional studies on CS are scarce and the mechanisms of initiation and development remain largely unknown. Whole-genome analysis of the C and S components reveals shared genomic alterations, thus emphasizing the clonal evolution of CS. Reconstructions of the evolutionary history of each tumor further reveal that C and S samples are composed of both ancestral cell populations and component-specific subclones, supporting a common origin followed by distinct evolutionary trajectories. However, while we do not find any recurrent genomic features associated with phenotypic divergence, transcriptomic and methylome analyses identify a common mechanism across the cohort, the epithelial-to-mesenchymal transition (EMT), suggesting a role for nongenetic factors in inflicting changes to cellular fate. Altogether, these data accredit the hypothesis that CS tumors are driven by both clonal evolution and transcriptomic reprogramming, essential for susceptibility to transdifferentiation upon encountering environmental cues, thus linking CS heterogeneity to genetic, transcriptomic, and epigenetic influences. Significance: We have provided a detailed characterization of the genomic landscape of CS and identified EMT as a common mechanism associated with phenotypic divergence, linking CS heterogeneity to genetic, transcriptomic, and epigenetic influences.

APCgermline pathogenic variants result in predisposition to familial adenomatous polyposis and extraintestinal tumours such as desmoid fibromatosis, medulloblastomas and thyroid cancers. They have also been recently involved in ovarian microcystic stromal tumours.APCinactivation has been described at the tumour level in epithelial ovarian cancers (EOCs). Here, we report the identification ofAPCgermline pathogenic variants in two patients diagnosed with premenopausal EOC in early 30s, with no other pathogenic variant detected in the known ovarian cancer predisposing genes. Subsequent tumour analysis showed neither a second hit ofAPCinactivation nor β-catenin activation. Both tumours did not have a homologous recombination (HR) deficiency, pointing towards the implication of other genes than those involved in HR.APCmay contribute to the carcinogenesis of EOC in a multifactorial context. Further studies are required to clarify the role ofAPCin predisposition to EOC.

The worldwide approval of the combination maintenance therapy of olaparib and bevacizumab in advanced high-grade serous ovarian cancer requires complex molecular diagnostic assays that are sufficiently robust for the routine detection of driver mutations in homologous recombination repair (HRR) genes and genomic instability (GI), employing formalin-fixed (FFPE) paraffin-embedded tumor samples without matched normal tissue. We therefore established a DNA-based hybrid capture NGS assay and an associated bioinformatic pipeline that fulfils our institution’s specific needs. The assay´s target regions cover the full exonic territory of relevant cancer-related genes and HRR genes and more than 20,000 evenly distributed single nucleotide polymorphism (SNP) loci to allow for the detection of genome-wide allele specific copy number alterations (CNA). To determine GI status, we implemented an %CNA score that is robust across a broad range of tumor cell content (25–85%) often found in routine FFPE samples. The assay was established using high-grade serous ovarian cancer samples for which BRCA1 and BRCA2 mutation status as well as Myriad MyChoice homologous repair deficiency (HRD) status was known. The NOGGO (Northeastern German Society for Gynecologic Oncology) GIS (GI-Score) v1 assay was clinically validated on more than 400 samples of the ENGOT PAOLA-1 clinical trial as part of the European Network for Gynaecological Oncological Trial groups (ENGOT) HRD European Initiative. The “NOGGO GIS v1 assay” performed using highly robust hazard ratios for progression-free survival (PFS) and overall survival (OS), as well a significantly lower dropout rate than the Myriad MyChoice clinical trial assay supporting the clinical utility of the assay. We also provide proof of a modular and scalable routine diagnostic method, that can be flexibly adapted and adjusted to meet future clinical needs, emerging biomarkers, and further tumor entities.

Heterogeneity of genomic rearrangements in carcinosarcoma.

Curated list of chromatin remodeling genes.

Genomic landscape of CS. A, Tissue origin and histological subtypes for each tumor (from P01 to P15). B, Main component and detailed percentage content for both samples (a and b) derived from the same tumor. C, CNA analysis: tumor purity (%), ploidy, FGA, and CNA breakpoints number. D, SV number, classified as deletion, duplication, inversion, and interchromosomal translocation. E, SNV and small indel number. F, Proportion of each of four mutational signatures deciphered in the whole cohort. G, MSI phenotype: MSI score (representing the percentage of unstable microsatellite loci), MLH1 promoter methylation (beta-value), and mRNA expression level (TPM – transcripts per million). H, Kataegis number illustrating APOBEC error-prone DNA repair process. I, BRCAness phenotype: number of LST, (white stars highlight samples with LST number ≥20), BRCA1 promoter methylation level (beta-value), BRCA1/2 genes alterations and TDP score, collectively summarized by the HRD status track.

List of differentially methylated gene promoters (paired analysis of 5 selected tumors).

Key gene alterations in uterine and ovarian CS. Oncoprint of alterations identified in the 50 most altered genes from TCGA cancer pathways and custom lists of CRG. The type of genomic alteration (deletion, amplification, fusion, broken by SV, missense, truncated) is described in the legend. Number and proportions of alteration types are summarized by gene (horizontal bars on right) and by sample (vertical bars on top). The samples are classified into MSI and CN-high according to TCGA endometrial carcinoma classification.

Evolutionary histories of CS. Clonal lineage histories of P02 MSI phenotype tumor (A), P12 BRCAness-phenotype tumor (B), and P15 ovarian bilateral tumor (C). Left, Clonal lineage inference for the whole tumor: subclonal populations (Ci), main cancer gene alterations, genome doubling events, and representative mutational signature proportions are represented. Right, Subclonal population frequencies projection in each tumor sample. Clones colored in gray are found in both derived samples, pink (or green) clones are specific to sarcoma- (or carcinoma-) derived samples, respectively.

Differential expression and methylation analysis (C vs. S). A, Heatmap and pathway overrepresentation analysis of differentially expressed genes between C and S (or undifferentiated) components of the 10 samples derived from the five selected CS tumors. Gene clustering method: Ward's; distance: Spearman. FDR: false discovery rate. B, Heatmaps of EMT and stemness pathways scores (left) and methylation level of EMT-related miRNAs (right). mRNA level: scaled TPM. Pathway scores: scaled ssGSEA scores. Methylation level: scaled beta-values.

List of differentially expressed genes (paired analysis of 5 selected tumors).

Objective This study aimed to clinically validate the performance of a multi-feature AI-based solution on detection of invasive versus in situ carcinomas and in situ ductal carcinoma high grade from atypical ductal hyperplasia/low grade DCIS compared to rigorous ground truth (GT) established by multiple blinded expert pathologists in breast biopsies. Design Performance of the AI solution was prospectively tested on breast biopsies from two medical institutions in different geographies. AI results were compared against the ground truth (GT) established by consensus of two subspecialist breast pathologists. The study endpoints were detection of invasive carcinoma (IDC, ILC) and DCIS/ADH, including differentiating between different DCIS grades and ADH. ADH and DCIS Low Grade were pooled together because of similar clinical management when diagnosed on a biopsy. Results Six pathologists participated in the study and reported on 436 breast biopsies (841 H&E slides), including 156 invasive (including 31 rare subtypes), 135 DCIS/ADH and 145 benign diagnoses. The AI solution demonstrated high performance when compared with the GT with an AUC of 0.99 for the detection of invasive carcinoma (specificity and sensitivity of 93.6% and 95.5% respectively) and with AUC of 0.95 for the detection of DCIS/ADH. The AI solution differentiated well between subtypes/grades of invasive and in-situ cancers with an AUC of 0.97 for IDC vs. ILC and AUC of 0.92 for DCIS high grade vs. low grade/ADH, respectively. Only 11 (7%) cases had discrepancies on invasive diagnosis, 4 of these between invasive versus benign diagnosis encompassing one case on which the invasive component was only represented by rare lympho-vascular invasion, two cases of ILC (one with a diffuse pattern and the second in a case with granulomatous mastitis with multinucleated giant cells and hemosiderin-laden macrophages) and one rare case of tubular carcinoma surrounded by flat epithelial atypia and columnar cell lesions. Fifteen cases (11%) had discrepancies on DCIS/ADH diagnosis between the two specialist pathologists necessitating a third assessment by a specialist to establish GT. Six of these cases also necessitated a review on a multihead microscope to reach a consensus decision, since there was no majority even after 3 reviews. Conclusion This blinded multi-site study reports the successful clinical validation of a multi-feature AI-based solution in detecting and automatically imparting clinically relevant diagnostic parameters regarding invasive and in situ breast carcinoma, offering an important tool for computer-aided diagnosis in routine pathology practice. Citation Format: Anne Vincent-Salomon, Guillaume Bataillon, Alona Nudelman, Judith Sandbank, Anat Albrecht Shach, Lucie Thibault, Lilach Bien, Rachel Mikulinsky, Ira Krasnitsky, Ronen Heled, Chaim Linhart, Manuela Vecsler, Daphna Laifenfeld. A multi-feature AI-based solution for cancer diagnosis in breast biopsies: A prospective blinded multi-site clinical study [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 PD11-04.

ObjectiveOur study aimed to evaluate the association between timing of cytoreductive surgery and pattern of presentation of the first recurrence in patients with advanced ovarian cancer. We also aimed to assess the impact of the pattern of recurrence on post-relapse overall survival according to surgical timing.MethodsThis retrospective multicenter study evaluated patients with International Federation of Gynecology and Obstetrics (FIGO) stage IIIC-IV ovarian cancer. Patients had undergone either primary debulking surgery, early interval debulking surgery after 3–4 cycles of neoadjuvant chemotherapy, or delayed debulking surgery after 6 cycles, with minimal or no residual disease, between January 2008 and December 2015. Survival analyses were conducted using the Log-rank test and the Cox model. Cumulative incidences of the different patterns of recurrence were estimated using a competing risks methodology.ResultsA total of 549 patients were included: 175 (31.9%) patients had primary, 224 (40.8%) early interval, and 150 (27.3%) delayed debulking surgery. The cumulative incidence of peritoneal recurrences at 2 years was higher with increasing neoadjuvant cycles (24.4%, 30.9% and 39.2%; p=0.019). For pleural or pulmonary recurrences, it was higher after early interval surgery (9.9%, 13.0% and 4.1%; p=0.022). Median post-relapse overall survival was 33.5 months (95% confidence interval (CI) (24.3 to 44.2)), 26.8 months (95% CI (22.8 to 32.6)), and 24.5 months (95% CI (18.6 to 29.4)) for primary, early interval, and delayed debulking surgery groups, respectively (p=0.025). The pattern of recurrence in a lymph node (hazard ratio (HR) 0.42, 95% CI (0.27 to 0.64)), delayed surgery (HR 1.53, 95% CI (1.11 to 2.13)) and time to first recurrence (HR 0.95, 95% CI (0.93 to 0.96)) were associated with post-relapse overall survival. For primary and early interval surgery, lymph node recurrences were associated with significantly longer post-relapse overall survival.ConclusionsThe pattern of first recurrence was associated with timing of surgery, with peritoneal recurrences being more frequent with the increasing number of cycles of neoadjuvant chemotherapy. Lymph node recurrences were associated with better prognosis, having higher post-relapse overall survival. This improved prognosis of lymphatic recurrences was not observed in patients who underwent delayed surgery.

The use of conventional methods (immunohistochemistry, pentaplex PCR) for detecting microsatellite instability (MSI), a predictive biomarker of immunotherapy efficacy, is debated for cancers with low MSI prevalence, such as breast cancer (BC). We developed two multiplex drop-off droplet digital PCR (ddPCR) assays targeting four microsatellites, initially identified from public BC whole-genome sequencing dataset. Performances of the assays were investigated and 352 tumor DNA and 28 circulating cell-free DNA from BC patients, with unknown MSI status were blindly screened. Cross-validation of ddPCR MSI status with other MSI detection methods was performed. We then monitored circulating tumor DNA (ctDNA) dynamics before and during pembrolizumab immunotherapy in one patient with MSI-high (MSI-H) metastatic BC. The assays showed high analytical specificity and sensitivity (limit of detection = 0.16%). Among N = 380 samples, seven (1.8%) were found as MSI-H by ddPCR with six of them confirmed by next-generation sequencing (NGS). Specificity was 100% in N = 133 microsatellite stable BC submitted to NGS. In the patient with MSI-H metastatic BC, ctDNA monitoring revealed an early decrease of microsatellite mutant allelic frequencies during immunotherapy. These results demonstrated MSI detection by ddPCR, a non-invasive, fast and cost-effective approach, allowing for large pre-screening of BC patients who may benefit from immunotherapy.

Homologous Recombination DNA-repair deficiency (HRD) is a well-recognized marker of platinum-salt and PARP inhibitor chemotherapies in ovarian and breast cancers (BC). Causing high genomic instability, HRD is currently determined by BRCA1/2 sequencing or by genomic signatures, but its morphological manifestation is not well understood. Deep Learning (DL) is a powerful machine learning technique that has been recently shown to be capable of predicting genomic signatures from stained tissue slides. However, DL is known to be sensitive to dataset biases and lacks interpretability. Here, we present and evaluate a strategy to control for biases in retrospective cohorts. We train a deep-learning model to predict the HRD in a controlled cohort with unprecedented accuracy (AUC: 0.86) and we develop a new visualization technique that allows for automatic extraction of new morphological features related to HRD. We analyze in detail the extracted morphological patterns that open new hypotheses on the phenotypic impact of HRD.

Uterine sarcomas represent a clinical challenge because of their difficult diagnosis and the poor prognosis of certain subtypes. The aim of this study was to evaluate the expression of the special AT-rich sequence-binding protein 2 (SATB2) in endometrial stromal sarcoma (ESS) and other types of uterine sarcoma by immunohistochemistry. We studied the expression of SATB2 on 71 full tissue sections of endometrial stromal nodule, low-grade ESS, uterine leiomyomas and leiomyosarcoma, undifferentiated uterine sarcoma, adenosarcoma, and carcinosarcoma samples. Nuclear SATB2 expression was then evaluated in an extended sample set using a tissue microarray, including 78 additional uterine tumor samples. Overall, with a cut-off of ≥10% of tumor cell staining as positive, the nuclear SATB2 score was negative in all endometrial stromal nodule samples (n=10) and positive in 83% of low-grade ESS samples (n=29/35), 40% of undifferentiated uterine sarcoma (n=4/10), 13% of leiomyosarcoma (n=2/16), 14% of adenosarcoma (n=3/22), and 8% carcinosarcoma (n=2/25) samples. Furthermore, in ESS patients, direct comparison of nuclear SATB2 scores with clinicopathologic parameters and other reported biomarkers such as progesterone receptor and estrogen receptor showed that nuclear SATB2 was associated with PR expression and a decreased risk of disease-specific death (odds ratio=0.06, 95% confidence interval=0.04-0.81, P=0.04). Our data suggest that SATB2 could be a marker with relative sensitivity (83%) for distinguishing between endometrial stromal nodule and ESS with potential prognostic value.

Background: The prognostic value of stromal tumor-infiltrating lymphocytes (TILs) as a biomarker for triple-negative breast cancer (TNBC) has been extensively demonstrated in patients (pts) receiving (neo)adjuvant systemic therapy. In addition, several small studies suggest that a subset of pts with early-stage TNBC and high TILs have excellent long-term outcomes, even in the absence of systemic therapy [1-3]. However, data on the absolute risk of TNBC recurrence according to TIL levels in the absence of systemic therapy are limited and critical to inform the design of future systemic therapy de-escalation clinical trials. Methods: We conducted an individual patient data pooled analysis of 12 international cohorts of pts with TNBC treated with locoregional therapy but no systemic therapy. TNBC was defined as tumors with estrogen and progesterone receptor of < 1% and HER2 negative (IHC 0, 1+ or IHC 2+ and FISH negative) per local evaluation. TILs were locally assessed in hematoxylin & eosin-stained slides according to the International Immuno-Oncology Biomarker Working Group guidelines (www.tilsinbreastcancer.org). We used the Kaplan-Meier method to assess survival outcomes according to prespecified TIL thresholds: 30% and 50%. Confidence intervals (CI) for survival probabilities were calculated using a percentile bootstrap method. The primary endpoint was invasive disease-free survival (iDFS, STEEP 2.0 definition). Key secondary outcomes included recurrence-free survival (RFS), distant disease-free survival (DDFS) and overall survival (OS). Results: 1,835 pts diagnosed with TNBC between 1982 and 2017 who did not receive systemic therapy were included. The median age at diagnosis was 56 (IQR 38-71). Menopausal status was known in 1,184 women, of whom 78% were post-menopausal. The median tumor size was 2.0 cm (IQR 1.2-2.6). Most pts (87%) had no axillary lymph node involvement (N0). Most tumors were invasive ductal carcinoma (74%) and grade 3 (70%). The median level of TILs was 15% (IQR 5-40). The median duration of follow-up was 30.4 years (95% CI 29.9, 31.1). A total of 950 (52%) iDFS, 828 (45%) RFS, 767 (42%) DDFS events, and 604 (33%) deaths were observed. In multivariable analyses, higher TILs were independently associated with improved iDFS, RFS, DDFS, and OS beyond clinicopathological factors (likelihood ratio p< 10e-6). Each 10% increment in stromal TILs was associated with an 8% (95% CI: 6-11), 10% (95% CI: 7-13), and 13% (95% CI: 10-15) reduction in the risk of experiencing an iDFS, RFS or DDFS event, and with a 12% (95% CI: 9-15) reduction in the risk of death. iDFS, RFS, DDFS and OS rates according to different TIL thresholds and nodal status are shown in the Table. Of note, the RFS estimates (which exclude second non-breast primaries and contralateral breast cancers) were consistently higher than the iDFS counterparts (which include both), consistent with a high rate of contralateral breast cancers and second primary tumors in this cohort. Notably, patients with node-negative—and especially stage I—TNBC with high TILs had excellent survival rates at 10-year follow-up. Conclusion: TILs are highly prognostic in pts with systemically untreated early-stage TNBC. Pts with pN0 (and especially stage I) TNBC with high TILs exhibited very favorable long-term outcomes even in the absence of systemic therapy. These data define the natural history of TIL-rich TNBC pts and are crucial to identifying the optimal patient population for future chemotherapy and immunotherapy de-escalation clinical trials. References: [1] Leon-Ferre et al, 2017, PMID: 28913760 [2] Park et al, 2019, PMID: 31566659 [3] de Jong et al, 2022, PMID: 35353548 Table 5 and 10-year survival endpoints according TIL level, nodal status, and stage Citation Format: Roberto A. Leon-Ferre, Sarah Flora Jonas, Roberto Salgado, Sherene Loi, Vincent De Jong, Jodi M. Carter, Torsten Nielson, Samuel Leung, Nazia Riaz, Giuseppe Curigliano, Carmen Criscitiello, Vincent Cockenpot, Matteo Lambertini, Vera Suman, Barbro Linderholm, John WM Martens, Carolien HM van Deurzen, Mieke Timmermans, Tatsunori Shimoi, Shu Yazaki, Masayuki Yoshida, Sung-Bae Kim, Hee Jin Lee, Maria Vittoria Dieci, Guillaume Bataillon, Anne Salomon, Fabrice Andre, Marleen Kok, Sabine Linn, Matthew P. Goetz, Stefan Michiels. Stromal tumor-infiltrating lymphocytes identify early-stage triple-negative breast cancer patients with favorable outcomes at 10-year follow-up in the absence of systemic therapy: a pooled analysis of 1835 patients [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr PD9-05.

Count data is becoming more and more ubiquitous in a wide range of applications, with datasets growing both in size and in dimension. In this context, an increasing amount of work is dedicated to the construction of statistical models directly accounting for the discrete nature of the data. Moreover, it has been shown that integrating dimension reduction to clustering can drastically improve performance and stability. In this paper, we rely on the mixture of multinomial PCA, a mixture model for the clustering of count data, also known as the probabilistic clustering-projection model in the literature. Related to the latent Dirichlet allocation model, it offers the flexibility of topic modeling while being able to assign each observation to a unique cluster. We introduce a greedy clustering algorithm, where inference and clustering are jointly done by mixing a classification variational expectation maximization algorithm, with a branch & bound like strategy on a variational lower bound. An integrated classification likelihood criterion is derived for model selection, and a thorough study with numerical experiments is proposed to assess both the performance and robustness of the method. Finally, we illustrate the qualitative interest of the latter in a real-world application, for the clustering of anatomopathological medical reports, in partnership with expert practitioners from the Institut Curie hospital., 34 pages, 11 figures, published in : Computational Statistics

Simple Summary Gynecologic carcinosarcomas are rare and aggressive diseases with a poor prognosis. The rarity of these tumors explains the lack of robust and specific data available in the literature. Using the data from the French National Rare Malignant Gynecological Tumors (TMRG) network, we conducted a multicentric cohort study to explore several burned questions. The main objective was to assess the outcome of patients with carcinosarcomas recorded in the network and to investigate the efficacy of initial adjuvant treatment and recurrent therapeutic strategies in a real-life setting. Four hundred and twenty-five patients were analyzed including 313 uterine and 112 ovarian carcinosarcomas. Our data suggest positive impact of adjuvant chemotherapy on survival in all stages (including FIGO IA uterine carcinosarcomas) and the importance of platinum-based combination for the treatment of relapse. In addition we report median PFS for various therapeutic strategies in the relapse setting. Abstract Background: Gynecological carcinosarcomas are rare and aggressive diseases, with a poor prognosis. The rarity of these tumors explains the lack of robust and specific data available in the literature. The objective of this study was to investigate the impact of initial adjuvant treatment and recurrent therapeutic strategies. Patients and methods: A multicentric cohort study within the French national prospective Rare Malignant Gynecological Tumors (TMRG) network was conducted. Data from all included carcinosarcomas diagnosed between 2011 and 2018 were retrospectively collected. Results: 425 cases of uterine and ovarian carcinosarcomas (n = 313 and n = 112, respectively) were collected and analyzed from 12 participating centers. At diagnosis, 140 patients (48%) had a FIGO stage III–IV uterine carcinosarcoma (UCS) and 88 patients (83%) had an advanced ovarian carcinosarcoma (OCS) (FIGO stage ≥ III). Two hundred sixty-seven patients (63%) received adjuvant chemotherapy, most preferably carboplatin-paclitaxel regimen (n = 227, 86%). After a median follow-up of 47.4 months, the median progression-free survival (mPFS) was 15.1 months (95% CI 12.3–20.6) and 14.8 months (95% CI 13.1–17.1) for OCS and UCS, respectively. The median overall survival for OCS and UCS was 37.1 months (95% CI 22.2–49.2) and 30.6 months (95% CI 24.1–40.9), respectively. With adjuvant chemotherapy followed by radiotherapy, mPFS was 41.0 months (95% CI 17.0–NR) and 18.9 months (95% CI 14.0–45.6) for UCS stages I–II and stages III–IV, respectively. In the early stage UCS subgroup (i.e., stage IA, n = 86, 30%), mPFS for patients treated with adjuvant chemotherapy (n = 24) was not reached (95% CI 22.2–NR), while mPFS for untreated patients (n = 62) was 19.9 months (95% IC 13.9–72.9) (HR 0.44 (0.20–0.95) p = 0.03). At the first relapse, median PFS for all patients was 4.2 months (95% CI 3.5–5.3). In the first relapse, mPFS was 6.7 months (95% CI 5.1–8.5) and 2.2 months (95% CI 1.9–2.9) with a combination of chemotherapy or monotherapy, respectively (p < 0.001). Conclusions: Interestingly, this vast prospective cohort of gynecological carcinosarcoma patients from the French national Rare Malignant Gynecological Tumors network (i) highlights the positive impact of adjuvant CT on survival in all localized stages (including FIGO IA uterine carcinosarcomas), (ii) confirms the importance of platinum-based combination as an option for relapse setting, and (iii) reports median PFS for various therapeutic strategies in the relapse setting.

The automatic analysis of stained histological sections is becoming increasingly popular. Deep Learning is today the method of choice for the computational analysis of such data, and has shown spectacular results for large datasets for a large variety of cancer types and prediction tasks. On the other hand, many scientific questions relate to small, highly specific cohorts. Such cohorts pose serious challenges for Deep Learning, typically trained on large datasets.In this article, we propose a modification of the standard nested cross-validation procedure for hyper-parameter tuning and model selection, dedicated to the analysis of small cohorts. We also propose a new architecture for the particularly challenging question of treatment prediction, and apply this workflow to the prediction of response to neoadjuvant chemotherapy for Triple Negative Breast Cancer.

Homologous recombination DNA-repair deficiency (HRD) is becoming a well-recognized marker of platinum salt and polyADP-ribose polymerase inhibitor chemotherapies in ovarian and breast cancers. While large-scale screening for HRD using genomic markers is logistically and economically challenging, stained tissue slides are routinely acquired in clinical practice. With the objectives of providing a robust deep-learning method for HRD prediction from tissue slides and identifying related morphological phenotypes, we first show that digital pathology workflows are sensitive to potential biases in the training set, then we propose a method to overcome the influence of these biases, and we develop an interpretation method capable of identifying complex phenotypes. Application to our carefully curated in-house dataset allows us to predict HRD with high accuracy (area under the receiver-operator characteristics curve 0.86) and to identify morphological phenotypes related to HRD. In particular, the presence of laminated fibrosis and clear tumor cells associated with HRD open new hypotheses regarding its phenotypic impact.

Purpose: Almost all cervical cancers are caused by human papillomavirus (HPV) and patients with advanced stage are at high risk for relapse. Circulating HPV DNA (HPV ctDNA) may serve as a residual tumor marker at the end of chemoradiation or to predict relapse during the follow-up period. Experimental Design: We analyzed serum samples from 94 HPV16- or HPV18-related CCs from the BioRAIDs prospective cohort. Samples were collected before and after treatment and during an 18-month follow-up period. Using digital droplet PCR (ddPCR), we assessed the relevance of circulating HPV E7 gene as a marker for residual disease compared to HPV integration site and PIK3CA mutations. Finally, the prognostic impact of circulating HPV E7 gene was assessed with its prediction value of relapse. Results: HPV E7 gene was the most sensitive tumor marker, superior to both HPV integration sites and PIK3CA mutations in serum. Circulating HPV DNA (HPV ctDNA) was detected in 63% (59/94) of patients, before treatment. HPV ctDNA detection in serum sample was associated with high FIGO stage (P = 0.02) and para-aortic lymph node involvement (P = 0.01). The level of HPV ctDNA was positively correlated with HPV copy number in the tumor (R = 0.39, P < 0.001). Complete clearance of HPV ctDNA by the end of treatment was significantly associated with a longer PFS (P < 0.0001). Patients with persistent HPV ctDNA in serum relapsed with a median time of 10 months (range, 2–15) from HPV ctDNA detection. Conclusions: HPV ctDNA detection is a useful marker to predict relapse in cervical cancer. See related commentary by Wentzensen and Clarke, p. 5733

Objective Tumors harboring a POLE pathogenic variant, associated with high tumor mutational burden, are good candidates for immunotherapy. However, POLE pathogenic variants are not currently screened in routine clinical practice. Can these tumors be identified by means of an already available test? Methods We describe seven tumors harboring a POLE pathogenic variant, among eight patients with tumors harboring multiple BRCA1/2 variants (from 4 to 20). All patients were managed at Institut Curie, Paris. Five patients were selected because of unexpected tumor BRCA testing results with multiple variants and another three patients were selected because of a POLE pathogenic variant detected by large tumor testing. We looked for other tumor variants by Next-Generation Sequencing in tumors harboring multiple BRCA1/2 variants, and for multiple BRCA1/2 variants in tumors harboring a POLE pathogenic variant. Results Four of the five tumors selected because of multiple BRCA1/2 variants exhibited a POLE pathogenic variant, and all three tumors selected for POLE pathogenic variants exhibited multiple BRCA1/2 variants. Conclusions Tumor BRCA testing could be a way to detect tumors harboring a highly mutagenic POLE pathogenic variant.

HER3, a member of the EGFR family of receptor tyrosine kinases coded by the ERBB3 gene, plays an important role in cancer, despite its lack of intrinsic kinase activity. As with genes coding for potential heterodimeric partners of HER3, EGFR, and HER2, oncogenic mutations of ERBB3 have been explored by several studies. In this review, we discuss the evidence presenting ERBB3 somatic mutations as potential tumoral drivers. We then show that ERBB3 mutations are not uncommon in many cancer types. Finally, we present the recent results of several studies evaluating different therapeutic approaches for treating patients with oncogenic ERBB3 mutations.

Background The role of chemotherapy in early luminal breast cancer remains challenged. The NEOPAL trial (NCT 02400567; Cottu et al, ESMO 2017 LBA09) compared sequential chemotherapy (CT) and letrozole-palbociclib (LP) as neoadjuvant treatment in PAM50 defined high-risk luminal breast cancer patients, showing that LP might be as efficient as CT with regard to breast conserving surgery and pathological response. We report here extended exploratory pathological results, focusing on tumor infiltrating lymphocytes (TILs), proliferative response and preoperative endocrine prognostic index (PEPI) scores. Material and Methods Tumor blocks from baseline biopsy and surgical specimens were available for centralized review from the 106 randomized patients (53 in each arm). TILs quantification, KI67 staining and counting, and ER quantification were performed according to standard methods. Residual proliferative cancer burden (RPCB) and PEPI scores were computed according to published algorithms. Wilcoxon rank sum test and Mann Whitney test were used to compare paired and unpaired data. The chi-square and Fisher exact tests were used for categorical variables. Results Overall, median TILs count did not differ between LP and CT patients, both at baseline (p=0.37) and at the end of treatment (p=0.42). Median TILs count climbed from 5% (0-60) to 10% (1-60) in the LP arm (p=0.0026) and from 2% (0-30) to 10% (0-60) in the CT arm (p=0.0023). Median Ki67 dropped sharply in both arms, from 30% (1-80) to 1% (0-30) in the LP arm (p=1.10e-8) and from 30% (2-80) to 5% (0-30) in the CT arm (p=3.10e-9). Decrease in the Ki67 geometric mean was as sharp. Of note, while baseline Ki67 was similar in both arms (p=0.315), decrease in the LP arm was significantly more profound than in the CT arm (p=0.00075). Pathological response according to RPCB were as follows, in the LP and CT arm, respectively: class 0: 9.6%/10.2%; class I: 84.6%/73.5%; class II: 5.8%/16.3%. The relapse free survival PEPI scores were as follow in the LP and CT arm, respectively: class I: 13.5%/16.3%; class II: 59.6%/46.9%; class III: 28.9%/36.8% (p=0.504). Breast cancer specific survival PEPI scores were as follow in the LP and CT arm, respectively: class I: 18.9%/8.2%; class II: 54.7%/40.8%; class III: 26.4%/51%. These results were significantly better in the LP arm (p=0.027). There was no correlation between final TILs quantification and the RPCB or PEPI scores. Conclusions In this prospective multicenter study with centralized pathological review, neoadjuvant letrozole-palbociclib combination generates impressive proliferative and endocrine specific response features. It compared well with chemotherapy. The LP combination also significantly increased lymphocytic infiltration. Its clinical significance and utility remain to be elucidated, but it potentially adds new prognostic and theranostic information. Citation Format: Vincent-Salomon A, Mathieu M-C, Bataillon G, Arnould L, Verrièle V, Ghnassia J-P, Haudebourg J, Penault-Llorca F, Lefebvre C, Maran-Gonzalez A, Guinebretière J-M, Duprez R, Berghian A, Blanc-Fournier C, Calès V, Galant C, Delrée P, Lemonnier J, Delaloge S, Cottu PH. TILs variations, proliferative response and PEPI scores in patients with luminal breast cancer receiving neoadjuvant letrozole-palbociclib or chemotherapy: An extended analysis of the NEOPAL trial [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 P4-15-02.

590 Background: Triple-Negative Breast Cancer (TNBC) is characterized by high metastatic potential and poor prognosis with limited treatment options. Neoadjuvant chemotherapy (NACT) is the standard of care in non-metastastic setting due to the ability to assess pathologic responses providing important prognostic information and guidance in adjuvant therapy decisions. However, the histological response heterogeneity is still poorly understood. We investigate the use of Machine Learning (ML) to predict from diagnosis Whole-Slide Images (WSI) of early TNBC the positive histological Complete Response (pCR) to NACT on surgical specimens. To overcome the known biases of small scale studies while respecting data privacy, we conduct a study in a multi-centric fashion behind hospitals’ firewalls using collaborative Federated Learning (FL). Thereby allowing access to enough TNBC data to sustain a complete response heterogeneity investigation. Methods: We collected in both comprehensive cancer centers: Centre Léon Bérard (A)(n=99) and Institut Curie (B) (n=420), WSI of biopsies performed at diagnosis and relevant clinical variables. We use traditional Multiple Instance Learning pipelines by tiling the matter on each WSI with a pre-trained Neural Network (NN). We train a second NN to predict the NACT pCR using the mean feature of each WSI. ML trainings are performed using either one cohort in isolation (NN Local) or both cohorts using FL. We compare the performance of this federated WSI based model to the best clinical model (Clin.) simulating clinical practice (using grade and Tumor-Infiltrating Lymphocytes (TILs) percentage) on both centers. Results: Performance of models to predict NACT pCR (AUC). All results are evaluated in 5 repeated 4-folds cross validations. Conclusions: The final ML model, that was trained in a privacy preserving fashion on both hospitals, provides better prediction of NACT pCR than current clinical standards. This study shows that 1. Not all relevant information is routinely extracted from WSI and 2. Non simulated FL is possible in Healthcare and gives better results than siloed studies on open medical questions. Additional interpretability results of the model show that it has re-discovered known biomarkers such as TILs and apocrine tumor cells without any tile-level annotation, and hints at potential new biomarkers. [Table: see text]

The french society of pathology (SFP) organized in 2020 its first data challenge with the help of Health Data Hub (HDH). The organisation of this event first consisted in recruiting almost 5000 slides of uterus cervical biopsies obtained in 20 pathology centers. After having made sure that patients did not refuse to include their slides in the project, the slides were anonymised, digitized and annotated by expert pathologists, and were finally uploaded on a data challenge platform for competitors all around the world. Competitors teams had to develop algorithms that could distinguish among four diagnostic classes in epithelial lesions of uterine cervix. Among many submissions by competitors, the best algorithms obtained an overall score close to 95%. The best 3 teams shared 25k€ prizes during a special session organised during the national congress of the SFP. The final part of the competition lasted only 6 weeks and the goal of SFP and HDH is now to allow for the collection to be published in open access. This final step will allow data scientists and pathologists to further develop artificial intelligence algorithms in this medical area.

Atypical lesions found on percutaneous breast biopsy raise specific management issues. The aim of this study was to validate the previous performance of a decision tree defined by Forgeard et al to select a subset of patients at low-risk of surgical diagnostic upgrade that would be eligible for surveillance.A consecutive series of 211 patients diagnosed with ADH on vacuum-assisted biopsy (VAB) of clustered microcalcifications alone, then operated in our institution, was reviewed. Histological findings on percutaneous cores were compared with definitive diagnoses on surgical specimens. The rate of cancer underestimation on VAB was analyzed in the four arms and two management attitudes defined in the scheme, using size and quality of microcalcification removal and the number of ADH foci.Ninety-eight women with ADH met the inclusion criteria. Overall, 20 cancers were diagnosed at surgery, showing a malignancy rate of 44% (17/39 patients) in the surgery group and of 5% (3/59 patients) in the surveillance group, which was not significantly different from the 2% rate in the monitored reference group (p 0.64). The malignancy rate increased significantly with the size of clustered microcalcifications (0% when 6mm, 17% when between 6mm and 21 mm, 48% when21 mm, p 0001) and the number of ADH foci on VAB (14% when ≤ 2, 45% when 2, p 0.005).Our results corroborate - within the limits of large confidence intervals - those obtained with the reference decision tree. Due to statistical uncertainty, however, they need to be prospectively validated in a broader series.

Simple Summary Immune checkpoint inhibitors (ICI) are now part of the therapeutical arsenal for cancers at several sites and in several settings. PD-L1 expression is assessed to predict treatment response. We used immunohistochemistry (E1L3N clone) to assess PD-L1 expression on tumor and immune cells from a cohort of 89 surgical specimens of T1-T3NxM0 triple-negative breast cancers (TNBC) from patients treated with neoadjuvant chemotherapy (NAC) with residual disease. PD-L1 expression levels were low in tumor and immune cells from post-NAC surgical specimens. PD-L1 positivity in tumor cells was significantly associated with aggressive post-NAC tumor characteristics. A small subset of TNBC patients displaying PD-L1 expression in the context of a more extensive post-NAC tumor burden could benefit from ICI treatment after NAC. Abstract The consequences of neoadjuvant chemotherapy (NAC) for PD-L1 activity in triple-negative breast cancers (TNBC) are not well-understood. This is an important issue as PD-LI might act as a biomarker for immune checkpoint inhibitors’ (ICI) efficacy, at a time where ICI are undergoing rapid development and could be beneficial in patients who do not achieve a pathological complete response. We used immunohistochemistry to assess PD-L1 expression in surgical specimens (E1L3N clone, cutoff for positivity: ≥1%) on both tumor (PD-L1-TC) and immune cells (PD-L1-IC) from a cohort of T1-T3NxM0 TNBCs treated with NAC. PD-L1-TC was detected in 17 cases (19.1%) and PD-L1-IC in 14 cases (15.7%). None of the baseline characteristics of the tumor or the patient were associated with PD-L1 positivity, except for pre-NAC stromal TIL levels, which were higher in post-NAC PD-L1-TC-positive than in negative tumors. PD-L1-TC were significantly associated with a higher residual cancer burden (p = 0.035) and aggressive post-NAC tumor characteristics, whereas PD-L1-IC were not. PD-L1 expression was not associated with relapse-free survival (RFS) (PD-L1-TC, p = 0.25, and PD-L1-IC, p = 0.95) or overall survival (OS) (PD-L1-TC, p = 0.48, and PD-L1-IC, p = 0.58), but high Ki67 levels after NAC were strongly associated with a poor prognosis (RFS, p = 0.0014, and OS, p = 0.001). A small subset of TNBC patients displaying PD-L1 expression in the context of an extensive post-NAC tumor burden could benefit from ICI treatment after standard NAC.