Hematologist-Level Classification of Mature B-Cell Neoplasm Using Deep Learning on Multiparameter Flow Cytometry Data.

The wealth of information captured by multiparameter flow cytometry (MFC) can be analyzed by recent methods of computer vision when represented as a single image file. We therefore transformed MFC raw data into a multicolor 2D image by a self-organizing map and classified this representation using a convolutional neural network. By this means, we built an artificial intelligence that is not only able to distinguish diseased from healthy samples, but it can also differentiate seven subtypes of mature B-cell neoplasm. We trained our model with 18,274 cases including chronic lymphocytic leukemia and its precursor monoclonal B-cell lymphocytosis, marginal zone lymphoma, mantle cell lymphoma, prolymphocytic leukemia, follicular lymphoma, hairy cell leukemia, lymphoplasmacytic lymphoma and achieved a weighted F1 score of 0.94 on a separate test set of 2,348 cases. Furthermore, we estimated the trustworthiness of a classification and could classify 70% of all cases with a confidence of 0.95 and higher. Our performance analyses indicate that particularly for rare subtypes further improvement can be expected when even more samples are available for training. © 2020 The Authors. Cytometry Part A published by Wiley Periodicals LLC. on behalf of International Society for Advancement of Cytometry.

Inhibition of PLK1 by capped-dose volasertib exerts substantial efficacy in MDS and sAML while sparing healthy haematopoiesis.

INTRODUCTION: Targeting the cell cycle machinery represents a rational therapeutic approach in myelodysplastic syndromes (MDS) and secondary acute myeloid leukemia (sAML). Despite substantial response rates, clinical use of the PLK inhibitor volasertib has been hampered by elevated side effects such as neutropenia and infections. OBJECTIVES: The primary objective was to analyse whether a reduced dose of volasertib was able to limit toxic effects on the healthy haematopoiesis while retaining its therapeutic effect. METHODS: Bone marrow mononuclear cells (BMMNCs) of patients with MDS/sAML (n = 73) and healthy controls (n = 28) were treated with volasertib (1 µM to 1 nM) or vehicle control. Short-term viability analysis was performed by flow cytometry after 72 hours. For long-term viability analysis, colony-forming capacity was assessed after 14 days. Protein expression of RIPK3 and MCL-1 was quantified via flow cytometry. RESULTS: Reduced dose levels of volasertib retained high cell death-inducing efficacy in primary human stem and progenitor cells of MDS/sAML patients without affecting healthy haematopoiesis in vitro. Interestingly, volasertib reduced colony-forming capacity and cell survival independent of clinical stage or mutational status. CONCLUSIONS: Volasertib offers a promising therapeutic approach in patients with adverse prognostic profile. RIPK3 and MCL-1 might be potential biomarkers for sensitivity to volasertib treatment.

Enfortumab Vedotin in Metastatic Urothelial Carcinoma: Survival and Safety in a European Multicenter Real-world Patient Cohort.

Background: Treatment options for patients with urothelial cancer (UC) refractory to platinum and immunotherapy are limited and survival is short. Enfortumab vedotin (EV) is a monoclonal anti-NECTIN4 antibody conjugated to monomethyl auristatin. It was recently approved because of superior survival in comparison to standard-of-care (SOC) chemotherapy. Real-world patients, however, often have worse characteristics than patients included in clinical trials. Objective: To analyze the efficacy and safety of EV in a cohort of real-world patients. Design setting and participants: Retrospective data were collected from 23 hospitals and private practices for patients with metastatic and previously treated UC who received EV either when reimbursed by their insurance company before European Medicines Agency (EMA) approval, within a compassionate use program, or as SOC treatment after EMA approval. Imaging and therapy management were in accordance with local standards. Outcome measurements and statistical analysis: Adverse events (AEs) were reported according to Common Terminology Criteria for Adverse Events (CTCAE) version 5.0 criteria. Objective responses were evaluated according to Response Evaluation Criteria in Solid Tumors version 1.1. Progression-free survival (PFS) and overall survival (OS) were estimated using the Kaplan-Meier method. Results and limitations: The median age for the 125 eligible patients was 66 yr (range 31-89). The Eastern Cooperative Oncology Group performance status (ECOG PS) was 0-1 for 76.0%, 2-4 for 13.6%, and unknown for 10.4% of patients. EV was administered in the fourth or later line for 44.8% of patients. The overall response rate was 41.6% (partial response 39.2%, complete response 2.4%). Median OS was 10.0 months (mo) (95% confidence interval 7.20-12.80) and median PFS was 5.0 mo (95% confidence interval 4.34-5.67). For patients with ECOG PS of 0-1, median OS was 14 mo. Any-grade AEs were observed in 67.2% and CTCAE grade ≥3 AEs in 30.4%. The most common AEs were peripheral sensory neuropathy and skin toxicity. Three fatal events (pneumonia, pneumonitis) occurred. Limitations include the retrospective design and short follow-up. Conclusions: Administration of EV for real-world patients was feasible with an acceptable toxicity profile. No new safety signals were reported. Antitumor activity in our cohort was comparable to data previously reported for trials. In summary, our results support the use of EV in patients with metastatic UC. Patient summary: Enfortumab vedotin is a medication that improved the survival of patients with bladder cancer in comparison to standard chemotherapy in clinical trials. However, patients included in clinical trials are highly selected and results for toxicities and improvements in survival do not always transfer to the real-world setting. We analyzed data for 125 patients who were treated with enfortumab vedotin. Our results are comparable to the outcomes from clinical trials regarding the safety and efficacy of this treatment.

Indeterminate and oncogenic potential: CHIP vs CHOP mutations in AML with NPM1 alteration.

In AML patients, recurrent mutations were shown to persist in remission, however, only some have a prognostic value and persistent mutations might therefore reflect a re-established premalignant state or truly active disease causing relapse. We aimed to dissect the nature of co-mutations in NPM1 mutated AML where the detection of NPM1 transcripts allows highly specific and sensitive detection of complete molecular remission (CMR). We analysed 150 consecutive patients who achieved CMR following intensive treatment by next generation sequencing on paired samples at diagnosis, CMR and relapse (38/150 patients). Patients with persistence or the acquisition of non-DTA (DNMT3A, TET2, ASXL1) mutations at CMR (23/150 patients, 15%) have a significantly worse prognosis (EFS HR = 2.7, p = 0.003; OS HR = 3.6, p = 0.012). Based on clonal evolution analysis of diagnostic, CMR and relapse samples, we redefine pre-malignant mutations and include IDH1, IDH2 and SRSF2 with the DTA genes in this newly defined group. Only the persistence or acquisition of CHOP-like (clonal hematopoiesis of oncogenic potential) mutations was significantly associated with an inferior outcome (EFS HR = 4.5, p = 0.0002; OS HR = 5.5, p = 0.002). Moreover, the detection of CHOP-like mutations at relapse was detrimental (HR = 4.5, p = 0.01). We confirmed these findings in a second independent whole genome sequencing cohort.

Knowledge transfer to enhance the performance of deep learning models for automated classification of B cell neoplasms.

Multi-parameter flow cytometry (MFC) is a cornerstone in clinical decision making for leukemia and lymphoma. MFC data analysis requires manual gating of cell populations, which is time-consuming, subjective, and often limited to a two-dimensional space. In recent years, deep learning models have been successfully used to analyze data in high-dimensional space and are highly accurate. However, AI models used for disease classification with MFC data are limited to the panel they were trained on. Thus, a key challenge in deploying AI into routine diagnostics is the robustness and adaptability of such models. This study demonstrates how transfer learning can be applied to boost the performance of models with smaller datasets acquired with different MFC panels. We trained models for four additional datasets by transferring the features learned from our base model. Our workflow increased the model's overall performance and, more prominently, improved the learning rate for small training sizes.

Molecular characterization of a second myeloid neoplasm developing after treatment for acute myeloid leukemia.

Therapy-related myeloid neoplasms (tMN) following successful treatment of acute myeloid leukemia (AML) are rare and poorly characterized. To evaluate the presence of a common ancestral clone, we performed whole-exome sequencing of 25 patients at AML diagnosis, tMN diagnosis (tMDS: 13; tAML: 12), and matched remission samples, identifying 607 mutations affecting 504 different genes (46 recurrently mutated). Number of mutations was higher in tAML vs. tMDS cases (median 19 vs 13 mutations, p = 0.05). Focusing on 24 genes commonly mutated in hematological malignancies, 19/25 (76%) patients were found to share mutations between AML and tMN, mostly affecting epigenetic modifiers (21/32; 66%), splicing factors (6/32; 19%), and chromatin modifiers (3/32; 9%). Analysis of remission samples identified 13 persisting mutations in 10/22 patients, affecting DNMT3A (n = 6), TET2 (n = 5), IDH1 and SRSF2 (n = 1, each). Comparison of cytogenetics revealed that 9/12 patients with a normal karyotype (NK) in AML harbored aberrations in tMN, four aberrant AML cases presented with NK in tMN, four other patients showed unrelated cytogenetic aberrations. Our study provides novel insights into the pathogenesis of tMN, hypothesizing the presence of a common ancestral clone in AML and tMN. Mutations mostly affected epigenetic modifiers, which have previously been linked to clonal hematopoiesis.

The combination of WGS and RNA-Seq is superior to conventional diagnostic tests in multiple myeloma: Ready for prime time?

The diagnosis and risk stratification of multiple myeloma (MM) is based on clinical and cytogenetic tests. Magnetic CD138 enrichment followed by interphase FISH (fluorescence in situ hybridisation) is the gold standard to identify prognostic translocations and copy number alterations (CNA). Although clinical implications of gene expression profiling (GEP) or panel based sequencing results are evident, those tests have not yet reached routine clinical application. We set up a single workflow to analyse MM of 211 patients at first diagnosis by whole genome sequencing (WGS) and RNA-Seq and validate the results by FISH analysis. We observed a 96% concordance of FISH and WGS results when assessing translocations involving the IGH locus and an overall concordance of FISH and WGS of 92% when assessing CNA. WGS analysis resulted in the identification of 17 additional MYC-translocations that were missed by FISH analysis. RNA-Seq followed by supervised clustering grouped patients in their expected genetically defined subgroup and prompted the assessment of WGS data in cases that were not congruent with FISH. This allowed the identification of additional IGH-translocations and hyperdiploid cases. We show the reliability of WGS an RNA-Seq in a clinical setting, which is a prerequisite for a novel routine diagnostic test.

Molecular characterization of AML with RUNX1-RUNX1T1 at diagnosis and relapse reveals net loss of co-mutations.

AML with RUNX1-RUNX1T1 fusion is a WHO entity with a favorable outcome following intensive chemotherapy. The absence of RUNX1-RUNX1T1 transcripts in remission defines complete molecular response and correlates with a superior survival. However, a significant proportion of patients still relapses and defining molecular risk factors that identify patients at diagnosis or at molecular remission that are at risk of relapse could help tailor treatment strategies for those high risk patients. Here, we analyze a cohort of 94 patients that reach a molecular remission (MR) following intensive treatment and identify 21 patients that relapse despite achieving MR. Using targeted sequencing of 63 genes implicated in hematologic malignancies we show that at diagnosis patients who relapse following MR have a higher burden of co-mutated genes than patients that do not relapse (median = 2 vs median = 0; P = 0.0156). This resulted in a relapse free survival rate of 65% vs 86% at 2 years, respectively (≥1 co-mutation vs no co-mutation, P = 0.02) with a trend for inferior overall survival (n.s.). Applying sensitive sequencing to reassess mutations at relapse in paired samples of 17/21 patients we demonstrate a net loss of co-mutations at relapse: median 2 (range 0-5) vs 1 (0-4) at diagnosis and relapse (P = 0.048). At relapse more patients had no detected co-mutation compared to diagnosis (47% vs 17%, P = 0.034). Co-mutations at diagnosis, therefore, might represent a general susceptibility of the AML clone to acquire mutations and the true nature of 2nd hit mutations that drive leukemia has to be defined for AML with RUNX1-RUNX1T1 fusion.

Complement and inflammasome overactivation mediates paroxysmal nocturnal hemoglobinuria with autoinflammation.

Patients with paroxysmal nocturnal hemoglobinuria (PNH) have a clonal population of blood cells deficient in glycosylphosphatidylinositol-anchored (GPI-anchored) proteins, resulting from a mutation in the X-linked gene PIGA. Here we report on a set of patients in whom PNH results instead from biallelic mutation of PIGT on chromosome 20. These PIGT-PNH patients have clinically typical PNH, but they have in addition prominent autoinflammatory features, including recurrent attacks of aseptic meningitis. In all these patients we find a germ-line point mutation in one PIGT allele, whereas the other PIGT allele is removed by somatic deletion of a 20q region comprising maternally imprinted genes implicated in myeloproliferative syndromes. Unlike in PIGA-PNH cells, GPI is synthesized in PIGT-PNH cells and, since its attachment to proteins is blocked, free GPI is expressed on the cell surface. From studies of patients' leukocytes and of PIGT-KO THP-1 cells we show that, through increased IL-1ß secretion, activation of the lectin pathway of complement and generation of C5b-9 complexes, free GPI is the agent of autoinflammation. Eculizumab treatment abrogates not only intravascular hemolysis, but also autoinflammation. Thus, PIGT-PNH differs from PIGA-PNH both in the mechanism of clonal expansion and in clinical manifestations.

NPM1 mutated AML can relapse with wild-type NPM1: persistent clonal hematopoiesis can drive relapse.

Acute myeloid leukemia (AML) with NPM1 mutation (NPM1 mut) defines a World Health Organization entity. Absence of minimal residual disease (MRD) following induction chemotherapy is associated with an excellent prognosis. Data are conflicting on NPM1 mut AML relapsing with wild-type NPM1 (NPM1 wt ). We analyzed 104 paired samples of NPM1 mut AML patients with relapse and identified 14/104 that relapsed with NPM1 wt AML. Blood counts at diagnosis differed significantly between patients with NPM1 mut and NPM1 wt relapse (median white blood cell count, 30 vs 3 × 109/L, P = .008; platelet count, 66 vs 128 × 109/l, P = .018). NPM1 mut relapse occurred significantly earlier than NPM1 wt relapse (14 vs 43 months, P = .004). At diagnosis, FLT3-ITD were more frequent in patients with NPM1 mut relapse (P = .029), whereas DNMT3A mutations were more frequent in patients with NPM1 wt relapse (P = .035). Sequencing analysis of paired samples at diagnosis, molecular remission, and NPM1 wt relapse identified cooccurring mutations that persist from diagnosis throughout remission and at relapse, suggestive of a preexisting clonal hematopoiesis. We provide evidence that AML relapsing with NPM1 wt is a distinct disease and that initial leukemia and relapse potentially arise from a premalignant clonal hematopoiesis.

CDK6 Antagonizes p53-Induced Responses during Tumorigenesis.

Tumor formation is a multistep process during which cells acquire genetic and epigenetic changes until they reach a fully transformed state. We show that CDK6 contributes to tumor formation by regulating transcriptional responses in a stage-specific manner. In early stages, the CDK6 kinase induces a complex transcriptional program to block p53 in hematopoietic cells. Cells lacking CDK6 kinase function are required to mutate TP53 (encoding p53) to achieve a fully transformed immortalized state. CDK6 binds to the promoters of genes including the p53 antagonists Prmt5, Ppm1d, and Mdm4 The findings are relevant to human patients: Tumors with low levels of CDK6 have mutations in TP53 significantly more often than expected.Significance: CDK6 acts at the interface of p53 and RB by driving cell-cycle progression and antagonizing stress responses. While sensitizing cells to p53-induced cell death, specific inhibition of CDK6 kinase activity may provoke the outgrowth of p53-mutant clones from premalignant cells. Cancer Discov; 8(7); 884-97. ©2018 AACR.This article is highlighted in the In This Issue feature, p. 781.