From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment.

Machine learning (ML) is increasingly used in clinical oncology to diagnose cancers, predict patient outcomes, and inform treatment planning. Here, we review recent applications of ML across the clinical oncology workflow. We review how these techniques are applied to medical imaging and to molecular data obtained from liquid and solid tumor biopsies for cancer diagnosis, prognosis, and treatment design. We discuss key considerations in developing ML for the distinct challenges posed by imaging and molecular data. Finally, we examine ML models approved for cancer-related patient usage by regulatory agencies and discuss approaches to improve the clinical usefulness of ML.

Atlas of clinically distinct cell states and ecosystems across human solid tumors.

Determining how cells vary with their local signaling environment and organize into distinct cellular communities is critical for understanding processes as diverse as development, aging, and cancer. Here we introduce EcoTyper, a machine learning framework for large-scale identification and validation of cell states and multicellular communities from bulk, single-cell, and spatially resolved gene expression data. When applied to 12 major cell lineages across 16 types of human carcinoma, EcoTyper identified 69 transcriptionally defined cell states. Most states were specific to neoplastic tissue, ubiquitous across tumor types, and significantly prognostic. By analyzing cell-state co-occurrence patterns, we discovered ten clinically distinct multicellular communities with unexpectedly strong conservation, including three with myeloid and stromal elements linked to adverse survival, one enriched in normal tissue, and two associated with early cancer development. This study elucidates fundamental units of cellular organization in human carcinoma and provides a framework for large-scale profiling of cellular ecosystems in any tissue.

Noninvasive Early Identification of Therapeutic Benefit from Immune Checkpoint Inhibition.

Although treatment of non-small cell lung cancer (NSCLC) with immune checkpoint inhibitors (ICIs) can produce remarkably durable responses, most patients develop early disease progression. Furthermore, initial response assessment by conventional imaging is often unable to identify which patients will achieve durable clinical benefit (DCB). Here, we demonstrate that pre-treatment circulating tumor DNA (ctDNA) and peripheral CD8 T cell levels are independently associated with DCB. We further show that ctDNA dynamics after a single infusion can aid in identification of patients who will achieve DCB. Integrating these determinants, we developed and validated an entirely noninvasive multiparameter assay (DIREct-On, Durable Immunotherapy Response Estimation by immune profiling and ctDNA-On-treatment) that robustly predicts which patients will achieve DCB with higher accuracy than any individual feature. Taken together, these results demonstrate that integrated ctDNA and circulating immune cell profiling can provide accurate, noninvasive, and early forecasting of ultimate outcomes for NSCLC patients receiving ICIs.

Dynamic Risk Profiling Using Serial Tumor Biomarkers for Personalized Outcome Prediction.

Accurate prediction of long-term outcomes remains a challenge in the care of cancer patients. Due to the difficulty of serial tumor sampling, previous prediction tools have focused on pretreatment factors. However, emerging non-invasive diagnostics have increased opportunities for serial tumor assessments. We describe the Continuous Individualized Risk Index (CIRI), a method to dynamically determine outcome probabilities for individual patients utilizing risk predictors acquired over time. Similar to "win probability" models in other fields, CIRI provides a real-time probability by integrating risk assessments throughout a patient's course. Applying CIRI to patients with diffuse large B cell lymphoma, we demonstrate improved outcome prediction compared to conventional risk models. We demonstrate CIRI's broader utility in analogous models of chronic lymphocytic leukemia and breast adenocarcinoma and perform a proof-of-concept analysis demonstrating how CIRI could be used to develop predictive biomarkers for therapy selection. We envision that dynamic risk assessment will facilitate personalized medicine and enable innovative therapeutic paradigms.

Distinct Hodgkin lymphoma subtypes defined by noninvasive genomic profiling.

The scarcity of malignant Hodgkin and Reed-Sternberg cells hampers tissue-based comprehensive genomic profiling of classic Hodgkin lymphoma (cHL). By contrast, liquid biopsies show promise for molecular profiling of cHL due to relatively high circulating tumour DNA (ctDNA) levels1-4. Here we show that the plasma representation of mutations exceeds the bulk tumour representation in most cases, making cHL particularly amenable to noninvasive profiling. Leveraging single-cell transcriptional profiles of cHL tumours, we demonstrate Hodgkin and Reed-Sternberg ctDNA shedding to be shaped by DNASE1L3, whose increased tumour microenvironment-derived expression drives high ctDNA concentrations. Using this insight, we comprehensively profile 366 patients, revealing two distinct cHL genomic subtypes with characteristic clinical and prognostic correlates, as well as distinct transcriptional and immunological profiles. Furthermore, we identify a novel class of truncating IL4R mutations that are dependent on IL-13 signalling and therapeutically targetable with IL-4Rα-blocking antibodies. Finally, using PhasED-seq5, we demonstrate the clinical value of pretreatment and on-treatment ctDNA levels for longitudinally refining cHL risk prediction and for detection of radiographically occult minimal residual disease. Collectively, these results support the utility of noninvasive strategies for genotyping and dynamic monitoring of cHL, as well as capturing molecularly distinct subtypes with diagnostic, prognostic and therapeutic potential.

Risk of Second Tumors and T-Cell Lymphoma after CAR T-Cell Therapy.

BACKGROUND: The risk of second tumors after chimeric antigen receptor (CAR) T-cell therapy, especially the risk of T-cell neoplasms related to viral vector integration, is an emerging concern. METHODS: We reviewed our clinical experience with adoptive cellular CAR T-cell therapy at our institution since 2016 and ascertained the occurrence of second tumors. In one case of secondary T-cell lymphoma, a broad array of molecular, genetic, and cellular techniques were used to interrogate the tumor, the CAR T cells, and the normal hematopoietic cells in the patient. RESULTS: A total of 724 patients who had received T-cell therapies at our center were included in the study. A lethal T-cell lymphoma was identified in a patient who had received axicabtagene ciloleucel therapy for diffuse large B-cell lymphoma, and both lymphomas were deeply profiled. Each lymphoma had molecularly distinct immunophenotypes and genomic profiles, but both were positive for Epstein-Barr virus and were associated with DNMT3A and TET2 mutant clonal hematopoiesis. No evidence of oncogenic retroviral integration was found with the use of multiple techniques. CONCLUSIONS: Our results highlight the rarity of second tumors and provide a framework for defining clonal relationships and viral vector monitoring. (Funded by the National Cancer Institute and others.).

Integrating genomic features for non-invasive early lung cancer detection.

Radiologic screening of high-risk adults reduces lung-cancer-related mortality1,2; however, a small minority of eligible individuals undergo such screening in the United States3,4. The availability of blood-based tests could increase screening uptake. Here we introduce improvements to cancer personalized profiling by deep sequencing (CAPP-Seq)5, a method for the analysis of circulating tumour DNA (ctDNA), to better facilitate screening applications. We show that, although levels are very low in early-stage lung cancers, ctDNA is present prior to treatment in most patients and its presence is strongly prognostic. We also find that the majority of somatic mutations in the cell-free DNA (cfDNA) of patients with lung cancer and of risk-matched controls reflect clonal haematopoiesis and are non-recurrent. Compared with tumour-derived mutations, clonal haematopoiesis mutations occur on longer cfDNA fragments and lack mutational signatures that are associated with tobacco smoking. Integrating these findings with other molecular features, we develop and prospectively validate a machine-learning method termed 'lung cancer likelihood in plasma' (Lung-CLiP), which can robustly discriminate early-stage lung cancer patients from risk-matched controls. This approach achieves performance similar to that of tumour-informed ctDNA detection and enables tuning of assay specificity in order to facilitate distinct clinical applications. Our findings establish the potential of cfDNA for lung cancer screening and highlight the importance of risk-matching cases and controls in cfDNA-based screening studies.

Concurrent pembrolizumab with AVD for untreated classic Hodgkin lymphoma.

Concurrent administration of pembrolizumab with chemotherapy in untreated classic Hodgkin lymphoma (CHL) has not been studied previously. To investigate this combination, we conducted a single-arm study of concurrent pembrolizumab with AVD (doxorubicin, vinblastine, and dacarbazine; APVD) for untreated CHL. We enrolled 30 patients and met the primary safety end point with no observed significant treatment delays in the first 2 cycles. Twelve patients experienced grade 3 or 4 nonhematologic adverse events (AEs), most commonly febrile neutropenia and infection/sepsis. Grade 3 or 4 immune-related AEs, including alanine aminotransferase elevation and aspartate aminotransferase elevation were observed in 3 patients. One patient experienced an episode of grade 2 colitis and arthritis. Six patients missed at least 1 dose of pembrolizumab because of AEs, primarily grade 2 or higher transaminitis. Among 29 response-evaluable patients, the best overall response rate was 100% and the complete response rate was 90%. With a median follow-up of 2.1 years, the 2-year progression-free survival (PFS) and overall survival were 97% and 100%, respectively. To date, no patient who has withheld or discontinued pembrolizumab because of toxicity has progressed. Clearance of circulating tumor DNA (ctDNA) was associated with superior PFS when measured after cycle 2 and at the end of treatment (EOT). None of the 4 patients with persistent uptake by fluorodeoxyglucose positron emission tomography (PET) at EOT yet negative ctDNA have relapsed to date. Concurrent APVD shows promising safety and efficacy but may yield spurious PET findings in some patients. This trial was registered at www.clinicaltrials.gov as #NCT03331341.

Anti-CD47 antibody synergizes with rituximab to promote phagocytosis and eradicate non-Hodgkin lymphoma.

Monoclonal antibodies are standard therapeutics for several cancers including the anti-CD20 antibody rituximab for B cell non-Hodgkin lymphoma (NHL). Rituximab and other antibodies are not curative and must be combined with cytotoxic chemotherapy for clinical benefit. Here we report the eradication of human NHL solely with a monoclonal antibody therapy combining rituximab with a blocking anti-CD47 antibody. We identified increased expression of CD47 on human NHL cells and determined that higher CD47 expression independently predicted adverse clinical outcomes in multiple NHL subtypes. Blocking anti-CD47 antibodies preferentially enabled phagocytosis of NHL cells and synergized with rituximab. Treatment of human NHL-engrafted mice with anti-CD47 antibody reduced lymphoma burden and improved survival, while combination treatment with rituximab led to elimination of lymphoma and cure. These antibodies synergized through a mechanism combining Fc receptor (FcR)-dependent and FcR-independent stimulation of phagocytosis that might be applicable to many other cancers.

Genomic profiling for clinical decision making in lymphoid neoplasms.

With the introduction of large-scale molecular profiling methods and high-throughput sequencing technologies, the genomic features of most lymphoid neoplasms have been characterized at an unprecedented scale. Although the principles for the classification and diagnosis of these disorders, founded on a multidimensional definition of disease entities, have been consolidated over the past 25 years, novel genomic data have markedly enhanced our understanding of lymphomagenesis and enriched the description of disease entities at the molecular level. Yet, the current diagnosis of lymphoid tumors is largely based on morphological assessment and immunophenotyping, with only few entities being defined by genomic criteria. This paper, which accompanies the International Consensus Classification of mature lymphoid neoplasms, will address how established assays and newly developed technologies for molecular testing already complement clinical diagnoses and provide a novel lens on disease classification. More specifically, their contributions to diagnosis refinement, risk stratification, and therapy prediction will be considered for the main categories of lymphoid neoplasms. The potential of whole-genome sequencing, circulating tumor DNA analyses, single-cell analyses, and epigenetic profiling will be discussed because these will likely become important future tools for implementing precision medicine approaches in clinical decision making for patients with lymphoid malignancies.

CD47 is an adverse prognostic factor and therapeutic antibody target on human acute myeloid leukemia stem cells.

Acute myeloid leukemia (AML) is organized as a cellular hierarchy initiated and maintained by a subset of self-renewing leukemia stem cells (LSC). We hypothesized that increased CD47 expression on human AML LSC contributes to pathogenesis by inhibiting their phagocytosis through the interaction of CD47 with an inhibitory receptor on phagocytes. We found that CD47 was more highly expressed on AML LSC than their normal counterparts, and that increased CD47 expression predicted worse overall survival in three independent cohorts of adult AML patients. Furthermore, blocking monoclonal antibodies directed against CD47 preferentially enabled phagocytosis of AML LSC and inhibited their engraftment in vivo. Finally, treatment of human AML LSC-engrafted mice with anti-CD47 antibody depleted AML and targeted AML LSC. In summary, increased CD47 expression is an independent, poor prognostic factor that can be targeted on human AML stem cells with blocking monoclonal antibodies capable of enabling phagocytosis of LSC.

Outcome prediction by interim positron emission tomography and IgM monoclonal gammopathy in diffuse large B-cell lymphoma.

In diffuse large B-cell lymphoma (DLBCL), a positive interim positron emission tomography (PET) scan predicts treatment failure, but the proportion of high-risk patients thus identified is small. To improve prediction, we combined the interim PET result with the presence or absence of an associated IgM gammopathy. Of 108 DLBCL patients participating in a prospective trial, nine (8%) were interim PET positive and 19 (18%) had an IgM gammopathy. The monoclonal protein was not associated with distinguishing genetic features, and its light chain restriction was not always concordant with the light chain restriction of the lymphoma. The information provided by interim PET and IgM gammopathy was combined to dichotomize the population into sizeable high-risk (1-2 adverse factors) and low-risk groups (no adverse factor) with widely different outcomes (population size, 25% vs. 75%; 3-year risk of progression, 51% vs. 10%; 3-year overall survival, 64% vs. 95%). Multivariable analyses including established risk factors revealed the interim PET result and the IgM gammopathy status to be the only factors significantly associated with outcome. Information about interim PET response and IgM gammopathy may be useful in studies testing risk-adapted treatment strategies.

Antigen presentation profiling reveals recognition of lymphoma immunoglobulin neoantigens.

Cancer somatic mutations can generate neoantigens that distinguish malignant from normal cells. However, the personalized identification and validation of neoantigens remains a major challenge. Here we discover neoantigens in human mantle-cell lymphomas by using an integrated genomic and proteomic strategy that interrogates tumour antigen peptides presented by major histocompatibility complex (MHC) class I and class II molecules. We applied this approach to systematically characterize MHC ligands from 17 patients. Remarkably, all discovered neoantigenic peptides were exclusively derived from the lymphoma immunoglobulin heavy- or light-chain variable regions. Although we identified MHC presentation of private polymorphic germline alleles, no mutated peptides were recovered from non-immunoglobulin somatically mutated genes. Somatic mutations within the immunoglobulin variable region were almost exclusively presented by MHC class II. We isolated circulating CD4+ T cells specific for immunoglobulin-derived neoantigens and found these cells could mediate killing of autologous lymphoma cells. These results demonstrate that an integrative approach combining MHC isolation, peptide identification, and exome sequencing is an effective platform to uncover tumour neoantigens. Application of this strategy to human lymphoma implicates immunoglobulin neoantigens as targets for lymphoma immunotherapy.

Circulating Tumor DNA Analysis for Detection of Minimal Residual Disease After Chemoradiotherapy for Localized Esophageal Cancer.

BACKGROUND & AIMS: Biomarkers are needed to risk stratify after chemoradiotherapy for localized esophageal cancer. These could improve identification of patients at risk for cancer progression and selection of additional therapy. METHODS: We performed deep sequencing (CAncer Personalized Profiling by deep Sequencing, [CAPP-Seq]) analyses of plasma cell-free DNA collected from 45 patients before and after chemoradiotherapy for esophageal cancer, as well as DNA from leukocytes and fixed esophageal tumor biopsy samples collected during esophagogastroduodenoscopy. Patients were treated from May 2010 through October 2015; 23 patients subsequently underwent esophagectomy, and 22 did not undergo surgery. We also sequenced DNA from blood samples from 40 healthy control individuals. We analyzed 802 regions of 607 genes for single-nucleotide variants previously associated with esophageal adenocarcinoma or squamous cell carcinoma. Patients underwent imaging analyses 6-8 weeks after chemoradiotherapy and were followed for 5 years. Our primary aim was to determine whether detection of circulating tumor DNA (ctDNA) after chemoradiotherapy is associated with risk of tumor progression (growth of local, regional, or distant tumors, detected by imaging or biopsy). RESULTS: The median proportion of tumor-derived DNA in total cell-free DNA before treatment was 0.07%, indicating that ultrasensitive assays are needed for quantification and analysis of ctDNA from localized esophageal tumors. Detection of ctDNA after chemoradiotherapy was associated with tumor progression (hazard ratio, 18.7; P < .0001), formation of distant metastases (hazard ratio, 32.1; P < .0001), and shorter disease-specific survival times (hazard ratio, 23.1; P < .0001). A higher proportion of patients with tumor progression had new mutations detected in plasma samples collected after chemoradiotherapy than patients without progression (P = .03). Detection of ctDNA after chemoradiotherapy preceded radiographic evidence of tumor progression by an average of 2.8 months. Among patients who received chemoradiotherapy without surgery, combined ctDNA and metabolic imaging analysis predicted progression in 100% of patients with tumor progression, compared with 71% for only ctDNA detection and 57% for only metabolic imaging analysis (P < .001 for comparison of either technique to combined analysis). CONCLUSIONS: In an analysis of cell-free DNA in blood samples from patients who underwent chemoradiotherapy for esophageal cancer, detection of ctDNA was associated with tumor progression, metastasis, and disease-specific survival. Analysis of ctDNA might be used to identify patients at highest risk for tumor progression.

High-throughput sequencing for noninvasive disease detection in hematologic malignancies.

Noninvasive monitoring of minimal residual disease (MRD) has led to significant advances in personalized management of patients with hematologic malignancies. Improved therapeutic options and prolonged survival have further increased the need for sensitive tumor assessment that can inform treatment decisions and patient outcomes. At diagnosis or relapse of most hematologic neoplasms, malignant cells are often easily accessible in the blood as circulating tumor cells (CTCs), making them ideal targets to noninvasively profile the molecular features of each patient. In other cancer types, CTCs are generally rare and noninvasive molecular detection relies on circulating tumor DNA (ctDNA) shed from tumor deposits into circulation. The ability to precisely detect and quantify CTCs and ctDNA could minimize invasive procedures and improve prediction of clinical outcomes. Technical advances in MRD detection methods in recent years have led to reduced costs and increased sensitivity, specificity, and applicability. Among currently available tests, high-throughput sequencing (HTS)-based approaches are increasingly attractive for noninvasive molecular testing. HTS-based methods can simultaneously identify multiple genetic markers with high sensitivity and specificity without individual optimization. In this review, we present an overview of techniques used for noninvasive molecular disease detection in selected myeloid and lymphoid neoplasms, with a focus on the current and future role of HTS-based assays.

Distinct patterns of B-cell receptor signaling in non-Hodgkin lymphomas identified by single-cell profiling.

Kinases downstream of B-cell antigen receptor (BCR) represent attractive targets for therapy in non-Hodgkin lymphoma (NHL). As clinical responses vary, improved knowledge regarding activation and regulation of BCR signaling in individual patients is needed. Here, using phosphospecific flow cytometry to obtain malignant B-cell signaling profiles from 95 patients representing 4 types of NHL revealed a striking contrast between chronic lymphocytic leukemia (CLL) and mantle cell lymphoma (MCL) tumors. Lymphoma cells from diffuse large B-cell lymphoma patients had high basal phosphorylation levels of most measured signaling nodes, whereas follicular lymphoma cells represented the opposite pattern with no or very low basal levels. MCL showed large interpatient variability in basal levels, and elevated levels for the phosphorylated forms of AKT, extracellular signal-regulated kinase, p38, STAT1, and STAT5 were associated with poor outcome. CLL tumors had elevated basal levels for the phosphorylated forms of BCR-signaling nodes (Src family tyrosine kinase, spleen tyrosine kinase [SYK], phospholipase Cγ), but had low α-BCR-induced signaling. This contrasted MCL tumors, where α-BCR-induced signaling was variable, but significantly potentiated as compared with the other types. Overexpression of CD79B, combined with a gating strategy whereby signaling output was directly quantified per cell as a function of CD79B levels, confirmed a direct relationship between surface CD79B, immunoglobulin M (IgM), and IgM-induced signaling levels. Furthermore, α-BCR-induced signaling strength was variable across patient samples and correlated with BCR subunit CD79B expression, but was inversely correlated with susceptibility to Bruton tyrosine kinase (BTK) and SYK inhibitors in MCL. These individual differences in BCR levels and signaling might relate to differences in therapy responses to BCR-pathway inhibitors.

Robust enumeration of cell subsets from tissue expression profiles.

We introduce CIBERSORT, a method for characterizing cell composition of complex tissues from their gene expression profiles. When applied to enumeration of hematopoietic subsets in RNA mixtures from fresh, frozen and fixed tissues, including solid tumors, CIBERSORT outperformed other methods with respect to noise, unknown mixture content and closely related cell types. CIBERSORT should enable large-scale analysis of RNA mixtures for cellular biomarkers and therapeutic targets (http://cibersort.stanford.edu/).