Luminal breast epithelial cells from wildtype and BRCA mutation carriers harbor copy number alterations commonly associated with breast cancer.

Cancer-associated mutations have been documented in normal tissues, but the prevalence and nature of somatic copy number alterations and their role in tumor initiation and evolution is not well understood. Here, using single cell DNA sequencing, we describe the landscape of CNAs in >42,000 breast epithelial cells from women with normal or high risk of developing breast cancer. Accumulation of individual cells with one or two of a specific subset of CNAs (e.g. 1q gain and 16q, 22q, 7q, and 10q loss) is detectable in almost all breast tissues and, in those from BRCA1 or BRCA2 mutations carriers, occurs prior to loss of heterozygosity (LOH) of the wildtype alleles. These CNAs, which are among the most common associated with ductal carcinoma in situ (DCIS) and malignant breast tumors, are enriched almost exclusively in luminal cells not basal myoepithelial cells. Allele-specific analysis of the enriched CNAs reveals that each allele was independently altered, demonstrating convergent evolution of these CNAs in an individual breast. Tissues from BRCA1 or BRCA2 mutation carriers contain a small percentage of cells with extreme aneuploidy, featuring loss of TP53 , LOH of BRCA1 or BRCA2 , and multiple breast cancer-associated CNAs in addition to one or more of the common CNAs in 1q, 10q or 16q. Notably, cells with intermediate levels of CNAs are not detected, arguing against a stepwise gradual accumulation of CNAs. Overall, our findings demonstrate that chromosomal alterations in normal breast epithelium partially mirror those of established cancer genomes and are chromosome- and cell lineage-specific.

Single-cell mtDNA dynamics in tumors is driven by coregulation of nuclear and mitochondrial genomes.

The extent of cell-to-cell variation in tumor mitochondrial DNA (mtDNA) copy number and genotype, and the phenotypic and evolutionary consequences of such variation, are poorly characterized. Here we use amplification-free single-cell whole-genome sequencing (Direct Library Prep (DLP+)) to simultaneously assay mtDNA copy number and nuclear DNA (nuDNA) in 72,275 single cells derived from immortalized cell lines, patient-derived xenografts and primary human tumors. Cells typically contained thousands of mtDNA copies, but variation in mtDNA copy number was extensive and strongly associated with cell size. Pervasive whole-genome doubling events in nuDNA associated with stoichiometrically balanced adaptations in mtDNA copy number, implying that mtDNA-to-nuDNA ratio, rather than mtDNA copy number itself, mediated downstream phenotypes. Finally, multimodal analysis of DLP+ and single-cell RNA sequencing identified both somatic loss-of-function and germline noncoding variants in mtDNA linked to heteroplasmy-dependent changes in mtDNA copy number and mitochondrial transcription, revealing phenotypic adaptations to disrupted nuclear/mitochondrial balance.

Allele-specific transcriptional effects of subclonal copy number alterations enable genotype-phenotype mapping in cancer cells.

Subclonal copy number alterations are a prevalent feature in tumors with high chromosomal instability and result in heterogeneous cancer cell populations with distinct phenotypes. However, the extent to which subclonal copy number alterations contribute to clone-specific phenotypes remains poorly understood. We develop TreeAlign, which computationally integrates independently sampled single-cell DNA and RNA sequencing data from the same cell population. TreeAlign accurately encodes dosage effects from subclonal copy number alterations, the impact of allelic imbalance on allele-specific transcription, and obviates the need to define genotypic clones from a phylogeny a priori, leading to highly granular definitions of clones with distinct expression programs. These improvements enable clone-clone gene expression comparisons with higher resolution and identification of expression programs that are genomically independent. Our approach sets the stage for dissecting the relative contribution of fixed genomic alterations and dynamic epigenetic processes on gene expression programs in cancer.

Weakly Supervised Deep Learning Predicts Immunotherapy Response in Solid Tumors Based on PD-L1 Expression.

Programmed death-ligand 1 (PD-L1) IHC is the most commonly used biomarker for immunotherapy response. However, quantification of PD-L1 status in pathology slides is challenging. Neither manual quantification nor a computer-based mimicking of manual readouts is perfectly reproducible, and the predictive performance of both approaches regarding immunotherapy response is limited. In this study, we developed a deep learning (DL) method to predict PD-L1 status directly from raw IHC image data, without explicit intermediary steps such as cell detection or pigment quantification. We trained the weakly supervised model on PD-L1-stained slides from the non-small cell lung cancer (NSCLC)-Memorial Sloan Kettering (MSK) cohort (N = 233) and validated it on the pan-cancer-Vall d'Hebron Institute of Oncology (VHIO) cohort (N = 108). We also investigated the performance of the model to predict response to immune checkpoint inhibitors (ICI) in terms of progression-free survival. In the pan-cancer-VHIO cohort, the performance was compared with tumor proportion score (TPS) and combined positive score (CPS). The DL model showed good performance in predicting PD-L1 expression (TPS ≥ 1%) in both NSCLC-MSK and pan-cancer-VHIO cohort (AUC 0.88 ± 0.06 and 0.80 ± 0.03, respectively). The predicted PD-L1 status showed an improved association with response to ICIs [HR: 1.5 (95% confidence interval: 1-2.3), P = 0.049] compared with TPS [HR: 1.4 (0.96-2.2), P = 0.082] and CPS [HR: 1.2 (0.79-1.9), P = 0.386]. Notably, our explainability analysis showed that the model does not just look at the amount of brown pigment in the IHC slides, but also considers morphologic factors such as lymphocyte conglomerates. Overall, end-to-end weakly supervised DL shows potential for improving patient stratification for cancer immunotherapy by analyzing PD-L1 IHC, holistically integrating morphology and PD-L1 staining intensity. SIGNIFICANCE: The weakly supervised DL model to predict PD-L1 status from raw IHC data, integrating tumor staining intensity and morphology, enables enhanced patient stratification in cancer immunotherapy compared with traditional pathologist assessment.

Identification of transcriptional programs using dense vector representations defined by mutual information with GeneVector.

Deciphering individual cell phenotypes from cell-specific transcriptional processes requires high dimensional single cell RNA sequencing. However, current dimensionality reduction methods aggregate sparse gene information across cells, without directly measuring the relationships that exist between genes. By performing dimensionality reduction with respect to gene co-expression, low-dimensional features can model these gene-specific relationships and leverage shared signal to overcome sparsity. We describe GeneVector, a scalable framework for dimensionality reduction implemented as a vector space model using mutual information between gene expression. Unlike other methods, including principal component analysis and variational autoencoders, GeneVector uses latent space arithmetic in a lower dimensional gene embedding to identify transcriptional programs and classify cell types. In this work, we show in four single cell RNA-seq datasets that GeneVector was able to capture phenotype-specific pathways, perform batch effect correction, interactively annotate cell types, and identify pathway variation with treatment over time.

Epigenetic dysregulation from chromosomal transit in micronuclei.

Chromosomal instability (CIN) and epigenetic alterations are characteristics of advanced and metastatic cancers1-4, but whether they are mechanistically linked is unknown. Here we show that missegregation of mitotic chromosomes, their sequestration in micronuclei5,6 and subsequent rupture of the micronuclear envelope7 profoundly disrupt normal histone post-translational modifications (PTMs), a phenomenon conserved across humans and mice, as well as in cancer and non-transformed cells. Some of the changes in histone PTMs occur because of the rupture of the micronuclear envelope, whereas others are inherited from mitotic abnormalities before the micronucleus is formed. Using orthogonal approaches, we demonstrate that micronuclei exhibit extensive differences in chromatin accessibility, with a strong positional bias between promoters and distal or intergenic regions, in line with observed redistributions of histone PTMs. Inducing CIN causes widespread epigenetic dysregulation, and chromosomes that transit in micronuclei experience heritable abnormalities in their accessibility long after they have been reincorporated into the primary nucleus. Thus, as well as altering genomic copy number, CIN promotes epigenetic reprogramming and heterogeneity in cancer.

Immune profiling after allogeneic hematopoietic cell transplantation in pediatric acute myeloid leukemia.

Although allogeneic hematopoietic cell transplant (allo-HCT) is curative for high-risk pediatric acute myeloid leukemia (AML), disease relapse remains the primary cause of posttransplant mortality. To identify pressures imposed by allo-HCT on AML cells that escape the graft-versus-leukemia effect, we evaluated immune signatures at diagnosis and posttransplant relapse in bone marrow samples from 4 pediatric patients using a multimodal single-cell proteogenomic approach. Downregulation of major histocompatibility complex class II expression was most profound in progenitor-like blasts and accompanied by correlative changes in transcriptional regulation. Dysfunction of activated natural killer cells and CD8+ T-cell subsets at relapse was evidenced by the loss of response to interferon gamma, tumor necrosis factor α signaling via NF-κB, and interleukin-2/STAT5 signaling. Clonotype analysis of posttransplant relapse samples revealed an expansion of dysfunctional T cells and enrichment of T-regulatory and T-helper cells. Using novel computational methods, our results illustrate a diverse immune-related transcriptional signature in posttransplant relapses not previously reported in pediatric AML.

Single-cell DNA replication dynamics in genomically unstable cancers.

Dysregulated DNA replication is both a cause and a consequence of aneuploidy, yet the dynamics of DNA replication in aneuploid cell populations remains understudied. We developed a new method, PERT, for inferring cell-specific DNA replication states from single-cell whole genome sequencing, and investigated clone-specific DNA replication dynamics in >50,000 cells obtained from a collection of aneuploid and clonally heterogeneous cell lines, xenografts and primary cancer tissues. Clone replication timing (RT) profiles correlated with future copy number changes in serially passaged cell lines. Cell type was the strongest determinant of RT heterogeneity, while whole genome doubling and mutational process were associated with accumulation of late S-phase cells and weaker RT associations. Copy number changes affecting chromosome X had striking impact on RT, with loss of the inactive X allele shifting replication earlier, and loss of inactive Xq resulting in reactivation of Xp. Finally, analysis of time series xenografts illustrate how cell cycle distributions approximate clone proliferation, recapitulating expected relationships between proliferation and fitness in treatment-naive and chemotherapeutic contexts.

Fast, light, and scalable: harnessing data-mined line annotations for automated tumor segmentation on brain MRI.

OBJECTIVES: While fully supervised learning can yield high-performing segmentation models, the effort required to manually segment large training sets limits practical utility. We investigate whether data mined line annotations can facilitate brain MRI tumor segmentation model development without requiring manually segmented training data. METHODS: In this retrospective study, a tumor detection model trained using clinical line annotations mined from PACS was leveraged with unsupervised segmentation to generate pseudo-masks of enhancing tumors on T1-weighted post-contrast images (9911 image slices; 3449 adult patients). Baseline segmentation models were trained and employed within a semi-supervised learning (SSL) framework to refine the pseudo-masks. Following each self-refinement cycle, a new model was trained and tested on a held-out set of 319 manually segmented image slices (93 adult patients), with the SSL cycles continuing until Dice score coefficient (DSC) peaked. DSCs were compared using bootstrap resampling. Utilizing the best-performing models, two inference methods were compared: (1) conventional full-image segmentation, and (2) a hybrid method augmenting full-image segmentation with detection plus image patch segmentation. RESULTS: Baseline segmentation models achieved DSC of 0.768 (U-Net), 0.831 (Mask R-CNN), and 0.838 (HRNet), improving with self-refinement to 0.798, 0.871, and 0.873 (each p < 0.001), respectively. Hybrid inference outperformed full image segmentation alone: DSC 0.884 (Mask R-CNN) vs. 0.873 (HRNet), p < 0.001. CONCLUSIONS: Line annotations mined from PACS can be harnessed within an automated pipeline to produce accurate brain MRI tumor segmentation models without manually segmented training data, providing a mechanism to rapidly establish tumor segmentation capabilities across radiology modalities. KEY POINTS: • A brain MRI tumor detection model trained using clinical line measurement annotations mined from PACS was leveraged to automatically generate tumor segmentation pseudo-masks. • An iterative self-refinement process automatically improved pseudo-mask quality, with the best-performing segmentation pipeline achieving a Dice score of 0.884 on a held-out test set. • Tumor line measurement annotations generated in routine clinical radiology practice can be harnessed to develop high-performing segmentation models without manually segmented training data, providing a mechanism to rapidly establish tumor segmentation capabilities across radiology modalities.

Exploiting allele-specific transcriptional effects of subclonal copy number alterations for genotype-phenotype mapping in cancer cell populations.

Somatic copy number alterations drive aberrant gene expression in cancer cells. In tumors with high levels of chromosomal instability, subclonal copy number alterations (CNAs) are a prevalent feature which often result in heterogeneous cancer cell populations with distinct phenotypes1. However, the extent to which subclonal CNAs contribute to clone-specific phenotypes remains poorly understood, in part due to the lack of methods to quantify how CNAs influence gene expression at a subclone level. We developed TreeAlign, which computationally integrates independently sampled single-cell DNA and RNA sequencing data from the same cell population and explicitly models gene dosage effects from subclonal alterations. We show through quantitative benchmarking data and application to human cancer data with single cell DNA and RNA libraries that TreeAlign accurately encodes clone-specific transcriptional effects of subclonal CNAs, the impact of allelic imbalance on allele-specific transcription, and obviates the need to arbitrarily define genotypic clones from a phylogenetic tree a priori. Combined, these advances lead to highly granular definitions of clones with distinct copy-number driven expression programs with increased resolution and accuracy over competing methods. The resulting improvement in assignment of transcriptional phenotypes to genomic clones enables clone-clone gene expression comparisons and explicit inference of genes that are mechanistically altered through CNAs, and identification of expression programs that are genomically independent. Our approach sets the stage for dissecting the relative contribution of fixed genomic alterations and dynamic epigenetic processes on gene expression programs in cancer.

Imaging and Molecular Annotation of Xenographs and Tumours (IMAXT): High throughput data and analysis infrastructure.

With the aim of producing a 3D representation of tumors, imaging and molecular annotation of xenografts and tumors (IMAXT) uses a large variety of modalities in order to acquire tumor samples and produce a map of every cell in the tumor and its host environment. With the large volume and variety of data produced in the project, we developed automatic data workflows and analysis pipelines. We introduce a research methodology where scientists connect to a cloud environment to perform analysis close to where data are located, instead of bringing data to their local computers. Here, we present the data and analysis infrastructure, discuss the unique computational challenges and describe the analysis chains developed and deployed to generate molecularly annotated tumor models. Registration is achieved by use of a novel technique involving spherical fiducial marks that are visible in all imaging modalities used within IMAXT. The automatic pipelines are highly optimized and allow to obtain processed datasets several times quicker than current solutions narrowing the gap between data acquisition and scientific exploitation.

Single-cell profiling reveals a memory B cell-like subtype of follicular lymphoma with increased transformation risk.

Follicular lymphoma (FL) is an indolent cancer of mature B-cells but with ongoing risk of transformation to more aggressive histology over time. Recurrent mutations associated with transformation have been identified; however, prognostic features that can be discerned at diagnosis could be clinically useful. We present here comprehensive profiling of both tumor and immune compartments in 155 diagnostic FL biopsies at single-cell resolution by mass cytometry. This revealed a diversity of phenotypes but included two recurrent patterns, one which closely resembles germinal center B-cells (GCB) and another which appears more related to memory B-cells (MB). GCB-type tumors are enriched for EZH2, TNFRSF14, and MEF2B mutations, while MB-type tumors contain increased follicular helper T-cells. MB-type and intratumoral phenotypic diversity are independently associated with increased risk of transformation, supporting biological relevance of these features. Notably, a reduced 26-marker panel retains sufficient information to allow phenotypic profiling of future cohorts by conventional flow cytometry.

Clonal somatic copy number altered driver events inform drug sensitivity in high-grade serous ovarian cancer.

Chromosomal instability is a major challenge to patient stratification and targeted drug development for high-grade serous ovarian carcinoma (HGSOC). Here we show that somatic copy number alterations (SCNAs) in frequently amplified HGSOC cancer genes significantly correlate with gene expression and methylation status. We identify five prevalent clonal driver SCNAs (chromosomal amplifications encompassing MYC, PIK3CA, CCNE1, KRAS and TERT) from multi-regional HGSOC data and reason that their strong selection should prioritise them as key biomarkers for targeted therapies. We use primary HGSOC spheroid models to test interactions between in vitro targeted therapy and SCNAs. MYC chromosomal copy number is associated with in-vitro and clinical response to paclitaxel and in-vitro response to mTORC1/2 inhibition. Activation of the mTOR survival pathway in the context of MYC-amplified HGSOC is statistically associated with increased prevalence of SCNAs in genes from the PI3K pathway. Co-occurrence of amplifications in MYC and genes from the PI3K pathway is independently observed in squamous lung cancer and triple negative breast cancer. In this work, we show that identifying co-occurrence of clonal driver SCNA genes could be used to tailor therapeutics for precision medicine.

Multimodal integration of radiology, pathology and genomics for prediction of response to PD-(L)1 blockade in patients with non-small cell lung cancer.

Immunotherapy is used to treat almost all patients with advanced non-small cell lung cancer (NSCLC); however, identifying robust predictive biomarkers remains challenging. Here we show the predictive capacity of integrating medical imaging, histopathologic and genomic features to predict immunotherapy response using a cohort of 247 patients with advanced NSCLC with multimodal baseline data obtained during diagnostic clinical workup, including computed tomography scan images, digitized programmed death ligand-1 immunohistochemistry slides and known outcomes to immunotherapy. Using domain expert annotations, we developed a computational workflow to extract patient-level features and used a machine-learning approach to integrate multimodal features into a risk prediction model. Our multimodal model (area under the curve (AUC) = 0.80, 95% confidence interval (CI) 0.74-0.86) outperformed unimodal measures, including tumor mutational burden (AUC = 0.61, 95% CI 0.52-0.70) and programmed death ligand-1 immunohistochemistry score (AUC = 0.73, 95% CI 0.65-0.81). Our study therefore provides a quantitative rationale for using multimodal features to improve prediction of immunotherapy response in patients with NSCLC using expert-guided machine learning.

Multimodal data integration using machine learning improves risk stratification of high-grade serous ovarian cancer.

Patients with high-grade serous ovarian cancer suffer poor prognosis and variable response to treatment. Known prognostic factors for this disease include homologous recombination deficiency status, age, pathological stage and residual disease status after debulking surgery. Recent work has highlighted important prognostic information captured in computed tomography and histopathological specimens, which can be exploited through machine learning. However, little is known about the capacity of combining features from these disparate sources to improve prediction of treatment response. Here, we assembled a multimodal dataset of 444 patients with primarily late-stage high-grade serous ovarian cancer and discovered quantitative features, such as tumor nuclear size on staining with hematoxylin and eosin and omental texture on contrast-enhanced computed tomography, associated with prognosis. We found that these features contributed complementary prognostic information relative to one another and clinicogenomic features. By fusing histopathological, radiologic and clinicogenomic machine-learning models, we demonstrate a promising path toward improved risk stratification of patients with cancer through multimodal data integration.

Fundamental immune-oncogenicity trade-offs define driver mutation fitness.

Missense driver mutations in cancer are concentrated in a few hotspots1. Various mechanisms have been proposed to explain this skew, including biased mutational processes2, phenotypic differences3-6 and immunoediting of neoantigens7,8; however, to our knowledge, no existing model weighs the relative contribution of these features to tumour evolution. We propose a unified theoretical 'free fitness' framework that parsimoniously integrates multimodal genomic, epigenetic, transcriptomic and proteomic data into a biophysical model of the rate-limiting processes underlying the fitness advantage conferred on cancer cells by driver gene mutations. Focusing on TP53, the most mutated gene in cancer1, we present an inference of mutant p53 concentration and demonstrate that TP53 hotspot mutations optimally solve an evolutionary trade-off between oncogenic potential and neoantigen immunogenicity. Our model anticipates patient survival in The Cancer Genome Atlas and patients with lung cancer treated with immunotherapy as well as the age of tumour onset in germline carriers of TP53 variants. The predicted differential immunogenicity between hotspot mutations was validated experimentally in patients with cancer and in a unique large dataset of healthy individuals. Our data indicate that immune selective pressure on TP53 mutations has a smaller role in non-cancerous lesions than in tumours, suggesting that targeted immunotherapy may offer an early prophylactic opportunity for the former. Determining the relative contribution of immunogenicity and oncogenic function to the selective advantage of hotspot mutations thus has important implications for both precision immunotherapies and our understanding of tumour evolution.

Tumor-associated antigen PRAME exhibits dualistic functions that are targetable in diffuse large B cell lymphoma.

PRAME is a prominent member of the cancer testis antigen family of proteins, which triggers autologous T cell-mediated immune responses. Integrative genomic analysis in diffuse large B cell lymphoma (DLBCL) uncovered recurrent and highly focal deletions of 22q11.22, including the PRAME gene, which were associated with poor outcome. PRAME-deleted tumors showed cytotoxic T cell immune escape and were associated with cold tumor microenvironments. In addition, PRAME downmodulation was strongly associated with somatic EZH2 Y641 mutations in DLBCL. In turn, PRC2-regulated genes were repressed in isogenic PRAME-KO lymphoma cell lines, and PRAME was found to directly interact with EZH2 as a negative regulator. EZH2 inhibition with EPZ-6438 abrogated these extrinsic and intrinsic effects, leading to PRAME expression and microenvironment restoration in vivo. Our data highlight multiple functions of PRAME during lymphomagenesis and provide a preclinical rationale for synergistic therapies combining epigenetic reprogramming with PRAME-targeted therapies.

Semisupervised Training of a Brain MRI Tumor Detection Model Using Mined Annotations.

Background Artificial intelligence (AI) applications for cancer imaging conceptually begin with automated tumor detection, which can provide the foundation for downstream AI tasks. However, supervised training requires many image annotations, and performing dedicated post hoc image labeling is burdensome and costly. Purpose To investigate whether clinically generated image annotations can be data mined from the picture archiving and communication system (PACS), automatically curated, and used for semisupervised training of a brain MRI tumor detection model. Materials and Methods In this retrospective study, the cancer center PACS was mined for brain MRI scans acquired between January 2012 and December 2017 and included all annotated axial T1 postcontrast images. Line annotations were converted to boxes, excluding boxes shorter than 1 cm or longer than 7 cm. The resulting boxes were used for supervised training of object detection models using RetinaNet and Mask region-based convolutional neural network (R-CNN) architectures. The best-performing model trained from the mined data set was used to detect unannotated tumors on training images themselves (self-labeling), automatically correcting many of the missing labels. After self-labeling, new models were trained using this expanded data set. Models were scored for precision, recall, and F1 using a held-out test data set comprising 754 manually labeled images from 100 patients (403 intra-axial and 56 extra-axial enhancing tumors). Model F1 scores were compared using bootstrap resampling. Results The PACS query extracted 31 150 line annotations, yielding 11 880 boxes that met inclusion criteria. This mined data set was used to train models, yielding F1 scores of 0.886 for RetinaNet and 0.908 for Mask R-CNN. Self-labeling added 18 562 training boxes, improving model F1 scores to 0.935 (P < .001) and 0.954 (P < .001), respectively. Conclusion The application of semisupervised learning to mined image annotations significantly improved tumor detection performance, achieving an excellent F1 score of 0.954. This development pipeline can be extended for other imaging modalities, repurposing unused data silos to potentially enable automated tumor detection across radiologic modalities. © RSNA, 2022 Online supplemental material is available for this article.

Harnessing multimodal data integration to advance precision oncology.

Advances in quantitative biomarker development have accelerated new forms of data-driven insights for patients with cancer. However, most approaches are limited to a single mode of data, leaving integrated approaches across modalities relatively underdeveloped. Multimodal integration of advanced molecular diagnostics, radiological and histological imaging, and codified clinical data presents opportunities to advance precision oncology beyond genomics and standard molecular techniques. However, most medical datasets are still too sparse to be useful for the training of modern machine learning techniques, and significant challenges remain before this is remedied. Combined efforts of data engineering, computational methods for analysis of heterogeneous data and instantiation of synergistic data models in biomedical research are required for success. In this Perspective, we offer our opinions on synthesizing complementary modalities of data with emerging multimodal artificial intelligence methods. Advancing along this direction will result in a reimagined class of multimodal biomarkers to propel the field of precision oncology in the coming decade.