Spatially aware deep learning reveals tumor heterogeneity patterns that encode distinct kidney cancer states.

Clear cell renal cell carcinoma (ccRCC) is molecularly heterogeneous, immune infiltrated, and selectively sensitive to immune checkpoint inhibition (ICI). However, the joint tumor-immune states that mediate ICI response remain elusive. We develop spatially aware deep-learning models of tumor and immune features to learn representations of ccRCC tumors using diagnostic whole-slide images (WSIs) in untreated and treated contexts (n = 1,102 patients). We identify patterns of grade heterogeneity in WSIs not achievable through human pathologist analysis, and these graph-based "microheterogeneity" structures associate with PBRM1 loss of function and with patient outcomes. Joint analysis of tumor phenotypes and immune infiltration identifies a subpopulation of highly infiltrated, microheterogeneous tumors responsive to ICI. In paired multiplex immunofluorescence images of ccRCC, microheterogeneity associates with greater PD1 activation in CD8+ lymphocytes and increased tumor-immune interactions. Our work reveals spatially interacting tumor-immune structures underlying ccRCC biology that may also inform selective response to ICI.

Spatially aware deep learning reveals tumor heterogeneity patterns that encode distinct kidney cancer states.

Clear cell renal cell carcinoma (ccRCC) is molecularly heterogeneous, immune infiltrated, and selectively sensitive to immune checkpoint inhibition (ICI). Established histopathology paradigms like nuclear grade have baseline prognostic relevance for ccRCC, although whether existing or novel histologic features encode additional heterogeneous biological and clinical states in ccRCC is uncertain. Here, we developed spatially aware deep learning models of tumor- and immune-related features to learn representations of ccRCC tumors using diagnostic whole-slide images (WSI) in untreated and treated contexts (n = 1102 patients). We discovered patterns of nuclear grade heterogeneity in WSI not achievable through human pathologist analysis, and these graph-based "microheterogeneity" structures associated with PBRM1 loss of function, adverse clinical factors, and selective patient response to ICI. Joint computer vision analysis of tumor phenotypes with inferred tumor infiltrating lymphocyte density identified a further subpopulation of highly infiltrated, microheterogeneous tumors responsive to ICI. In paired multiplex immunofluorescence images of ccRCC, microheterogeneity associated with greater PD1 activation in CD8+ lymphocytes and increased tumor-immune interactions. Thus, our work reveals novel spatially interacting tumor-immune structures underlying ccRCC biology that can also inform selective response to ICI.

Integrative clinical and molecular characterization of translocation renal cell carcinoma.

Translocation renal cell carcinoma (tRCC) is a poorly characterized subtype of kidney cancer driven by MiT/TFE gene fusions. Here, we define the landmarks of tRCC through an integrative analysis of 152 patients with tRCC identified across genomic, clinical trial, and retrospective cohorts. Most tRCCs harbor few somatic alterations apart from MiT/TFE fusions and homozygous deletions at chromosome 9p21.3 (19.2% of cases). Transcriptionally, tRCCs display a heightened NRF2-driven antioxidant response that is associated with resistance to targeted therapies. Consistently, we find that outcomes for patients with tRCC treated with vascular endothelial growth factor receptor inhibitors (VEGFR-TKIs) are worse than those treated with immune checkpoint inhibitors (ICI). Using multiparametric immunofluorescence, we find that the tumors are infiltrated with CD8+ T cells, though the T cells harbor an exhaustion immunophenotype distinct from that of clear cell RCC. Our findings comprehensively define the clinical and molecular features of tRCC and may inspire new therapeutic hypotheses.

Building Tools for Machine Learning and Artificial Intelligence in Cancer Research: Best Practices and a Case Study with the PathML Toolkit for Computational Pathology.

Imaging datasets in cancer research are growing exponentially in both quantity and information density. These massive datasets may enable derivation of insights for cancer research and clinical care, but only if researchers are equipped with the tools to leverage advanced computational analysis approaches such as machine learning and artificial intelligence. In this work, we highlight three themes to guide development of such computational tools: scalability, standardization, and ease of use. We then apply these principles to develop PathML, a general-purpose research toolkit for computational pathology. We describe the design of the PathML framework and demonstrate applications in diverse use cases. PathML is publicly available at www.pathml.com.

Integrative molecular characterization of sarcomatoid and rhabdoid renal cell carcinoma.

Sarcomatoid and rhabdoid (S/R) renal cell carcinoma (RCC) are highly aggressive tumors with limited molecular and clinical characterization. Emerging evidence suggests immune checkpoint inhibitors (ICI) are particularly effective for these tumors, although the biological basis for this property is largely unknown. Here, we evaluate multiple clinical trial and real-world cohorts of S/R RCC to characterize their molecular features, clinical outcomes, and immunologic characteristics. We find that S/R RCC tumors harbor distinctive molecular features that may account for their aggressive behavior, including BAP1 mutations, CDKN2A deletions, and increased expression of MYC transcriptional programs. We show that these tumors are highly responsive to ICI and that they exhibit an immune-inflamed phenotype characterized by immune activation, increased cytotoxic immune infiltration, upregulation of antigen presentation machinery genes, and PD-L1 expression. Our findings build on prior work and shed light on the molecular drivers of aggressivity and responsiveness to ICI of S/R RCC.

Checkpoint Blockade Immunotherapy Induces Dynamic Changes in PD-1-CD8+ Tumor-Infiltrating T Cells.

An improved understanding of the anti-tumor CD8+ T cell response after checkpoint blockade would enable more informed and effective therapeutic strategies. Here we examined the dynamics of the effector response of CD8+ tumor-infiltrating lymphocytes (TILs) after checkpoint blockade therapy. Bulk and single-cell RNA profiles of CD8+ TILs after combined Tim-3+PD-1 blockade in preclinical models revealed significant changes in the transcriptional profile of PD-1- TILs. These cells could be divided into subsets bearing characterstics of naive-, effector-, and memory-precursor-like cells. Effector- and memory-precursor-like TILs contained tumor-antigen-specific cells, exhibited proliferative and effector capacity, and expanded in response to different checkpoint blockade therapies across different tumor models. The memory-precursor-like subset shared features with CD8+ T cells associated with response to checkpoint blockade in patients and was compromised in the absence of Tcf7. Expression of Tcf7/Tcf1 was requisite for the efficacy of diverse immunotherapies, highlighting the importance of this transcriptional regulator in the development of effective CD8+ T cell responses upon immunotherapy.

Induction and transcriptional regulation of the co-inhibitory gene module in T cells.

The expression of co-inhibitory receptors, such as CTLA-4 and PD-1, on effector T cells is a key mechanism for ensuring immune homeostasis. Dysregulated expression of co-inhibitory receptors on CD4+ T cells promotes autoimmunity, whereas sustained overexpression on CD8+ T cells promotes T cell dysfunction or exhaustion, leading to impaired ability to clear chronic viral infections and diseases such as cancer1,2. Here, using RNA and protein expression profiling at single-cell resolution in mouse cells, we identify a module of co-inhibitory receptors that includes not only several known co-inhibitory receptors (PD-1, TIM-3, LAG-3 and TIGIT) but also many new surface receptors. We functionally validated two new co-inhibitory receptors, activated protein C receptor (PROCR) and podoplanin (PDPN). The module of co-inhibitory receptors is co-expressed in both CD4+ and CD8+ T cells and is part of a larger co-inhibitory gene program that is shared by non-responsive T cells in several physiological contexts and is driven by the immunoregulatory cytokine IL-27. Computational analysis identified the transcription factors PRDM1 and c-MAF as cooperative regulators of the co-inhibitory module, and this was validated experimentally. This molecular circuit underlies the co-expression of co-inhibitory receptors in T cells and identifies regulators of T cell function with the potential to control autoimmunity and tumour immunity.

Single-cell RNA-seq supports a developmental hierarchy in human oligodendroglioma.

Although human tumours are shaped by the genetic evolution of cancer cells, evidence also suggests that they display hierarchies related to developmental pathways and epigenetic programs in which cancer stem cells (CSCs) can drive tumour growth and give rise to differentiated progeny. Yet, unbiased evidence for CSCs in solid human malignancies remains elusive. Here we profile 4,347 single cells from six IDH1 or IDH2 mutant human oligodendrogliomas by RNA sequencing (RNA-seq) and reconstruct their developmental programs from genome-wide expression signatures. We infer that most cancer cells are differentiated along two specialized glial programs, whereas a rare subpopulation of cells is undifferentiated and associated with a neural stem cell expression program. Cells with expression signatures for proliferation are highly enriched in this rare subpopulation, consistent with a model in which CSCs are primarily responsible for fuelling the growth of oligodendroglioma in humans. Analysis of copy number variation (CNV) shows that distinct CNV sub-clones within tumours display similar cellular hierarchies, suggesting that the architecture of oligodendroglioma is primarily dictated by developmental programs. Subclonal point mutation analysis supports a similar model, although a full phylogenetic tree would be required to definitively determine the effect of genetic evolution on the inferred hierarchies. Our single-cell analyses provide insight into the cellular architecture of oligodendrogliomas at single-cell resolution and support the cancer stem cell model, with substantial implications for disease management.

A Distinct Gene Module for Dysfunction Uncoupled from Activation in Tumor-Infiltrating T Cells.

Reversing the dysfunctional T cell state that arises in cancer and chronic viral infections is the focus of therapeutic interventions; however, current therapies are effective in only some patients and some tumor types. To gain a deeper molecular understanding of the dysfunctional T cell state, we analyzed population and single-cell RNA profiles of CD8(+) tumor-infiltrating lymphocytes (TILs) and used genetic perturbations to identify a distinct gene module for T cell dysfunction that can be uncoupled from T cell activation. This distinct dysfunction module is downstream of intracellular metallothioneins that regulate zinc metabolism and can be identified at single-cell resolution. We further identify Gata-3, a zinc-finger transcription factor in the dysfunctional module, as a regulator of dysfunction, and we use CRISPR-Cas9 genome editing to show that it drives a dysfunctional phenotype in CD8(+) TILs. Our results open novel avenues for targeting dysfunctional T cell states while leaving activation programs intact.