Acute Lymphoblastic Leukemia with Myeloid Mutations Is a High-Risk Disease Associated with Clonal Hematopoiesis

Myeloid neoplasms arise from preexisting clonal hematopoiesis (CH); however, the role of CH in the pathogenesis of acute lymphoblastic leukemia (ALL) is unknown. We found that 18% of adult ALL cases harbored TP53, and 16% had myeloid CH-associated gene mutations. ALL with myeloid mutations (MyM) had distinct genetic and clinical characteristics, associated with inferior survival. By using single-cell proteogenomic analysis, we demonstrated that myeloid mutations were present years before the diagnosis of ALL, and a subset of these clones expanded over time to manifest as dominant clones in ALL. Single-cell RNA sequencing revealed upregulation of genes associated with cell survival and resistance to apoptosis in B-ALL with MyM, which responds better to newer immunotherapeutic approaches. These findings define ALL with MyM as a high-risk disease that can arise from antecedent CH and offer new mechanistic insights to develop better therapeutic and preventative strategies.

SIGNIFICANCE

CH is a precursor lesion for lymphoblastic leukemogenesis. ALL with MyM has distinct genetic and clinical characteristics, associated with adverse survival outcomes after chemotherapy. CH can precede ALL years before diagnosis, and ALL with MyM is enriched with activated T cells that respond to immunotherapies such as blinatumomab. See related commentary by Iacobucci, p. 142.

High-dimensional deconstruction of pancreatic cancer identifies tumor microenvironmental and developmental stemness features that predict survival

Numerous cell states are known to comprise the pancreatic ductal adenocarcinoma (PDAC) tumor microenvironment (TME). However, the developmental stemness and co-occurrence of these cell states remain poorly defined. Here, we performed single-cell RNA sequencing (scRNA-seq) on a cohort of treatment-naive PDAC time-of-diagnosis endoscopic ultrasound-guided fine needle biopsy (EUS-FNB) samples (n = 25). We then combined these samples with surgical resection (n = 6) and publicly available samples to increase statistical power (n = 80). Following annotation into 25 distinct cell states, cells were scored for developmental stemness, and a customized version of the Ecotyper tool was used to identify communities of co-occurring cell states in bulk RNA-seq samples (n = 268). We discovered a tumor microenvironmental community comprised of aggressive basal-like malignant cells, tumor-promoting SPP1+ macrophages, and myofibroblastic cancer-associated fibroblasts associated with especially poor prognosis. We also found a developmental stemness continuum with implications for survival that is present in both malignant cells and cancer-associated fibroblasts (CAFs). We further demonstrated that high-dimensional analyses predictive of survival are feasible using standard-of-care, time-of-diagnosis EUS-FNB specimens. In summary, we identified tumor microenvironmental and developmental stemness characteristics from a high-dimensional gene expression analysis of PDAC using human tissue specimens, including time-of-diagnosis EUS-FNB samples. These reveal new connections between tumor microenvironmental composition, CAF and malignant cell stemness, and patient survival that could lead to better upfront risk stratification and more personalized upfront clinical decision-making.

Targeting Solid Cancers with a Cancer-Specific Monoclonal Antibody to Surface Expressed Aberrantly O-glycosylated Proteins

The lack of antibodies with sufficient cancer selectivity is currently limiting the treatment of solid tumors by immunotherapies. Most current immunotherapeutic targets are tumor-associated antigens that are also found in healthy tissues and often do not display sufficient cancer selectivity to be used as targets for potent antibody-based immunotherapeutic treatments, such as chimeric antigen receptor (CAR) T cells. Many solid tumors, however, display aberrant glycosylation that results in expression of tumor-associated carbohydrate antigens that are distinct from healthy tissues. Targeting aberrantly glycosylated glycopeptide epitopes within existing or novel glycoprotein targets may provide the cancer selectivity needed for immunotherapy of solid tumors. However, to date only a few such glycopeptide epitopes have been targeted. Here, we used O-glycoproteomics data from multiple cell lines to identify a glycopeptide epitope in CD44v6, a cancer-associated CD44 isoform, and developed a cancer-specific mAb, 4C8, through a glycopeptide immunization strategy. 4C8 selectively binds to Tn-glycosylated CD44v6 in a site-specific manner with low nanomolar affinity. 4C8 was shown to be highly cancer specific by IHC of sections from multiple healthy and cancerous tissues. 4C8 CAR T cells demonstrated target-specific cytotoxicity in vitro and significant tumor regression and increased survival in vivo. Importantly, 4C8 CAR T cells were able to selectively kill target cells in a mixed organotypic skin cancer model having abundant CD44v6 expression without affecting healthy keratinocytes, indicating tolerability and safety.

Dual Targeting of Apoptotic and Signaling Pathways in T-Lineage Acute Lymphoblastic Leukemia

PURPOSE

Relapsed T-acute lymphoblastic leukemia (T-ALL) has limited treatment options. We investigated mechanisms of resistance to BH3 mimetics in T-ALL to develop rational combination strategies. We also looked at the preclinical efficacy of NWP-0476, a novel BCL-2/BCL-xL inhibitor, as single agent and combination therapy in T-ALL.

EXPERIMENTAL DESIGN

We used BH3 profiling as a predictive tool for BH3 mimetic response in T-ALL. Using isogenic control, venetoclax-resistant (ven-R) and NWP-0476-resistant (NWP-R) cells, phosphokinase array was performed to identify differentially regulated signaling pathways.

RESULTS

Typical T-ALL cells had increased dependence on BCL-xL, whereas early T-precursor (ETP)-ALL cells had higher BCL-2 dependence for survival. BCL-2/BCL-xL dual inhibitors were effective against both subtypes of T-lineage ALL. A 71-protein human phosphokinase array showed increased LCK activity in ven-R cells, and increased ACK1 activity in ven-R and NWP-R cells. We hypothesized that pre-TCR and ACK1 signaling pathways are drivers of resistance to BCL-2 and BCL-xL inhibition, respectively. First, we silenced LCK gene in T-ALL cell lines, which resulted in increased sensitivity to BCL-2 inhibition. Mechanistically, LCK activated NF-κB pathway and the expression of BCL-xL. Silencing ACK1 gene resulted in increased sensitivity to both BCL-2 and BCL-xL inhibitors. ACK1 signaling upregulated AKT pathway, which inhibited the pro-apoptotic function of BAD. In a T-ALL patient-derived xenograft model, combination of NWP-0476 and dasatinib demonstrated synergy without major organ toxicity.

CONCLUSIONS

LCK and ACK1 signaling pathways are critical regulators of BH3 mimetic resistance in T-ALL. Combination of BH3 mimetics with tyrosine kinase inhibitors might be effective against relapsed T-ALL.

Uncertainty-informed deep learning models enable high-confidence predictions for digital histopathology

A model's ability to express its own predictive uncertainty is an essential attribute for maintaining clinical user confidence as computational biomarkers are deployed into real-world medical settings. In the domain of cancer digital histopathology, we describe a clinically-oriented approach to uncertainty quantification for whole-slide images, estimating uncertainty using dropout and calculating thresholds on training data to establish cutoffs for low- and high-confidence predictions. We train models to identify lung adenocarcinoma vs. squamous cell carcinoma and show that high-confidence predictions outperform predictions without uncertainty, in both cross-validation and testing on two large external datasets spanning multiple institutions. Our testing strategy closely approximates real-world application, with predictions generated on unsupervised, unannotated slides using predetermined thresholds. Furthermore, we show that uncertainty thresholding remains reliable in the setting of domain shift, with accurate high-confidence predictions of adenocarcinoma vs. squamous cell carcinoma for out-of-distribution, non-lung cancer cohorts.

The impact of site-specific digital histology signatures on deep learning model accuracy and bias

The Cancer Genome Atlas (TCGA) is one of the largest biorepositories of digital histology. Deep learning (DL) models have been trained on TCGA to predict numerous features directly from histology, including survival, gene expression patterns, and driver mutations. However, we demonstrate that these features vary substantially across tissue submitting sites in TCGA for over 3,000 patients with six cancer subtypes. Additionally, we show that histologic image differences between submitting sites can easily be identified with DL. Site detection remains possible despite commonly used color normalization and augmentation methods, and we quantify the image characteristics constituting this site-specific digital histology signature. We demonstrate that these site-specific signatures lead to biased accuracy for prediction of features including survival, genomic mutations, and tumor stage. Furthermore, ethnicity can also be inferred from site-specific signatures, which must be accounted for to ensure equitable application of DL. These site-specific signatures can lead to overoptimistic estimates of model performance, and we propose a quadratic programming method that abrogates this bias by ensuring models are not trained and validated on samples from the same site.