Faculty Bibliography

Identifying new drug targets and developing safe and effective drugs is both challenging and risky. Furthermore, characterizing drug development risk, the probability that a drug will eventually receive regulatory approval, has been notoriously hard given the complexities of drug biology and clinical trials. This inherent risk is often misunderstood and mischaracterized, leading to inefficient allocation of resources and, as a result, an overall reduction in R&D productivity. Here we argue that the recent resurgence of Machine Learning (ML) in combination with the availability of data can provide a more accurate and unbiased estimate of drug development risk.

In lung cancer, enrichment of the lower airway microbiota with oral commensals commonly occurs and ex vivo models support that some of these bacteria can trigger host transcriptomic signatures associated with carcinogenesis. Here, we show that this lower airway dysbiotic signature was more prevalent in group IIIB-IV TNM stage lung cancer and is associated with poor prognosis, as shown by decreased survival among subjects with early stage disease (I-IIIA) and worse tumor progression as measured by RECIST scores among subjects with IIIB-IV stage disease. In addition, this lower airway microbiota signature was associated with upregulation of IL-17, PI3K, MAPK and ERK pathways in airway transcriptome, and we identified Veillonella parvula as the most abundant taxon driving this association. In a KP lung cancer model, lower airway dysbiosis with V. parvula led to decreased survival, increased tumor burden, IL-17 inflammatory phenotype and activation of checkpoint inhibitor markers.

During the germinal center (GC) reaction, B cells undergo extensive redistribution of cohesin complex and three-dimensional reorganization of their genomes. Yet, the significance of cohesin and architectural programming in the humoral immune response is unknown. Herein we report that homozygous deletion of Smc3, encoding the cohesin ATPase subunit, abrogated GC formation, while, in marked contrast, Smc3 haploinsufficiency resulted in GC hyperplasia, skewing of GC polarity and impaired plasma cell (PC) differentiation. Genome-wide chromosomal conformation and transcriptional profiling revealed defects in GC B cell terminal differentiation programs controlled by the lymphoma epigenetic tumor suppressors Tet2 and Kmt2d and failure of Smc3-haploinsufficient GC B cells to switch from B cell- to PC-defining transcription factors. Smc3 haploinsufficiency preferentially impaired the connectivity of enhancer elements controlling various lymphoma tumor suppressor genes, and, accordingly, Smc3 haploinsufficiency accelerated lymphomagenesis in mice with constitutive Bcl6 expression. Collectively, our data indicate a dose-dependent function for cohesin in humoral immunity to facilitate the B cell to PC phenotypic switch while restricting malignant transformation.

Epilepsy is a heterogenous group of disorders defined by recurrent seizure activity due to abnormal synchronized activity of neurons. A growing number of epilepsy cases are believed to be caused by genetic factors and copy number variants (CNV) contribute to up to 5% of epilepsy cases. However, CNVs in epilepsy are usually large deletions or duplications involving multiple neurodevelopmental genes. In patients who underwent seizure focus resection for treatment-resistant epilepsy, whole genome DNA methylation profiling identified 3 main clusters of which one showed strong association with receptor tyrosine kinase (RTK) genes. We identified focal copy number gains involving epidermal growth factor receptor (EGFR) and PDGFRA loci. The dysplastic neurons of cases with amplifications showed marked overexpression of EGFR and PDGFRA, while glial and endothelial cells were negative. Targeted sequencing of regulatory regions and DNA methylation analysis revealed that only enhancer regions of EGFR and gene promoter of PDGFRA were amplified, while coding regions did not show copy number abnormalities or somatic mutations. Somatic focal copy number gains of noncoding regulatory represent a previously unrecognized genetic driver in epilepsy and a mechanism of abnormal activation of RTK genes. Upregulated RTKs provide a potential avenue for therapy in seizure disorders.

PURPOSE/OBJECTIVE:Several biomarkers of response to immune checkpoint inhibitors (ICI) show potential but are not yet scalable to the clinic. We developed a pipeline that integrates deep learning on histology specimens with clinical data to predict ICI response in advanced melanoma. EXPERIMENTAL DESIGN/METHODS:We used a training cohort from New York University (New York, NY) and a validation cohort from Vanderbilt University (Nashville, TN). We built a multivariable classifier that integrates neural network predictions with clinical data. A ROC curve was generated and the optimal threshold was used to stratify patients as high versus low risk for progression. Kaplan-Meier curves compared progression-free survival (PFS) between the groups. The classifier was validated on two slide scanners (Aperio AT2 and Leica SCN400). RESULTS:= 0.03 for the Leica SCN400). CONCLUSIONS:Histology slides and patients' clinicodemographic characteristics are readily available through standard of care and have the potential to predict ICI treatment outcomes. With prospective validation, we believe our approach has potential for integration into clinical practice.

Linker histone H1 proteins bind to nucleosomes and facilitate chromatin compaction¹, although their biological functions are poorly understood. Mutations in the genes that encode H1 isoforms B-E (H1B, H1C, H1D and H1E; also known as H1-5, H1-2, H1-3 and H1-4, respectively) are highly recurrent in B cell lymphomas, but the pathogenic relevance of these mutations to cancer and the mechanisms that are involved are unknown. Here we show that lymphoma-associated H1 alleles are genetic driver mutations in lymphomas. Disruption of H1 function results in a profound architectural remodelling of the genome, which is characterized by large-scale yet focal shifts of chromatin from a compacted to a relaxed state. This decompaction drives distinct changes in epigenetic states, primarily owing to a gain of histone H3 dimethylation at lysine 36 (H3K36me2) and/or loss of repressive H3 trimethylation at lysine 27 (H3K27me3). These changes unlock the expression of stem cell genes that are normally silenced during early development. In mice, loss of H1c and H1e (also known as H1f2 and H1f4, respectively) conferred germinal centre B cells with enhanced fitness and self-renewal properties, ultimately leading to aggressive lymphomas with an increased repopulating potential. Collectively, our data indicate that H1 proteins are normally required to sequester early developmental genes into architecturally inaccessible genomic compartments. We also establish H1 as a bona fide tumour suppressor and show that mutations in H1 drive malignant transformation primarily through three-dimensional genome reorganization, which leads to epigenetic reprogramming and derepression of developmentally silenced genes.

Aims/UNASSIGNED:Histology, the microscopic study of normal tissues, is a crucial element of most medical curricula. Learning tools focused on histology are very important to learners who seek diagnostic competency within this important diagnostic arena. Recent developments in machine learning (ML) suggest that certain ML tools may be able to benefit this histology learning platform. Here, we aim to explore how one such tool based on a convolutional neural network, can be used to build a generalizable multi-classification model capable of classifying microscopic images of human tissue samples with the ultimate goal of providing a differential diagnosis (a list of look-alikes) for each entity. Methods/UNASSIGNED:We obtained three institutional training datasets and one generalizability test dataset, each containing images of histologic tissues in 38 categories. Models were trained on data from single institutions, low quantity combinations of multiple institutions, and high quantity combinations of multiple institutions. Models were tested against withheld validation data, external institutional data, and generalizability test images obtained from Google image search. Performance was measured with macro and micro accuracy, sensitivity, specificity, and f1-score. Results/UNASSIGNED:In this study, we were able to show that such a model's generalizability is dependent on both the training data source variety and the total number of training images used. Models which were trained on 760 images from only a single institution performed well on withheld internal data but poorly on external data (lower generalizability). Increasing data source diversity improved generalizability, even when decreasing data quantity: models trained on 684 images, but from three sources improved generalization accuracy between 4.05% and 18.59%. Maintaining this diversity and increasing the quantity of training images to 2280 further improved generalization accuracy between 16.51% and 32.79%. Conclusions/UNASSIGNED:This pilot study highlights the significance of data diversity within such studies. As expected, optimal models are those that incorporate both diversity and quantity into their platforms.s.

Introduction: T-cell Acute Lymphoblastic Leukemia (T-ALL) is an aggressive leukemia with a high incidence in children, adolescents and young adults. Although multiple therapeutic options are available, almost one fifth of patients affected by T-ALL eventually succumb to the disease, suggesting an unrecognized biological complexity that might contribute to drug resistance. To better understand the differences between T-ALL subtypes and their clinical course, we systematically analyzed cohorts of patients with different risk profiles at the genetic and epigenetic level. We recently demonstrated that differences in three-dimensional (3D) chromatin architecture can influence the integrity of topologically associating domains (TADs) and rewire specific enhancer-promoter interactions, impacting gene expression and leading to disease. As an example, we focused in particular on the Myc family of oncogenes, revealing disease-specific patterns of enhancer-promoter interactions.
Method(s): We initially profiled a large cohort of T-ALL patients falling under different risk categories in order to identify differences in their genetic and epigenetic features. We then systematically integrated matched in situ Hi-C, RNA-seq and CTCF ChIP-seq datasets to reveal widespread differences in intra-TAD chromatin interactions and TAD boundary insulation in patients affected by TALL.
Result(s): Using primary acute leukemia patient samples, we have, for the first time, identified recurrent TAD disruptions in leukemia involving key oncogenes (e.g. NOTCH1, c-Myc) and their targets. For example, we identified a recurrent disruption of 3D chromatin topology in the c-Myc locus at a previously uncharacterized non-coding CTCF-bound region that insulates MYC from a downstream super-enhancer. This disruption enables chromatin interactions between the c-Myc oncogene and the downstream super-enhancer leading to an increase in c-Myc expression. In parallel, while focusing patients falling into a higher risk category, namely the early T cell progenitor acute lymphoblastic leukemia (ETP-ALL), we discovered a previously uncharacterized region of high activity encompassing a novel lncRNA interacting with the proto-oncogene N-Myc.
Conclusion(s):With the current study we demonstrated an inherent difference between subtypes of T-ALL based on their epigenetic profile, which in turn influences the expression of key oncogenes. By focusing on a new methods of regulating a known family of transcription factors, we provide a new mechanism which could open interesting ways for targeted therapy of patients at different risk levels