Faculty Bibliography

In response to nutrient deprivation, the cell mobilizes an extensive amount of membrane to form and grow the autophagosome, allowing the progression of autophagy. By providing membranes and stimulating LC3 lipidation, COPII (Coat Protein Complex II) promotes autophagosome biogenesis. Here, we show that the F-box protein FBXW5 targets SEC23B, a component of COPII, for proteasomal degradation and that this event limits the autophagic flux in the presence of nutrients. In response to starvation, ULK1 phosphorylates SEC23B on Serine 186, preventing the interaction of SEC23B with FBXW5 and, therefore, inhibiting SEC23B degradation. Phosphorylated and stabilized SEC23B associates with SEC24A and SEC24B, but not SEC24C and SEC24D, and they re-localize to the ER-Golgi intermediate compartment, promoting autophagic flux. We propose that, in the presence of nutrients, FBXW5 limits COPII-mediated autophagosome biogenesis. Inhibition of this event by ULK1 ensures efficient execution of the autophagic cascade in response to nutrient starvation.

Abdominal aortic aneurysms (AAA) are characterized by extensive extracellular matrix (ECM) fragmentation and inflammation. However, the mechanisms by which these events are coupled thereby fueling focal vascular damage are undefined. Here we report through single-cell RNA-sequencing of diseased aorta that the neuronal guidance cue netrin-1 can act at the interface of macrophage-driven injury and ECM degradation. Netrin-1 expression peaks in human and murine aneurysmal macrophages. Targeted deletion of netrin-1 in macrophages protects mice from developing AAA. Through its receptor neogenin-1, netrin-1 induces a robust intracellular calcium flux necessary for the transcriptional regulation and persistent catalytic activation of matrix metalloproteinase-3 (MMP3) by vascular smooth muscle cells. Deficiency in MMP3 reduces ECM damage and the susceptibility of mice to develop AAA. Here, we establish netrin-1 as a major signal that mediates the dynamic crosstalk between inflammation and chronic erosion of the ECM in AAA.

Visual inspection of histopathology slides is one of the main methods used by pathologists to assess the stage, type and subtype of lung tumors. Adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) are the most prevalent subtypes of lung cancer, and their distinction requires visual inspection by an experienced pathologist. In this study, we trained a deep convolutional neural network (inception v3) on whole-slide images obtained from The Cancer Genome Atlas to accurately and automatically classify them into LUAD, LUSC or normal lung tissue. The performance of our method is comparable to that of pathologists, with an average area under the curve (AUC) of 0.97. Our model was validated on independent datasets of frozen tissues, formalin-fixed paraffin-embedded tissues and biopsies. Furthermore, we trained the network to predict the ten most commonly mutated genes in LUAD. We found that six of them-STK11, EGFR, FAT1, SETBP1, KRAS and TP53-can be predicted from pathology images, with AUCs from 0.733 to 0.856 as measured on a held-out population. These findings suggest that deep-learning models can assist pathologists in the detection of cancer subtype or gene mutations. Our approach can be applied to any cancer type, and the code is available at https://github.com/ncoudray/DeepPATH .

Visual inspection of histopathology slides is one of the main methods used by pathologists to assess the stage, type and subtype of lung tumors. Adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) are the most prevalent subtypes of lung cancer, and their distinction requires visual inspection by an experienced pathologist. In this study, we trained a deep convolutional neural network (inception v3) on whole-slide images obtained from The Cancer Genome Atlas to accurately and automatically classify them into LUAD, LUSC or normal lung tissue. The performance of our method is comparable to that of pathologists, with an average area under the curve (AUC) of 0.97. Our model was validated on independent datasets of frozen tissues, formalin-fixed paraffin-embedded tissues and biopsies. Furthermore, we trained the network to predict the ten most commonly mutated genes in LUAD. We found that six of them-STK11, EGFR, FAT1, SETBP1, KRAS and TP53-can be predicted from pathology images, with AUCs from 0.733 to 0.856 as measured on a held-out population. These findings suggest that deep-learning models can assist pathologists in the detection of cancer subtype or gene mutations. Our approach can be applied to any cancer type, and the code is available at https://github.com/ncoudray/DeepPATH .

Homologous recombination (HR) is a crucial pathway for the repair of DNA double-strand breaks. BRCA1/2 breast cancer proteins are key players in HR via their mediation of RAD51 nucleofilament formation and function; however, their individual roles and crosstalk in vivo are unknown. Here we use super-resolution (SR) imaging to map the spatiotemporal kinetics of HR proteins, revealing the interdependent relationships that govern the dynamic interplay and progression of repair events. We show that initial single-stranded DNA/RAD51 nucleofilament formation is mediated by RAD52 or, in the absence of RAD52, by BRCA2. In contrast, only BRCA2 can orchestrate later RAD51 recombinase activity during homology search and resolution. Furthermore, we establish that upstream BRCA1 activity is critical for BRCA2 function. Our analyses reveal the underlying epistatic landscape of RAD51 functional dependence on RAD52, BRCA1, and BRCA2 during HR and explain the phenotypic similarity of diseases associated with mutations in these proteins.

BACKGROUND:Although large-scale, next-generation sequencing (NGS) studies of cancers hold promise for enabling precision oncology, challenges remain in integrating NGS with clinically validated biomarkers. METHODS:To overcome such challenges, we utilized the Database of Evidence for Precision Oncology (DEPO) to link druggability to genomic, transcriptomic, and proteomic biomarkers. Using a pan-cancer cohort of 6570 tumors, we identified tumors with potentially druggable biomarkers consisting of drug-associated mutations, mRNA expression outliers, and protein/phosphoprotein expression outliers identified by DEPO. RESULTS:Within the pan-cancer cohort of 6570 tumors, we found that 3% are druggable based on FDA-approved drug-mutation interactions in specific cancer types. However, mRNA/phosphoprotein/protein expression outliers and drug repurposing across cancer types suggest potential druggability in up to 16% of tumors. The percentage of potential drug-associated tumors can increase to 48% if we consider preclinical evidence. Further, our analyses showed co-occurring potentially druggable multi-omics alterations in 32% of tumors, indicating a role for individualized combinational therapy, with evidence supporting mTOR/PI3K/ESR1 co-inhibition and BRAF/AKT co-inhibition in 1.6 and 0.8% of tumors, respectively. We experimentally validated a subset of putative druggable mutations in BRAF identified by a protein structure-based computational tool. Finally, analysis of a large-scale drug screening dataset lent further evidence supporting repurposing of drugs across cancer types and the use of expression outliers for inferring druggability. CONCLUSIONS:Our results suggest that an integrated analysis platform can nominate multi-omics alterations as biomarkers of druggability and aid ongoing efforts to bring precision oncology to patients.

Transposable elements (TEs) represent a substantial fraction of many eukaryotic genomes, and transcriptional regulation of these factors is important to determine TE activities in human cells. However, due to the repetitive nature of TEs, identifying transcription factor (TF)-binding sites from ChIP-sequencing (ChIP-seq) datasets is challenging. Current algorithms are focused on subtle differences between TE copies and thus bias the analysis to relatively old and inactive TEs. Here we describe an approach termed "MapRRCon" (mapping repeat reads to a consensus) which allows us to identify proteins binding to TE DNA sequences by mapping ChIP-seq reads to the TE consensus sequence after whole-genome alignment. Although this method does not assign binding sites to individual insertions in the genome, it provides a landscape of interacting TFs by capturing factors that bind to TEs under various conditions. We applied this method to screen TFs' interaction with L1 in human cells/tissues using ENCODE ChIP-seq datasets and identified 178 of the 512 TFs tested as bound to L1 in at least one biological condition with most of them (138) localized to the promoter. Among these L1-binding factors, we focused on Myc and CTCF, as they play important roles in cancer progression and 3D chromatin structure formation. Furthermore, we explored the transcriptomes of The Cancer Genome Atlas breast and ovarian tumor samples in which a consistent anti-/correlation between L1 and Myc/CTCF expression was observed, suggesting that these two factors may play roles in regulating L1 transcription during the development of such tumors.