Debashis Sahoo

Background Adenoid cystic carcinoma (ACC) is a lethal malignancy of exocrine glands, characterized by the coexistence within tumor tissues of 2 distinct populations of cancer cells, phenotypically similar to the myoepithelial and ductal lineages of normal salivary epithelia. The developmental relationship linking these 2 cell types, and their differential vulnerability to antitumor treatments, remains unknown. Methods Using single-cell RNA sequencing, we identified cell-surface markers (CD49f, KIT) that enabled the differential purification of myoepithelial-like (CD49fhigh/KITneg) and ductal-like (CD49flow/KIT+) cells from patient-derived xenografts (PDXs) of human ACCs. Using prospective xenotransplantation experiments, we compared the tumor-initiating capacity of the 2 cell types and tested whether one could differentiate into the other. Finally, we searched for signaling pathways with differential activation between the 2 cell types and tested their role as lineage-specific therapeutic targets. Results Myoepithelial-like cells displayed higher tumorigenicity than ductal-like cells and acted as their progenitors. Myoepithelial-like and ductal-like cells displayed differential expression of genes encoding for suppressors and activators of retinoic acid signaling, respectively. Agonists of retinoic acid receptor (RAR) or retinoid X receptor (RXR) signaling (all-trans retinoic acid, bexarotene) promoted myoepithelial-to-ductal differentiation, whereas suppression of RAR/RXR signaling with a dominant-negative RAR construct abrogated it. Inverse agonists of RAR/RXR signaling (BMS493, AGN193109) displayed selective toxicity against ductal-like cells and in vivo antitumor activity against PDX models of human ACC. Conclusions In human ACCs, myoepithelial-like cells act as progenitors of ductal-like cells, and myoepithelial-to-ductal differentiation is promoted by RAR/RXR signaling. Suppression of RAR/RXR signaling is lethal to ductal-like cells and represents a new therapeutic approach against human ACCs.

Deep learning is highly pervasive in today's data-intensive era. In particular, convolutional neural networks (CNNs) are being widely adopted in a variety of fields for superior accuracy. However, computing deep CNNs on traditional CPUs and GPUs brings several performance and energy pitfalls. Several novel approaches based on ASIC, FPGA, and resistive-memory devices have been recently demonstrated with promising results. Most of them target only the inference (testing) phase of deep learning. There have been very limited attempts to design a full-fledged deep learning accelerator capable of both training and inference. It is due to the highly compute- and memory-intensive nature of the training phase. In this article, we propose LiteCON , a novel analog photonics CNN accelerator. LiteCON uses silicon microdisk-based convolution, memristor-based memory, and dense-wavelength-division-multiplexing for energy-efficient and ultrafast deep learning. We evaluate LiteCON using a commercial CAD framework (IPKISS) on deep learning benchmark models including LeNet and VGG-Net. Compared to the state of the art, LiteCON improves the CNN throughput, energy efficiency, and computational efficiency by up to 32×, 37×, and 5×, respectively, with trivial accuracy degradation.