Research Areas

Fluorescently labeled and bright field images of spindle-DU145 and tight-DU145 in culture. In-vivo experiment and H&E stained tissue.

Prostate Cancer Biology and Targeted Therapeutics

CDCP1-Regulated Cancer Cell Adhesion and Survival and Therapeutic Vulnerability During Prostate Cancer Progression
The CUB-domain containing protein 1, CDCP1, is a transmembrane glycoprotein that is able to sequester SRC and PKCδ kinases into unique microdomains within the plasma membrane. CDCP1 can either promote or suppress tumor metastasis, dependent on the cancer type and experimental system. The molecular causes that lead to the controversial function of CDCP1 are not well understood. We demonstrated, for the first time in patients with prostate cancer, a significant reduction of CDCP1 protein expression in metastatic relative to primary tumor tissues, concordant with gene expression levels in ONCOMINE and TCGA data sets. To investigate how the loss of CDCP1 facilitates prostate cancer metastasis, we determined the consequences of CDCP1 loss in nonadherent cancer cells, which provide an experimental model system of circulating tumor cells (CTC). Via functional assays, co-immunoprecipitation and drug studies, we discovered a new mechanism of regulation of CDK5 in prostate cancer cells, which leads to the activation of focal adhesion kinase (FAK) independent of β1-integrin. The potential biological and clinical consequences of this mechanism are (1) a switch from cell-matrix to cell-cell adhesion, (2) increased sensitivity of CTCs to FAK inhibitors, and (3) improved adaptation and survival of cancer cells in circulation and at the metastatic sites.

Pathway downstream of CDCP1 loss in prostate cancer cells that have lost adhesion.

Molecular and Computational Pathology for Clinical Biomarker Development

Multiplex Immunohistochemistry for Analysis of DNA Damage Repair in Tumor Cells
The prognosis for men with castration-resistant prostate cancer (CRPC) is dismal. A subset of patients presents with anaplastic carcinoma, characterized by low serum PSA levels, minimal response to androgen deprivation therapy (ADT) and a large visceral tumor burden. These patients respond transiently to platinum-based chemotherapy, suggesting a defect in DNA damage-repair mechanisms. Multiple DNA damage-repair deficiencies have been identified through genomic sequencing. However, clinical studies demonstrate that mutations and genomic rearrangements cannot identify all patients who respond to cisplatinum or PARP1 inhibitors, revealing a need for functional assays. In collaboration with investigators at the University of Arizona and at Jefferson University, we are developing multiplex antibody-based immunofluorescence and immunohistochemistry methods to measure the amount of homologous recombination, nonhomologous end-joining, and PARP1 activity in cancer tissues. This work is supported by a UCLA SPORE Developmental Research Award.

Digital Image Analysis and Computational Pathology
The overall goal of the digital image analysis program is to extract numerical information from digital images that are derived from human tissues and integrate measurements with genomic, gene expression and proteomic data. A rich archive of cases in the pathology department allows the training of algorithms that can be deployed to TCGA and other cohorts of digital images for biomarker extraction.

In collaboration with Dr. Arkadiusz Gertych, who oversees the image processing and software development for digital image analysis in the BioImage Informatics Lab, we train machine-learning algorithms that recognize cell types in hematoxylin and eosin (H&E)-stained tissue sections based on their specific protein expression and analyze the organization and connectivity of these cells. As a proof-of-principle project, this approach has led to the extraction of vascular network biomarkers in clear cell renal cancer.

Another goal of the digital image analysis program is to map and quantify growth patterns of cancers using targeted and handcrafted approaches, as well as deep learning approaches. The output of this analysis has versatile applications. It is used to generate tumor masks in slides stained with multiple antibodies, delineate cancer regions for subsequent analysis of nuclear morphology and integrate tumor architecture with clinical parameters to predict the danger of cancer recurrence.