Interpretable deep learning uncovers cellular properties in label-free live cell images that are predictive of highly metastatic melanoma

Deep learning has emerged as the technique of choice for identifying hidden patterns in cell imaging data but is often criticized as "black box." Here, we employ a generative neural network in combination with supervised machine learning to classify patient-derived melanoma xenografts as "efficient" or "inefficient" metastatic, validate predictions regarding melanoma cell lines with unknown metastatic efficiency in mouse xenografts, and use the network to generate in silico cell images that amplify the critical predictive cell properties. These exaggerated images unveiled pseudopodial extensions and increased light scattering as hallmark properties of metastatic cells. We validated this interpretation using live cells spontaneously transitioning between states indicative of low and high metastatic efficiency. This study illustrates how the application of artificial intelligence can support the identification of cellular properties that are predictive of complex phenotypes and integrated cell functions but are too subtle to be identified in the raw imagery by a human expert. A record of this paper's transparent peer review process is included in the supplemental information. VIDEO ABSTRACT.

Targeting Oncogene mRNA Translation in B-Cell Malignancies with eFT226, a Potent and Selective Inhibitor of eIF4A

The PI3K/AKT/mTOR pathway is often activated in lymphoma through alterations in PI3K, PTEN, and B-cell receptor signaling, leading to dysregulation of eIF4A (through its regulators, eIF4B, eIF4G, and PDCD4) and the eIF4F complex. Activation of eIF4F has a direct role in tumorigenesis due to increased synthesis of oncogenes that are dependent on enhanced eIF4A RNA helicase activity for translation. eFT226, which inhibits translation of specific mRNAs by promoting eIF4A1 binding to 5'-untranslated regions (UTR) containing polypurine and/or G-quadruplex recognition motifs, shows potent antiproliferative activity and significant in vivo efficacy against a panel of diffuse large B-cell lymphoma (DLBCL), and Burkitt lymphoma models with ≤1 mg/kg/week intravenous administration. Evaluation of predictive markers of sensitivity or resistance has shown that activation of eIF4A, mediated by mTOR signaling, correlated with eFT226 sensitivity in in vivo xenograft models. Mutation of PTEN is associated with reduced apoptosis in vitro and diminished efficacy in vivo in response to eFT226. In models evaluated with PTEN loss, AKT was stimulated without a corresponding increase in mTOR activation. AKT activation leads to the degradation of PDCD4, which can alter eIF4F complex formation. The association of eFT226 activity with PTEN/PI3K/mTOR pathway regulation of mRNA translation provides a means to identify patient subsets during clinical development.

Design of Development Candidate eFT226, a First in Class Inhibitor of Eukaryotic Initiation Factor 4A RNA Helicase

Dysregulation of protein translation is a key driver for the pathogenesis of many cancers. Eukaryotic initiation factor 4A (eIF4A), an ATP-dependent DEAD-box RNA helicase, is a critical component of the eIF4F complex, which regulates cap-dependent protein synthesis. The flavagline class of natural products (i.e., rocaglamide A) has been shown to inhibit protein synthesis by stabilizing a translation-incompetent complex for select messenger RNAs (mRNAs) with eIF4A. Despite showing promising anticancer phenotypes, the development of flavagline derivatives as therapeutic agents has been hampered because of poor drug-like properties as well as synthetic complexity. A focused effort was undertaken utilizing a ligand-based design strategy to identify a chemotype with optimized physicochemical properties. Also, detailed mechanistic studies were undertaken to further elucidate mRNA sequence selectivity, key regulated target genes, and the associated antitumor phenotype. This work led to the design of eFT226 (Zotatifin), a compound with excellent physicochemical properties and significant antitumor activity that supports clinical development.

Abstract 2698: eFT226, a potent and selective inhibitor of eIF4A, is efficacious in preclinical models of lymphoma

Dysregulated messenger RNA (mRNA) translation drives the pathogenesis of multiple hematological malignancies. In lymphoma this includes the upregulation of key driver oncogenes and anti-apoptotic proteins (e.g., MYC, CCND1/3, BCL2 and MCL1) that contain a highly structured 5’-untranslated region (UTR) in their mRNA requiring enhanced eIF4A helicase activity for translation. eIF4A is a component of the eIF4F translation initiation complex and catalyzes the ATP-dependent unwinding of RNA duplexes and facilitates 43S ribosome scanning within the 5’-UTR. The activation of oncogenic signaling pathways, including RAS and PI3K, enhance eIF4A activity through phosphorylation of eIF4B, eIF4G and PDCD4 which facilitates formation of eIF4F and full activation of eIF4A. The PI3K/AKT/mTOR pathway is frequently activated in lymphoma, promoting the translation of oncogenes with complex 5’-UTRs that are required for tumor cell proliferation, survival and metastasis.

eFT226 is a potent and sequence selective eIF4A1 inhibitor that promotes eIF4A1 binding to specific 5’-UTR polypurine and/or G-quadraplex recognition motifs leading to a selective block in ribosome mRNA scanning. The sequence dependency of eFT226 translational inhibition was evaluated in cell-based reporter assays demonstrating >100-fold greater sensitivity for reporter constructs containing a polypurine motif in the 5’-UTR (IC50 ~2 nM). Direct binding studies also confirmed the formation of a stable ternary complex with increased drug residence time between eFT226, eIF4A1 and RNA oligonucleotides containing polypurine motifs. The ability of eFT226 to inhibit MYC or MCL1 expression was found to be dependent on the presence of their respective 5’-UTR supporting a translational regulation mechanism dependent on recognition elements within the 5’-UTR.

eFT226 shows potent anti-proliferative activity (GI50 < 15 nM) against a panel of B-cell lymphoma cell lines. Treatment with eFT226 leads to coordinated inhibition of MYC, CCND1/3, BCL2 or MCL1 protein expression resulting in significant anti-tumor activity. eFT226 has good pharmacokinetic properties and exhibits significant in vivo activity across a panel of diffuse large B cell lymphoma (DLBCL), and Burkitt lymphoma tumor models with ≤1 mg/kg/week IV administration. Further evaluation of predictive markers of sensitivity or resistance has shown that tumors with mTOR mediated activation of eIF4A are most sensitive to eFT226. In addition, tumors with PTEN mutations do not exhibit activated eIF4A and are generally resistant to induction of apoptosis by eFT226, resulting in reduced in vivo efficacy. The association of eFT226 activity with PI3K/mTOR pathway activation and mutational status provides a means to identify patient subsets during clinical development. Clinical trials in patients with lymphoma and other malignancies are planned.

Citation Format: Peggy A. Thompson, Boreth Eam, Nathan P. Young, Sarah Fish, Joan Chen, Maria Barrera, Haleigh Howard, Eric Sung, Ana Parra, Jocelyn Staunton, Gary G. Chiang, Christopher J. Wegerski, Andres Nevarez, Jeff Clarine, Samuel Sperry, Alan Xiang, Chinh Tran, Christian Nilewski, Garrick K. Packard, Theodore Michels, Paul A. Sprengeler, Justin T. Ernst, Siegfried H. Reich, Kevin R. Webster. eFT226, a potent and selective inhibitor of eIF4A, is efficacious in preclinical models of lymphoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 2698.