8th Annual Conference
2018

Presentation Abstract

Flow Cytometry and Small Molecule Drug Development in Pharma: if Flow is so Good, How is Pharma using it?

Pharma is facing increasing challenges: the cost to develop and market new small molecule drugs is steadily growing, increasing numbers of currently marketed pharmaceuticals are coming off-patent, and there is a rising societal pressure to lower the price of prescription drugs. A recent paper (DOI: 10.1056/NEJMp1500848) indicates that the current cost to develop a single new drug is approaching $2.6 billion, up from $800 million estimated in 2003. In this same time period, most large Pharma companies have significantly reduced R&D funding and decreased their research staff. Anecdotal evidence suggests that the majority of cell-based assays performed during drug development employ image analysis and that many of these image-based assays use whole field/population signal averaging rather than single-cell analyses. Published data also show that the majority of image-based tests used by Pharma operate measuring 2-3 independent features. The status of flow cytometry in Pharma is less clear, as the industrial R&D groups may not publish their most valuable data. That said, published studies show the use of up to 4-6 dimensions in flow cytometry based drug-development assays. With limited exception, computational analysis of the results is generally limited to uncomplicated univariate or bivariate techniques.

During the past five years, I worked with a start-up CRO (AsedaSciences) to develop multi-parametric flow-based assays for early assessment of compound risk. The screen employs state-of-the-art hardware such as auto-samplers, robotic liquid handling systems, and multiparametric flow cytometry. Simultaneously measuring multiple cellular responses across a range of compound concentrations, the measured biological features are organized in tensors of numerical values, jointly describing dissimilarities between controls and measured samples in biological feature space. The resultant tensors characterize the tested compounds and can be compressed, compared, and used as inputs for predictive machine learning algorithms. This approach departs from the tradition of representing compound “toxicity” as EC50 (logarithm of half maximal effective concentration) for each individual phenotypic marker. Instead, the method integrates all measurements to produce single values representing the likelihood of specific cellular stress. Abandoning the univariate cell-stress/compound toxicity metric via EC50 and adopting a genuinely multivariate representation of cell responses facilitates assignment of probabilistic scores to compound-cell interaction fingerprints. This provides a framework for quantitative evaluation and validation of assay performance via sensitivity and specificity measures typically used for diagnostic tests. Using this approach, the assay recognizes ~50% of compounds present in the test database which “failed” (included compounds that failed for multiple reasons in more advanced stages of development , e.g. organ-specific toxicities, DILI, toxic metabolites, efficacy, etc.). The assay also offers remarkable precision (positive predictive value) of 98%. The described work demonstrates an example of a broader philosophy of cellular-stress testing, emphasizing simplicity and reproducibility paired with sophisticated computational analysis and machine learning. I will argue that the future of drug development will depend on the broader use of massively parallel and machine-learning-aided screening systems, coupled to well validated and reproducible assays. This philosophy emphasizing single cell analysis implicitly relies on cell population heterogeneity to characterize compounds in early phases of drug development and differs dramatically from costly and complex tissue and animal models.

Presentation Abstract

Imaging Non-conical New Roles for Natural Killers in Health and Disease

Natural killer (NK) cells are conventionally valued for their ability to make interferon-gamma and to kill virus-infected or cancerous cells. Our work reveals new functional roles for NK cells in regulating the duration, robustness, and character of immune responses. On one hand, we show that NK cells respond to inflammation and immunopathology during chronic virus infection by up-regulating expression of BAFF (B-cell activating factor belonging to the tumor necrosis factor family) and concomitantly providing support to maintain marginal zone macrophages and B cells. Chronic infection in mice devoid of NK cells results in complete and sustained loss of the marginal zone, which is associated with increased susceptibility to secondary bacterial infection. These data reveal a crucial and previously unappreciated role for NK cells in sustaining immune function during chronic inflammation. In contrast, we find that NK cells can also suppress adaptive immune function via perforin-dependent cytolytic targeting of activated CD4 T cells. This activity constrains development of follicular helper T cell and germinal center B cell responses following immunization. The reduced magnitude of germinal center responses is linked to decreased vaccine-elicited antibody titers and limited affinity maturation of antibodies. NK cell suppression of the germinal center is associated with transient localization of NK cells in the splenic white pulp. This discovery reveals that targeting of NK cell immunoregulatory function may enhance efforts to promote vaccine-elicited production of high affinity antibodies against pathogens like HIV, for which we do not have effective vaccine regimens.