February 1, 2017 (Vol. 37, No. 3)

Richard A. A. Stein M.D., Ph.D.

Advances Have Made It Possible to Generate Vast Datasets While Decreasing Costs and Reducing Timeframes

One of the most significant advances that has occurred in life sciences over the last few decades is the development of novel technologies and platforms that have made it possible to generate vast datasets while decreasing costs and reducing timeframes. As part of this process, fields that were historically viewed as unrelated have become indispensable components of vibrant interdisciplinary platforms. These advances catalyzed both novel frameworks to interrogate the molecular bases of human development and disease, and the search for therapeutic compounds.

A potent combination of disciplines—human biology plus chemistry plus chemical biology—could advance our understanding of disease, as well as our ability to treat it. This possibility motivates the work of Stuart L. Schreiber, Ph.D., the Morris Loeb Professor of Chemistry and Chemical Biology at Harvard University and a founding core member of the Broad Institute.

In a recent multidisciplinary effort, Dr. Schreiber and colleagues sought to identify new antimalarial compounds with novel mechanisms of action. The urgent need for new compounds to treat malaria stems from the scarcity of the existing therapeutic options and from the high and increasing resistance rates. As part of this work, Dr. Schreiber and colleagues interrogated a collection of about 100,000 synthetic compounds that have been underrepresented in routinely used databases.

“Our approach differed from that of most other efforts in this area,” says Dr. Schreiber. “Rather than acquiring by purchase a compound collection to screen, we ‘handmade’ every individual compound. We gave them special chemical features that we imagined would be useful for binding novel targets.”

The 3D structures that were included in this collection were inspired not by specific natural compounds, but by the diversity of the repertoire of natural products. This approach helped identify novel potential antimalarial compounds with new mechanisms of action, and validated the use of this collection as a valuable resource for drug discovery.

“In the past, we assembled massive collections of compounds while maintaining the reasonable expectation that some of them would do what we wanted,” comments Dr. Schreiber. “But the vast majority of compounds people use today do nothing at all, and those that do something, do the same thing over and over again.”

Chemistry and chemical biology are ideally positioned to speed the discovery of compounds with new mechanisms of action. These efforts require interdisciplinary undertakings in which investigators from several fields collaborate and bring expertise that spans multiple techniques and approaches.

“There is so much to learn from human biology to understand what needs to be fixed in some of the most daunting diseases,” concedes Dr. Schreiber. “But I am confident that with the current pace of progress, we will have our hands full of precise blueprints for achieving good answers in medicine.”


Cell Modeling Gets Organized

“The biggest thing that needs to happen in the next few years is a more extensive interoperability of the information that is obtained from high-content screening and analysis,” insists Robert F. Murphy, Ph.D., the Ray and Stephanie Lane Professor of Computational Biology and professor of biological sciences, biomedical engineering, and machine learning at Carnegie Mellon University. Dr. Murphy’s laboratory is focusing on developing models of subcellular organization from images and incorporating them into complex simulations of cellular behaviors.

Cells use a significant amount of energy to maintain their 3D organization and compartmentalization. All the expense and upkeep presumably has a purpose. Accordingly, cell models that fail to account for organizational and architectural details will most likely miss the complexity of cellular dynamics and the intricacy of perturbations occurring in response to external or internal factors.

Initial efforts to describe the cellular architecture have been based on descriptive models, which take into account the features and the state of a particular cell. However, in order to capture variation and to provide predictive power, models have to be generative—they must allow new parameters to be incorporated.

When descriptive features are used in the analysis of microscopy images, one of the challenges is to compare and integrate data across experiments, particularly when specific features captured using different experimental platforms may have different meanings for different investigators. “One potential way to address this is to make the features interpretable,” suggests Dr. Murphy. “But that can be impossible if people use different microscopes, conditions, and objectives—and often different cells.”

Efforts to develop generative models of cellular organization and protein distribution from fluorescence microscopy images have been undertaken in Dr. Murphy’s laboratory. These efforts have led to the development of an open-source platform, the CellOrganizer project.

“What is critical in any imaging experiment,” asserts Dr. Murphy, “is to try to convert the image into a model of what was visualized in that experiment.” Generative models provide the opportunity to compare models created with different microscopes, under different conditions, and in different labs.

“We want the ability to combine information from different experiments,” declares Dr. Murphy. “The larger goal, however, is to be able to compare, in a meaningful way, information collected from experiments on different cell types.”

The CellOrganizer can capture heterogeneities in the spatial distribution, size, and quantity of different parameters within a cellular population. Based on such measurements, the platform can generate quantitatively realistic synthetic images that reflect the underlying cell population. This provides interoperability or transferability, which enables comparisons and integration across different labs, cell types, and tissue types.

“While we pursue this goal,” informs Dr. Murphy, “we are also focusing on the ability to fine-tune different measurements.” For example, even though imaging can reveal that lysosomes differ among cells, it can still miss details such as differences in number, size, or spatial organization. “The basic idea is not only to compare better, but also to interpret in terms of specific processes that are being affected,” explains Dr. Murphy.

Another challenge that can be addressed through generative models is that, with some exceptions, it is very difficult to simultaneously image many proteins, organelles, or markers in the same cell. “As of right now,” complains Dr. Murphy, “we cannot measure more than a couple of things at the same time in the same living cell.” Some multiplexed or cyclic techniques allow tens or hundreds of things to be measured at the same time. However, these techniques, Dr. Murphy points out, have “some very significant limitations.”

Generative models of an entire cell require information from different experiments to be combined and integrated. This involves measuring the components in the same cell, developing a model of out how they work together, performing the same measurements for other components, and assembling the model.

The ability to combine models, advises Dr. Murphy, is another example of interoperability. “A big challenge is understanding how components that were measured separately interact with each another.”

These principles also allow a model to be transferred from one cell type to another. For example, after building a model of mitochondria in one cell type, one can tentatively make assumptions about them for another cell type. In addition, after incorporating the model into the new cell type, one can make certain predictions and test them. “And if those predictions pan out,” contends Dr. Murphy, “the model is more broadly usable.”


At Carnegie Mellon University, the laboratory of Robert F. Murphy, Ph.D., focuses on developing models of subcellular organization from images and incorporating them into complex simulations of cellular behavior. The laboratory is developing CellOrganizer, a platform for the automated learning of models for cell shape; nuclear shape; chromatin texture; vesicular organelle size, shape, and position; and microtubule distribution.

Systems Biology Meets Pharmacology

“We are applying quantitative systems pharmacology (QSP) to programs in drug discovery and development,” says D. Lansing Taylor, Ph.D., Allegheny Foundation Professor of Computational and Systems Biology at the University of Pittsburgh and director of the University of Pittsburgh Drug Discovery Institute (UPDDI). QSP, a field that developed in recent years at the interface between systems biology and pharmacology, promises to provide a new platform for drug discovery and development.

The definition of this novel interdisciplinary field is somewhat fluid. “In our definition,” details Dr. Taylor, “QSP is the use of iterative and integrative computational and experimental methods to determine the mechanisms of disease progression and the mechanisms of the action of drugs on multiscale systems.”

The need for this novel framework became clear when investigators began to recognize the complexities inherent in the relationship between between genes and diseases. This relationship is shaped by processes that unfold at different levels, that is, at the genetic and epigenetic levels.

Experimentation remains a fundamental component of this discipline, though it may be incorporated in novel ways. “One can jump ahead and make some computational predictions, and then test them using experimental methods, and that is the fundamental basis of everything that we do,” explains Dr. Taylor.

In the QSP work performed in the UPDDI, phenotypic discovery represents the key platform in terms of experimental approaches. “The overarching concept,” comments Dr. Taylor, “is that we want to perform phenotypic discovery by building sophisticated human models of disease that incorporate metrics for heterogeneity.”

Using patient-derived induced pluripotent stem cells, Dr. Taylor and colleagues are generating microfluidic 3D liver models to study nonalcoholic fatty liver disease. A phenotypic hallmark of this highly prevalent condition is macrosteatosis, or the accumulation of lipid vesicles inside hepatocytes.

In phenotypic screening experiments, Dr. Taylor and colleagues seek to identify compounds that can reverse this phenotype. “Because we use phenotypic discovery, we do not preselect for a molecular target,” asserts Dr. Taylor. This model provides the opportunity to perform cell-based measurements and quantify heterogeneity, which historically has been underinvestigated. One of the challenges is that if an existing drug or a novel compound reverses the phenotype, the information is still not informative about the molecular targets.

To address the underlying mechanisms, Dr. Taylor’s laboratory worked with Nathan Yates, Ph.D., the scientific director of the University of Pittsburgh’s Biomedical Mass Spectrometry Center, to develop a chemical proteomics platform. According to Dr. Taylor, the platform can “predict the molecular targets with which the validated hit compounds are interacting.”


Modeling Takes in Heterogeneity

 “We need more relevant and more predictive cell models, and this is why we are moving toward more complex experimental systems,” says Christophe Antczak, Ph.D., laboratory head, Center for Proteomic Chemistry, Novartis Institutes for BioMedical Research.

While the 2D cell and tissue cultures that have been used for over a century have laid the foundations of cell biology, one of their shortcomings is that they do not recapitulate the complex 3D architecture of tissues. Consequently, the cultures fail to represent processes that occur in the 3D milieu, such as the bidirectional signaling between cells and the extracellular matrix that has been shown to shape development, homeostasis, and disease.

3D or organoid cultures, which recapitulate specific features of tissue architecture and/or function, have been developed to address this gap, and they have become valuable tools for drug discovery. However, despite providing a wealth of information, their readout can be difficult due to the multiple cell types that are present.

Dr. Antczak and colleagues, including Shanni Chen, Vincent Yu, Sandeep Daya, Kevin White, and Carla Bauer from Novartis Institutes for Biomedical Research, used an in vitro model of the respiratory alveoli to understand cellular heterogeneities in the lung. “This model managed to recapitulate tissue organization and function up to a point,” details Dr. Antczak. “It was much more heterogeneous than our traditionally models, which grow cells of single type in monolayers.”

Using the new model, Dr. Antczak and colleagues developed confocal image analysis methods to characterize the heterogeneity that is intrinsic to organotypic cultures derived from pluripotent cells. “Thanks to image quantification,” explains Dr. Antczak, “we found that depending on where in the well spheroids are located, they may mature in different ways.” The more mature spheroids, which constitute the population of interest for these studies, seemed to be on the top of the well.

“To confirm this, we had to systematically perform unbiased image quantification,” recalls Dr. Antczak. These proof-of-principle experiments have provided a framework for similar studies that may facilitate the interrogation of heterogeneity in other tissues.

“Many people think of high content and image quantification as something that is performed during screening,” observes Dr. Antczak. “But in this case, it is something that we performed very early on, before even further developing the assay for screening.”




























This site uses Akismet to reduce spam. Learn how your comment data is processed.