NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Institute of Medicine (US) Forum on Drug Discovery, Development, and Translation. Emerging Safety Science: Workshop Summary. Washington (DC): National Academies Press (US); 2008.

Cover of Emerging Safety Science

Emerging Safety Science: Workshop Summary.

Show details

3Screening Technologies I: Human Cell–Based Approaches

Much of the emerging science described at the workshop centers on ways to screen drugs for potential safety problems as early as possible in the development process. In introducing the session on human cell–based approaches, David Jacobson-Kram, Associate Director for Pharmacology and Toxicology, Office of New Drugs in the U.S. Food and Drug Administration’s Center for Drug Evaluation and Research, noted that existing preclinical models and paradigms often fail to predict toxicology that appears later in development. Developing the tools to select better candidates is a challenging task. Therefore, much of the workshop was focused on how to find new and more effective ways to screen or predict toxicological outcomes of drugs. Speakers described a number of approaches to improve screening for candidates that could be applied at various points in the process, from basic research and discovery through clinical development. The first topic in this series was emerging screening technologies that are based on human cells.


Dr. Butcher outlined the characteristics of an ideal screen for drug evaluation:

  • Quantitative, reproducible, robust, and high-throughput—These characteristics would make it possible to carry out informatics correlations with clinical data.
  • Highly standardized—Highly standardized assays would be well suited to database generation and archiving. Standardized screens would make it possible to perform multiple comparisons among test groups, as well as to make comparisons over time. This is a particularly important feature, one that is missing in current efforts to model human biology.
  • Based on human biology—As it is not possible to use human beings, the next best option is to use human primary cells.
  • Of broad interest to many people—Assays should cover a wide range of biology and a large number of mechanisms of toxicity, including various targets, pathways, and diseases.
  • Integrative—Integrative assays would attract the interest of scientists from multiple disciplines, including, for example, biologists, chemists, clinicians, and safety scientists.
  • Predictive—Assays should predict safety, toxicology, efficacy, and clinical indications.

Although no such ideal screens exist today, researchers should keep this vision in mind as they work to develop new screens. The value of a screen will depend in large part on how closely it approaches this ideal.


Elaborating on BioSeek’s own efforts to develop an ideal screen, Butcher described the long-term goal as developing in vitro models that can predict in vivo biology. By developing a database that connects drug biology to clinical responses, BioSeek’s BioMAP system, based on human cells, can be used to provide an early prediction of which drug candidates are most likely to be developed as safe and effective therapeutics.

System Overview

Although BioMAP uses an artificial cell culture system, it is based on human cells placed in complex environments designed to reflect key aspects of the natural environments the cells would experience in the human body. Using information from the literature, the company’s scientists strive to create environments that mirror real situations in which multiple pathways are active at the same time—pathways similar to those believed to work together in different disease states. One cannot model in vitro biology by assaying a large number of individual targets because one cannot predict network biology from target and pathway biology. The BioSeek approach involves taking advantage of systems biology principles and engineering complexity into the system so as to model as closely as possible the biology that occurs in vivo in different disease physiological settings.

To create these complex environments, the BioMAP system combines a number of cells, sometimes of a single type and sometimes a defined mixture, such as peripheral blood mononuclear and endothelial cells. The purpose of using multiple cells is to capture cell–cell interactions and begin to model at a simple level what is going on at the tissue level. These systems are then stimulated by agents such as cytokines and growth factors to create an environment where a number of disease-relevant pathways are simultaneously active.

The development of these models is informed by in vivo data from the literature. The goal is to develop a model system that responds in a particular way to particular drugs because that is how the tissue responds in vivo. The process is an iterative one, with lessons learned at each step being applied to modify the system in useful ways.

In particular, the focus of the BioMAP system is on factors that reflect and control biology in vivo and that mediate disease, such as small receptors, cytokines, chemokines, enzymes, and growth factors. Because it is more cost effective, the system uses standard enzyme-linked immunosorbent assay (ELISA)–based or morphologic readouts rather than microarray data.

The experimental process follows a standard pattern. The assays are set up and stimulated, and the researchers then read off the various parameters in which they are interested. Next, the system is perturbed with a drug to generate a profile that is entered into a database. Through numerous iterations of this process, BioSeek has accumulated a database containing thousands of BioMAP profiles that catalogue the effects of thousands of different drugs in a variety of disease model systems.

BioSeek has developed at least 30 of these complex cell systems designed to model different aspects of disease and is actively working to develop additional systems. The types of cells used include, for example, primary endothelial cells, monocytes, lymphocytes, macrophages, mast cells, smooth muscle cells, keratinocytes, bronchial epithelial cells, and smooth muscle cells.

Table 3-1 illustrates the kinds of cell systems BioMAP employs. The first two systems contain primary endothelial cells. In the first, the cells are stimulated in a TH-1 environment with three cytokines added, creating an environment similar to one that might be seen in psoriasis or rheumatoid arthritis. In the second system, there is a TH-2 environment with IL-4 (interleukin-4) and histamine, resulting in an environment more relevant to asthma. The two other systems combine peripheral blood mononuclear and endothelial cells, stimulated through either (1) a selector that would selectively activate through a monocyte cascade that would activate many pathways, or (2) the T cell receptor.

TABLE 3-1 Examples of BioMAP Systems.


TABLE 3-1 Examples of BioMAP Systems.

These four systems alone, because they encompass most of the targets and mechanisms and pathways involved in inflammation, allow the BioSeek researchers to detect and discriminate compounds of basically every important immunomodulatory agent, as well as many molecules that are important in other biological and therapeutic areas, including cardiovascular disease, metabolism, and cancer. This broad range of applications is not surprising because even though these four cell systems were created to model inflammatory and cardiovascular states, the cells they contain express the receptors that cancer cells adapt and use for their own purposes, as well as many of the receptors involved in controlling metabolism and lipid biology. The resulting breadth of coverage of targets and pathways provides BioSeek with a unique opportunity to assess effects across a broad array of human biology in a common format.

BioMAP Profiles

Butcher provided an overview of how the BioMAP system is used in the development process. Once a compound has been received, it is run through a set of cell-based assays, all of which are performed using robotics and micro titer plates. The resulting profiles are reviewed for quality control and then archived. BioSeek researchers have developed visualization and knowledge management tools that allow them to correlate the profile data of a compound with any other information of interest. Chemical informatics is used for structure–activity relationship (SAR) analysis, as well as for literature mining.

An example is a profile of a p38 inhibitor that was tested in multiple cell systems (see Figure 3-1). All of the systems were normalized to a control level without the presence of the drug, and by stimulating each system with the test compound, the researchers established dose–response curves. Because these assays are performed by ELISA using robotics, it is possible to test hundreds of compounds in multiple assays each week, to repeat the tests a number of times, and therefore to collect enough data that the statistical analysis can be quite sophisticated. These statistics allow the researchers to calculate a 99 percent significance envelope, which is indicated in the figure by a gray background; anything outside of this envelope is a highly significant response.

FIGURE 3-1 Example of a BioMAP profile developed using a reference p38 inhibitor


FIGURE 3-1 Example of a BioMAP profile developed using a reference p38 inhibitor. The figure displays data from multiple experiments in which a p38 inhibitor was tested in several BioMAP cell systems. Readout parameters are listed at the bottom of the (more...)

With thousands of profiles, it is possible to look for similarities among them and thus identify compounds with similar responses in the cell systems. In particular, the BioSeek scientists analyze their profiles using a form of clustering known as multidimensional scaling. An example of such a clustering analysis is shown in Figure 3-2 (see p. 20). Each dot represents a profile; some profiles end up close together, indicating similarity, while others end up far apart. Since the graph is formed by collapsing 40-dimensional space into 2 dimensions, two dots being positioned close together on the graph does not always imply that their profiles are similar. To make the similarities clear, lines are drawn between compounds whose profiles are statistically similar within the data set, so that only compounds connected by those lines have similar profiles.

FIGURE 3-2 (facing page) Example of a computational analysis of BioMAP profiles of various compounds


FIGURE 3-2 (facing page) Example of a computational analysis of BioMAP profiles of various compounds. This example of a clustering analysis illustrates the power of BioMAP to classify compounds by their mechanism of action. On the figure, each dot represents (more...)

The analysis makes it possible to cluster compounds with similar mechanisms of action rapidly and to identify secondary off-target activities quickly. For example, Figure 3-2 shows two MEK inhibitors, PD098059 and UO126, that have the same primary activity but do not cluster together because of their strong secondary activities. In a similar way, the clustering analysis has separated two PPAR inhibitors, Rezulin and avandia, which have very different biological activities. The same approach can also be used for pathway analysis.

When a new molecule is submitted for analysis, the BioSeek researchers run its profile and then compare this against the thousands of pro files in the database, looking for homology, similarity, and function. The database returns a list of compounds that are most similar to the test compound, ranked in order of similarity and with accompanying statistics. An example is an analysis of a MAP kinase inhibitor. When screened against the database, the top two hits returned previous runs performed with the same compound. However, the next dozen hits were other p38 MAP kinase inhibitors. This example demonstrates that analysis of the BioMAP profiles makes it possible to rapidly link chemistry to biology and identify mechanisms of action.

Applications to Safety Science

Although BioSeek is focusing on techniques for application to drug development, lead identification, and optimization, techniques based on the BioMAP system are equally applicable to safety evaluation. For example, BioSeek was examining a candidate drug for which there were compelling animal data in models of inflammatory bowel disease, but the development had stalled because the mechanism of the drug’s action was unknown. When the researchers compared the compound’s profile against the database, they identified a potential match, and this match suggested a mechanism of action. After confirming the target with a biochemical assay, the researchers were able to reject the program because that particular target had known target-specific toxicities.

In another case, a BioSeek partner was screening compounds that had been identified in a simple inflammatory assay. The BioMAP profiles were performed and run against the database, and the database comparison identified a cancer target as the mechanism in one of the compounds. The researchers at the partner company were at first skeptical of the results because the chemistry was incorrect. However, the target was confirmed, and the toxicity expected from the compound made it unacceptable for the desired indication.

BioSeek has also found that it can look at compounds at higher doses, induce toxicity, and then classify those compounds by the mechanisms by which the toxicity is induced. Many toxic compounds, for example, have a common final pathway for inducing apoptosis but different mechanisms of action leading up to the apoptotic event. BioMAP clustering analysis can separate compounds according to these varying mechanisms. This ability is of interest to partners in a variety of areas. For example, BioSeek recently undertook a collaboration with the Environmental Protection Agency to characterize biologically the mechanisms of toxicity of a wide array of both drugs and environmental chemicals.

Yet another application of the BioMAP system is to identify off-target activities. For instance, ibuprofen was recently found to cross-react with nuclear hormone receptors of the PPAR family, a result that would have been immediately apparent from BioMAP analyses. In comparing the ibuprofen profile against the BioSeek database, a number of nonsteroidal anti-inflammatory drugs (NSAIDs) closely related to ibuprofen would appear, but one would also pick up various PPAR agonists. Thus ibuprofen’s tendency to activate PPARs would be apparent. This ability to identify off-target activities can also be used in comparing and differentiating among similar drugs. Butcher showed a figure with profiles of several p38 inhibitors. As expected, the profiles were similar, but there were noticeable differences, and in many cases these differences could be associated with specific secondary targets. Even if an off-target activity cannot be identified, the fact that such activity exists in a new compound should lead researchers to think carefully about what the compound might be doing differently in biological terms.

Finally, BioMAP profiles can be applied to clinical prediction and establishment of biomarkers. Indeed, the BioMAP system was purposely developed with clinically relevant readouts. The system focuses on molecules, selected for their information content, that mediate disease, are sensitive to many targets, have high predictive value, and have potential suitability as clinical biomarkers.

Butcher described BioSeek’s goal of developing a comprehensive BioMAP database connecting drug biology to clinical responses. Such a database could help identify drug candidates with safe and effective therapeutic profiles. Accomplishing this goal will demand accumulating a great deal of clinical data about what drugs are doing in people in addition to BioMAP profiles, conventional toxicological data, and models of human biology and disease. It will be an iterative process in which biological information and statistical and informatics correlations will be used to develop predictors; the predictions will be made; and they will be improved over time, with clinical outcomes being fed back to inform the further development of the BioMAP system and interpretations.


Dr. Westwick described a second analytical technique based on human cells. The ultimate goal is conceptually similar to that of the BioMAP process: to develop profiles of responses to various compounds that can then be used to analyze new drugs, pinpointing their mechanisms of action and predicting potential toxicities. However, the technique used to develop the profiles is very different.

Biologists often draw pathways as linear or circuit diagrams, but that is not how signal transduction happens or how drugs act on their targets. Instead, drug effects occur on protein complexes containing dozens of protein components—not on isolated proteins in a test tube, and not even on isolated proteins in the cell. Furthermore, in real biological pathways, the proteins and protein complexes move around. To gain a better understanding of what is happening on the cellular level, it is necessary to look in the cell because the effect of a drug will be dependent on the localization of the drug target. To better understand these mechanisms, Odyssey Thera developed a method for observing the actions of drugs in the context of living human cells.

High-Content Chemical Biology

The profiling method developed by Odyssey Thera relies on two main techniques. The first is a high-content cellular analysis that employs automated, high-throughput confocal microscopy. Excellent instrumentation is available in this area, and automated image analysis has also improved dramatically and can be combined effectively with the automated microscopy. The second technique is a proprietary process based on protein-fragment complementation assay (PCA). With this technique, two fragments of a rationally dissected reporter protein are attached to proteins of interest, for example, a kinase and a substrate. When those two proteins come into close proximity, this allows the spontaneous refolding of the reporter protein and the generation of a signal, which can be enzymatic, fluorescent, or luminescent. With a microscope, one can observe these protein complexes in live cells—not just their existence, but also their positions within the cells.

Westwick showed a real-time film of the types of interactions that can be observed by this method. Reporter protein fragments had been attached to two protein kinases, AKT and PDK-1, so that when they came together, a fluorescent signal was generated, and a greenish glow marked their presence. At the beginning, the protein complexes could be seen in the cytoplasm of the cells. When the cells were stimulated with a growth factor, the complexes moved to the membrane. Finally, when an inhibitor of an upstream kinase was added, it inhibited the membrane localization, and the complexes moved back to the cytoplasm.

According to Westwick, the system is a cell biologist’s dream as it offers the opportunity to generate a tremendous amount of data about what is going on inside a cell, and where. At the same time, it also poses a number of major challenges. Westwick focused on two challenges in particular: the informatics challenge, and the challenge of developing diverse biological assays for use in identifying the numerous pathways in a cell.

Work on the informatics challenge is the more advanced of the two. The company has five automated microscopes capturing more than half a million images per day, 6 days a week. With hundreds of compounds being analyzed every few weeks with several hundred assays, the company’s researchers are generating several terabytes of data per week. Only in the past year did the company finally solve the information technology infrastructure challenge by developing a novel strategy, and Odyssey Thera is now able to handle this avalanche of data effectively.

The challenge of developing diverse biological assays is much greater. Good instruments are available, as are the engineered platforms, but there are relatively few assays for use on these platforms. Moreover, the sort of profiling done by Odyssey demands a wide range of assays. Westwick stressed the importance of being “agnostic as to target class and pathway”; one must cast a broad net when looking for off-target effects.

To overcome these challenges, Odyssey devoted much of the past few years to expanding its assay panel, trying to cover as many pathways and as much cellular space as possible. The company’s assays now encompass a wide variety of targets in the cell: GPCRs, kinases, cytoskeletal proteins, GTPases, G proteins, transcription factors such as p53, and nuclear receptors such as PPAR , as well as some less common targets such as protein ubiquitination and the proteasome. The assays can also be used to look at apoptotic machinery, heat shock proteins, ion channels, and protein complexes involved in chromatin remodeling. In short, the company’s assays cover a wide range of target classes and processes.

To create profiles, a large and varied panel of assays is chosen—at least 100 are essential to encompass diverse pathways and target classes. These assays are then used to screen the compounds of interest at multiple time points. The researchers choose doses that are efficacious for the compound on its target in cell-based models. After the assays have been performed, the plates are scanned, and the signals are processed at the subcellular level to generate quantitative data. Various data classification strategies are then applied for data summary and display (see Figure 3-3).

FIGURE 3-3 (facing page) Strategy for pharmacological profiling of compounds with high-content PCAs


FIGURE 3-3 (facing page) Strategy for pharmacological profiling of compounds with high-content PCAs. (1) Pathways of interest are selected and high-content PCA assays are created (pathway represented as red spheres connected by arrows). Assays measure (more...)

The resulting data are then expressed as a heat map that looks something like a gene expression heat map. From this heat map it is possible to identify structure–activity relationships. The statins cluster together, for instance. Beyond this clustering, the other important thing about the heat maps is that each of these of these compounds has a unique fingerprint; no two look the same. It is simple to use the underlying data to understand the effect of a specific chemical substitution on a compound’s activity within the cellular networks.

Application to Safety Science

Application of this technology for safety and toxicology analysis of new compounds requires the generation of fingerprints or signatures for toxicants as well as for efficacious drugs. Following the generation of a signature (a subset of assays that appear to be characteristic for a particular compound), automated algorithms are used to search through all the other thousands of compounds in the company’s database for those that have a signature similar to that of the compound of interest. Comparison of compound signatures within a target class can lead to interesting discoveries. To illustrate, Figure 3-4 shows signatures from a number of different statins. From these data, the following types of information can be derived:

FIGURE 3-4 Exploring global mechanistic differences within multiple statin compounds


FIGURE 3-4 Exploring global mechanistic differences within multiple statin compounds. Each of the points on this graph represents the results of a single assay, and there are several hundred for each compound (the figure does not show all of them). The (more...)

  • Looking at the statins, one can see that there are characteristic activities within cellular networks. Each of the points on a graph represents the results of a single assay, and there are several hundred for each compound (the figure does not show all of them).
  • The signatures of the statins are similar, which would be expected, but there are also differences. Throughout one indicated region, for instance, pravastatin looks distinctly different from simvastatin. One thing that can be done is to generate an average profile for the statins across the whole assay panel and then compare the various drugs in the database with this average profile. Not surprisingly, most of the statins tend to resemble this average profile, but pravastatin is an exception that deviates notably from the average.
  • One can also ask whether other drugs have similarities to the statins. Some do, of course, and the analysis identifies them. It is then interesting to note exactly what similarities they have. By examining the statins more closely, as well as the differences among them and their similarities to other drugs, one can draw conclusions about how the various statins are functioning. For example, cerivastatin is moderately similar to the other statins, but it also has similarities to rotenone and other compounds that disrupt the mitochondrial energy chain, as well as to some antibiotics and toxicants, such as n-nitrosodiethylamine. In that same vein, the comparisons show that rosuvastatin is most similar to the “core” statin signature. In some respects it is the average statin, but it also has profile matches to microtubule modulators and DNA binding compounds, which could explain some of its off-target activities.
  • Atorvastatin, or Lipitor, is very similar to the other statins, but it also matches closely both FK506 and rapamycin, both of which are immunosuppressive drugs used to prevent rejection of transplanted organs. In particular, atorvastatin has activity on the S6 kinase pathway, which is why it matches up with the two immunosuppressive compounds, suggesting that it, too, will have immunomodulatory activity. This is important information because the anti-inflammatory activity of statins is essential to how they work.
  • Finally, pravastatin’s global profile is notably different from the profiles of all the other statins. The part of its profile relating to its activity on HMG-CoA reductase is similar to that of the other statins, since all statins are inhibitors of this enzyme; over the rest of its profile, however, pravastatin is similar to cyclooxygenase inhibitors, indicating that it may have unique anti-inflammatory properties. This is a testable hypothesis, one that is supported by the literature.

A global profile of cholesterol was also run to identify drugs with similar profiles. A number of hits were returned, including rotenone, β-laphachone, Ketek, and nefazodone. All of these compounds exhibit some toxicity, and all can be toxic to hepatocytes. Because high levels of cholesterol are also toxic to hepatocytes, these results make sense. Although this profile would not necessarily disqualify Ketek from consideration, it would prompt researchers to investigate further its effects on hepatocytes in vitro and in vivo.


The profiles generated by this technology can offer a number of insights into the potential toxicity of compounds, as well as into desirable drug mechanisms. BioSeek is working to develop a comprehensive BioMAP database connecting drug biology to clinical responses. Odyssey Thera’s current strategy is to rigorously define training sets based on toxicants as well as desirable drug classes and then to match test compounds to these profiles. In this way, the researchers hope to be able to enable a deeper understanding of cellular networks and drug targets and to facilitate more informed discovery and development decisions.



This section is based on the presentation of Eugene Butcher, cofounder and Chair of the Scientific Advisory Board, BioSeek, and Professor, Department of Pathology, Stanford University School of Medicine.


This section is based on the presentation of Dr. Butcher.


This section is based on the presentation of John Westwick, President and Chief Scientific Officer, Odyssey Thera, Inc.

Copyright © 2008, National Academy of Sciences.
Bookshelf ID: NBK4054


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (2.9M)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...