Meeting Report: High-Throughput Technologies for In Vivo Imaging Agents

Abstract

Combinatorial chemistry and high-throughput screening have become standard tools for discovering new drug candidates with suitable pharmacological properties. Now, those same technologies are starting to be applied to the problem of discovering novel in vivo imaging agents. Important differences in the biological and pharmacological properties needed for imaging agents, compared to those for a therapeutic agent, require new screening methods that emphasize those characteristics, such as optimized residence time and tissue specificity, that make for a good imaging agent candidate.

Introduction

Although the discovery of therapeutic and imaging agents often follow similar tracks, the use of combinatorial chemistry methods and high-throughput screening technologies has not been as widely adopted among those seeking new imaging agents as it has been by the world's drug discovery laboratories. Envisioning that these “molecular discovery” tools could indeed make a positive impact on the rate at which potential new imaging agents are identified, the National Cancer Institute (NCI) organized a 2-day workshop on November 9–10, 2004, titled “High-throughput Technologies for In Vivo Imaging Agents,” to explore how such technologies could be used to discover imaging agents for cancer indications, and to solicit ideas for how the NCI could accelerate such activities. This article summarizes the presentations and some of the discussions of the workshop.

Session I: Overviews

Ralph Weissleder discussed the differences between a diagnostic agent and a therapeutic agent. Diagnostic agents, including imaging agents, are usually given in a single dose, in trace amounts, whereas therapeutic agents are usually administered multiple times at much higher doses. Although some toxicity is tolerated for therapeutic agents, particularly an anticancer agent, diagnostic agents must have minimal or no toxicity. On the other hand, while it is acceptable for a small amount of a therapeutic agent to accumulate in nontarget tissues, this is not a good characteristic for an imaging agent because it produces high background readings that can confound detection. Indeed, target specificity is a critical, make-or-break characteristic for a diagnostic agent. So, too, is the circulating lifetime for a diagnostic agent—while a long therapeutic lifetime is often desirable in a therapeutic agent, the best imaging agents will have circulating lifetimes on the order of minutes or hours versus days for a therapeutic agent.

Weissleder then discussed the subject of how one might plan a combinatorial chemistry/high-throughput screening (HTS) approach to developing new imaging agents. The traditional approach would be to screen libraries of small molecules, peptides, carbohydrates, and other types of compounds for their ability to bind to some target molecule. Any hits identified in this screen would then have to be turned into an imaging agent before further testing ensued. An alternative approach is to start with a molecular library of imaging agents (MLIA) that already contain the other chemical entities, such as linkers, targeting molecules, and metal chelates. His group, for example, has synthesized small molecule libraries in which each chemical in the library is attached to a 20-nm magnetic particle. This nanoparticle not only serves as a magnetic contrast agent for MRI, but helps with purification and separation steps involved in library synthesis and screening. Using this approach, the Weissleder group has been able to categorize both the small molecules in the library and the various linkers and conjugation strategies that might be applied to creating nanoparticulate, targeted imaging agents. These MLIA were then screened for desirable traits, such as an ability to detect angiogenesis or apoptosis, and for the absence of undesirable characteristics, such as macrophage uptake. He noted that it typically takes approximately 2 weeks to develop a library and 2 days to run a screen. One area that he believes needs work is on developing HTS methodology that is organism-based. Such models, possibly using zebrafish, C. elegans, and other small animals, would help increase the odds that a “hit” would function in the body as desired. He added that there is still a paucity of available libraries and more information is needed on target and pathway identification for imaging agents, as opposed to drug molecules.

Richard Houghton discussed his group's use of a technique known as positional scanning, which when used to analyze HTS data on mixtures of peptide libraries rapidly provides information on individual compounds in a given mixture. His laboratory has used this approach to develop naturally fluorescent rhodamine-derivatized peptides that bind tightly and specifically to the kappa receptor. A similar experiment led to the development of a decapeptide library capable of defining T-cell epitopes. This library has yielded a specific T-cell clone that is specific for a tumor-associated antigen. He noted that newer methods for analyzing HTS results on libraries of mixtures, particularly a technique known as biometrical analysis, should provide new insights into what a T-cell actually sees when it recognizes a given antigen.

Recognizing that peptides are not the ideal substances for further development as imaging or therapeutic agents, Houghton's group has started making libraries of small molecules based on peptidomimetic libraries. For example, his group has now developed libraries in which the peptide bond between amino acids has been changed into a less labile group. Other chemical modifications made to initial libraries can yield multiple libraries with extensive chemical diversity for further analysis. Although such libraries provide dense coverage of specific parts of chemical space, they still fall short of providing broad coverage of all of chemical space. Nonetheless, such an approach can quickly yield promising results. Recent work targeting the enzyme caspase used both peptide and small molecule libraries containing over one million compounds in 600 mixtures.

Ken Wertman discussed efforts to create and screen libraries of novel small molecules designed for a limited number of target classes and then use a software-based mapping process that relates chemical space to biological space. This approach is based on the premise that a combinatorial library can be viewed as a statistical experiment that can provide data with which to create large virtual libraries. Through an iterative process in which these libraries are analyzed using genomics and other statistical tools, new scaffold ideas are generated upon which highly refined libraries are built and then tested using virtual screening methods. The end result is a focused in silico library of a manageable number of compounds that can generate a high proportion of hits. This approach has yielded several compounds that are now in clinical and preclinical testing.

Session 2: Chemistry

Kit Lam discussed the application of one-bead, one-compound (OBOC) combinatorial libraries and chemical microarrays in the development of targeting agent for cancers. OBOC libraries can be very large and can be subjected to numerous different screening methods. Lam's group routinely uses four different screening methods: an enzyme-linked colorimetric assay, whole cell binding assay, fluorescent quenching protease substrate assay, and a radiolabeled protein kinase substrate assay. Using these techniques, Lam has developed and identified peptides that bind to ovarian cancer cells. Using the sequence of these peptides as a starting point, Lam and his colleagues then prepared a secondary, focused library of some 3 million peptides in which some of the amino acids were fixed, based on data from the first-generation library. Screening of this library identified a number of peptides that bound specifically to tumor cells. Modifying these hits with the copper-labeled chelate DOTA yielded constructs that readily identified tumors in animal models using PET. Using similar methods, Lam's group has also identified a peptide-mimetic motif that binds specifically to lymphoma cells and renders them visible using near-infrared imaging. Looking toward the future, Lam envisions a day when there will be a host of DOTA-labeled tumor-binding peptide available in every clinical oncology laboratory. Upon receiving a tumor sample from a given patient, these peptides, perhaps immobilized on a microarray, would be used to classify the tumor according to the specific peptide to which it bound. A ⁶⁴Cu-labeled version of that peptide could then be used to image the tumor, while a ⁹⁰y-labeled version, using the same DOTA moiety for chelation, would be used as a therapeutic agent.

Young-Tae Chang discussed his laboratory's work on creating diversity-directed fluorescence libraries for biological imaging. These libraries are built with a fluorophore attached to all library components so that any hits identified do not have to be modified further for screening as imaging agents. In addition, Chang's group uses chemical methods that do not require purification after the synthetic steps, further shortening the time that it takes to go from library construction to lead identification. Using these methods, Chang and his team have identified fluorescent agents capable of targeting and imaging specific organelles inside cells. Current efforts are aimed at modifying the linkers used in the library construction to create molecules useful with other imaging methods.

Irwin Kuntz discussed his group's efforts in developing high-throughput computational methods for imaging agent discovery and development. This work relies on the large and growing number of databases of target structures that cover a large number of protein families and combines the data from such databases with the ever increasing number of good computer screening and docking tools that evaluate putative ligands in silico. He stressed that while most combinatorial libraries will not have drugs or imaging agents in them, analysis of these libraries can provide solid leads for further development. Indeed, with the availability of automated docking programs that can fit millions of compounds into the active sites of known structures, existing libraries can be a valuable source of ideas for creating new directed libraries that yield 100-fold better results than when using a random library. As an example, Kuntz showed how his group developed molecules capable of labeling cathepsin D with binding constants on the order of 500 pmol based on cell-based assay results.

Garland Marshall discussed efforts at designing privileged scaffolds for targeting and imaging. This approach focuses on designing scaffolds in silico by introducing conformational constraints into starting scaffolds as a way of decreasing the entropy of binding, and thus, enhancing binding affinity. In particular, Garland spoke of developing molecular scaffolds that mimic critical structural features found in peptides, including β-sheets, α-helices, and reverse turns. He noted that the overall objective is to determine the recognition motif of a given receptor or other target and to develop a variety of confirmation templates as molecular probes. These probes will have relatively rigid scaffolds that satisfy at least three requirements: they will posses limited three-dimensional structures, be readily synthesized, and uniquely orient the side chains that are believed to transfer most of the information about peptide-receptor interactions.

Renata Pasqualini discussed combinatorial mapping of protein interactions for the development of targeted imaging agents. Her group uses phage display technology to generate combinatorial libraries aimed at mapping cell surface proteins. She noted that this versatile and functional technology enables one to target a wide variety of molecules in a fast and flexible manner. Using phage display and bioinformatics clustering tools, she and her group developed ligand-directed clustering data for the NCI-60 panel that generates a one-step target identification profile by denoting target-peptide combinations for further study. Pasqualini also discussed a new technology that uses phage libraries that selectively distribute to specific tissues and organs, and in particular, the various parts of the vasculature. She noted that phage display, when used in conjunction with laser-capture microdissection, could provide a fast method of identifying potential imaging agents for fingerprinting tumors.

Robert Ladner discussed efforts using phage-generated cyclic and linear peptides to identify potential in vivo imaging agents. He detailed work done to optimize chelation conditions, improve serum stability of library peptides, and characterize biodistribution properties, and then discussed the development of a VEGF-R2-targeted tumor angiogenesis imaging agent. He noted that while there are several drugs targeting this receptor that are in human clinical trials, there is little work being done on imaging this target. His group identified two peptides of different structure and combined them into one heterodimer that was vastly superior at binding to the receptor than either peptide alone.

John Babich discussed the application of a novel chelate system for either technetium or rhenium that would enable radionuclide and fluorescence imaging of biological targets using essentially the same molecular probe. He discussed the successful efforts to improve chelation chemistry to the point where it has become irrelevant as far as creating combinatorial libraries is concerned. The solution to this problem was to create an amino acid analog that could act as a chelator for both technetium and rhenium and that can be easily incorporated into peptide libraries using standard solid phase synthesis methods. In one example, he showed how peptides containing this chelator can be used with technetium to perform whole-animal imaging in a monkey, and to then conduct follow-up fluorescence microscopy to image the rhenium-containing peptide to examine target binding at the cellular level. It was clear from these examples that the rhenium-labeled peptides could prove particularly useful for rapidly screening peptides for their usefulness as in vivo imaging agents.

Session 3: Assay Development

Robert Gillies discussed his group's work on developing lanthanide-containing molecules that can be incorporated in combinatorial libraries and be used in time-resolved fluorescence measurements. Lanthanides have the advantage of having a very large Stokes shift, but their extinction coefficient is very low, almost immeasurable, which requires surrounding the metal ion with molecular antennas that basically funnel photons into the system. The other advantage lanthanides offer for screening and imaging applications that the excited states of these elements last as long as milliseconds, reducing background to virtually nil as all other fluorescence has decayed by then. Upon developing a suitable chelator, Gillies' group found that they could detect receptor-bound, europium-labeled peptides at the level of 200 zeptomolar, low enough to detect endogenous receptor expression. In other words, this level of sensitivity means that the limit of detection is confined by ligand affinity, not receptor numbers. He then showed examples of using Eu-labeled peptides in cell surface binding assays with either engineered of natural cell lines and how the results can be validated using confocal fluorescence microscopy. He also noted that it is a simple matter to replace the lanthanide with an imaging metal ion, such as indium-111 or technetium-99m, or with a radiotherapeutic metal ion, without any change in the binding properties of the imaging agent.

David Piwnica-Worms discussed strategies for real-time imaging of signal transduction and protein-protein interactions as an HTS assay for new drugs and imaging agents. He presented several examples of genetically encoded reporters for studying proteasome targeting, the ligand-induced IkB/NF-kB pathway, and mTOR signaling pathways and showed how the results of a 96-well plate formatted system that uses split luciferase bioluminescence imaging for screening protein-protein interactions could be readily translated into a platform for identifying an in vivo imaging agent.

John Dunne discussed using flow cytometry to image human immune responses in what he referred to as ex vivo immunology. He began by reviewing how flow cytometry has evolved into a widely and easily used assay system that was amenable to automation for HTS applications. He showed how flow cytometry could be used with a wide variety of cells and biomarkers, and presented several examples, including one that followed T-cell differentiation into multiple subpopulations. He noted that flow cytometry can be used to measure a wide variety of responses to a wide variety of signals over many time scales, and that its ability to track dynamic processes could be particularly useful in cancer biology research because tumors are not static isolated structures, but rather interact dynamically with their environment.

Alice Ting discussed efforts to use genetically encoded, small-molecule reporters for studying protein function and trafficking in a low-throughput assay system. Her group uses an enzyme-catalyzed, protein labeling system built using biotin ligase, a bacterial enzyme with no substrate in mammalian cells, and utilizing a modified biotin derivative capable of labeling cell surface proteins. She presented an example in which her group used this probe to label the EGF receptor without affecting its function, as is the case with larger labels, such as green fluorescent protein. She also presented data from a study involving the AMPA receptor, which is involved in memory formation and some disease processes.

Session 4: Screening Models

Callum McRae discussed the use of zebrafish as model organisms for HTS, highlighting its advantages: Its embryos are transparent, it undergoes external fertilization, and develops rapidly—development is largely complete by 72 hr. Zebrafish also have a diploid, divergent, and tractable genome, which should be sequenced by the middle of 2005, and a short breeding cycle that produces 300–500 embryos per mating. Most importantly, this fish is amenable to HTS assays and chemical screening is doable in fairly easy manner and on a scale of 12,000 measurements per day. He showed an example of using the zebrafish in an HTS that measures drug-induced repolarization, a rhythm disturbance in the heart that accounts for many drug failures and even postapproval drug withdrawals. The zebrafish HTS was able to identify 22 out of 23 drugs tested that are known to induce this problem. McRae also discussed work developing imaging techniques capable of following real-time vascular development in the zebrafish embryo. Using fluorescence angiography combined with transgenic reporters vasculogenesis, his group was able to follow individual gene expression in developing blood vessels.

Ross Cagan discussed work using the fly eye as a model system for cancer-based HTS assays. The adult fly, he explained, is composed of 750 identical units, each comprising 14 cells, and it is perhaps the best understood tissue in terms of development and signal transduction. The fly eye is a good model system for HTS assays because disturbed pathways are often easily detected simply by looking at readily characterized, physical changes in the gross structure of the eye. In addition, targeting the eye does not affect the animal's survival in the laboratory. The fly can also be grown in 96-well plates and be used to screen up to a million compounds in a few months. Using the fly eye, Cagan's group has been studying pathways involved in triggering type 2 multiple endocrine neoplasia, an unusually well-defined cancer syndrome that involves an activating mutation in Ret, a receptor tyrosine kinase. Flies have Ret, and Cagan's group was able to clone the gene, make mutations, and fuse it to a fly eye-specific promoter. He noted that it is easy to see the effect of the mutation, and in a sense, the fly eye acts as an amplifier of small changes. In this case, the fly HTS identified an already-existing drug that is now beginning clinical trials as a therapy for type 2 multiple endocrine neoplasia. Cagan also presented his laboratory's work on a major regulator of cell death known as Src and its effect on two pathways that either trigger or suppress apoptosis.

Session 5: Informatics and Laboratory Management

Christopher Austin discussed NIH's molecular libraries roadmap initiative, which falls under the major theme of catalyzing new pathways to discovery. This initiative has two components: the Molecular Libraries Screening Center Network and the Small Molecule Repository, which will contain some 100,000–500,000 compounds chosen by an advisory panel. He noted that a major thrust of these programs is to create imaging probes for research accessible to the scientific community. To accomplish this, the NIH has established the Imaging Probe Development Center, which will be on the NIH campus and will open in July 2005. This center will synthesize precursors for published and novel imaging probes on request from the scientific community. Another key part of this initiative is the PubChem database (http://pubchem.ncbi.nlm.nih.gov), a comprehensive public sector chemical database that is fully linked to other NCBI Entrez databases through PubMed. An imaging probe database, set to debut at the end of 2004 and overseen by NCI's Cancer Imaging Program, will function as a subset of PubChem. The initiative will also focus on technology development, with efforts in chemical diversity, assay development, instrumentation, and predictive ADME and toxicology. In particular, the initiative will fund projects aimed at developing new probes that will achieve a 1–2 orders of magnitude improvement in the ability to detect and image specific molecular events in vivo and that will have potential for clinical development. The initiative will also help develop chemoinformatics software for public use.

Michel Lebl discussed developing automated methods for combinatorial library synthesis. He began by reviewing the history of automated library synthesis, noting that it took 30 years from the invention of solid-phase peptide synthesis for the field to realize the power of solid-phase methods for creating large, diverse libraries of chemicals. To date, two drugs first created via combinatorial chemistry are now on the market and some 74 are in human clinical trials. He then reviewed a host of different synthetic methods and equipment for creating combinatorial libraries using automated methods, concluding that there are techniques available for synthesizing libraries of all scales and types. The future, he said, will be to develop techniques for ultra-high-density synthesis and screening of compound and to move toward the rational development of libraries.

Alex Kiselyov discussed focused diversity and its ability to accelerate hit identification. He discussed starting with a feasible chemistry space and winnowing this initial virtual library using a variety of filters—active versus reactive compounds, pharmacokinetic properties, intellectual property concerns, and others—to create a virtual drug-like set of molecules. This list is then pared further through a variety of means to generate subsets of molecules that will then be synthesized.

Paul Clemons spoke about multidimensional screening for small-molecule and cell-state profiling. He demonstrated this approach in which the multidimensional assay varied the cell state (using known modulators of biological function), chemical identity, and activity over time. Analysis of this multidimensional data, using bioinformatics clustering tools, revealed clusters of compounds that showed the influence of structural properties of the various small molecules on measured the activity outcomes. In discussing another example, he showed how it was possible to obtain metabolic pro-Qfiling of lung carcinoma cells and identify compounds capable of affecting metabolic state.

Barry Bunin discussed efforts to create a database platform that will allow collaborators to archive, mine and selectively share data from HTS assays. The idea behind this effort is to tap into the collective intelligence of all collaborators and to use this to translate novel discoveries into drug candidates. The initial focus of this database was structure-activity relationship data, taken largely from public sources. Once the data are assembled. it should be possible to build a community of researchers who will apply their models to the data and develop a set of filters for parsing the data.