The NCGC Pharmaceutical Collection: A comprehensive resource of clinically approved drugs enabling repurposing and chemical genomics

Assembling a comprehensive enumeration

Using the definitions above, each category was enumerated using data first from the FDA, then from regulatory agencies outside the US. In this step, inclusion and completeness was the goal, with redundancy to be eliminated later (see following section). Lists of drug names approved for human use were obtained from the FDA official publications including the Orange Book, the National Drug Code (NDC), the Drugs@FDA webpage, the Over-the-Counter (OTC) listings from the Office of Nonprescription Products, and its Substance Registration System’s Unique Ingredient Identifier (UNII). After assigning structures and Chemical Abstracts Service (CAS) numbers to these names wherever applicable and removing duplicate entries, we found that FDA has over 100,000 Drug Products registered. These Drug Products have in them over 10,000 Drugs. However, this latter number, though formally correct, is misleading since the majority of these 10,000 are different brands of the same API, different APIs of the same ME, or chemically undefined substances (e.g. allergenic extracts) (Supplementary Table 1).

We considered the possibility that additional Drugs might exist that are not listed by the FDA, if they were in use prior to the relevant statutes took effect in 1938 (37). We could find no evidence of such additional drugs, and FDA considers such drugs unlikely (38). FDA has made efforts to evaluate drugs in use prior to 1938, and has taken proactive steps to exclude them from the market if the science was found lacking, as was the case with ethyl nitrite (“sweet spirit of nitre”) (39, 40).

Two categories of compounds not currently approved for human use were then added to this enumeration due to their potential for human application: veterinary products listed in the Green Book, and drugs previously approved for human use but subsequently withdrawn from the market. Ivermectin is a broad-spectrum antiparasitic that was first approved for veterinary use, and subsequently repurposed for treatment of human helminthic diseases, particularly onchocerciasis (41). Thalidomide is a prominent example of a previously withdrawn drug being repurposed for another indication and re-approved (10). Drug withdrawal may occur either because FDA or another regulatory agency withdraws marketing approval (42, 43), or because the manufacturer voluntarily ceases production (e.g., mesoridazine); in the latter case, the drug may remain listed in regulatory publications. Several resources that list and monitor drug withdrawals (44–49) were included in the NPC. However, the designation “withdrawn” is not unambiguous, since drugs are frequently withdrawn for one indication while remaining approved for others, or withdrawn in one country/market but while remaining on the market in others. The veterinary drugs and drugs withdrawn for certain indications are labeled as such in the NPC browser (http://tripod.nih.gov/npc/).

Drugs are frequently approved in other countries but not approved by the U.S. FDA; we wished to capture these Drugs for the NPC as well. We therefore performed analogous definition of terms, enumeration of categories, and definition of structures for Drugs approved by regulatory agencies in other countries (Table 1). We obtained listings from the Dictionary of Medicines and Devices published by the U.K. National Health Service Information Authority (NHS), Health Canada’s (HC) Drug Products Database, the European Medicines Agency (EMEA), and an English translation of the Japanese Pharmacopeia, Fourteenth Edition.

While currently approved, or previously approved, drugs may be those most amenable to repurposing, compounds that have been registered for human testing but not necessarily approved by any regulatory agency, represent potentially attractive starting points for further testing and/or medicinal chemistry optimization, so were also included in the NPC. This category includes unapproved Drugs registered by the US Drug Enforcement Agency (DEA), compounds listed in the World Health Organization (WHO) International Nonproprietary Names (INN) and United States Adopted Names (USAN) registries, and compounds listed on the US tariff schedule (Table 1(b)). These listings include Drugs which have an approved Investigational New Drug (IND) application with the FDA or analogous approval by regulatory agencies outside the U.S., and those that are being tested or have been tested in clinical trials in humans. Importantly, inclusion in any of these latter registries does not indicate that the drug has in fact been tested in humans, much less the stage of testing (e.g., Phase I, II, III). The latter information is impossible to determine systematically since it is generally not disclosed by the company doing the trials; ClinicalTrials.gov, for instance, requires prospective registration of the trial but does not standardize disclosure of the API(s) being tested. While not immediately useful in repurposing applications, these USAN/INN drugs may be considered partially developed drugs and therefore require less effort to achieve regulatory approval than compounds in preclinical stages of drug development. Detailed information on the number of drug records obtained from each source can be found in Supplementary Table S1.

Generating a non-redundant list of molecular entities

Having produced an aggregate enumeration that was complete, we then devised a process to eliminate redundancy to arrive at a list that was non-redundant – i.e., in which each ME was represented only once. There are multiple mechanisms by which a single molecular entity may be listed more than once in drug listings, including by simple duplication within or between countries, or by listing of MEs as distinct APIs, Drugs, or Drug Products. Since this redundancy is the source of much of the confusion in the literature about how many MEs exist, this process of redundancy elimination is described here in some detail.

Different regulatory agencies often assign different names for the same API (e.g., paracetamol/acetaminophen), do not adhere to standard ways of listing active ingredients (e.g., terazosin, terazosin hydrochloride, and terazosin hydrochloride anhydrous are all used to refer to the same API); idiosyncratic and inconsistent naming made synonym identification difficult. Several heuristics were created to reliably eliminate redundancy. Since the only completely unambiguous identifier was chemical structure, but the regulatory agencies did not supply structural information, APIs were matched to chemical structures via names, synonyms, and/or CAS numbers. Structures were primarily derived from ChemIDPlus’s PubChem deposited substances (50), Prous’ PubChem deposited substances (search for “Prous Science Drugs of the Future”[SourceName] under PubChem Substances) (23), FDAMDD (51), commercial supplier catalog structures, SciFinder searches, or ChemOffice’s name-to-structure utility (52) if an IUPAC name was available, and were manually checked against existing literature including drug labels (53). CAS identifiers are included in the Japanese Pharmacopeia, and were also obtained from ChemIDPlus via PubChem, and manually from SciFinder. We relied on SciFinder to find the correct mapping from name to CAS and to structure once an inconsistency was identified. As a structure source, commercial supplier catalogs (Supplementary Tables S3 and S4) were found to be less reliable (i.e., more error prone) than ChemIDPlus and FDAMDD. An initial scan identified 1770 inconsistencies (different structures having the same name) from 12,800 vendor record indicating an error rate of approximately 14%. Based on the incorrect structures identified during the manual curation process, ChemIDPlus, of which we curated 12,300 records linked to approved drugs, appeared to be a reliable source for structures with an estimated error rate of 1.3% (155 errors found) and FDAMDD, with 990 of its 1,217 structures curated, was of similar quality with an error rate of 2.0% (20 errors found).

Many compound structures included salt and solvent. Common salts were removed, and remaining mixtures separated into component MEs, utilizing automated software followed by manual curation to verify the results. As structure representations of heavy metal containing compounds are frequently problematic, a special set of heuristics were applied to these drugs, where any fragment without a carbon or nitrogen and with less than 6 atoms was removed, and the rest of the molecule treated as one molecular entity. Structures were then canonicalized to facilitate molecular entity matching using the IUPAC International Chemical Identifier or InChI hash key (54) and a NCGC software package for structure standardization (http://tripod.nih.gov/?p=61) (55). To accomplish this, API records were first merged by canonical molecular entities, then by CAS number, and finally by names and synonyms. Each unique Molecular Entity was assigned a unique ID. If more than one unique ID shared the same CAS numbers or name, an alert flag was added indicating a potential error in the structure, name or CAS. Manual curation was then performed to correct such mistakes (Supplementary Table 2). Since a unique ID was assigned to each unique molecular entity, mixtures (i.e., drugs made up of multiple molecular entities) had more than one unique ID assigned.

Comparison to previous estimates of drug numbers

The NPC is the most comprehensive and accurate exposition to date of molecular entities registered or approved for human or veterinary use worldwide. There have been many previous efforts at compiling drug lists, but upon examination it was evident that all have suffered from substantial overcounting, undercounting, and/or misclassification. Much of the confusion derives from different definitions of what a “drug” is, the redundancy of drug name listings, the often opaque nature of regulatory agency databases, and lack of connection of drug names to unambiguous chemical identifiers such as structures or SMILES. In addition, the term “approved” has previously been used in different ways. A compound may be “approved” for listing in a database (such as the INN), “approved” for use in experimental settings only (as indicated by IND approval by FDA), “approved” by one or more regulatory agencies for specific clinical uses and marketing, or may have been previously “approved” but subsequently withdrawn from the market. Compounds in all of these categories have been included in previous listings of “approved” drugs, though only a fraction can actually be used in medical practice. In the NPC, “approved” means that marketing and use in medical practice, for the prevention or treatment of one or more disease indications, is currently allowed by one or more regulatory agencies worldwide.

These ambiguities, and the lack of methodological detail in many previous estimates of drug listings, make comparison to our results difficult. Previous estimates of the number of “FDA-approved” drugs for screening have ranged from 1382 (56) to 6534 (31) and of approved drugs worldwide up to 9990 (19). Scrutiny of these databases revealed undercounting of approved drugs, misdesignation of tested drugs as approved, and inclusion of the same ME more than once due to naming ambiguities. Finally, some reports have incorrectly indicated that compound listing in the USAN or INN indicates either entry into Phase II clinical trials or approval for regulatory use outside the U.S., greatly inflating the reported number of “approved” drugs available for repurposing. USAN/INN listing in fact only indicates registration by a sponsor of intention to file for human use at some point, but not approval for any such use (57).

How many “drugs” are there?

This rigorous process of definition, enumeration, and redundancy elimination allowed us to arrive at definitive numbers of MEs, APIs, Drugs, and Drug Products, approved for human and/or veterinary use, in the US and/or other countries; these are summarized in Figure 2 and listed in detail in Supplementary Table S1. Since for scientific purposes the term “drug” is most properly used to describe MEs, the most accurate answer to the question, “How many drugs are there?” is in reality “How many approved molecular entities are there?” The answer to this question is that 2356 MEs are approved for human use by the US FDA, and 3936 MEs are approved for human use in major markets worldwide including the US. These will be the richest source for repurposing applications, since they are already approved for human use and thus approval of new indications will be most straightforward.

When molecular entities approved for veterinary use are included, the numbers of MEs increase to 2508 FDA, and 4034 worldwide. Finally, there are 4935 unique molecular entities included in the USAN, INN, US Tariff schedule, WHO, DEA, KEGG drugs and FDAMDD database listings of compounds registered for experimental human use but not approved by any regulatory agency for marketing (Supplementary Table 1). While not immediately useable for human applications, these MEs would provide a “jump start” to new drug approvals, so are of interest and importance as well. Taken together, the 4034 unique worldwide approved MEs and 4935 unique worldwide registered MEs sum to a total unique 8969 molecular entities which are represent the universe of compounds for repurposing, advanced drug development, and chemical genomics.

Acquisition of all approved drugs

Acquisition of physical samples of the 2750 worldwide approved ME/APIs on the NPC Screening Resource list was surprisingly challenging, principally because chemical vendors generally list their inventory by structure, IUPAC name, or CAS number, and none of this information is routinely available from the regulatory agencies. Therefore, we used intermediary data sources to individually connect ME/APIs with vendor entries (Figure 3). In addition, different vendors frequently represented a given ME with different structures, so software was written to detect discrepancies and resolve them automatically whenever possible. When resolving ambiguities, ChemIDPlus (http://chem.sis.nlm.nih.gov/chemidplus) was particularly accurate and useful.

Compound acquisition was prioritized by approval status, ease of acquisition, and cost. Currently approved drugs were assigned a higher priority for acquisition than investigational drugs, and drugs registered in the U.S. were assigned a higher priority than drugs registered in other countries. Drugs were procured from commercial bioactive compound collections (e.g. LOPAC) and bulk chemical suppliers (e.g. Sigma) first, from which large numbers of compounds were available at relatively low cost, with structures provided for all compounds. Procurement aggregators such as ChemNavigator and specialty chemical vendors (Supplementary Table S4) that generally supply compounds at higher costs were utilized next. If no commercial supplier could be found for a drug, the drug product was obtained from pharmacies and the API purified. For compounds not available commercially, drugs were custom synthesized either by NCGC chemists or via outsourcing; cost for custom synthesis depended on structural complexity and number of synthetic steps, and ranged from $1,000 to $40,000 for 100 mg of compound.

The current acquisition status of the NPC Screening Resource is summarized in Table 2. The majority (64%) of the NPC Screening Resource approved drugs, totaling 1767 compounds, were obtained from major suppliers, including Sigma-Aldrich (St. Louis, MO), Tocris Bioscience (Ellisville, Missouri), MicroSource Discovery Systems (Gaylordsville, CT), Enzo Life Sciences International, Inc. (Formerly BIOMOL International, L.P., Plymouth Meeting, PA), Prestwick Chemical (Illkirch, France), the United States Pharmacopeia (USP), the National Institute on Drug Abuse (NIDA), and and the National Cancer Institute (NCI) (Supplementary Table 3). Controlled substances were mainly procured from the NIDA and Sigma, after licensing of the NCGC by the U.S. Drug Enforcement Administration (DEA). These suppliers were willing to provide compounds in 96-well plate or 96-tube rack formats, making them also the easiest to prepare for screening. Approximately 15% of the collection (404 compounds) were sourced as individual compounds from over 70 smaller chemical suppliers (Supplementary Table 4), either directly from the supplier or through procurement aggregators such as ChemNavigator (San Diego, CA). This was a time-intensive process, requiring the iterative manual compilation of a master list of compound names and structures. Continual changes in vendor offerings led us to create a custom structure comparison tool (MolOverlap v1.0), which compares SD files from all vendors to a master list of structures, and outputs a text file of matching structures to be obtained; this tool is freely available at http://tripod.nih.gov/moloverlap/. This allowed us to rapidly, accurately, and ongoingly extract updated catalog items and procure them for the collection. APIs not commercially available were sourced as drugs from pharmacies and purified. Approximately 21% of the collection (579 compounds) were not available from any vendor, so required custom synthesis, either by NCGC chemists or via contract synthesis. As of this publication, syntheses have completed for 220 of these compounds; synthesis of the remaining 359 will be completed over the next six months.

Acquisition of compounds registered but not approved for human use

Of the 4881 MEs identified as appropriate for inclusion in the NPC Screening Resource that are registered only by the World Health Organization (WHO) International Nonproprietary Names (INN), and the United States Adopted Names (USAN), or compounds listed on the US tariff schedule, only a small proportion are available commercially. Currently 928 (19%) of these compounds have been procured or synthesized. Approximately 20% of the remaining MEs are obtainable from chemical vendors, and the remaining 60%, totaling nearly 3000 compounds, will require synthesis at the NCGC or via contract synthesis (Table 2). Given the cost and time required for custom synthesis, we expect the expansion of the NPC Screening Resource to include all registered MEs will be a long-term effort, but a critically important one. When considering starting points for chemical optimization for a new drug, these 3000 compounds may be considered advanced leads with likely attractive activity, physicochemical, and ADME properties, which may therefore allow considerable time saving compared to leads generated from conventional HTS of diversity collections.

Recently, we have begun actively procuring or synthesizing drugs approved in countries other than the US, UK, Europe, Canada and Japan, and active metabolites of approved drugs. The building of the NPC Informatics and Screening resources will be ongoing, and as new compounds are added to regulatory databases, and physical samples are obtained, the Informatics and Screening Resource pages will be updated on our website. For immediate use by the community, we have listed all compounds in the NPC Informatics and Screening Resources by regulatory agency and supplier in Figure 4 and Table S4, and at the NPC website, http://tripod.nih.gov/npc/.

QC of the Screening Resource

Ensuring the correct identity and purity of compounds in screening collections is a critical aspect of drawing reliable conclusions from HTS data (58). This is particularly critical for the NPC, since the data generated on these compounds may be used to advance new clinical applications and draw conclusions about the universe of targets affected by clinically approved drugs, and will be made publically available. Therefore though we have received suppliers’ Certificates of Analysis, each sample has also been subjected to independent QC at the NCGC to ensure identity and >90% purity by LC/MS. Three types of detectors are used for the analysis. The primary analytical technique for assessing compound identity is mass spectrometry. Identification is based on the expected nominal mass being detected. The primary technique for assessing analytical purity is an Evaporative Light Scattering Detection (ELSD). The secondary technique for accessing compound purity is UV absorbance at a wavelength of 220 nanometers. UV detection becomes important for samples which give a poor response to the ELSD (58).

Repurposing

Expansion of a drug’s use to diseases other than that for which it was originally intended is commonly referred to as “repurposing”. While many individual examples exist of successful repurposing, only recently has the concept of large-scale, even comprehensive, examination of the disease applications of clinically used drugs been considered (19, 20). In order to be maximally reliable and useful, the collection being screened must be truly comprehensive, the screening paradigm must minimize false positives and false negatives, confirmatory testing should be done, and the data should be made publically available. The substantial infrastructure and diverse disease expertise required for such an effort has until now prevented comprehensive repurposing from being implemented. The NPC, in the context of the collaborative mission of the NCGC and the NIH Therapeutics for Rare and Neglected Diseases (TRND) program (64) makes this comprehensive approach to drug repurposing feasible, and the enormous unmet medical need – over 6000 rare and neglected diseases currently have no treatment – makes it urgent. Such repurposing will not only provide the possibility of rapid therapeutic advances, but also obviate the need for NME development, a long and expensive process. Ultimately, application of the NPC to a large number of diseases will help determine the proportion of human diseases are amelioratable by a drug in the current pharmacopeia; this question has both theoretical and practical importance, informing questions of common disease mechanisms and helping determine the scope of the problem of therapeutic development of the thousands of diseases currently without treatment.

Virtual screening of the NPC Informatics Resource can be performed by any investigator worldwide with an internet connection, and we encourage researchers worldwide to do so. To enable the research community and build the knowledge base of drug activities, we encourage researchers to inform us of their successes (and failures) using the Resource, and contribute their results to PubChem. Laboratory-based screening of the NPC Screening Resource is done at the NCGC via collaboration with any researcher who has a disease-relevant assay. The screening requirements for the NPC are much less demanding than a typical HTS campaign given the small number of compounds (3,500, compared to >350,000 for a typical HTS); in our experience the assays that produce results most directly applicable to clinical applications utilize primary patient cells. We encourage any researcher to contact us with their interest. Given the expense of building and maintaining the NPC Screening Resource, we cannot send copies of the collection to collaborators (100 screens can be performed at the NCGC with the amount of compound required to send to one collaborator), but we routinely have collaborators bring their assays to the NCGC, or send them to us, for collaborative screening. A solicitation for development and screening of rare/neglected disease assays for repurposing applications will be released by the TRND program shortly (http://trnd.nih.gov). For those researchers who prefer to reproduce all or part of the NPC Screening Resource in their own laboratories, information on suppliers of all compounds can be found in Table S4 and on our website at http://tripod.nih.gov/npc/.

Toxicology

While unanticipated biological activities of known drugs may be therapeutically beneficial for repurposing, those activities may also be responsible for unanticipated toxicological effects of drugs. Drug toxicity is one the major reason for failure of new drug development programs (65), and approved drugs are regularly removed from the market because of adverse effects; frequently, the mechanism by which the toxicity occurs is not known. To improve the reliability and mechanistic understanding of toxicity of chemicals, the NPC will be screened across a very broad range of pathways and cellular phenotypes relevant to toxicity as part of the Tox21 program, a collaboration between the NCGC, the National Toxicology Program, the US Environmental Protection Agency, and the US Food and Drug Administration (66).

Chemical Genomics

Ultimately, improvements in the efficiency of drug development and application to disease will rely on improved understanding, and therefore predictability, of the general principles by which small molecules interact with their biological targets. This long-term goal will be greatly advanced by the broad and rigorous profiling to which the NPC will be subjected. All data generated by the NPC will be placed into the NCGC’s publically available relational browser (NPC browser v2.0.0) (67) that will allow relationships between targets, pathways, diseases, and drugs to be queried in a user-defined fashion. This browser is currently available at http://tripod.nih.gov/npc/, and improvements will be made regularly in data richness and analysis capabilities.