By Douglas Livingston & Tim Lease
Discovery Partners International

The Molecular Libraries Small Molecule Repository (MLSMR) is a central component of the NIH Roadmap Molecular Libraries Initiative (MLI; see interview with Chris Austin in the pre-conference issue of SBS News). Over the next decade, the MLSMR is intended to host up to a million compounds comprising the central screening compound collection of the MLI. It is responsible for compound acquisition, curation, and the distribution of formatted samples to specific centers established within the MLI. It is not feasible within the current budget to distribute samples of the repository screening collection more broadly to the research community outside the scope of the MLI; however, all of the compounds in the collection are being registered in the publicly accessible database PubChem (http://pubchem.ncbi.nlm.nih.gov), including searchable structures and other relevant information about each compound, and all screening data as it is generated by the MLI (search on “SMR” in PubChem to extract the current set of Repository compounds).

MLSCN Ups the Ante
The Repository is administered under contract by our company, Discovery Partners International (DPI; www.discoverypartners.com), and is physically located in a newly renovated 4,000 square foot facility at the DPI site in South San Francisco. During the first year of the project (beginning August 2004), we were fortunate to work closely with the NIH Chemical Genomics Center (NCGC) staff.

More recently, our scope of interaction took a quantum leap in complexity with the completion of the Molecular Libraries Screening Center Network (MLSCN) in June, when three-year grants were awarded by NIH to establish extramural screening centers at nine academic institutions spanning the continental US (www.sbsonline.org/nih/archi/2005.06.php for more information). Frequently controversial, often misunderstood, the MLSCN is the publicly accessible screening consortium of the MLI, set up specifically to identify innovative chemical tools that will enable the next era of fundamental medical, biological and chemical genomics discovery. None of these activities are intended to capture or compromise intellectual property in a manner that would harm pharma or other private sector interests. The unexpected amount of concern and wary interest that has been expressed to date is not surprising in hindsight; the 10 (including NCGC) centers of the MLSCN now collectively represent an impressive arsenal of bioassay technologies. And to us, they collectively represent a daunting appetite for plates of screening samples that need to be supplied in a timely manner across a range of capacities and formats.

Two key aspects of the MLSMR project are unique. First, the components of the MLI are typically distributed across extramural centers or are a combined effort among multiple NIH Institutes. Because of the advantages in centralizing the stewardship of a large compound collection, the MLSMR is an exception. For analogous reasons, PubChem has also become an intramural nexus of the MLI. The practical consequence for us is that an unusually large number of interactions need to be supported, particularly during the initial phases. This will continue to increase as more screens are implemented, and the number of compound suppliers expands.

Second, the MLSMR is funded under a contract, not a grant; the latter is by far the more common extramural funding instrument used by NIH. DPI is thus obligated to specific deliverables and must follow the rules required of any US Government contractor. It also means that vendors of compounds and equipment must become subcontractors to the DPI contract, with much the same obligations as DPI. Many of the available compounds must be sourced from companies based in former Eastern Bloc nations. Establishing these companies as US Government subcontractors for the first time has been an interesting, but surprisingly straightforward process. The interest and enthusiasm of potential compound suppliers to the project has been very high. However, establishing vendors properly is a sequential process, determined by available resources. We encourage everyone to update us on your offerings regularly. Given the large appetite for diversity on this project, we expect that a large percentage of the available vendors will be represented in the library as we move forward.

In summary, our objectives under the MLSMR contract are to identify and acquire appropriate compounds; do quality control and store the compounds; create and distribute compound arrays for screening; monitor compound inventory and re-supply or re-synthesize as appropriate; implement an inventory database, interface with PubChem, and create a public website; deliver various reports to NIH; and interact with other components of the Molecular Libraries Roadmap Initiative.

Building the Repository
So what is the composition of the collection, and how was that determined? Our first step was to identify the classes of compounds to put into the repository. In part due to budget constraints, this goal is framed by what compounds we can gather from a combination of commercial purchases and donations from government, academic, and other sources. To jump start collection of the first 100,000 compounds, we focused on the most readily available compounds—those from commercial suppliers. We created a database of approximately four million commercial HTS compounds from approximately 25 international suppliers.

We grouped compounds into four categories: Specialty Sets (SS) comprising known bioactives, drugs, toxins, and others; Natural Products (NP); Targeted Libraries (TL), subcategorized into protease, kinase, GPCR, ion channel, and nuclear receptor sets; and Diversity Compounds (DC) comprising compounds that didn’t fit into the other three categories. DPI accepted TLs as advertised by the suppliers and did not assess the validity of the suppliers’ claims. For the first 100,000 compounds, MLSMR targeted 80,000 DC, 15,000 TL, 3,000 NP, and 2,000 SS compounds.

MLSMR selected compounds for acquisition based on multiple criteria, which were established based on input from a panel of advisors convened by NIH. Compounds were required to be at least 90% pure according to the suppliers (MLSMR tested the purity of each compound when received - vide infra) and needed to be available in 10 mg quantity. MLSMR further filtered DC and TL compounds to meet the Lipinski Rule of Five and calculated water solubility of at least 20 micrograms/mL. Since the MLI is geared toward discovery of molecular probes and not drugs, these requirements were simplifying assumptions for the first 100,000 and weren’t intended for the entire collection. Any bias introduced to the collection by these filters will be corrected in later stages of compound collection. We also filtered DC and TL compounds for chemical substructures representing reactive functional groups or moieties known to interfere with HTS.

Finally, MLSMR used Daylight fingerprints to select diverse sets from the DC and TL compounds passing the physicochemical and excluded functionality filters. We clustered each diverse compound with up to four nearest neighbors of between 85% and 99% Tanimoto similarity. MLSMR collected these micro-clusters to provide instant structure activity relationship (SAR) data for screening hits. Having near neighbors was optional for the first 100,000 compounds and only 34% of the diverse compounds had at least one close analog as defined. Based on input from the screening centers, greater emphasis will be placed on selecting these micro-clusters in future acquisitions.

Once we identified—and NIH approved—the desired compounds, MLSMR commenced sample acquisition. In our standard process, we send bar-coded, pre-tared 4 mL glass vials to suppliers, who fill the vials and return them to MLSMR. Of 125,000 selected compounds, 90,000 were available in the desired 10 mg amount. We process received compounds in a standard, high-throughput quality control regimen that checks the weight, solubility, identity, and purity of the samples. MLSMR first checks the sample weight to ensure we receive minimum amount required from suppliers; then, it dissolves the samples in a volatile solvent mixture, 90:10 chloroform/methanol, and removes aliquots LC-MS analysis and our DMSO Working Store solution. We completely remove the volatile solvent from soluble samples, along with the LC-MS and Working Store aliquots; insoluble samples are rejected.

We check sample identity and purity by LC-MS using two Waters eight-channel MUX LC-MS instruments with combined capacity to analyze one million samples per year under a variety of conditions. MLSMR accepts samples where the molecular ion is identified and Area Under the Curve (AUC) is at least 90% by Evaporative Light Scattering (ELS) or UV (at 214 nm) detection.

Of 90,000 samples collected by MLSMR to date, 67,000 passed all the QC criteria and are available for distribution to the MLSCN screening centers. Samples that pass QC are stored in automated storage equipment for fast retrieval. We have a three-tier storage system. The bulk sample is stored in what we call our Long Term Store: the sample is neat at -20 °C under N2 in glass screw cap vial. The long term store samples are designed to be stored for five to ten years. Our long term store vials are stored in a DPI Universal Store, an automated sample storage and retrieval system. The universal store will store up to 500,000 of the 4 mL vials, and can load and retrieve up to 60,000 vials in eight hours.

An approximately 3 micromole sample of each compound is contained in a working store: a 10 mM DMSO solution in a 0.5 mL polypropylene tube. We use aliquots from the working store samples to create the microtiter plates shipped to the MLSCN screening centers. DMSO solutions of the compounds are always stored and handled under N2 atmosphere to prevent water absorption. We’ve validated that our process minimizes water absorption and haven’t observed compound precipitation from the DMSO solutions.

Distributing Compounds
The current compound quota is 450 nanomoles per screening center per year, not including material for hit follow up. MLSMR delivers the full repository collection for standard screening to each center up to two times per year. To date, MLSMR has distributed starter sets of approximately 3,000 compounds to nine of the 10 MLSCN centers, including two locations for one center. The current 67,000 compound set is available to the centers and we’ve begun delivering the full set upon request.

The MLSMR project requires extensive data tracking. DPI created ComIT (Compound Inventory Tracking), an Oracle based application to track, manage, and report all sample operations from ordering from our suppliers through shipment to the screening centers. ComIT is a customization of CambridgeSoft’s Inventory Enterprise application that coordinates what physically happens in the repository and communicates it to users inside and outside DPI. Repository workers access the application by a standard web browser interface and use ComIT to automatically guide and record their work.

MLSMR deposits the structures of compounds accepted into the repository into PubChem. The MLSCN screening centers will link to these substances when depositing their HTS data with PubChem. In tandem, DPI is building a public website specifically oriented to MLSMR activities that includes descriptions of the repository and its compounds. Eventually, the MLSCN screening centers will use the website to order their compounds and the public will be able to search and browse all compounds available in the repository.

In closing, we would like to share an observation. We opened this article by giving a sense of the diversity of opinion that has existed regarding the nature of this program. Admittedly, we at first found this wide diversity of opinions to be disconcerting. But over time, we have learned to see the juxtaposition of such diverse opinions as a core strength of the NIH culture, and now fully appreciate NIH’s impressive ability to bring together passionate individuals with widely differing opinions in order to extract the best thinking of our nation’s scientists. Consensus is not a necessary outcome.