High-Throughput Purification in Drug Discovery: Scaling New Heights of Productivity

Abstract

With the “low hanging fruit” of early drug discovery gone, pharmaceutical companies are increasingly turning to developing high-throughput synthetic platforms capable of greatly shortening the design–make–test cycle of new drugs. Purification has long been considered the bottleneck of this procedure; however, new technologies and systems are now being integrated into these high-throughput synthetic workflows, providing compounds of high purity capable of being used directly in biological screening.

Drug discovery is a slow and expensive process.¹ The quest for greater efficiencies through the use of laboratory automation has been the goal of most research organizations over the past 25 years, especially during the boom years of combinatorial chemistry when compound libraries expanded from low thousands to low millions in a very short period of time.² However, for most research groups, the rapid increase in compound synthesis rates was not matched by increased compound purification capacity; consequently. as compound numbers rocketed, quality control plummeted. With the advent of newer synthetic approaches, such as multi-component reactions (MCR), it was relatively easy to generate multiple libraries, each of tens of thousands of compounds, as brightly colored dimethyl sulfoxide (DMSO) solutions stored in newly constructed (at great costs), robotically controlled stock rooms. This, combined with costly high-throughput screening programs, often resulted in poor false positive/negative data and low follow-on hit confirmation success rates. We may be guilty, when thinking back, of exaggerating the downside of this approach, but hopefully there is now consensus that increased synthetic capability without appropriate quality control is a false economic strategy that can lead research projects down a very long and very expensive wrong track.

Over the past 5 to 10 years the new paradigm is to extract high-level data from early-stage projects, including structure–activity relationship and metabolic stability profiles. Hence the need for highly purified compound libraries, to feed the machine, is crucial to the success of each project.

Preparative HPLC is still the go-to method of choice for compound purification, the gold standard being prep-LCMS for targeted peak selection³ to achieve the holy grail of “one sample injected–one fraction collected”.

Advances in automated liquid–liquid extractions,^4,5 including microdroplets,⁶ and solid phase extraction (SPE)/resin cleanup methods⁷ offer potentially faster and cheaper alternatives—but they cannot match the generic capabilities of C18 HPLC, with a wide compound structure range and high purity levels. Sample preparation, using SPE, for crude samples and advanced techniques of “trap and release” of “heart-cut” samples,⁸ combined with LCMS, offer the ultimate purification solution.

Significant progress is also being made in supercritical fluid chromatography (SFC)-MS purification.⁹ This technique has been used for the successful purification of synthesized compound libraries.¹⁰ Direct comparison between HPLC-MS and prep-SFC has found comparable success rates;¹¹ however, for most applications prep-LCMS, alone, is easier to setup and maintain.

So, what progress has been made with high-throughput purification systems over the past couple of years?

This Microperspective is focused on three stand-out themes of recent publications:

• 1.Scale
• 2.Accurate quantification of target compound
• 3.Integrated systems

Scale

The challenge, when planning and implementing compound purification capacity, is to develop systems to cope with the full gamut of R&D activities from milligrams to hundreds of grams of compounds.

For early-stage projects and corporate compound collection expansion activities, the target has typically been 10s of mg of compounds as solids and mM DMSO stock solutions. Even at this relatively low scale, the purification resources required in terms of a) laboratory equipment, b) laboratory bench/FC space, c) consumables and d) waste generation/environmental impact are significant—as are the costs of setting up and maintaining the systems.

For example, an average target amount of 10 mg of material, assuming low to moderate synthesis yields, would typically require a 21 mm column running at 20–30 mL/min for approximately 10 min, injection to injection. The result would be relatively low numbers of sample purifications per day, the bottleneck drying of high milliliter volumes organic/aqueous fractions and very large volumes of HPLC waste—not to mention the potential complication of pooling fractions and audit control.

So, in terms of compound quantities, “how much is enough?”

Most screening methods require low microliter volumes of DMSO stock solutions using 96-, 384-, and 1536-plate formats. Also, advances in microfluidics and sonic sampling techniques enable the use of picoliter quantities of sample.^12,13

Recent publications by teams from Merck¹⁴ and Bristol Myers Squibb¹⁵ neatly demonstrate the advances in scaling down reaction quantities and purifying compounds at the low mg to sub-mg scale.

Key to the success of both these microscale preparative systems was recent progress made in the development of <5 μm fully porous particles. Columns with smaller particle sizes are able to achieve better separation and improve analysis times, enabling the step down to shorter columns and mobile flow rates as low as 1 to 5 mL/min.

Equally important is the considerable reduction in the use of chemical reagent inputs and solvent/HPLC mobile phase waste generation. We are all increasingly aware, quite rightly, of the need to reduce the environmental impact of our research activities.

Working at this scale presents far more opportunities for fully integrated automation to purify and isolate high numbers of compounds per day—generating enough material to provide fast turnaround support to active projects and, in many cases, surplus stock DMSO solutions for the corporate compound collection. One of the major advantages is the much-reduced fraction volumes that can be directly collected into 96-well plate tubes—this greatly reduces the dry-down bottleneck, allows “plate to plate” sample transfer using standard liquid handling robotics and simplifies sample tracking.

Quantification

Reducing compound quantities to low mg and sub-mg levels raises new concerns and challenges about the accuracy of gravimetric measurements. Even for compounds submitted to the stock room, presumably of reasonable physical state, i.e., salts rather than free base oils, weighing <5 mg in glass vials requires careful handling to stay within acceptable error margins.

For small-scale library synthesis, typically using 96-well polypropylene tubes, the opportunity for weighing errors is exacerbated, mainly due to static build-up during handling. It is not uncommon for repeated tare weighing of a 96-well polypropylene tube to show errors of ±2 mg. Add to those issues of long drying times required for HPLC fractions—Is the sample really dry? Is there water/solvent/base trapped in the oil product? There is, of course, the option of repeated, hours-long drying, but as compound molecular weights are being driven down (all for good chemistry reasons), i.e. fragment compounds—does prolonged drying cause problems with small molecular weight/more volatile compounds?

Thankfully there is an alternative quantitative technique that has gained widespread acceptance within the industry over recent years: charged aerosol detection (CAD).^16,17 Several groups have reported impressively accurate quantitative data for the analysis of compounds, even at microgram levels.^18,19 Using this technology, the accuracy of quantitation of the actual target compound is greatly improved compared to weighing the sample containing the target compound plus salts, water etc.

This also has a positive impact on the workflow and cycle times. For example, it has been demonstrated that two repeated dry-down steps can be replaced with a single dry-down, saving considerable time and increasing overall sample capacity. The ability to accurately quantify the target compound in a) crude reaction mixtures and b) isolated HPLC fraction solutions, combined with LCMS data, enables the user to select only viable samples, at an early stage, to progress to the next steps in the process, to maximize throughput and success rates.

Compared to its older cousin, evaporative light scattering detection (ELSD), CAD has a wider dynamic range and lower sensitivity. Both techniques are touted as “universal detectors” by the instrument manufacturers, but practitioners will be quick to point out the merits and limitations of each and the validity of such grand claims. However, the drive toward CAD, within the drug discovery sector, is gaining pace.²⁰ In terms of library synthesis, this is largely driven by much less intercompound response variability; i.e., accuracies of 7% to 11% have been reported²¹ for the mg/mL quantification of 50 diverse compounds.

Integrated Systems

So, you have developed a high-throughput purification system capable of generating 100s of high-quality samples per day. But that is only part of the story—the compounds need to be dissolved in DMSO at 10 mM in 96- or 384-well plates ready for screening and storage in the corporate collection (see Figure 1).

Figure 1.

Integrated reaction/purification/storage workflow.

Without automation of the whole process—from a) fraction collection/pooling, b) dry-down, c) quantitative analysis (gravimetric and/or CAD analysis), d) dissolving in DMSO to 10 mM to e) cherry-picking/plating for a specific project—then all you are doing is moving the bottleneck further down the line.

Recent publications demonstrate slick, high-throughput systems that are capable of high numbers, in a fully automated system running 24 h a day and 7 days a week.²² These systems can sometimes even include the testing of the purified material in biological assays. An example from AbbVie gives a turnaround time of between 24 and 36 h from starting materials to assay result.²³

However, such systems require considerable investments of time and money and a diverse set of skills including robotics, bespoke system integration and scheduling software and, typically, bespoke hardware components—not forgetting analytical chemistry skills and experience to optimize performance and throughput. Each organization will form their own strong opinions as to whether such a resource should be provided as a dedicated service support or open to all research groups as a “walk-up” facility.²⁴

The rapid growth in Artificial Intelligence (AI) will bring a whole new meaning to “integrated systems”—with the potential to drastically transform drug discovery²⁵ using enhanced target intelligence and more focused libraries—but we still need the tools to rapidly make, purify and test. In such a fast feedback loop, tackling the rate-limiting step of purification is paramount.

Future Developments

It is perhaps surprising that recently published papers refer only to single analytical column and single prep-HPLC column systems. Efforts to further increase capacity seem to be limited to the purchase of multiple systems. For many research groups this is cost prohibitive, plus the issue of premium limited bench and fume cupboard space.

But, 20 years ago at the height of combinatorial chemistry mania, we saw the introduction of four and eight HPLC column systems, including the Waters MUX and Gilson 215-8 channel nebula system, capable of parallel SPE workup, LCMS analysis and purification.²⁶ But, like combinatorial chemistry, they fell out of fashion.

Perhaps it is time to revisit the multi-column approach (see Figure 2): the combination of sub-mg scale chemistry using four- or eight-column HPLC prep systems, mass spectrometer-directed fraction collection, with CAD quantitative analysis and fully integrated to deliver screen-test-ready plates. Such systems would offer the throughput and capacity to turn virtual compound ideas into real compound libraries in a cost-effective manner with reduced environmental impact.

Figure 2.

Multi-column prep LCMS system. In-house developed system based on four independent columns and gradient pumps, using a single Microsaic MiD mass spectrometer for selective mass ion fraction collection. Combined with XYZ liquid handling robotics with 2 × 96 well autosampler and fraction collection capacity per run.

The authors declare no competing financial interest.

Special Issue

Published as part of the ACS Medicinal Chemistry Letters virtual special issue “New Enabling Drug Discovery Technologies - Recent Progress”.