One day, machines will do all our work for us—even our drug discovery work. What’s more, the machines will do our work better than we ever could. In drug discovery, the machines will eliminate the human inconsistencies and errors that limit the quantity and quality of our drug candidates. They will perceive and exploit efficiencies beyond our understanding. And they will help us develop drug candidates so economically that even the rarest diseases will attract drug discovery investments.

For such a drug discovery utopia to occur, a great change will be needed. Of course, it will happen gradually. Indeed, it has already begun, now that enabling technologies are maturing and becoming more widely available. These technologies include quantum-inspired molecular optimization, target deconvolution via protein painting and advanced mass spectrometry, and end-to-end automation of drug discovery workflows. When these technologies are implemented, small molecule design and lead optimization are accelerated, shortening the drug discovery process by years and saving untold millions of dollars.

Exploring vast chemical libraries

“We want to create as many small molecule drug leads as possible,” says Shahar Keinan, PhD, CEO and co-founder, Polaris Quantum Biotech (PQB), “[by] combining quantum computing, artificial intelligence (AI), and precision medicine.” The company expects to produce up to 100 “drug blueprints” per year while lowering drug development costs and bringing successful leads to market more quickly.

PQB plans to sift through large chemical libraries to identify small molecules that exhibit the properties a drug would need to change the course of disease. This approach poses an optimization problem. To solve it, PQB partnered with Fujitsu to develop a molecular optimization platform that significantly improves the speed and chemical diversity in small molecule lead discovery.

Fujitsu’s quantum-inspired Digital Annealer solves combinatorial optimization problems 10,000 times faster than other currently available alternatives. In less than five minutes, the Digital Annealer can search a virtual library of a billion molecules that are relevant to a protein-binding pocket associated with a particular disease.

“The library is combinatorial and grows very fast,” notes Keinan. “The problem was searching it. If you can take a three- to four-year process and reduce it to eight months, then addressing smaller patient groups becomes financially feasible.”

After the molecules are searched using the quantum-inspired technology to identify a couple of thousand molecules that satisfy basic design criteria, the next set of candidates goes through more elaborate quantum mechanics and molecular mechanics calculations to define the 10–20 best options. These leads are synthesized and tested to identify the optimal candidate.

Since the platform is fast, it needn’t be confined to investigations of known diseases, such as dengue fever. It can also be used to tackle emerging diseases, such as COVID-19, as well as diseases that have exploited mutations to become nonresponsive to current drugs.

“We are open to collaborations with organizations focused on specific diseases to pursue small molecule therapeutics for their area of interest,” Keinan stresses. “We are faster and less expensive, and we are in this business to find a solution for diseases that are currently overlooked.”

Identifying protein-protein interfaces

Using conventional technologies to identify targetable protein-protein interfaces can be difficult and time consuming. One of these conventional technologies is protein crystallography. It provides unparalleled resolution of protein structures, but it confronts users with crystal structures that can take many years to solve. Another conventional technology, hydrogen deuterium exchange mass spectrometry (HDX-MS), poses experimental challenges, such as low pH digestion, as well as interpretive problems that may be impossible to solve without specialized software.

An alternative technology is protein painting. Developed in 2014 by Alessandra Luchini, PhD, Lance A. Liotta, PhD, and Virginia Espina, PhD, as an outgrowth of their dye chemistry work, protein painting uses noncovalent dyes to selectively label solvent-accessible regions of protein complexes—that is, regions other than the solvent-inaccessible protein-protein interface—in native protein conditions.

The dyes block trypsin cleavage, essentially making the solvent-accessible regions “invisible” to mass spectrometry, thus allowing selective identification of the undyed regions. Protein painting uses recombinant proteins similar to crystallography and HDX but takes only a few days to complete.

“We had some experience using dyes as ‘bait’ for proteins in other applications, and we decided to see if we could directly label proteins,” explains Amanda Haymond, PhD, research assistant professor, School of System Biology, Center for Applied Proteomics and Molecular Medicine (CAPMM), George Mason University. “One of the biggest challenges was identifying appropriate dyes or combinations for the reactions.”

To develop a technique that would be accessible, the investigators had to show that it worked with dyes that were either commercially available or simple to synthesize. A dye that Haymond and colleagues described recently interacted primarily with lysine and tyrosine residues, bound nonspecifically to a range of proteins, bound in high numbers to a range of proteins, and protected bound regions of the protein from urea denaturation.¹

“We probed the interface between cytokine IL-1B, its receptor IL-1R1, and accessory protein IL-1RAcP,” Haymond reports. “The protein-painting data allowed design of an interfering peptide that disrupted IL-1B signaling, which has therapeutic applications for osteoarthritis. Most recently, we identified a key residue in the PD-1/PD-L1 interface.¹ We designed a peptide that mimicked this region and disrupted PD-1/PD-L1 complex formation.”

“We are working on analogs as potential oncology immunotherapeutic agents,” she continues. “We are excited that other groups have adopted protein painting.”^2,3

Deconvoluting target strategies

Accurate target deconvolution and understanding on- and off-target effects is of vital importance to the drug discovery process. According to Diarmuid Kenny, PhD, group leader, Integrated Biology, Charles River Laboratories, mass spectrometry–based proteomics stands at the forefront of most unbiased target deconvolution strategies.

The company has developed Capture Compound mass spectrometry (CCMS), a technology that utilizes a photoaffinity label (PAL) to capture and subsequently identify proteins interacting with a small molecule. The PAL generates a covalent bond between the small molecule and the protein of interest, allowing very stringent washing and identification of weak but specific binding proteins. Using a combination of different PALs maximizes the probability of identifying specific binding partners.

Other strategies include the proteome integral solubility alteration (PISA) assay, which incorporates the principles behind thermal proteome profiling to identify specific protein binding partners with an altered temperature stability profile after treatment with a small molecule.

“CCMS is ideal when working with compounds that have good structure-activity relationship (SAR) data,” says Kenny. “We need to modify the compound to incorporate a PAL without abolishing the compound’s activity.

“The PISA assay, which does not rely on modifying the parent compound, is more suitable for in-cell target deconvolution as it can also be used to potentially identify downstream treatment effects. Both CCMS and PISA are functionally agonist, and therefore they can identify both primary and off-target proteins bound by a compound.”

Shotgun proteomics, or data-dependent acquisition, is the traditional methodology for unbiased proteomics workflows. Although robust and widely suitable, this methodology may miss data values. More recently, there has been a movement toward an alternative methodology, namely, sequential window acquisition of all theoretical mass spectra (SWATH-MS), or data-independent acquisition. It is suited to projects that process many samples and need to quantify many proteins in each sample.

Other advanced mass spectrometry technologies include ion mobility–mass spectrometry (IM-MS). It allows for the discrimination of isomers based on differences in their mobility through a gas-filled region while subjected to an electric field. Such technologies can overcome the limitations of conventional technologies to distinguish analytes of the same mass.

Harnessing automation

When transitioning from clinical work to laboratory research, Martin-Immanuel Bittner, MD, PhD, co-founder and CEO, Arctoris, was surprised to learn how much time was spent on highly repetitive manual experimentation. Simultaneously, he realized that the low productivity in drug discovery could be attributed mainly to low data quality, which also led to research failures and, consequently, additional costs. Reports showed that less than 25% of published results were reproducible.^4,5

“Scientific protocols are highly ambiguous, which impacts reproducibility and makes pooling data from different sources near impossible,” Bittner states. But protocols could be more definite. To illustrate this point, Bittner suggests that a protocol that says “mix the sample” could add details, such as the settings used on an automated mixer, where mixing is “clearly defined as x minutes at x speed at x temperature for a fully compliant, highly rigorous, standardized approach. If each protocol step is clearly defined and automated, it brings data to new quality levels.”

Typically, automation is thought of as a tool that accomplishes high-throughput screening by enabling a very small number of assays to be run over and over again. At Arctoris, however, automation is something that can be more versatile, provided it is implemented throughout discovery. Full automation, the company asserts, can run assays down to single-plate experiments, resulting in better quality data and shorter cycle times.

Arctoris uses its automation approach to help biotech companies manage projects more efficiently. Also, the company can perform a range of experiments, including experiments that might be inaccessible to some researchers. Both standardized assays and customized assays (for example, target-based assays) are available.

The fully automated Arctoris research facility can perform a selection of cell biology, molecular biology, and biochemistry experiments in a nonbiased, reproducible way, and it can run concurrent experiments around the clock seven days a week. The facility, which is modular and incorporates large robotic work cells within sterile enclosures, interconnects scientific instruments via a network of conveyor belts and robotic arms.

“There is no lengthy onboarding process,” Bittner asserts. “Instead, [there is] complete transparency about the entire process. Access to raw data is in real time. Price is per experiment and depends on reagent cost as well as degree of customization.”

“In the move toward AI-driven drug discovery,” he continues, “the quality of data becomes even more important because with machine learning, the ‘garbage in, garbage out’ mantra applies. Our automated platform produces high-quality, consistent, and structured data, resulting in quality inputs for the next generation of drug discovery efforts.”

Histological analysis of 3D cultures

Three-dimensional (3D) cell culture techniques are becoming more important as model platforms for drug discovery and biology. To facilitate handling of 3D spheroids and organoids for high-throughput histological analysis, a Purdue University team, consisting of engineering professors Thomas Siegmund, PhD, Bumsoo Han, PhD, and George T.C. Chiu, PhD, has developed a new technology, the collapsible basket array (CBA).

According to Siegmund, in the CBA, the 3D cultures reside in fluid-permeable baskets attached to a flexible grid that is submerged in microplate wells containing the culture media. After culture, the CBA is removed from the well plate and released from the carrier structure. The CBA then collapses, allowing the array to be fitted to a standard histology cassette for microscopy analysis of the 3D cultures. The grid can conform to a microplate of any size. The histology cassette is the limiting factor.

A U.S. patent application has been filed. Now that the technology is attracting interest, the Purdue Office of Technology Commercialization and the inventors hope to license it to an industry partner.

Making the CBA compatible with automated pipetting and robotics systems for end-to-end automation is a top priority, says Siegmund. The Purdue team believes the CBA will address obstacles with laborious, mostly manual, and low-throughput handling and analysis of 3D cultures, thereby streamlining the process of drug development and reducing errors occurring during transfer. Currently, 3D printed, larger scale manufacturing is under development.

References
1. Haymond A, Dey D, Carter R, et al. Protein painting, an optimized MS-based technique, reveals functionally relevant interfaces of the PD-1/PD-L1 complex and the YAP2/ZO-1 complex. J. Biol. Chem. 2019; 294(29): 11180–11198. DOI: 10.1074/jbc.RA118.007310.
2. Lin X, Ammosova T, Choy MS, et al. Targeting the Non-catalytic RVxF Site of Protein Phosphatase-1 With Small Molecules for Ebola Virus Inhibition. Front. Microbiol. 2019; 10: 2145. DOI: 10.3389/fmicb.2019.02145.
3. Rosa B, Marchetti M, Paredi G, et al. Combination of SAXS and Protein Painting Discloses the Three-Dimensional Organization of the Bacterial Cysteine Synthase Complex, a Potential Target for Enhancers of Antibiotic Action. Int. J. Mol. Sci. 2019; 20(20): 5219. DOI: 10.3390/ijms20205219.
4. Prinz F, Schlange T, Asadullah K. Believe it or not: How much can we rely on published data on potential drug targets? Nat. Rev. Drug Discov. 2011; 10: 712. DOI: 10.1038/nrd3439-c1.
5. Begley, C, Ellis L. Raise standards for preclinical cancer research. Nature 2012; 483: 531–533. DOI: 10.1038/483531a.