Closing the Loop: Developing an Integrated Design, Make, and Test Platform for Discovery

Abstract

The relatively slow cycle time within medicinal chemistry from synthesis to assay is constantly being challenged to help improve the efficiency of the discovery process. While both synthesis and assay have been automated to varying degrees, there has, until recently, been limited focus on the complete design, make, and test process. This Innovations article outlines the development of Cyclofluidic from inception through to the commercialization of a fully integrated closed loop design, synthesis, and screen platform.

One of the highlights of working as a medicinal chemist is receiving the latest set of biological data associated with both new and existing compounds. Did the structure–activity relationship (SAR) continue to hold up as expected? Are there new trends emerging? Which compounds are suitable for further investigation and how is the latest data going to influence the design of future compounds? It is this constant feedback of data for new molecules that satiates the appetite of the medicinal chemist ultimately resulting in new compounds entering into the clinic. Frustratingly, the lag between synthesis and assay results was and still is frequently measured in weeks.

It was during early 2007 that an intriguing proposition was being circulated around pharma companies to build a Design Make Test platform for small molecule discovery that would slash the cycle time of days and weeks to just hours. It would be achieved through setting up and financing a new company to undertake the technology development and build a fully integrated “closed loop” platform encompassing the design, synthesis, and assay of new molecules with immediate feedback of the results into the next design cycle–closing the loop. The company would be able to build on the pioneering work¹ undertaken by the GSK technology development department in the UK under Brian Warrington’s direction. A significant incentive was the benefit of potential eligibility for a grant from the UK government, which would provide support for a collaborative project in the areas of microfluidics and life sciences and derisk the partners investment.

While there was initial interest from a large number of potential participants in the project, it was only UCB and Pfizer who signed up to the 5 year staged equity investment in 2008 to initiate the project. This met the collaborative requirements of the UK DTI (later to become InnovateUK), Cyclofluidic was incorporated, and the grant was secured in the summer of 2008. Up until this point my role had been to champion the proposal and gain management support for the investment within UCB; I had not foreseen my personal future involvement being only later in 2008 amid changes within UCB that I successfully applied for a position within Cyclofluidic and joined at the beginning of 2009 as the first employee.

Starting out at the helm of a new company on day one is both incredibly exciting and daunting in equal measure. The challenges were numerous, while both a business and technical plan were in place as part of the funding process that still left many open questions. I was very comfortable with the scientific and technical aspects to be undertaken having been increasingly involved with the evolution of the proposal but may have taken a slightly different stance on some of the more challenging technical milestones had I known I was going to be responsible for their delivery. Running a business from day one, however, was an entirely new challenge! Company location, suitable facilities, and how best to identify and recruit staff were very much at the top of the list alongside the more mundane but no less important business related aspects including insurance, solicitors, accountants, etc. A career to date in a research environment did ensure that I was well equipped to find answers to the questions and rapidly move the business forward.

The business plan was to undertake a research and development program to build the platform (later to be christened CyclOps—Cyclofluidic Optimisation Platform) and then to move to revenue generation through the sale of hardware. The investors had certain preferential rights in terms of both technology access and purchase.

The first challenge was to assemble a suitable team with the breadth of skills required to deliver on the business plan. A very wide range of expertise was required from synthetic chemistry ideally with flow experience, biochemistry and assay development, engineering, software development, analytical chemistry, and systems integration. Recruitment is never an easy challenge; however, the ongoing churn in the UK pharmaceutical sector around this time did ensure that great colleagues were to become available. A Chief Technology Officer was successfully recruited during the first half of 2009 bringing considerable practical experience in chemistry technologies and integration alongside a keen desire to realize the approach.

The technical vision was a fully integrated design–make–test platform utilizing the flow paradigm for both the make and test components while maintaining the high data quality required by medicinal chemistry (Figure 1). This prompted an important early strategic decision: does the company develop bespoke hardware specifically for the platform or utilize a commercial products where available?

Figure 1.

Deceptively simple schematic in the business plan summarized the technical vision for the project.

The first focus was on the flow synthesis component—at its simplest, just the ability to pump a number of reagents into a reaction tube at a defined rate with heating and cooling options. It was quickly agreed that using a commercial product was the most effective use of the internal resources rather than trying to “re-invent” and build from scratch. This strategy of utilizing existing hardware where available and innovating when required became a core strategy for the company; with the advantage of vendor maintenance and software support.

The scale of synthesis was an important consideration for the project; the technical plan foresaw a microfluidic scale platform (a strict definition of microfluidic is still elusive; however, diameters of <1 mm and volumes measured in μL remain good starting approximations). Small scale offered the advantage of low reagent usage; however, as the reaction tubing diameter is reduced, the propensity to block increases and higher pumping pressures are required. The initial choice was driven by both pragmatism and existing experience settling on a glass chip based flow synthesis platform. Reagents were delivered through a bespoke pressure driven reagent dispensing “carousel” housing up to 24 samples.

The output from synthesis is rarely pure enough for direct submission to assay, and therefore, the integration of some form of analysis and purification capability was a core requirement. Building upon work undertaken at Pfizer, it was possible to rapidly integrate an HPLC system. Quantitation was required to derive meaningful assay results, and ELSD was the preferred option from a limited range. The product of interest was captured by sampling the middle of the HPLC peak of interest, the so-called heart cut, which provided material of both highest purity and concentration. A simple dilution assembly driven by differential flow rates between HPLC product and diluent streams proved adequate to provide a diluted sample suitable for assay.

Alongside the “make” components, attention was also being applied to “test” the biological assay component. The vision was to explore biological assays in a flow environment to complement the chemistry paradigm; however, there was no suitable commercial biochemical assay platform available at the time. The company was fortunate to gain access to a flow biology platform originally developed in Cambridge U.K. Exploring drug discovery assays in this flow based environment was an interesting challenge and a steep learning curve for the company underlining the need for not only excellent scientists but very practical individuals.

The assay platform initially utilized a custom glass chip for the assay with channel dimensions of approximately ca. 80 μm; early observations of laminar flow with colored aqueous dyes rapidly opened my eyes to the challenges of working in this environment. For practical reasons (lead time, cost, and ease of replacement) the glass chip was replaced with 75 μm ID capillary tubing, a simple switch that proved remarkably effective. Pumping at the very low flow rates with the required relatively high pressures determined by the assay chip or capillaries was not so straightforward—this required a nano-HPLC pump. A particular challenge was in achieving a reliable gradient at the very low flow rates of one channel at the extremes of the assay gradient. This was of particular concern in examples where the IC₅₀ value was close to the start or end of the gradient; to ensure reliable data adjustments in reagent concentration would be made to move the curve more toward the center of the gradient.

Much was learned along the way including the challenges of minimizing the adherence of reagents to glass surfaces, accurate pumping to achieve a gradient of reagents into the assay chip with time, and ensuring solubility of the reagents at all times. The data obtained was compelling, a rapidly sampled read out at a single point on the assay chip gave a very rich data set including at the all important assay curve inflection point (Figure 2).² There was also the additional unexplored opportunity to sample the read out at multiple points in the flow path to provide further information on the assay with respect to time.

Figure 2.

Representation of the data from flow biochemical assays of three known thrombin inhibitors. The data is from sequential runs and superimposed to a single graph. The red line represents inhibitor concentration gradient increasing with time, and the inhibitor data is shown as the three colored lines where the rank order is clearly demonstrated. Argatroban is the most potent compound and is the furthest left representing inhibition at the lower end of the inhibitor concentration gradient. Reprinted with permission from ref (2). Copyright 2017 Elsevier.

This flow biochemical assay component was readily integrated into the synthesis and purification front end, and some straightforward synthesis protocols were utilized to prepare and assay a small number of known thrombin inhibitors (Figure 3). The results generated were secondary to the overall platform performance whereby 14 compounds were successfully prepared and assayed in a seamless process in less than 24 h. This represented an important technical milestone for the company achieving the demonstration of an integrated make and test platform for discovery achieved within 2 years of day one.

Figure 3.

Schematic of the first implementation of a synthesis and screen prototype within Cyclofluidic. Syringes represent individual pumps, chemistry reagents were introduced using an in-house developed reagent carousel, and there was a continuous fluidic path from reagent through to biochemical assay. The entire process was controlled through a combination of in-house and proprietary software.

This prototype platform had provided an excellent learning experience and a strong foundation to enable the rapid progression to the next generation platform: Cyclofluidic Optimisation Platform truncated to CyclOps. This was a significant step forward in complexity requiring much more software engineering to enable the fully automated and closed loop SAR generation. This was of course an ongoing evolution of the prototype and much of the work continued seamlessly as the experience within the company was built up.

The synthesis requirements were revisited in the light of experience, and it was agreed to move away from the glass chip reaction vessel platform. While this format offered excellent chemical compatibility the range of scales that could be undertaken and the flexibility of configuration, e.g., to merge additional reagent streams or to readily change effective reactor volume was more limited than that possible with a tube-based reaction platform. The ability to rapidly tailor tube based reaction vessels to the chemistry and platform through material, diameter, and length was considered advantageous, and a commercial platform was selected to provide this with good modularity and flexibility of configuration. A growing body of literature methods on this platform was also a valuable resource to the company.

Walk away operation dictated that a reagent delivery system was required and would need to accommodate a reasonable number of reagents to enable the exploration of a useful area of chemistry space. This number was hard to define; however, 300 was a reasonable starting point enumerating to 1 million potential molecules (i.e., 100 cubed). A number of suitable liquid handlers are available, and one was selected based on its simplicity, reliability, and ease to control and proved an excellent choice. It was also felt that once integration had been demonstrated, it should be straightforward (although not necessarily trivial) to change to a larger capacity reagent deck if and when required.

The scope of chemistry that could be undertaken on the platform was of paramount importance, and considerable resources were expended on developing synthetic methodologies. Single step synthesis was straightforward, whereas multiple steps presented new challenges. It was readily possible to telescope together two synthetic steps for many transformations simply by introducing a second reagent set at a suitable point in the flow reactor. This did of course require considerable planning for reagent compatibility and dilution effects. Perhaps one of the more notable was sequential Suzuki and Buchwald–Hartwig coupling transformations. All reagents needed to be compatible and soluble in the appropriate solvent system and due consideration needed to be given to the dilution effect of introducing a second reagent stream on product concentration. Three steps were possible particularly if the last was a relatively straightforward deprotection such as tert-butoxycarbonyl removal. Longer sequences would require some form of intermediate work up, which was never explored by us; however, others have done this successfully.^3,4

Where metal catalysts were used, it is known that any residual metal ions can interfere in assays; a simple silica plug was used in line prior to any chromatography to trap polar material, and while no analysis was undertaken to ensure this was fully effective, the assay results proved consistent and reliable, including on retest with material prepared by alternate means. If contamination had been suspected, a range of metal specific scavenger resins were available that could be readily inserted into the fluidic line for their removal.

One area of synthesis that was considered particularly attractive in the context of the CyclOps platform was the scope for reagent free transformations removing the need to introduce further reagents and dilution. Work was undertaken in collaboration⁵ with Boston University to explore the use of a photolytic protecting group with promising initial results. Both photochemistry and electrochemistry are potentially very versatile techniques for multistep flow synthesis as neither necessarily introduce new reagents and may result in only limited benign side products, the more ideal scenario.

Solubility in flow synthesis is frequently a challenge, and the chemistry team developed considerable expertise in maintaining solubility on the platform. Simple modifications to the hardware proved invaluable, keeping reagents at a very slightly elevated temperature and avoiding thermal shock as reactions exited from higher temperature reaction tubing proved to be remarkably effective. Solubility, while sometimes a challenge, very rarely proved to be totally insurmountable particularly in the light of the requirement for new compounds to have good physical properties.

The initial HPLC integration work confirmed this an excellent option for purification and analysis utilizing ELSD for the quantitation although the choice of hardware changed. What was required was a more elegant solution to the reformatting of the purified material within the captured peak ready for a range of assays.

While the vendor software was used for the control of the HPLC and mass spectroscopic capability, the ELSD quantitation and heart cut peak capture was under the platform control. The timing of the peak exit from the HPLC is key, and only a small window was available to catch the all-important heart cut. Full control of this process including the pivotal ELSD integration curve enabled the ongoing refinement that consistently delivered the material with good accuracy. ELSD proved to be suitable for the vast majority of drug-like molecules and determining concentration of material immediately prior to the assay was considered an advantage—no question about dissolution as the sample was always maintained in solution.

A robust HPLC method also ensured the desired material could be recovered from poorly yielding reactions—as little as 2 or 3% on occasion, which proved a real benefit in multistep synthesis, and equivalent performance has also been reported in a similar context.⁶

The purified product typically exited the HPLC in the millimolar concentration range, while the assay required inhibitor solutions in the 10s micromolar range. The amount of material required for biochemical assays is vanishingly small (to this chemist’s eye). Only microgram amounts of material are utilized in individual assays, so working with standard 4.6 mm × 150 mm analytical columns provides more than sufficient material for a number of assays to be completed. The dilution required was an advantage, and a next generation reformatting station utilized an elegantly simple microfluidic protocol that proved consistently effective at delivering solutions of the desired concentration with low organic (typically <1%) concentration. This was a dilution driven process using a fixed dilution ratio with the result that different products were isolated at differing concentrations due to the variation in the synthesis. The total variability for a run could extend to an order of magnitude; however, this never proved an obstacle in the assay: to ensure an IC₅₀ value could be returned, a 3-fold dilution sequence was used for assays. Exploration of the use of a system that could deliver similar product concentrations was frequently discussed; however, no attempts were ever made to reduce to practice. We remained very mindful of the potential for residual organic to influence biochemical assays; however, at no time did this prove to be an insurmountable issue, although there were occasions where HPLC solvents were changed for this reason. This capability was the key enabler to join the synthesis and assay processes and protected as a trade secret rather than pursuing a patent.

The CyclOps platform required an alternate biological assay, and having demonstrated the effective delivery of an assay ready sample from HPLC, there was considerable scope to explore the options. The ubiquitous microtiter plate (MTP) format remained a natural choice given the range of hardware available both for reagent dispensing and reading of the data. Cyclofluidic chose to work with a European automation company to customize a single dispensing probe liquid handling platform. This accommodated a number of reagent reservoirs including cooling, the reformatting hardware that delivered the assay ready sample, and an arm to load the assay plate into the reader. This was a robust platform and the liquid and plate handling operations were operated by the proprietary control software with triggering from the in-house control software. Cyclofluidic was able to demonstrate robust biological data from a very wide range of assay types in both 96- and 384-well MTP formats always with good correlation with data obtained independently either by the company or a collaborator.⁷

A flexible optical cartridge based plate reader was selected for the platform, which enabled a wide range of optical read out assays to be undertaken. This included a wide range of fluorescence based techniques (including fluorescence polarization (FP) and homogeneous time-resolved fluorescence (HTRF), simple luminescence) all of which were effectively utilized. Direct measurement of binding is an area of considerable interest, and while Cyclofluidic did undertake some surface plasmon resonance methods, these were never integrated into the CyclOps platform; the flow based nature of these assays did, however, hold a particular attraction.

The optimization of an assay for the platform required a different emphasis compared to either a high throughput screening (HTS) type assay or a routine project assay. Data generated needed to be reproducible and comparable through an extended number of cycles, i.e., a number of days without intervention placing increased focus on ensuring reagent stability over this time period.

With the platform delivering new compounds and biochemical assay results, extension to include some form of physical chemistry measurement was highly desirable. Lipophilicity is ideally measured using a shake flask method; however, useful values can be determined using chromatographic methods. To this end, an adaption of a chromatographic Log D method⁸ was implemented on a second HPLC system taking samples from the purified material that went on to enter the assay platform. This provided a Log D value and opened up the option for not only optimizing on potency or selectivity but also metrics such as lipophilic ligand efficiency. This second independent HPLC method also provided valuable confirmation of product purity post-HPLC and an approximate concentration from the UV detector.

The software requirements for the CyclOps environment were significant and were broken down into three discrete areas: control, informatics (data capture and reporting), and design (Figure 4).

Figure 4.

Overall architecture of the platform software, while simple, belies the significant resource commitment to establish the working CyclOps platform and continually optimize the capabilities.

All control and scheduling software was written in house, a core requirement for an evolving, robust, and flexible platform control environment. A software engineer was an essential member of the team familiar with both the code and the hardware providing an invaluable resource in the ongoing development, troubleshooting, and operation of the platform. A proprietary software environment was used as the programming environment providing excellent control (via multichannel control boards) and data processing capabilities. As indicated earlier where possible the original vendor software was utilized and controlled either through an API or a simple trigger (e.g., contact closure) to initiate a defined protocol on the hardware.

Data extraction was not always straightforward, however, a range of options allowed effective operation and data capture; the simple fallback position was to let the hardware control software write the output files to defined folders and then poll for new data with the correct file name at suitable time points. Data processing was then undertaken in the programming environment or proprietary software packages giving full control as to how raw data was processed and then written to the database. A detailed description of the code is outside of the scope of this Innovations article and certainly beyond the capabilities of the author; however, it is worth noting that in total more than 10 000 lines of code made up the control and scheduling software!

Informatics and design were very different requirements, and it was agreed early on that these would be better outsourced given that both were anticipated to be shorter term engagements that would deliver the components for the platform. Informatics utilized a commercial software environment and included the building of a straightforward chemical registration system for reagents and products alongside a database for all results and platform information. A reporting capability to enable data access and review was also enabled within this platform. One of the benefits of automation is the ease in ensuring that all details about an experiment are readily recorded. For all the experiments run on CyclOps, over 200 parameters including results were stored: all the details from synthesis, purification and product capture, reformatting, biological assay CHI LogD, and purity alongside the desired output of SAR data. Automation also allows for an excellent level of reproducibility, which was demonstrated internally on numerous occasions.

The algorithm was a significant challenge as it involved developing a much more novel approach to compound design and, at the time, was well ahead of the current interest in all things AI. It also needed to be relatively fast to enable the rapidly evolving SAR results to be included in the next iteration on an ongoing iterative basis—ideally completing within minutes to drive the desired serial optimization with a rapid cycle time. The company outsourced this element to a very capable consultant to undertake this work on a contract basis. This worked well through a couple of relatively short full time engagements; the first covering a research phase to establish the best method to use and a second focused on the considerable testing required alongside integration with the platform software and informatics.

A number of methodologies were investigated including neural networks, Baysian, support vector networks, and random forest before the latter was selected and optimized for the CyclOps process. The chosen method utilized a number of molecular descriptors (including AlogP, H bond donors and acceptors, PSA) alongside ECFP-6 fingerprints to build the predictive model. The algorithm was also able to undertake full multiparameter optimization albeit with the requirement to fully define the nature and weighting for the various parameters. Evaluation of methodologies was facilitated considerably through a combinatorial library of MMPi13 inhibitors,⁹ which also provided a benchmark for a genetic algorithm method for compound selection and optimization. The shareholders were also forthcoming in making some sets of SAR data available to explore how the algorithm might perform utilizing the sparse matrix data set typical of a medicinal chemistry project. In this latter case, the algorithm proved to be particularly effective at efficiently identifying the most potent compound from the data set provided when implemented in a serial compound by compound sequence.

SAR optimization is multiparameter, and the algorithm needed to include this capability enabling, for example, optimization based on selectivity between two or more assays and/or calculated molecular properties. This ensured that the approach would be applicable to the types of SAR optimization undertaken by the medicinal chemist, however, at the cost of additional complexity at configuration, e.g., the weighting of each parameter. Outside of the integrated platform, the algorithm proved useful in exploring existing SAR and in reagent selection.

Key to delivering the value of the platform was the effective use of the algorithm. For any particular SAR elucidation, an area of chemical space was defined by the reagents available on the platform. Two strategies were available to the algorithm, an exploration of the breadth of the SAR dimensions or an exploitation of the SAR to select the compounds that best met the objective. Over a number of projects, the team built up useful guidelines on how best to operate the algorithm combining these two strategies. At the start of an experiment the focus would be on exploration, and as the SAR data was built up, a switch could be made to exploit this for the optimal predicted compounds. A combination of the methods proved most effective and defining when and how many times to switch strategy was built upon experience.

The CyclOps capability was built over three years or so and had been robustly tested during its implementation; however, now validation was required to demonstrate the capability, which the team felt would be best achieved through publication in a leading medicinal chemistry journal. A simple description of the platform and its capabilities would not be sufficient for publication and would need to include some new medicinal chemistry. Choice of chemistry and target was important; demonstration of real medicinal chemistry type synthesis on the platform was very important as was showing that the entire platform output was providing useful SAR data. A novel target would have been both exciting and challenging; however, there would be limited comparable SAR to provide context. A well precedented target would enable the reader to rapidly see the context of the results generated and judge for themselves the value or otherwise of the approach, and it was the latter that was adopted.

The chosen target was the kinase ABL; two assays were to be run in parallel, ABL1 and 2, alongside the LogD data that was generated by the platform. CyclOps was never intended as a hit generation tool, and so a defined area of chemical space was required to be explored utilizing the automated closed loop. Help was enlisted in computational chemistry to define a potentially novel area of chemistry, and the key building blocks were prepared in house. The algorithm was configured to increase potency against ABL1 activity for this key experiment. The platform ran in fully automated mode over a long weekend, completing 72 cycles without any intervention, a significant achievement! New compounds and valuable SAR was uncovered, and the work was promptly published¹⁰ demonstrating the potential for the approach.

Key elements of the approach were identified during this work: the medicinal chemist defined the chemistry to be explored and would clearly be the inventor on any resulting IP. The generation of synthetic reagents was a significant investment (and probably underestimated at the inception of the work), although perhaps not that dissimilar to many medicinal chemistry projects where advanced intermediates are prepared in bulk for subsequent derivatization. The biological data obtained proved robust with a number of compound results being reconfirmed offline; the platform software provided an excellent level of detail for experimental details.

One of the key capabilities of the platform is the ability to take a small amount of a crude synthesis product and rapidly deliver approximately 150 μL of an assay ready sample of known concentration in less than 30 min and to include lipophilicity from the integral chromatographic logD method. The platform was readily modified to enable this functionality in isolation requiring an alternate injection port for crude sample submission, defining a rack location for delivery of the assay sample and some considerable effort by the software engineer. This also had a positive impact on the scale of chemistry that could be undertaken, amounts of material required for an analytical scale purification are small, and it became possible to undertake reactions on the order of 100 μg and less to obtain purified material ready for assay. This opened up a number of opportunities including rapid access to assay data while optimizing challenging chemistry and SAR generation from very limiting reagents, e.g., the derivatization and assay of natural products and in late stage functionalization¹¹ of molecules of interest.

Moving on from the flow based bioassay prototype platform coincided with the start of a shift in strategy. When Cyclofluidic was conceived, the business plan was to build a prototype platform and then sell the platform or its component parts to pharma and biotech companies. During the early years of the company, the pharma sector tended to be moving away from large internal technology platforms and not making significant investments in internal capabilities. With input from the shareholders and more widely, it was agreed that the business should look at alternative options for revenues based upon the rapidly evolving capabilities. This came down to a choice of two options, either become a technology integrator providing the expertise to tackle complex automation and integration projects in the life sciences or to provide services based upon the envisaged capabilities of the Cyclofluidic platform. The latter was adopted based upon the perceived markets of both options and the company’s ability to compete effectively.

With the strategic direction agreed upon, investment was made in building the business development capability for the company. Recruiting a chief business officer spearheaded the efforts within the company, and a US subsidiary (Cyclofluidic Inc.) was incorporated. An experienced business development professional was recruited and proved to be an able and willing tutor about sales and marketing to the company.

Selling services to pharma and biotech is a significant challenge particularly for a small company with limited resources and capabilities. There exists a very significant market for medicinal chemistry services; however, it is also intensely competitive on a global scale and lead times for contracts are typically very long. The CyclOps paradigm of rapid design make and test cycles was viewed as an attractive opportunity, and many organizations were interested in the approach and how it may be applied. Cyclofluidic completed a number of collaboration projects, which demonstrated the capability to deliver new SAR data in a very efficient manner although only a small number were able to be published.¹² The ongoing translation of interest into long-term collaborations remained a challenge and probably no different to many other companies starting out.

The CyclOps platform capabilities were in place by 2013, and while no major changes to functionality were made after this time, considerable evolution and optimization of the process continued in parallel with the ongoing collaboration projects (Figures 5 and 6). During this time there has been an increasing interest in the opportunities for better integration of the synthesis and screening within medicinal chemistry.¹³⁻¹⁸

Figure 5.

Schematic of the CyclOps platform indicating the key components, fluidic path, and the optional functionality (yellow boxes) to omit either or both of synthesis and assay.

Figure 6.

Image of the final format of the CyclOps platform in the Cyclofluidic laboratories. The reagents are positioned to the left of the flow synthesis hardware within the fume cupboard on a single probe autosampler; HPLC and mass spectrometer are under the bench, below ELSD and VDU.

It is interesting to compare the approach taken by Cyclofluidic to some of the others in this arena. CyclOps was conceived to support the single iterative approach with the ability to integrate the latest assay results into the evolving training set as a core requirement; others have looked more at the automation of existing workflows. Lilly developed the automated synthesis lab to effectively industrialize the parallel medicinal chemistry process. A large footprint production line approach enabled parallel synthesis through a series of workstations. This approach offered considerable flexibility in the chemistries that could be undertaken (including microwave heating, flow based hydrogenation, and workup including SPE) at ca. 1 g scale. A viable alternative to the outsourcing approach that many companies undertook for routine parallel synthesis with full internal control over both timelines and IP.

AbbVie demonstrated considerable success in the utilization of segmented flow chemistry for parallel synthesis coupled with mass triggered purification to efficiently deliver compound arrays at ca. 10 mg scale. This is reported to have been very effective at reducing the timeline for array production to less than 5 days. Subsequently AbbVie established the integration of synthesis with assay using parallel chemistry automation in batch, LC-MS based purification, sample dry down, and dissolution prior to biochemical assay in a single process. This enabled the delivery of both compounds and biological data in ca. 24 h, a considerable improvement on the more normally quoted timelines in pharma companies.

Roche demonstrated the integration of flow based synthesis, LC-MS based purification and a microfluidic biochemical assay to deliver compound and data in the impressive time scale of 1 h. This approach was noteworthy in the use of a Taylor dispersion based approach to deliver a compound gradient for IC₅₀ determination.

Cyclofluidic was the only group to include a design element into the process to effectively close the loop between cycles. This introduces both increased complexity and also the requirement to have access to a wide range of reagents available to the platform. The latter has a significant impact on both set up time and costs particularly given that dependent on the evolution of the SAR many reagents may not be utilized in the experiment. The flexibility offered by the synthesis platform did offer the ability to mix synthesis types within a single experiment compared to a parallel synthesis based approach.

Cyclofluidic was committed to the flow chemistry front end from the commencement of the project. While considerable success was achieved, it did impose some constraints, and there was never the opportunity to fully explore reactions where flow offers considerable advantage, e.g., photochemistry and electrochemistry. Batch chemistry was never implemented; however, the ability to directly inject a crude synthesis sample to deliver either an assay ready sample or the assay result demonstrated the flexibility.

The AbbVie approach to integrating assays offers considerable flexibility in both the chemistry and the biology that can be undertaken. It may be viewed as more modular than the Cyclofluidic platform which should enable more complex parallel synthesis protocols to be undertaken and processed through to assay results. For smaller compound arrays both AbbVie and CyclOps would demonstrate similar throughput, however, for larger arrays around fixed chemistry AbbVie offers much improved throughput. The different objectives of these platforms was also reflected in their scale of synthesis.

Core to the Cyclofluidic approach was the integration of compound design and serial iterative operation. The ability to efficiently move through an area of SAR space and home in on the compounds of most interest is a very attractive goal. Whether this is best achieved through a single compound iterative approach or by the use of multiple compound arrays remains a challenging question and it would certainly take considerable more data to be available before a definitive answer could be determined. Both approaches are probably best suited to the earlier stages of the discovery process where medicinal chemistry is focused on the rapid generation of compounds with SAR for a relatively small number of assays.

Flexibility is certainly key to the success of these approaches, Lilly offered this in the breadth of chemistry that could be undertaken; however, there was no integration through to the assay. AbbVie described a platform somewhat modular in nature that should allow for a wide range of synthesis and assays to be incorporated into a similar workflow whereas Roche had focused on the speed element. Cyclofluidic was more constrained given the choice of flow synthesis, and the serial iterative workflow, however did demonstrate the potential impact that rapid processing of crude synthesis product to assay ready sample might deliver.

As with any technology development program there exists considerable scope for further innovation and optimization especially with hindsight. One distinct area originally flagged was automated route optimization, which Cyclofluidic did not pursue however progress is now being made including a commercial product for route selection.¹⁹

The platform configuration for single compound iterations had downsides; at any one time a large part of the functionality was not in use. Throughput could be effectively doubled by scheduling the next synthesis to commence while the previous sample is in assay and while this impacts on the design algorithm it is considered unlikely to offset the improvements in productivity.

The scale of synthesis undertaken was well in excess of the small amount purified and reformatted for assay, it is straightforward to envisage how parallel preparative HPLC could be introduced to provide milligram quantities of product in a broadly similar timeline. The collection of the crude reaction mixture would enable rapid access to further sample without the need for resynthesis, and this was successfully carried out during a number of projects. A significant reduction in scale was not explored but remains feasible given the core capabilities demonstrated and the small amounts of material required for even a number of assays in parallel.

The majority of the assays undertaken on CyclOps was straightforward fluorescence based readouts, which provided considerable scope for SAR generation. The ability to extend this further into other assay platforms, e.g., mass spectroscopy and surface plasmon resonance, would expand the scope of biology and binding measurements that can be obtained on the platform. Incorporation of ADME assays to the platform was an attractive option; rapid optimization of not only primary assays but also metabolic stability data would certainly add significant value and extend applicability of the approach. Introduction of cell based assays would significantly extend the biology capability; a number of automated cell screening platforms are available and incorporation into an integrated platform is conceivable. It will, though, add considerable complexity and require the infrastructure to provide cells in a timely manner.

There also remains considerable scope to review how the design component impacts on the overall performance of the platform and to assess alternative techniques and modes of operation. The computational methods continue to evolve, and given the current interest in automated design, make, test approaches alongside AI, and machine learning in general, this would be expected to catalyze further developments in the technology. The value of a closed loop platform to support AI methods in drug discovery has also recently been highlighted.²⁰

Medicinal chemistry involves many different processes and groups to execute effectively, and as such, its industrialization will remain a considerable challenge. Cyclofluidic committed to a well-defined specification at the outset and fully met these expectations. Looking forward a more modular approach would offer greater flexibility in all aspects of the workflow. What was pivotal to the successful delivery of CyclOps was the truly multidisciplinary team all entirely focused on the platform delivery and its full utilization to the benefit of collaborators.

Transitioning from an R&D focused technology company to a sustainable services business was going to be a challenge for any company. In discussions with potential partners, there was always a very encouraging level of interest in the technology, speed of iteration, and the advantages that this could deliver, although the approach did raise questions. Working with collaborators dictated a lead time to develop the platform methodology for both synthesis and assay prior to data generation which detracted from some of the speed advantages. The value of the algorithm led iterative approach was readily understood however more challenging to demonstrate. A true comparative experiment of algorithm driven SAR generation versus a medicinal chemistry team would have been both interesting and valuable; however, whatever the outcome, more than just a single example would be desirable. While the company did successfully complete a good number of collaborations with very positive feedback from a range of partners from biotech, pharma, and academia, unfortunately it never reached the momentum to become a sustainable business.

Cyclofluidic achieved significant technical progress in demonstrating the potential for rapid SAR generation utilizing a fully automated design make and test process. The algorithm falls within the definition of AI and the company was initially somewhat ahead of the curve in its utilization in discovery; however, it probably cannot take much credit for the current level of interest. There was a considerable amount learned and disseminated through the project, and this will have undoubtedly contributed to the significant ongoing interest in reducing cycle times through automation and integration.

Glossary

ABBREVIATIONS

ELSD

evaporative light scattering detector

HPLC

high performance liquid chromatography

MS

mass spectrometry

SAR

structure–activity relationship

SPE

solid phase extraction

Author Present Address

^† ATW Regulatory Services, Royston, Hertfordshire SG8 0LA, U.K.

Financial support was received from InnovateUK (Grant 300028) and the shareholders UCB and Pfizer.

The author declares no competing financial interest.