By Chau-Chyun Chen, AspenTech

A new and easy to use solubility model has been developed, which represents a significant advance in the accuracy of solubility prediction over currently available models. The value benefit from applying this modelling and prediction capability to characterisation of New Chemical Entities’ (NCEs’) solubility properties in the “Lead Optimisation” stage of drug discovery is potentially very significant to the pharmaceutical industry. Several leading practitioners in the domains of NCE solubility, best practice crystallization, and crystal polymorph stability screening, to name but a few, are already using the model to drive efficiency, cost reduction, and better decision making into their drug development programmes. The financial benefits may run into millions of dollars. Leading companies are already exploring the model in other areas of interest and challenge: excipient selection, solubility across pH ranges, solubility of New Biological Entities (NBEs), and selection of solvents for optimum chromatographic performance.

Figure 1 (right): High level drug development timeline, noting area of application for solubility modelling and prediction (yellow).

The Solubility Challenge:
Many questions in drug discovery revolve around the issue of solubility. However, scant NCE material in late drug discovery means that opportunity for experimental determination of solubility is rare, at best, and often rate limiting. This may for instance delay the development of solubility characterisation upon which valuable and far reaching business decisions can be made, such as decisions relating to which NCE to take forward into formal development, or how to process or formulate for manufacture.

Such “partially blind development” has the potential to lead to huge research and development spend on candidate drugs that may turn out to be “unprocessable” further down the track, or yield physical properties late in development (or even after launch!) that may ultimately delay or suspend market supply. Even in the recent history of drug research and development, there are precedents for such events. The HIV protease inhibitor Ritonovir suffered a late emerging polymorph - ultimately a function of solubility and crystallisation - that necessitated a reformulation after launch, temporarily impacting market availability.

As the number of NCEs is whittled down to a handful at the Lead Optimisation stage (see Figure 1.), solubility characterisation becomes increasingly critical. Whereas aqueous solubility is a key and high level determinant at preceding drug discovery stages, the impact of solubility at the “Lead Optimisation” stage fans out into myriad critical areas. These are usually investigated by chemists operating at the earliest stages of API (bulk) and DP (formulation) process development. These critical areas are summarised in Figure 2, split into those areas under current and published investigation (blue) and those for which solubility modelling and prediction may also add considerable value (red).

Figure 2: Critical areas of solubility characterisation in early stages of process development.

Most of the solubility prediction and modelling value opportunities typically arise at that stage of drug discovery where there is considerable overlap of activity between late stage Lead Optimisation and early stage Process Development. This alignment of overlap is illustrated in Figure 3.

Figure 3: Solubility modelling and prediction at the lead optimisation stage.

The Solubility Model:
To carry out solubility prediction a scientifically sound and thermodynamically consistent mathematical model is required, and this needs to be embedded in a usable software tool (including relevant solvent property databases, associated calculation tools, display graphics, user interfaces, and user workflows.) The model described here is the NRTL-SAC (Non Random Two Liquid – Segmentation Activity Coefficient) model1,2 and its related model eNRTL-SAC (for salts). These models represent a significant advance in the accuracy of prediction over other currently available models. Its novel “conceptual segmentation” approach to predicting solubility makes it uniquely positioned to handle the complex NCEs and solvent mixtures that are typically used in today’s challenging pharmaceutical research and development.

The model has been developed by, with patent applications filed on behalf of, AspenTech Inc. It will be commercially available as a set of comprehensive software tools in 2009. Fundamentally, its design enables solubility predictions based on as few as four solubility experiments (on four solvents of varying hydrophobicity and polarity) using only tiny amounts of precious NCE and, on running the results through the model, will predict solubility profiles in myriad other solvents and solvent mixtures. As such, it represents a quantum step forward in the accuracy of prediction of NCE solubility. Its rigorous thermodynamic framework underpins its leading capability in predicting the NCE solubility profiles required at the earliest stages of process development, and thereafter throughout ongoing process development.

The application potential of this capability is extensive and some companies have already been active in applying the model to some of their most challenging areas, with exciting results:

• Eli Lilly scientists have applied NRTL-SAC to screen solvents for processing steps with the requirement to maximize solubility and to reduce solvent usage3.  The NRTL-SAC solubility predictions were first identified from ten data points in six solvents at four temperatures. They then developed a protocol to automatically evaluate solubility in 120 pure solvents and 122,000 binary combinations. These “virtual” experiments, which took only 5 CPU hours to complete, were then repeated at different temperatures and pressures. The NRTL-SAC predictions identified promising solvent candidates and conditions which were then validated in the physical laboratory. A binary green solvent system, which has much enhanced solubility over the original solvent used in the lab, was chosen for scale-up. The study ably demonstrated the effectiveness of this solubility modelling technique. Eli Lilly have also carried out extensive work on modelling drug solubility to identify optimal solvent systems for crystallisation4, and are standardizing its use as part of their work flows in this area.

• Design of crystallization processes for the manufacture of API is a significant technical challenge to process research and development groups, and an equally rich seam of value to pharma. AstraZeneca have examined the role of solubility modelling and its application within the crystallization process design framework5.  NRTL-SAC has been demonstrated through the case study on Cimetidine as a valuable aid to solubility data assessment and targeted solvent selection for crystallization process design. The model is becoming their standard way of selecting the right solvent for optimal performance in the process steps associated with API manufacture and crystallisation in particular.  

• Bristol-Myers Squibb researchers have reported a modelling strategy for optimal solvent composition selection in the design of a new API process6.  They have developed a modelling strategy for solvent selection and process optimization for API processes, including: reaction, extraction, distillation, and crystallization.  This modelling strategy helps them identify a solvent composition “sweet spot” for the design of their API processes. 

• GSK researchers have developed an exciting methodology, utilising NRTL-SAC, for high-throughput crystal form screening, with a view to understanding and characterising the right solvent conditions to entice the most stable crystal polymorph to appear early on in process development. 7

So What Is The Benefit?
At the highest level, and for any R&D centric industry, the sooner improvements to new products (or the quality of decisions surrounding them) can be made, the greater the overall value potential. Value potential here is not just measured by value delivered to a product or process, but by “redundant cost avoidance” too. It is this that hits pharma’s “value sweet spot” square on, because pharma research is fundamentally much more “selection” than “instruction” in its nature, and so the biggest value impact may be felt in cost avoidance. This is where modelling and prediction comes into its own, and can yield millions of dollars of accrued value in the course of subsequent R&D and throughout the life-cycle of the new drug thereafter. This is illustrated in Figure 4: a value plot against timeline to launch a new drug.

Figure 4: Value potential and benefit plot against timeline for a new drug, noting area of application for solubility modelling and prediction (yellow).

Given the recent development of the NRTL-SAC model, and the long timelines for product development required for a new drug, value benefit can only be a qualitative estimation, based on current applications and anticipated capabilities. Notwithstanding the inevitably “estimated” nature of value benefit, any potential benefit should also be seen against the back drop of apparently ever increasing R&D costs.

In the 80’s and 90’s, relative R&D spending represented approximately 15% -17% of revenue for the average drug company. Today, that average is approaching 20%, and for some companies may exceed that level. Estimates have placed the cost of bringing an NCE successfully to market to be anywhere from $700 - $1200 million, over the course of 9 to 12 years of R&D. Some companies estimate that getting as far as completion of Lead Optimisation requires spend of some $300 million, over the first 4 - 5 years of research. With the discovery and development of high value medicines becoming harder and harder, and NDA annual submissions on the decline, the time is fast approaching where dramatic operational efficiency improvements in R&D will be as much a central plank for competitive advantage in pharma as it already is for manufacturing operations.

With the above in mind, the application of solubility modelling and prediction should add value in four major ways across any pharma R&D organisation:

A.    Efficiency Improvement: by driving up the efficiency with which NCE solubility can be fully characterised, and all the potentially advantageous effects that this can confer in the Lead Optimisation space. Literally hundreds of “experimental hours” could be reduced to just a few through the use of modelling and prediction software. This value may manifest as a reduction in cost through headcount reduction or an increase in throughput rate of NCEs in late discovery / early development. The latter is the likelier benefit route for companies with healthy pipelines of NCEs.

B.    Risk Management / Better Decision Making: by exploiting the predictive power of the model to drive more informed and earlier decisions relating to selecting the candidate drug to best progress with respect to its “processability” downstream in API and DP manufacture. This ultimately enables a more informed and better investment focus. Delaying or dropping candidate drugs exhibiting very significant process challenges could save time, money and resource, or direct attention to solving “knock-out” issues first, before devoting more investment. Additionally, this should augment an “eyes open” approach to portfolio management of NCEs in early development with respect to their risk profile for manufacturability.

C.    Speed To Market Launch / Continuity of Supply Post Launch: by enabling aspects of process development activity (often delayed owing to insufficient NCE material) to proceed earlier. This can be achieved by using the modelling and predictive power of the software to sidestep this common cause of delay by using prediction to replace what would otherwise be experimentally derived process design data. This may translate into earlier clinical trials, and just possibly faster to market. Further value may manifest by avoiding or reducing the emergence of unforeseen disasters downstream, which may severely compromise launch times or continuity of supply after launch. (Late emerging crystal polymorphs are a good example here, in which solubility characteristic of the active drug are permanently changed, and can impact dose form stability, and even bioavailability.)

D.    API & DP Manufacturing Process Performance and Cost Profiles: by enabling the informed design of many aspects of the API and DP manufacturing processes, such that the final developed process is: better characterised, optimised, greener, and higher in yield, thereby reducing cost of goods from the outset of launch. Areas of application that could yield significant improvement are listed in Figure 2.

Looking Ahead:
In Lipinski’s thought leading article on computational approaches to solubility in drug discovery 8, he states that, “the knowledge of the thermodynamic solubility of drug candidates is of paramount importance in assisting the discovery, as well as the development, of new drug entities at later stages.” Going forward there will undoubtedly be numerous opportunities and efforts to open up other areas of application for solubility prediction with NRTL-SAC. One example of particular interest under investigation is solubility modelling and prediction of biologically-derived or engineered macromolecules, such as monoclonal antibodies and genetically engineered proteins. This could be particularly exciting, as it appears that the segmentation nature of the NRTL-SAC model lends itself well to more and more complex chemical / biochemical entities. Another area of interest is solubility prediction in body fluid cavities of highly complex make up and therefore, potentially, of bioavailability? This may well be within the model’s grasp too, and is particularly valuable to discovery chemists who may have only a few micrograms of NCE available, but may need solubility profiles in multiple fluid cavity types.

The NRTL-SAC model is already demonstrating its value to the pharmaceutical industry. It is certain to become one of the key tools in the prediction of solubility in early process development and thereafter in the development workflow. As such, it should drive efficiency, speed, informed risk management, and cost reduction into the increasingly complex process of new drug discovery, research and development.

References:
1.    Tung, Tabora, Variankaval & Baken (Merck) and Chen (ApsenTech Inc.): “Prediction of Pharmaceutical Solubility via NRTL-SAC and COSMO-SAC.” Journal of Pharmaceutical Sciences, 2008, 97, 1813-1820.

8.    Lipinski C.A. et al., ”Experimental and Computational Approaches to Estimate Solubility and Permeability in Drug Discovery and Development Settings,”  Adv. Drug Delivery Reviews, 23, 3-25 (1997)