Solubility Matters in Drug Discovery
The first point to note is that some aqueous solubility is absolutely critical for compound progression. There is little point in running in vitro assays with compounds sitting at the bottom of the tube. This applies just as much to potency assays as to ADME assays. The other thing that soon becomes apparent when considering solubility effects more widely is the need for careful design of dilution protocols. This is clearly true for potency assays where frequently an IC50 is being determined but, equally, I have often seen rather subtle mixing and (re-)dissolution effects in microsomal stability assays resulting in unexpected concentration-time profiles, typically an initial rise in concentration superimposed on the metabolism phase.
How much aqueous solubility is required? Well, it is difficult to give absolutes as every company will have its own protocols but, if we consider microsomal stability to be a critical assay, then typically this will be performed at around 1µM and so this becomes a hard cut-off for compound progression. Below this level, any microsomal data is unreliable and either compounds should not go into the assay or the data should not be reported. The former is obviously preferred to avoid wasting analyst time and reagents. Indeed, poorly soluble compounds will end up taking a disproportionate amount of time to process and interpret the data if it is not known that solubility is the issue, so these compounds are certainly worth screening out from that point of view. Another example is CYP inhibition assays run at a single fixed concentration, say 10µM. Similarly there is little point progressing less soluble compounds into this assay. Due to the indirect nature of the readout of these assays, i.e. the analyte is not the test compound, there will be no indication of a problem in the assay but as for microsomal stability assays the risk is in reporting false negatives.
With ADME assays there is the complication of protein binding but whether the compound is in the form of a solid particle, bound to plastic surfaces or protein amounts to the same thing, i.e. it is protected from (or not available to) the metabolising enzyme and the free or soluble concentration is too low. For Plasma Protein Binding assays there is still an aqueous solubility requirement as the compound has to remain in free solution on the buffer side (assuming measurement by equilibrium dialysis).
Types of solubility assays
High throughput kinetic solubility assays involve starting from a (typically 10mM) DMSO stock solution and adding this to the buffer of interest. Hence solubility will be measured in the presence of a certain amount of solvent, say 2% DMSO in the event of adding 1 volume of DMSO stock to 49 volumes of buffer. The physical processes that may be involved in the assay are mixing (dilution), precipitation, crystallisation and re-dissolution. Samples are left for a period of time to equilibrate, say 18-24h, and are then filtered prior to analysis to quantify the soluble concentration. The assay is easily compatible with, and would normally be run in, a 96 well plate format. Analysis by LC-UV (or alternative detection method such as ELSD if required) is most appropriate with MS used to confirm peaks of interest. Fast gradient (e.g. 1.5min) LC methods are appropriate – the same methods essentially as would also be used in open access systems and for high throughput QC. As a high throughput assay, single point calibration with a single standard dissolved in something like 100% DMSO or acetonitrile:water [50:50] is adequate. The standard concentration can be chosen to give best accuracy around the range of interest although it would probably not be advisable to go far below 10µM, and 50µM represents a good compromise to cover the range of the assay. Bear in mind that, without going to extreme lengths, 1µM is getting close to the limit of quantification for many compounds with “average” chromophores. Much below 1µM is the preserve of LCMS. Also, this assay as described here has a top concentration of 200µM above which solubilities can not be measured and compounds can not be differentiated. However, this is generally not relevant information as the focus of the assay is to identify poorly soluble compounds as discussed previously and a working range of 1 to 200µM is fit for purpose. More standards can be included for calibration to improve accuracy but there will be a cost in analysis time, so there is a judgement to be made as to whether that is worthwhile or relevant.
Although I have only discussed LC methods, other techniques based on light scattering (nephelometry) or microscopy evaluation can be used and are also high throughput although generally less accurate techniques most useful for binning solubilities which is, admittedly, what is very likely done with LC derived figures in any case.
Thermodynamic solubility refers to the situation where buffer is added to solid and, after an adequate period of time mixing or agitating in some way, say 18-24h, the sample is filtered and the test compound concentration in the sample determined usually by LC. Here the process involved is dissolution until a saturated solution is achieved that is in equilibrium with the solid present. The weight of solid initially taken is not used in calculation of solubility so doesn't need to be accurately known. However, it is useful information in the case of particularly soluble compounds as it can be used to indicate when a saturated solution has not been achieved and the compound has simply all dissolved in the added buffer and the measured concentration will not represent the solubility limit. In such cases, the experiment should be repeated with a greater ratio of solid to buffer or the result reported as > the measured value if that is sufficient information.
In terms of practical details, filter vials are very useful and convenient devices for these measurements. Typically 1-2mg solid may be added to the vial and 400-500µl buffer added, although this can be miniaturised using micro versions of the vials. The filter can be put in place so as to form a seal but leaving an adequate air gap to allow mixing. Vortex mixing or some sort of end over end rotational mixing is then required to ensure sufficient agitation of the sample mixture.
Analysis is generally by LC-UV and, as this is a low throughput assay, some method optimisation may be carried out e.g. to optimise the detector wavelength for low solubility compounds. Compared to a kinetic solubility assay with a range of interest of 1-200µM, thermodynamic solubility assays can present samples varying far more widely, say up to in excess of 10mM (mg per ml levels). Certainly, the filtrates for compounds at the higher end of solubility will require dilution (e.g. 10X in buffer) to bring the peaks within the linear range of the detector. Compounds at the lower end can be challenging. The problems relate to impurities in the sample and the likelihood of these being much more soluble than the compound of interest. Hence they become vastly more concentrated than the analyte in the filtrate. In such instances, the chromatogram can bear little resemblance to a QC of the compound in question. A fully dissolved sample analysed in an appropriate QC system to give a peak height within the linear detector range may show the compound to be “100% pure” and, at a first glance at the solubility analysis you may question if you have the right compound looking at the forest of peaks in front of you! There are two main options here. You can work on the LC method e.g. by extending the gradient, comparing low and high pH mobile phases etc until the analyte peak is resolved from the impurities and you have a useable method. An MS is certainly not required for this work but a diode array is almost indispensable to optimise the detection wavelength (favour the analyte compared to impurities) as well as giving confidence in the final method through comparison of spectra between sample and standard, and for peak tracking. Alternatively (additionally) you can make the problem much easier by repeating the experiment with less solid. This should yield a sample with the same test compound concentration since it was at its solubility limit but greatly reduced levels of impurities. You will soon develop a feel for how to approach each project and for what works best for your own workflows.
NMR can be used to measure solubility and has the advantage of not requiring a filtration step as only soluble compound is detected, nor dilution nor an individual standard for each compound. However, it is low throughput, generally much less sensitive than LC and will tie up an NMR instrument and operator. Nevertheless in certain circumstances where concentrations are high it is worth consideration if you have the capability.
Buffer (media) choices
Given that the purpose of a rapid screening solubility assay is to flag compounds with such poor solubility that further assays, including potency and ADME, on that compound will be compromised, it makes sense for the kinetic solubility assay to approximate the conditions of those other assays and so a phosphate buffer at pH 7.4 would be a generally sound choice.
If the objective is to mimic more closely intestinal conditions then a slightly lower pH of 6.8 might be chosen as a model of the upper GI tract. Clearly it is possible to run a range of pH values and generate a pH-solubility profile and others have included pH values of 5 as well as gastric pH at say 1.2 (HCl/KCl). Obviously any pH can be chosen but we are getting away from the original object and increasing the complexity and workload.
Thermodynamic solubility is much more tailored towards DMPK objectives such as predicting absorption. Hence media at gastric and intestinal pH are both relevant, e.g. 1.2 and 6.8. As bile salts in the gut have an important solubilising effect for lipophiliccompounds, it is also important to include a medium to model this such as incorporating 20mM sodium glycochenodeoxycholate (GCDC) in the pH 6.8 buffer, and comparing measured solubility in its presence and absence. Alternatively, simulated gastric fluid (SGF) and simulated intestinal fluids (SIF) are commercially available as convenient powders for preparation of these biorelevant media (available from Biorelevant). SIF is available in both fasted and fed state (so-called FaSSIF and FeSSIF) which differ in concentration, plus Version 2 FeSSIF has been updated to include digestive components (sodium oleate and glycerol monooleate). It certainly seems to make good sense to make use of these materials.
Spotting solubility issues via HPLC reproducibility
It is my experience that stability is a less common issue than solubility. I have often been informed that “Compound X is going off” and asked to check it out. Accordingly I would prepare a standard (in DMSO or acetonitrile:water [50:50]) and sample (in buffer) and begin to analyse by LC at intervals but what is actually seen is poor reproducibility rather than a steady decline in the sample. Shaking the sample will also dramatically affect the results. I am assuming here that an obvious precipitate is not visible at least without close observation or with a magnifier. The co-efficient of variation on peak area (%CV or RSD) is a very strong indicator of incomplete dissolution. For a well-functioning LC system, assuming a reasonable signal, the CV should certainly be less than 1%. Anything higher and the first thing to suspect is a solubility issue. For problematic compounds this may even be seen with the standard. Frequently, close inspection of the samples reveals a very fine suspension or slight opaqueness compared to the standard in pure solvent. The compound is “on the edge” of being in solution but this is being picked up by the intrinsically highly reproducible LC system. The other thing to check, of course, is the structure and clogP/pKa values as almost invariably aqueous solubility problems go hand in hand with highly lipophilic compounds. A magnifying loupe (10X) is an indispensable tool for examining samples in HPLC vials.
Some additional observations
Other important solubility experiments in discovery include formulations. Both solution and suspension formulations are of interest but I will cover that, along with related topics such as dissolution testing and formulation stability separately. Dealing with suspension formulations is particularly interesting.
Organic solubility is another related request that comes in occasionally from Chemists or others. One reason is to avoid DMSO in either in vivo or in vitro testing. Depending on the concentration this might only require a visual evaluation. One example from my recent experience was to support CYP inhibition assays from acetonitrile rather than DMSO stocks.
I have also been asked to quantify compounds in solutions before and after they have been passed through lengths of plastic tubing to mimic an in vitro infusion experiment. Binding to plastics is a well-known phenomenon related to lipophilicity, and well within the remit of the Analytical Chemist to carry out.
Solubility assays are notoriously fickle and deceptively simple. That they can be difficult to reproduce between (even within) laboratories is well documented. The devil is in the detail and frequently in the unrecorded detail at that. There are so many factors to control in terms of media preparation, temperature, physical scale, equilibration and timescales that are all too easily overlooked. I can offer one pertinent example from the time I first set up a kinetic solubility assay. I diluted DMSO stocks in DMSO to prepare standards, vortexed thoroughly (I thought) and analysed by HPLC. The solubility samples were left to equilibrate overnight. The next day I re-analysed the standards and found increases in peak area of up to 25%. This was not due to a change in the HPLC system. The point here is that if mixing DMSO with DMSO is not a trivial or rapid process then nothing can really be taken for granted! This was a salient lesson that I have not forgotten since...
If any of the issues discussed in these articles are of interest to you, please feel free to contact me directly on firstname.lastname@example.org for further information.