In activity-based HTS campaigns, secondary counterscreens frequently are applied in order to exclude artifacts, such as compounds that inactivate the substrate or detection method rather than the target of interest. One advantage of an affinity-based HTS strategy is that there are no additional reagents, substrates, or cofactors in the assay to increase the potential for false positives due to such artifacts. Additionally, any false positive ligands that do occur should be similar for all targets because the assay format and detection method is identical for every target screened. Aggregated compounds that cannot pass through the 10 000 Da molecular weight cutoff filters are an example. Interestingly, because of the requirement in ASMS for ligands to be dissociated before MS detection, ASMS will not detect reactive compounds bound covalently to a target protein (a common source of false-positive hits in activity-based screens). Compounds that have promiscuous or non-specific affinity for a variety of proteins are another potential problem for HTS, including ASMS. Furthermore, while certain chemotypes seem to recur as non-specific hits in HTS (activity or affinity), some individual members within a class can have just enough selectivity to hit in very few screens due to the relatively high stringency. This can lead to a significant waste of time trying to optimize these into quality lead compounds, as the series can rarely attain drug-like selectivity.
After screening and deconvolution of primary hits from dozens of targets by our ASMS technique it became apparent that the frequency of compound overlap between targets was high but aggregated compound occurrences were very low. Aggregated compounds could be detected in two ways: (i) by showing apparent binding to every target in the primary screen, and (ii) by showing a very high retention in the absence of protein. Non-selective but non-aggregated ligands were discovered as expected, and exhibited a range of KD values as measured in decon-volution experiments. Therefore, simply adjusting the stringency or rejecting hits above a certain KD threshold cannot easily eliminate these non-selective ligands. Many distinct structural classes or chemotypes were observed, but a phenylsulfo-namide series represented by compounds 1-3 appeared most often.
Compound 1 has been resolved as a ligand for ten distinct protein targets out of 16 target screens run, compound 2 as a ligand for 13 out of 40 screens, and compound 3 as a ligand for 12 out of 45 screens. The apparent KD value depends on the particular target, with compound 2, for example, having affinities ranging from 1 mM to 30 mM for different targets.
We reasoned that a low stringency ASMS screening campaign might allow identification of non-selective ligands, thereby enabling prioritization of hits that are most likely to be useful leads by virtue of their relatively selective affinity. We also observed that different protein targets varied widely in their propensity to bind promiscuous ligands. This suggests that targets could be profiled for selective chemical tractability. Once a compound library is profiled for the non-specific ligands, the information can be used for all other screens to prioritize compounds for follow-up against specific targets and to prioritize targets based on the likelihood that they can be bound selectively by small molecules. Blood serum, whose principal component is albumin, is known for its ability to bind reversibly to a very large variety of ligands. For this reason, serum was employed as a model target for general non-specific binding. While serum protein binding can be engineered out in a medicinal chemistry campaign on a given series, and serum protein binding is not intrinsically a criterion for deprioritizing a particular lead compound, in the case of affinity-based screening, the leads most likely to be optimized into drugs should be those with selective affinity for a target. Subsequent medicinal chemistry for potency, pharmacokinetic, and other properties, will likely result in some degree of binding to serum proteins which needs to be considered in the context of the rest of the properties, but candidates with target-selective binding interactions make easier starting points for a medicinal chemistry campaign than do non-selective ligands, since non-selective hydrophobicity tends to increase during optimization [41-43].
Our goal was to divide the screening library into two populations, a set of compounds with extremely low probability of promiscuous binding and as small a set as possible that would contain all serum protein ligands under our ASMS conditions; the latter is called the promiscuous compound filter (PCF) list. To do this, screening was carried out at several serum concentrations and in several replicates in order to gather sufficient data for analysis. Figure 4.2 shows statistics from running 45 mixtures containing a total of 123 405 compounds in duplicate
Was this article helpful?