So apparently I’m a critic of the PAINS concept so maybe it’s a good idea to state my position. Firstly, I don’t know exactly what is meant by ‘PAINS concept’ so, to be quite honest, it is difficult to know whether or not I am a critic. Secondly, I am fully aware that many compounds are observed as assay hits for any of a number of wrong reasons and completely agree that it is important to understand the pathological behavior of compounds in assays so that resource does not get burned unnecessarily. At the same time we need to think more clearly about different types of behavior in assays. One behavior is that the compound does something unwholesome to a protein and, when this is the case, it is absolutely correct to say, ‘bad compound’ regardless of what it does (or doesn't) do to other proteins. Another behavior is that the compound interferes with the assay but leaves the target protein untouched and, in this case, we should probably say ‘bad assay’ because the assay failed to conclude that the protein has emerged unscathed from its encounter with the compound. It is usually a sign of trouble when structurally-related compounds show activity in a large number of assays but there are potentially lessons to be learned by those prepared to look beyond hit rates. If the assays that are hit are diverse in type then we should be especially worried about the compounds. If, however, the assays that are hit are of a single type then perhaps the specific assay type is of greater concern. Even when hit rates are low, appropriate analysis of the screening output may still reveal that something untoward is taking place. For example, a high proportion of hits in common may reflect that a mechanistic feature (e.g catalytic cysteine) is shared between two enzymes (e.g. PTP and cysteine protease)
While I am certainly not critical of attempts to gain a greater understanding of screening output, I have certainly criticized over-interpretation of data in print ( 1 | 2 ) and will continue to do so. In this spirit, I would challenge the assertion, made in the recent Nature PAINS article that “Most PAINS function as reactive chemicals rather than discriminating drugs” on the grounds that no evidence is presented to support it. As noted in a previous post, the term ‘PAINS’ was introduced to describe compounds that showed frequent-hitter behavior in a panel of six AlphaScreen assays and this number of assays would have been considered a small number even two decades ago when some of my Zeneca colleagues (and presumably our opposite numbers elsewhere in Pharma) started looking at frequent-hitters. After reading the original PAINS article, I was left wondering why only six of 40+ screens were used in the analysis and exactly how these six screens had been selected. The other point worth reiterating is that only including a single type of assay in analysis like this makes it impossible to explore the link between frequent-hitter behavior and assay type. Put another way, restricting analysis to a single assay type means that the results of the analysis constitute much weaker evidence that compounds interfere with other assay types or are doing something unpleasant to target proteins.
I must stress that I’m definitely not saying that the results presented in the original PAINS article are worthless. Knowledge of AlphaScreen frequent-hitters is certainly useful if you’re running this type of assay. I must also stress that I’m definitely not claiming that AlphaScreen frequent hitters are benign compounds. Many of the chemotypes flagged up as PAINS in that article look thoroughly nasty (although some, like catechols, look more ‘ADMET-nasty’ than ‘assay-nasty’). However, the issue when analyzing screening output is not simply to be of the opinion that something looks nasty but to establish its nastiness (or otherwise) definitively in an objective manner.
It’s now a good time to say something about AlphaScreen and there’s a helpful graphic in Figure 3 of the original PAINS article. Think of two beads held in proximity by the protein-protein interaction that you’re trying to disrupt. The donor bead functions as a singlet oxygen generator when you zap it with a laser. Some of this singlet oxygen makes its way to the acceptor bead where its arrival is announced with the emission of light. If you disrupt the protein-protein interaction then the beads are no longer in close proximity and the (unstable) singlet oxygen doesn’t have sufficient time to find an acceptor bead before it is quenched by solvent. I realize this is a rushed explanation but I hope that you’ll be able to see that disruption of the protein-protein interaction will lead to a loss of signal because most of the singlet oxygen gets quenched before it can find an acceptor bead.
I’ve used this term ‘quench’ and I should say a bit more about what it means. My understanding of the term is that it describes the process by which a compound in an excited state is returned to the ground state and it can be thought of as a physical rather than chemical process, even though intermolecular contact is presumably necessary. The possibility of assay interference by singlet oxygen quenchers is certainly discussed in the original PAINS article and it was noted that:
“In the latter capacity, we also included DABCO, a strong singlet oxygen quencher which is devoid of a chromophore, and diazobenzene itself”
An apparent IC50 of 85 micromolar was observed for DABCO in AlphaScreen and that got me wondering about what the pH of the assay buffer might have been. The singlet oxygen quenching abilities of DABCO have been observed in a number of non-aqueous solvents which suggests that the neutral form of DABCO is capable of quenching singlet oxygen. While I don’t happen to know if protonated DABCO is also an effective quencher of singlet oxygen, I would expect (based on a pKa of 8.8) the concentration of the neutral form in an 85 micromolar solution of DABCO buffered at neutral pH to be about 1 micromolar. Could this be telling us that quenching of singlet oxygen in AlphaScreen assays is possibly a bigger deal than we think?
Compounds can also react with singlet oxygen and, when they do so, the process is sometimes termed ‘scavenging’. If you just observe the singlet oxygen lifetimes, you can’t tell whether the singlet oxygen is returned harmlessly to its ground state or if a chemical reaction occurs. Now if you read enough PAINS articles or PAINS-shaming blog posts, you’ll know that there is a high likelihood that, at some point, The Great Unwashed will be castigated for failing to take adequate notice of certain articles deemed to be of great importance by The Establishment. In this spirit, I’d like to mention that compounds with sulfur doubly bonded to carbon have been reported ( 1 | 2 | 3 | 4 | 5 ) to quench or scavenge singlet oxygen and this may be relevant to the ‘activity’ of rhodanines in AlphaScreen assays.
The original PAINS article is a valuable compilation of chemotypes associated with frequent-hitter behavior in AlphaScreen assays although I have questioned whether or not this behavior represents strong evidence that compounds are doing unwholesome things to the target proteins. It might be prudent to check the singlet oxygen quencher/scavenger literature a bit more carefully before invoking a high hit rate in a small panel of AlphaScreen assays in support of assertions that literature has been polluted or that somebody’s work is crap. I’ll finish the post by asking whether tethering donor and acceptor beads covalently to each other might help identify compounds that interfere with AlphaScreen by taking out singlet oxygen. Stay tuned for the next blog post in which I’ll show you, with some help from Denis Healey and the Luftwaffe, how to pollute the literature (and get away with it).