We took several approaches to identify actives in the SIE-bla qHTS. Use of both the 460 and Ratio readings were instrumental in determining active compounds. About 70% of the 1330 declared actives displayed Class 1-3 curves in both the measurements. An additional 25% showed activity in either reading (). In many of these cases, particularly for Class 1 and 2 curves, the discrepant activity assignments corresponded to compounds of low efficacy, near the threshold level assigned for background (). Having two means to measure activity allowed us to recover actives that would not have been possible using either measurement alone. Furthermore, 92 samples (7% of actives) were originally scored as inactive by our automated curve-fit program but showed high-quality curve fits of low efficacy and were therefore reassigned as Class 3 actives. While these approaches helped ensure recovery of many biological actives, some declared actives, especially compounds associated with Class 3 curves, may be false positives.
While qHTS is well suited to determine comprehensively the activity of a chemical library, it is critical to consider how, and with what level of confidence, activity is scored. This is important when profiling a compound’s activity across many assays as inaccurate assignments may lead to incorrect assessment of target selectivity. As shown here, the scoring of compounds as inactive can be less certain in cell assays because compounds of insufficient efficacy but well-fit titration-response curves may not be scored as active (). We found that a combination of curve-fit parameters (P value <0.05, x-intercept between −5 and 5 % activity and ≤2 masked points) applied to both 460 and Ratio readings were useful in recovering some of these low efficacy compounds.
The SIE-bla qHTS identified 11% of the screened compounds having some degree of activity. However, this percentage falls to ~2.5 % when only actives associated with high confidence curve classes (Class 1 and 2.1) are counted (). Because our screen was configured to recover actives in two modes, agonist and antagonist, the high confidence percentage decreases to 1.6% and 0.9% of the library for the agonist and antagonist modes, respectively. A substantial component of library activity was derived from compounds displaying lower potency and lower curve classes. The low potency actives were recovered because our qHTS included 9 uM and 17 uM library concentrations. Such actives are valuable for evaluating nascent SAR of chemical series. An additional factor contributing to the overall yield of actives was the occurrence fluorescent compounds within the library. The follow up counterscreen identified 374 as fluorescent, thus indicating that at least 28% of library actives were fluorescent artefacts.
Having identified these compounds as active, what level of confidence should be applied to the activity assignments? Curve Classes 1 and 2 provide reasonable confidence of indicating a compound as active. Such curves show activity above background at two or more concentrations and have goodness of fit values (r2
) of at least 0.8. Indeed, the highest quality curves, Class 1, 2.1 and sometimes Class 2.2, are all used for the initial clustering of actives for our SAR analysis 3, 4
. Class 3 curves are less certain indicators of an assay response, as these curves display activity at only one concentration, typically the highest tested, or have two or more points of activity but poor curve-fits (r2
<0.8). As well, the reclassified inactives mentioned above were categorized as Class 3. For the SIE-bla
qHTS, 24 Class 3 compounds, all of which showed two or more points of activity, were retested and 83% confirmed as active compared to 90% of Class 1 and 2 actives. This result indicated that this category of Class 3 was a reliable predictor of activity, though not as well as Class 1 and 2. For profiling purposes, compounds associated with Class 3 curves are conservatively viewed as inconclusive until subsequent retesting confirms the initial result and/or the activity is supported by SAR derived from Class 1-2.1 actives.
Our retrospective analyses have shown that titration-based screening identifies more actives than by use of a single concentration and therefore decreases the number of false positives and false negatives 3, 42
. Indeed, the retrospective analysis of the SIE-bla
qHTS indicated that 35% to 71% of Class 1 and 2 qHTS actives were scored as false negatives when the 1.5, 7.7 or 17 uM concentration and a 3 σ threshold of activity were used ().
Several factors contributed to the high percentage of false negatives. First, while sub-threshold activity may be detected at one concentration, other concentrations can show activity that exceed the threshold, particularly when multiple concentrations are fit to a curve (). Second, qHTS data can use activity thresholds lower than 3 σ and generate well-fit curves; for this assay, 30% was used as the activity threshold whereas 3 σ of background activity was 40%. Importantly, the qHTS method can make use of titration curves derived from the 460 and 530 readings in addition to ratiometric determination. This allowed us to identify compounds that showed sub threshold titration-response curves in the Ratio determination but exceeded threshold activity in the 460 measurement as well as assess potential cytotoxicity. In contrast, the individual 460 and 530 readings are difficult to use with single concentration data because of well-to-well variations in cell number and reagent dispenses, leaving only the ratiometric data to assess activity.
qHTS did not identify two reported JAK inhibitors as active; Tyrphostin AG 490 43
and SD-1029 44
were screened but determined as inactive. However, the SD-1029 sample that was screened failed QC, having an incorrect mass, indicating it was likely degraded or a different compound. While it is difficult to estimate what percentage of SIE-bla
qHTS was false negative without rescreening the collection, we expect a value of less than one percent. Indeed, if the 92 ‘rescued actives’ are considered as false negatives, the percentage of false negatives from the screen is about 0.8%.
A general degree of connectivity among the STAT, AP-1 and hypoxia signalling pathways can be estimated by comparing the number of SIE-bla actives shared with or distinct from AP-1-bla and/or HRE-bla actives. Actives were defined as having curve class 1, 2.1 or 2.2 (with >50% efficacy) in the ratio reading and class 1-3 in the 460 nm reading. Using these criteria, 20% of the SIE-bla actives were active in both AP-1-bla and HRE-bla assays, 45% were active in either the AP-1-bla or HRE-bla assay, and 35% were active in neither. This result indicates significant overlap between the three assays, possibly because of the commonality in the assay technology used (beta-lactamase gene reporter), common components utilized in the pathways, or cross-talk among signalling components between pathways. In addition, many of the small molecule actives likely modulate multiple targets, whether closely related members within a protein family (for example, kinases) or structurally unrelated targets.
qHTS found a number of selective compounds that will provide useful reagents for studying STAT signaling mediated by IL-6 and perhaps the biology of cervical cancer as well. Of the 32 active series identified from the qHTS, six were selective for SIE-bla
, showing opposite or no activity in AP-1-bla
assays. Three other series were active in SIE-bla
assays while five were active in SIE-bla
assays. While our comparative approach identified these series as STAT-selective, this interpretation should be held with some caution. Some selective actives may in fact be nonselective, because the promoter-gene reporters in the AP1-bla
may be less sensitive or efficient in recording transcriptional activation. Additional evidence of STAT selectivity can be sought using secondary assays involving JAK or STAT activation such as those described in 45
. The molecules reported here are likely to be important tool compounds for investigating this important pathway.