Single-point SUPREX is an abbreviated version of the SUPREX technique, which exploits the H/D exchange properties of globally protected amide protons in a protein to measure the increase in a protein’s thermodynamic stability upon ligand binding in solution. In a full SUPREX analysis, the extent to which a protein, or protein complexed with ligand, undergoes amide H/D exchange during a specified amount of time is determined as a function of denaturant concentration (see ). Ultimately, the denaturant concentration at the transition midpoint is used to ascertain the thermodynamic stability of the protein or protein-ligand complex. Proteins complexed with ligand have a SUPREX transition midpoint that is shifted to a higher denaturant concentration than that of the protein alone, and the magnitude of the shift is related to the binding affinity.
In single-point SUPREX the extent to which a protein, or a protein in the presence of ligand, undergoes amide H/D exchange during a specified amount of time is determined at a single denaturant concentration. The denaturant concentration is chosen such that there is a maximum difference between the ΔMass values expected for the protein and for the protein complexed with ligands having the minimum binding affinity desired in the selection. Based on the SUPREX curves expected for CypA in the presence and absence of a hypothetical ligand that binds CypA with a Kd value of 10 µM (see ), a 1.5 M denaturant concentration was chosen for the single-point SUPREX experiments in this work. This allowed for the selection of library compounds with Kd values ≤ 10 µM. We reasoned that ligands with Kd values up to ~10 µM would be useful lead compounds in our search for CypA-targeted diagnostic and imaging agents.
The 1280 and 9600 compounds in the LOPAC and DIVERSet libraries, respectively, were screened for binding to CypA using the protocol outlined in . The protocol involved pooling ten library compounds per well of a microtiter plate prior to initiating the single-point SUPREX experiment. The library pools were prepared in 10 µl of deuterated buffer containing 1.5 M GdmCl and 10% DMSO, and the library compounds in each pool were present at a concentration of ~100 µM per compound. The single-point SUPREX protocol was initiated upon addition of CypA to each library pool. The CypA was allowed to undergo H/D exchange for 35 min in the presence of each library pool. Ultimately, the H/D exchange reactions were quenched upon the addition of a MALDI matrix solution, which also prepared the CypA protein in each sample for a MALDI-TOF analysis. The MALDI-TOF analysis was used to determine the number of amide protons in CypA that were exchanged with solvent deuterons during the 35-min H/D exchange reaction (i.e., the ΔMass value). The magnitude of the ΔMass value was used as described below to determine whether or not a given pool of library compounds contained binding ligand(s).
Figure 2 Schematic representation of the screening methodology used in this work. Library compounds were screened as 10-compound pools to improve throughput, and CypA was added and allowed to exchange. The mass increase was then used to determine whether a binding (more ...)
The pools of 10 library compounds, along with two sets of control samples, were screened for CypA binding using the single-point SUPREX protocol described above. The distributions of ΔMass values obtained in the analysis of the 1089 pools and 251 control samples are shown in . The control samples included 126 positive controls that contained CsA, which is a known tight binding ligand to CypA, and 125 negative controls that did not contain a ligand. The data obtained on these control samples, which were analyzed with every set of 10 pools, was used to generate the cut-off values for hit selection and to help establish the false positive and false negative rates of the strategy employed in this work. shows the ΔMass values obtained for the positive and negative controls obtained during the analysis of each library.
Figure 3 Distribution of ΔMass values recorded for the 1089 library pools (grey bars), 126 positive controls (black bars), and 125 negative controls (open bars) analyzed in the 1st tier screening of the two libraries used in this work. The solid lines (more ...)
Figure 4 The ΔMass values from the positive (●) and negative (○) control samples analyzed with the LOPAC library compounds (A) and the DIVERSet library compounds (B). The solid lines represent the average ΔMass values of the control (more ...)
In the LOPAC library screening, a cut-off value for the selection of hits in the first screening was set at 49.0 Da, which was 2.5 standard deviations below the average ΔMass value obtained for the negative controls. The ΔMass value of one negative control in the LOPAC library (see ) was determined using the Q-test at the 95% level to be an outlier18
, and it was not used in the cut-off value calculation. The ΔMass value cut-offs used for the selection of hits in the DIVERSet library screening were calculated by subtracting 2.5 standard deviations from a 9-point moving average of the negative control data. Thus, a specific ΔMass value cut-off was determined for each set of ten pools that were analyzed in the DIVERSet library screening. We previously showed that the use of such a moving average in the calculation of ΔMass cut-off values is useful in single-point SUPREX screening experiments that are performed over the course of multiple days, as was done in the DIVERSet library screening.5
Ultimately, a list of hit-pools (i.e., those that yielded ΔMass values below the cut-off values) was generated. A total of 8 out of the 131 LOPAC library pools and 13 out of 958 DIVERSet Library pools were identified “hit-pools” in this first tier of screening. The hit-pools from each library were re-screened in a second tier of screening. We have previously shown that such a two-tier screening strategy significantly reduces the false positive rate using the single-point SUPREX assay.5
The two-tier screening strategy is useful because the assay is subject to random error, which is reflected in the Gaussian distributions observed for the ΔMass values recorded here (see ). The main sources of random error include the mass measurement uncertainty and the differential back-exchange (see below). After the second tier of screening, five hit-pools were identified from the LOPAC library, and five hit-pools were identified from the DIVERSet library.
The pooling strategy described here ultimately requires that separate analyses be performed on the individual compounds in the selected pools in order to determine which compound(s) yielded the “hit.” (i.e, were responsible for the low ΔMass value). Thus, the 100 compounds from the ten hit-pools identified from the LOPAC and DIVERSet libraries were individually screened for binding to CypA using the single-point SUPREX protocol. Ultimately, nine compounds were identified as CypA binding ligands in these individual screens (see ). This included one compound from each of the ten hit-pools with the exception of one hit-pool that did not yield an individual hit. The binding affinity of these nine newly identified CypA ligands were then measured using the conventional SUPREX protocol.
Summary of the nine CypA ligands selected in this work.
Full SUPREX analyses were performed on CypA in the presence of the selected compounds in order to determine their affinity for CypA (). The individual data points in a full SUPREX curve are subject to the same ΔMass value uncertainties as those in the single-point SUPREX protocol. However, protein-ligand binding analyses using the full SUPREX protocol are much less sensitive to this error because multiple ΔMass values are used to define a full SUPREX curve (see ). The full SUPREX analyses indicated that 8 of the 9 selected compounds yielded measurable binding interactions with CypA, i.e., a shift in the transition midpoint of the SUPREX curves. The SUPREX-derived Kd values of these 8 compounds ranged from 0.200–37 µM, with the majority being in the low micromolar range ().
Figure 5 Full SUPREX curves obtained for CypA in the absence (●) and presence (○) of the nine selected hit-compounds. The solid and dashed arrows indicate the transition midpoints obtained in the absence and presence of ligand, respectively. Eight (more ...)
Dissociation constants for two hit compounds, 5 and 8 from the DIVERSet library, could not be determined due to difficulties associated with obtaining MALDI ion signals for the CypA in the presence of these compounds at high [GdmCl] concentrations. At these elevated concentrations of denaturant, a MALDI ion signal for CypA was not observed and multiple peaks at m/z 200–1000 Da higher than that expected for CypA were detected. However, the ΔMass values obtained at the lower denaturant concentrations used in our full SUPREX analyses of CypA in the presence of these individual compounds were consistent with those expected for a pre-transition baseline, suggesting that binding was, in fact, detected, and that the transition midpoint of the curve was at least 1.6 and 2.3 M for 5 and 8, respectively. Binding was not detected in our full SUPREX analyses of CypA in the presence of one of the selected ligands, 6 (i.e., the SUPREX transition midpoints for CypA were similar in the presence and absence of this ligand), which appeared to be a false positive.
The nine newly identified CypA ligands were also tested for their ability to inhibit CypA’s peptidyl-prolyl isomerase activity. Initial velocities of the chymotrypsin-catalyzed hydrolysis of the peptide substrate N-succinyl-Ala-Ala-Pro-Phe-p-nitroanilide were determined in the presence and absence of CypA and then in the presence of CypA and each newly identified CypA ligand. The substrate peptide is cleaved by chymostrypsin when the Ala-Pro peptide bond is in the trans configuration. CypA increases the apparent rate that chymotrypsin hydrolyzes the scissile bond in this substrate by catalyzing the conversion of the cis isomer to the trans isomer. The degree to which CsA and the nine hit compounds identified in this study impact the rate of this hydrolysis reaction in the presence of CypA was determined from initial rate measurements (see ).
Figure 6 Results obtained in CypA inhibition studies. The closed bars and left y-axis represent the initial velocities determined for the chymotrypsin-catalyzed hydrolysis of the peptide substrate N-succinyl-Ala-Ala-Pro-Phe-p-nitroanilide in the presence and absence (more ...)
The chymotrypsin-coupled enzyme assays revealed that seven of the nine compounds (including Compounds 1–5, 7, and 9) inhibited CypA’s isomerase activity, and the degree to which each of these ligands inhibited CypA was generally consistent with the amount of free protein calculated based on each ligand’s measured Kd value, with the exception of Compound 8 that appears to bind CypA with relatively high affinity but have minimal effect on CypA’s isomerase activity. It is not surprising that such a compound was identified in this work, as the single-point SUPREX assay does not select for compounds with specific activities, rather just binding interactions. The variety of compounds that can be selected in the single-point SUPREX assay (e.g., both those that alter a specific protein function and those that do not) can be an advantage of the methodology over other HTS approaches for certain applications, such as the one described here to identify molecular imaging agents. Interestingly, Compound 6, which was not found to have a measurable CypA binding interaction in our full SUPREX analysis (Kd value > 40 µM), did not significantly inhibit the isomerase activity of CypA isomerase. Compound 6 is most likely a false positive.
Throughput and Efficiency
The first tier of the 1280-compound LOPAC library screening was accomplished by researchers working over the course of one day for 3.5 hours, which was the time it took to perform the H/D exchange reactions and collect the mass spectral data. This equates to approximately 1.5 minutes per pool or less than 10 seconds/compound. The first tier of the 9600-compound Chembridge DIVERSet library screening was accomplished by two researchers working over the course of four days for a total of 22 hours, which again was the time it took to perform the H/D exchange reactions and collect the mass spectral data. This equates to a screening rate of approximately 1 min/pool or about 6 seconds/compound.
The overall throughput achieved in this work, 6 seconds/compound, is about 30 times faster than that previously reported for the single-point SUPREX protocol.5
In theory, the pooling strategy used here should increase the throughput of the original single-point SUPREX protocol by 10-fold. The additional 3-fold increase in throughput realized in this work is likely due to the use of a high-throughput MALDI-TOF instrument, which was equipped with a high repetition rate laser.
The false positive and false negative rates of a high-throughput screening assay can provide a measure of its efficiency. A total of 126 positive controls and 125 negative controls were analyzed during the single-point SUPREX screening experiments described here. Using the ΔMass cut-off values described above, false positive and false negative rates of 0% and 9% were determined from the control data, which are both identical to that previously observed for the control data in our Prestwick Chemical Library screening, which were 0 and 9%, respectively.5
During the screening of the pooled library samples, false negatives go undetected, and false positives are only detected after subsequent screens and/or analyses. A total of 10 hit-pools were initially identified after our analysis of the 1,089 10-compound pools. Subsequent analyses of the individual ligands from these 10 hit-pools identified hit-compounds from all but two pools, suggesting a false positive rate of 20%. Interestingly, one of the two false positive hit pools consistently produced low ΔMass values upon re-analysis, even though it failed to produce a hit-compound when the ligands from the pool were individually analyzed for CypA binding. This suggests that the higher occurrence of false positives in the pooled samples (i.e., ~20%) compared to the false positives in the controls (i.e., 0%), which were not pooled, may be a result of the pooling strategy. This is also supported by the observation that the 20% false positive rate determined from the pooled library compounds in this work was larger than the 0% false positive rate determined from the library compounds in our earlier screen of the Prestwick Chemical Library, which did not employ a pooling strategy.5
The 20% false positive rate in this work is in line with typical pharmaceutical screens, which have been estimated to have false positive rates of about 40%.19
A complicating issue in the pooling strategy is the potential for higher false positive rates due to promiscuous aggregation and non-specific binding of library compounds.20, 21
There have been reports of small molecules aggregating into particles with diameters ranging from 95 to 400 nm.21
Separate studies have shown that enzymes can be inhibited through reversible adsorption of enzyme onto the surface of such molecular aggregates.20
This phenomenon is worse at higher concentrations of small molecules relative to protein concentration. Sequestration of the protein of interest in a screening assay by aggregate particles would thus increase the protein’s stability due to non-specific binding. This could explain why some pools (with total small molecule concentrations of 1 mM in the analysis buffers) were selected as positive hits, but then none of the individual compounds from the pools would be selected as binding ligands when they were analyzed individually (with the total small molecule concentrations on the order of 100 µM).
A measure of assay robustness is the Z’-factor. The Z’-factor is maximized by increasing the difference between the ΔMass values of hits and non-hits and by minimizing the standard deviations of the ΔMass values. The difference between hit and non-hit ΔMass values is largely a property of the specific protein under study, and it is directly related to the number of amide protons that are globally protected in the protein’s three-dimensional structure. The standard deviations of the ΔMass values are limited by the mass spectrometer. The standard deviations obtained using the MALDI TOF mass spectrometer in this work averaged 3.7 Da during the screening of the LOPAC library and 3.4 Da in the DIVERSet. Generally, HTS assays with Z’-factors ≥ 0 are useful for screening large combinatorial libraries.13
The distribution of Z’ –factors determined from the positive and negative controls in this work is shown in .
The Z’ –factors values varied in this work mainly because the amplitude of the assay signal (i.e., ΔMass value difference between the positive and negative controls) varied over the course of the screening (see ). This was a result of differential back-exchange (a random effect of not being able to determine the ΔMass values in the MALDI-TOF experiment at exactly the same time after they are quenched) and the fact that the standard deviations for the mass measurements of some controls were large and clearly outliers. While the moving average procedure accounted for this differential back-exchange to set the cut-off values in the screen, it is clear that the back-exchange reaction negatively impacts the Z’-value. While there was a large range of Z’-values (i.e., −1.3–0.7), the large majority of Z’-values were positive with most clustered between 0.1 and 0.5 (see ) using the protocol described here. The distribution of Z’-factors observed here using our pooling strategy is very similar to that observed in our previous application of the single-point SUPREX assay that did not utilize a pooling strategy.5
In theory, the Z’-factors in the single-point assay should be normally distributed. The higher than expected number of low Z’-factors in here (), and in our previous work,5
is likely due to the back-exchange reaction, which compromise Z’-values by reducing the separation between hits and non-hits. These problems associated with the back-exchange reaction can be minimized by recording ΔMass values of the H/D exchanged protein samples in the MALDI-TOF experiment as soon after they are quenched as is possible.
Identification of Novel CypA Ligands
Full SUPREX analyses of the nine hit-compounds selected in our screen () revealed that eight of the nine selected compounds bound CypA with Kd
values in the low micromolar to high nanomolar range (). These eight hit-compounds are all novel CypA ligands. It is interesting to find that CypA binds to 4
(cisplatin) since this drug is a commonly used anti-tumor agent. CypA has been found to be overexpressed in many cancer types including non-small cell lung cancer1, 2
, pancreatic cancer22, 23
, endometrial carcinoma24
, and oral cancer.25
In addition, CypA has been shown to be associated with the progression of breast cancer.26
The mode of action for cisplatin-induced apoptosis is not completely understood but is at least partly through the generation of reactive oxygen species.27
It has previously been observed that cells overexpressing CypA have been more resistant to cisplatin-induced death by minimizing stress-induced apoptosis, while offering no additional resistance to other anti-cancer drugs.28
We hypothesize that this may be the result of CypA binding to cisplatin as shown in our results, thus decreasing the amount of cisplatin available to the cell to generate reactive oxygen species.
It has also been shown that Compound 2
(Bay 11–7085) inhibits the tumor necrosis factor-α (TNFα)-induced phosphorylation of IκBα, the inhibitor form of nuclear factor kappa B (NFκB), which serves to decrease expression of the inflammatory molecules ICAM-1, VCAM-1, and E-selectin.29
The specific protein target of Bay 11–7085 was not determined in the study. However, it was speculated that the phosphorylation inhibition might be due to inhibition of a signaling protein upstream of the IκBα. It has been demonstrated that CypA has the ability to induce phosphorylation of IκBα in inflammation sites caused by rheumatoid arthritis30
, and thus it is feasible to hypothesize that the mode of action of Bay 11–7085 involves CypA being inhibited by Bay 11–7085, which in turn decreases the phosphorylation of IκBα and in turn the activation of NFκB-activated expression of these inflammatory molecules. The other selected compounds from the LOPAC library have known roles. Compound 3
(ZM 39923 HCl) is a selective Janus Kinase 3 (JAK3) inhibitor, binding to JAK3’s ATP binding site.31
It has also been found to inhibit transglutamase during a HTS assay.32
(α-NETA) is a selective, fluorescent inhibitor of choline acetyltransferase and is used for the investigation of acetylcholine synthesis.33
To date, the hits from the Chembridge DIVERSet library have no previously known function to the best of our knowledge.
This study demonstrated the use of single-point SUPREX for small (<10,000 compounds) libraries. Using a pooling strategy, the throughput was increased more than ten-fold compared to what has been previously reported. However, one drawback of the pooling strategy was a higher false positive rate of 20%. While the pooling strategy did not specifically impact Z’-factors of the single-point SUPREX assay, it is important to recognize that the Z’-values determined in this work were generally between 0 and 0.5, with approximately one-third of the values being less than 0. As a result of these relatively low Z’ values, the assay requires at least a two-tier screening strategy. This was feasible for the relatively small (1000–10,000) compound libraries studied here. However, the relatively low Z’ values of the assay may limit applications to larger compound libraries. The results of this study confirmed the assay’s capacity to identify novel ligands to cyclophilin. In this case, eight novel CypA ligands were discovered, with one compound estimated to have a Kd of less than 200 nM. Additionally, the use of a traditional biochemical assay showed that seven of the eight novel ligands also inhibited the isomerase activity of CypA. The assay is unique among high-throughput screening assays in its generality (i.e., ligand selection is based solely on the introduction of structural stability and not necessarily on the modulation a protein’s specific biological activity).