Info

>50

a Value obtained from average of triplicate analyses in both positive and negative ion modes.

b Values obtained from dose response curves, with the exception of Compounds 9 and 10, which were extrapolated from a single 10 mM dose in duplicate (Included with permission from SAGE publications. [10])

a Value obtained from average of triplicate analyses in both positive and negative ion modes.

b Values obtained from dose response curves, with the exception of Compounds 9 and 10, which were extrapolated from a single 10 mM dose in duplicate (Included with permission from SAGE publications. [10])

Fig. 4.3 Primary screening and deconvolution stage mass spectra. (A) The region of the negative ion mass spectra containing the ions of interest is shown for two compounds, in two replicate primary screening experi-ments. Full-scale y-axis intensity values are normalized to 308 counts per second for compound 4 (m/z @ 436) and 162 counts per second for compound 5 (m/z @ 498).

Fig. 4.3 Primary screening and deconvolution stage mass spectra. (A) The region of the negative ion mass spectra containing the ions of interest is shown for two compounds, in two replicate primary screening experi-ments. Full-scale y-axis intensity values are normalized to 308 counts per second for compound 4 (m/z @ 436) and 162 counts per second for compound 5 (m/z @ 498).

itive, and the logistics of manipulating a larger number of smaller mixtures would be difficult. While larger mixtures of compounds result in an increase in mass redundancy and therefore a concomitant increase in the number of compounds that need to be deconvoluted and retested, the overall efficiency is greatly increased.

There are approximately 2700 compounds per primary screening mixture, and the readout is in essence multiplexed; the ligands are individually ionized and identified in the mass spectrometer according to their exact mass positions. The readout, however, does not unambiguously identify compounds, as multiple compounds in a single mixture may have the same mass, i.e., a particular peak may correspond to as many as 31 compounds with closely related masses. The protein excess over individual compounds coupled with the rarity of potent ligands within a randomly assembled library minimizes competition between ligands for

174 4 Library Screening Using Ultrafiltration and Mass Spectrometry

B + Protein - Protein

Fig. 4.3 (B) Deconvolution experiments for compound 4. The compound is screened in a much smaller mixture than in primary screening and with no mass redundancy. Both round 0 data (before affinity selection) and round 3 data (after three rounds of affinity selection) are shown, where round 0 represents a sampling prior to any ultrafiltration. Round 0 and round 3 samples undergo identical denaturation/solvent extraction procedures. Data were generated

Fig. 4.3 (B) Deconvolution experiments for compound 4. The compound is screened in a much smaller mixture than in primary screening and with no mass redundancy. Both round 0 data (before affinity selection) and round 3 data (after three rounds of affinity selection) are shown, where round 0 represents a sampling prior to any ultrafiltration. Round 0 and round 3 samples undergo identical denaturation/solvent extraction procedures. Data were generated both with (+protein) and without any protein present (-protein), in order to observe whether compound retention by ultrafiltration is protein-dependent. Compounds shown here are observed in mass spectra in round 0 regardless of the presence or absence of protein in the starting sample and are observed in round 3 only in the presence of protein. Spectra are normalized to an intensity of 16000 counts per second.

available sites. In the theoretical case where the number of ligands overwhelms the number of target sites the apparent affinity of ligands will be reduced. Caution must be used in assembling libraries of either biomolecular or combinatorial origin because these could have problems with weak binding as a class [32, 33]. This could result in significant competition, making individual higher affinity ligands undetectable. With a sufficiently diverse collection of compounds this is not a concern.

Fig. 4.3 (C) Deconvolution experiments for compound 5, performed as in (B). Round 0 spectra in (B) and (C) likely are more intense in the presence of protein than in its absence due to protein preventing compound binding to the ultrafiltration membrane. Included from [10] with permission from SAGE Publications.

Fig. 4.3 (C) Deconvolution experiments for compound 5, performed as in (B). Round 0 spectra in (B) and (C) likely are more intense in the presence of protein than in its absence due to protein preventing compound binding to the ultrafiltration membrane. Included from [10] with permission from SAGE Publications.

The retesting/deconvolution phase of screening utilizes small mixtures of non-mass-redundant compounds. A balance of stringent rejection criteria and emphasis on reduction of false negatives is maintained during this phase. While in theory 90% of the compound is lost in each round of selection in the absence of protein, an actual protein-free selection is carried out for each mixture in order to increase accuracy and decrease false positives. At one extreme, compounds that form large aggregates [44, 45] could appear to be ligands in the protein-containing selection even if they cannot, in fact, bind to the target, but such compounds will also demonstrate an equivalent fraction retained per round of selection in the absence of protein. When this occurs, the fraction bound [Eq. (3)] calculates approximately to zero, and no KD estimate is made. Additionally, the set of four spectra [R0(+protein), R0(—protein), R3(+protein), and R3(—protein)]

for each putative deconvoluted ligand is scored visually for verification of appropriate ligand behavior. Based on a survey of several thousand randomly chosen compounds, approximately 80% of the library compounds are visible under the experimental conditions (data not shown). The remainder may be poorly extracted, show poor sensitivity to electrospray ionization, or have an incorrectly assigned structure and formula due to degradation or rearrangement during storage. Although a limitation of the method is its bias toward generally more MS visible compounds, compound structural series that are identified through traditional high-throughput screening techniques, such as fluorescence polarization assay, are also discovered in ASMS screens [37]. The total time required for the screen from primary screening through retesting and deconvolution is under three weeks, and it can be further reduced by automation.

The concept of promiscuous compound filtering was implemented for ASMS screening as a means to prioritize hits based on their potential value as drug leads, but it also may be used to prioritize targets. Note that 65% of the MurF hits resulted in overlap with the total combined PCF list. In 34 ASMS screens run against targets across several areas of pharmaceutical research, we have observed that 36-92% of primary hits for individual protein targets overlap with compounds on the PCF list. Since targets vary widely in their tendency to bind compounds on the PCF list, it is tempting to believe that targets with higher frequency of overlap with the PCF list will pose a more difficult challenge in drug discovery either because of a similarity to serum proteins or a binding site that is ideal for promiscuous compounds in general. Medicinal chemistry directed at these kinds of targets, even with initial leads that show some binding selectivity, may result in optimized compounds that have undesired binding affinity for other proteins if the nature of the active site on the target is inherently similar to other proteins in the ability to bind promiscuous classes of small molecules. Targets that result in hit lists with very high overlap with the PCF list may be considered less desirable even if a few selective hits are discovered, though this is only speculation at this point. The difference in the propensity of various proteins to bind to the major classes of promiscuous compounds is one of the more interesting results of these experiments and would require more study in order to fully understand all of the ramifications on the drugability of different kinds of targets.

Although there must be ''innocent bystanders'' present in the PCF list, the mass redundancy of the primary screening mixtures makes this unavoidable. The time cost of deconvoluting all serum binders would be prohibitive. Our strategy still costs the time spent in screening the 45 compound mixtures, six times (Fig. 4.2). However, in addition to allowing prioritization of hits and targets, the PCF list filtering also reduces the cost of retesting and deconvolution if the hits overlapping the PCF list are not pursued. For MurF, the initial list of 1147 matched compounds was reduced to 402 compounds by application of the promiscuous compound filter. This meant that only 30 mixtures of 13 compounds needed to be tested in the deconvolution step instead of 86 mixtures. At this rate, the investment in upfront promiscuous compound filtering is realized after screening just five targets. The research cost of attempting to optimize the chem ically intractable compounds and targets that application of promiscuous compound filtering may eliminate, however, is likely to be much higher. Importantly, the PCF list is used to electronically filter hits as a means of prioritizing hits, but no information is lost. One can also choose to deconvolute those hits that overlap with the PCF list, with the expectation that many, but not all, of the hits identified will subsequently be shown to exhibit non-specific protein binding, e.g., compounds 1-3.

ASMS is applicable to combinatorial and traditional libraries of small molecules, peptides, and carbohydrates, although with libraries that may share some non-specific affinity for particular targets, large mixtures should be tested to ensure that there is not significant aggregate binding of the mixture (discussed above). No protein tag or protein molecular weight constraints are required. Like other affinity techniques for HTS, ASMS identifies compounds that bind to a target without regard to function, and its speed, efficiency, and applicability to all soluble targets makes it appropriate for genomics and proteomics targets. For example, we screened the inactive form of a given kinase with the intent of identifying a non-active site binder that would prevent target activation required for downstream activity. In doing so, we isolated a small molecule that bound to an extraneous site that exhibits kinase specific and selective inhibition (data not shown). Of note is the ligand confirmation efficiency built into the system. In most HTS screen formats, chemical matter showing activity or binding must be independently confirmed for structural integrity [46]. In ASMS, ligands are identified from their mass spectrometric peak position, so the only opportunity for misidentification is via a structural isomer. ASMS can be complementary to activity screening, but also can be useful in identifying ligands for targets with particularly difficult or expensive activity assays. While one novel class of MurF ligands discovered here clearly was optimizable for in vitro potency, no whole-cell antibacterial activity has been demonstrated for this series, even after steps were taken to address potential issues of cellular permeability and active transport of compounds out of the cell [36]. The discovery of the MurF ligands demonstrates the utility and advantages of the lead discovery methods described here.

Was this article helpful?

0 0

Post a comment