Considerable biological evidence has accumulated in support of nominating the Class I PI3Ks (phosphoinositide 3-kinases) as excellent targets for the development of novel pharmaceuticals to treat cancer and inflammatory disease. Although it remains a goal to deliver compounds with precise PI3K isoform selectivity in order to minimize safety risks, it is not yet certain that this approach will deliver suitable benefit against disease when tested in the clinic. The UCB strategy, therefore, has been to generate a range of compounds covering a broad spectrum of PI3K isoform inhibition. Scaffold diversity has been accomplished by identifying hits using both pharmacophore search and high-throughput screening campaigns, while modulation of potency and isoform selectivity has been achieved through exploratory medicinal chemistry. Simple, high-throughput cell assays relevant to either inflammation or cancer have then been employed to establish a blueprint for defining how isoform selectivity affects biological potency. I will focus on two compounds from our collection: a pan-PI3K inhibitor and UCB1311236, a compound with significant potency against only the PI3Kγ isoform. These examples will be used to illustrate the extent to which isoform selectivity informs on compound potency against other kinases and to highlight the risks and benefits of developing compounds with limited isoform selectivity.
- DNA-dependent protein kinase (DNA-PK)
- isoform selectivity
- phosphoinositide 3-kinase (PI3K)
- phosphoinositide 3-kinase inhibitor
Modulation of the intracellular PIP3 (phosphatidylinositol 3,4,5-trisphosphate) concentration is well established as a mechanism by which eukaryotic cells are able to respond to changes in their environment. A family of lipid kinases, of which the Class I PI3K (phosphoinositide 3-kinase) isoforms (α, β, γ and δ) are the most extensively studied, catalyse the final step in the synthesis of PIP3. An association can be made between increased PIP3 levels and a drive towards increased cell activation or proliferation, both of which are hallmarks of the progression of inflammatory disease and cancer. A significant advance in confirming the relevance of this association came with the discovery that tumour biopsies from patients with glioblastomas and gastric or breast cancers often contained activating mutations within the PI3Kα isoform . In addition, mice possessing catalytically dead PI3K γ or δ are unable to mount a full inflammatory response, yet they are otherwise healthy [2,3]. This information is buttressed by numerous publications, which describe the actions of the rapidly growing collection of chemical entities able to inhibit the catalytic activity of the Class I PI3Ks. In most examples, these inhibitors are able to elicit an effect that supports the hypothesis that Class I PI3K activity is intimately connected to the onset or progression of inflammatory disease and cancer. It must not be ignored, however, that these same inhibitors also alter the behaviour of numerous cell types that are responsible for physiological rather than pathological processes, warning us that we might encounter unwanted side effects if these compounds are applied to use in humans. For developing PI3K inhibitors as drugs, it is therefore important that three key questions are answered.
(i) Can Class I PI3K inhibitors with precise isoform selectivity be made?
(ii) Do these inhibitors retain sufficient efficacy to alter disease progression?
(iii) Are any side effects commensurate with the disease being treated?
The discovery of isoform-selective PI3K inhibitors
To date the only reported inhibitor with mono-isoform selectivity is IC87114, which its inventors at ICOS Pharmaceuticals demonstrated had significant activity against the PI3K δ isoform, marginal potency against γ and little or no activity against both PI3K α and β . An examination of how it bound in the PI3Kγ crystal provided a rationale for the selectivity as it was revealed that it causes a conformational change to the ATP-binding site not present in the ATP-ligated structure . This is an enormously significant result because the relatively conserved amino acid sequence around the ATP-binding pocket of the four PI3K isoforms had given little hope that it would be easy to generate isoform-selective inhibitors. Perhaps this plasticity exists at other sites allowing different compound classes to generate similar conformational shifts on the other isoforms. The implicit unpredictability of this phenomenon suggests that the screening of large compound collections against the PI3K isoforms is a valid approach for finding selective inhibitors and recent developments in high-throughput PI3K enzyme assays has made this increasingly possible .
PI3K enzyme assays must still be performed on each compound in isolation, however, making investment in technological advances that allow more extensive sampling of compound binding to the four isoforms worthwhile. To address this problem we have employed AS-MS (affinity selection MS; Figure 1). In this screening format, mixtures of ligands are combined with purified recombinant protein and unbound ligands separated away by gel filtration. Ligands that bind to the protein of interest are identified by their mass spectrometric signature and rescreened in isolation . The attraction of AS-MS lies not just in the high number of compounds that can be screened simultaneously in an automated fashion but also in its ability to isolate compounds that bind to inhibitory sites on the protein that might be missed in standard enzyme assays. In the case of the PI3Ks, for example, this might include the Ras or membrane-binding domains. The screen also depends on the ligands having relatively slow off-rates, which in theory lends itself to selecting compounds with complex, multisite binding modes often associated with increased selectivity. When PI3Kγ was employed as the test protein we positively identified four distinct compound series, one of which displayed sub-micromolar IC50 values in assays measuring the production of PIP3 by the γ isoform but no measurable activity in the same assay with PI3Kβ. Simple modification around the core structure led to UCB1311236, which had an IC50 of 258 nM against PI3Kγ but in excess of 20 μM against the other three isoforms (Table 1). Encouragingly, this compound series was also able to inhibit the production of ROS (reactive oxygen species) by human neutrophils stimulated with the chemotactic peptide fMet-Leu-Phe (N-formylmethionyl-leucylphenylalanine). This cellular assay has been extensively studied with regard to its dependence on the catalytic activity of the Class I PI3Ks and unequivocally assigned as a reliable readout of intracellular PI3Kγ activity .
Where possible, the AS-MS screen is preceded by establishing optimized buffer conditions with a series of standard ligands, which in the case of PI3Kγ could include compounds whose site of interaction has previously been determined by X-ray crystallography, for example LY294002 . Competition assays can then determine whether the newly discovered PI3Kγ ligand binds like LY294002 in the ATP pocket. This was the case for the series that led to UCB1311236, so we hoped that, like IC87114, these compounds would help to define a novel binding mode unique to the γ isoform. The X-ray structure of a selective compound closely related to UCB1311236 revealed that it did indeed bind to the ATP-binding site of PI3Kγ, but disappointingly caused little or no perturbation to the protein's conformation. Furthermore, its points of interaction were to amino acids perfectly conserved among the four isoforms making it impossible to determine the origin of its spectacular selectivity. Perhaps this is governed instead by subtleties in the overall three-dimensional shape of the compound, a point highlighted by the fact that relatively simple changes to UCB1311236 radically improved the potency against the other PI3K isoforms.
It is a commonly held view that compounds with selectivity against closely related enzymes such as the Class I PI3Ks will naturally show selectivity against all other enzyme targets. Unfortunately this need not be so and UCB1311236 provides a case in point. A screen of this compound against 50 protein kinases in separate ATP competition assays revealed no significant inhibition but we found it was able to potently inhibit the phosphorylating activity of DNA-PK (DNA-dependent protein kinase; Table 1). This enzyme is a member of the PI3K-related kinase family and shares significant sequence homology within its catalytic domain to the same region of the Class I PI3Ks . As DNA-PK activity has been associated with the repair of radiation-induced DNA damage, we believe this off-target activity of UCB1311236 may be a serious impediment to its development as a drug for the treatment of inflammatory disease. Further work is required to examine why DNA-PK is such a frequent off-target activity of otherwise selective PI3K inhibitors and how this contamination can be eliminated. Additional off-target activities of UCB1311236 cannot be discounted as we have found that compounds from this series show activity in cell assays believed to be driven by PI3K isoforms other than γ such as the inhibition of cell proliferation (results not shown). How might this occur given that we found this series had little or no potency against the other isoforms? The answer may lie in the recognition that, similar to other kinases, within their cellular context the PI3Ks are subject to an array of post-translational modifications and binding events that serve to alter their catalytic activity. It is already known, for example, that, through interaction with activated Ras, the PI3Kγ isoform displays an altered affinity to ATP, suggesting that the binding of ATP-competitive inhibitors could also be compromised . Perhaps the subtleties of UCB1311236 selectivity are lost when faced with PI3Ks in their natural setting. Building on from these ideas it might be prudent to employ more than one activation state of the PI3Ks when screening compounds in catalytic assays or binding screens such as AS-MS. This approach might even deliver more examples of isoform-selective inhibitors than would otherwise be found.
A proposal for inhibitors with limited PI3K isoform selectivity
Clearly the techniques used for identifying selective inhibitors of the PI3Ks are still in their infancy and in order to encourage their development we require further evidence that this approach is merited. One way to obtain this is to thoroughly investigate the biological properties of non-selective PI3K inhibitors. How safe might they be if used in the treatment of human disease? The most widely employed non-selective PI3K inhibitors are LY294002, wortmannin and the recently profiled PI103 . These compounds have proved invaluable to researchers attempting to define the PI3K dependence of physiological and pathological processes but it is also known that they display activity against a range of other enzymes and, in the case of LY294002, receptors and ion channels. It would be unwise, therefore, to use them as agents to define the limits for the clinical use of compounds that inhibit two, three or all four of the PI3K isoforms. Perhaps pan-PI3K inhibitors will be suitable anyway for the treatment of acute, life-threatening diseases such as late stage cancers where moderate side effects such as the alteration of insulin-dependent events could be tolerated. It is also likely that humans employ a different array of PI3K isoforms to perform tasks previously assigned to one isoform in test organisms such as rodents. This opens the possibility that isoform-selective inhibitors will not transfer their potency seen in animal tests when used in the clinic. This was highlighted recently in a study of the fMet-Leu-Phe-stimulated activation of mouse and human neutrophils . Careful dissection of the process using genetically altered rodents and a panel of PI3K inhibitors revealed that, whereas in mice the γ isoform was the only contributing PI3K, in human neutrophils both γ and δ isoforms played distinct roles. It is certainly true that human disease is rarely the result of one single pathological process. Rheumatoid arthritis, for example, is a disease that relies on numerous inputs from the host immune system but disease progression may also be promoted by pro-angiogenic stimulation of the endothelium . It could be argued, therefore, that a less selective inhibitor acting on a wider variety of cellular processes might be more beneficial in these multifactorial diseases. Finally, in these cost-conscious times, where delivering one drug is reported to require a US$800m investment , the possibility that a single, pan-PI3K inhibitor could treat several different diseases if proven safe enough starts to negate the financial risks associated with its development.
These considerations have led us to also generate a compound series exemplified by compound X, which display only limited PI3K isoform selectivity. This particular compound shows no activity when tested against a panel of 240 protein kinases, no activity against a panel of 50 receptor or ion channels and significantly lower relative potency against the DNA-PK than UCB1311236 (Table 2). This does not exclude the possibility that it inhibits enzymes not included in these test panels or inactive versions of the kinases, but to help demonstrate that this is unlikely, its inhibitory activity in a series of pro-inflammatory cell assays was shown to be significantly greater than compound Y, a closely related compound with reduced PI3K potency. The limited isoform selectivity of this inhibitor series raises its potential for development as a treatment for other diseases such as cancer but, of equal importance, we have the opportunity to examine what the safety consequences of inhibiting several of the Class I PI3K actually are. We know, for example, that compound X is able to inhibit the uptake of insulin-stimulated glucose by rat adipocytes and we are now examining how this relates to the alteration of blood glucose or insulin levels when it is present at and above therapeutic doses. These tests will be supported by other physiological readouts believed to be controlled by PI3K activity in addition to the normal routine of extensive safety tests employed in the selection of compounds that may one day be used in the treatment of human disease.
We believe that the Class I PI3Ks are exciting new targets for the pharmaceutical industry. The task of making drug-like isoform-selective inhibitors is difficult but it first needs to be proved worthwhile through a careful examination of the properties of compounds displaying potency across a wider range of the Class I PI3Ks.
I thank all my colleagues at UCB who have supplied the data and ideas that help shape this paper.
3rd Focused Meeting on PI3K Signalling and Disease: Biochemical Society Focused Meeting held at Bath Assembly Rooms, U.K., 6–8 November 2006. Organized and Edited by B. Hemmings (Friedrich Miescher Institute for Biomedical Research, Switzerland), B. Vanhaesebroeck (Ludwig Institute for Cancer Research, U.K.), S. Ward (Bath, U.K.) and M. Welham (Bath, U.K.).
Abbreviations: AS-MS, affinity selection MS; DNA-PK, DNA-dependent protein kinase; fMet-Leu-Phe, N-formylmethionyl-leucylphenylalanine; PI3K, phosphoinositide 3-kinase, PIP3, phosphatidylinositol 3,4,5-trisphosphate; ROS, reactive oxygen species
- © 2007 The Biochemical Society