Auditing the Pharmacological Accounts for Hsp90 Molecular Chaperone Inhibitors: Unfolding the Relationship between Pharmacokinetics and Pharmacodynamics¹

Defining the precise relationship between PK³ and PD should be a mandatory component of the development of all new drugs (1, 2). In this issue, Chiosis et al. (3) report interesting new data on the unusual cellular uptake properties of the geldanamycin class of natural product inhibitors of the Hsp90 molecular chaperone and discuss the impact of this behavior on the pharmacological effects of these potentially important new agents. The present commentary will place the new results within the broader context of the need to develop a detailed understanding of the connections between molecular target status, PK, and PD, and the biological and clinical outcome. Most of the points to be discussed are generic to the development of all new drugs.

We teach our students that PK is what the body does to the drug and PD is what the drug does to the body (4). In drug development, it is extremely valuable to establish a well-understood pharmacological “audit trail” (5) that links all of the essential parameters of drug action, from the molecular target to the clinical effect. This encourages the setting of performance criteria for all of the key properties of the drug. It is especially important to monitor the connections between the following parameters: (a) the expression or status of the molecular target, biological pathway, or other prognostic biomarkers; (b) the achievement of active blood and preferably tissue concentrations of the drug; (c) the demonstration of activity against the intended molecular target (e.g., inhibition of an enzyme or antagonism of a receptor) ideally in cancer cells or, alternatively, in surrogate normal cells; (d) the modulation of the biochemical pathway in which the molecular target functions (e.g., the Ras to Erk1/2 and PI3-kinase pathways); (e) the induction of a downstream biological effect (e.g., the inhibition of proliferation, cell cycle progression, survival, invasion, or angiogenesis); and (f) the achievement of a clinical response. By making measurements at each of these hierarchical levels of drug action, it is possible to describe and to understand the performance of the drug at each stage.

The concept of the pharmacological audit trail is a particularly useful one because it allows us to answer two absolutely key questions: “How much gets there and what does it do?” (5). It compels the investigator to ask a more detailed series of difficult and searching questions about how the drug is shaping up and leaves nowhere to hide if the desired performance is not achieved. It also allows the identification of situations in which the relationships between the various parameters are not consistent. Additional studies can then be carried out to identify the causes of any inconsistencies, and this can greatly help the rational development of the drug. Such is the case in the article by Chiosis et al. (3), as will be discussed later.

Critically important questions must be addressed as early as possible. For example, at the target level, what is the expression in the cancer cells of the molecular target, is it wild type or mutant, and what is the status of the biochemical pathway in which the target operates? What is the expression of genes and proteins that may be involved in cellular sensitivity or act as biomarkers of response? Is it possible to administer the drug in such a way as to achieve active concentrations in the circulation and at the tumor site? If so, is the molecular target sufficiently modulated in surrogate tissue (e.g., peripheral blood lymphocytes, buccal mucosa, or skin) and, preferably, is the target acted upon in the tumor itself? And does this then result in an appropriate impact on the cognate biochemical pathway and the desired downstream biological effect? Finally, does this all translate into clinical benefit for the patient?

A generic pharmacological audit trail is illustrated on the left hand side of Fig. 1. There are both intellectual and practical advantages to using this audit trail approach. From a scientific point of view, clinical trials of new therapies become more rigorous, and the ability to demonstrate proof of concept and to test a mechanistic hypothesis is greatly enhanced. PK/PD-driven clinical trials may be challenging to conduct because of the measurements that need to be made, but the quality of the decision-making is much improved, and the selection of, for example, the optimal dose and schedule can be made with greater reliability and confidence. When critical performance criteria set for the drug cannot be met, for example, the achievement of an active drug concentration in the plasma or the ability to modulate the molecular target, pathway, or biological effect, then appropriate action may be taken. This might, for example, involve terminating the development of the drug. Or switching to a follow-up or back-up drug candidate that has the potential to overcome the particular limitations identified. Such early “no go” decisions are extremely important for pharmaceutical companies because they can save the many millions of dollars that would otherwise have been spent on later-stage efficacy trials that have no hope of showing success. Patients are also spared unnecessary exposure to drugs that are unlikely to be effective. In certain instances, a drug may be killed off because active exposures cannot be achieved or because the molecular target cannot be inhibited. In such cases, the limitations may be overcome with alternative inhibitors in the class. In other situations, one might discover that the molecular target or pathway can, indeed, be modulated by the drug, but no therapeutic benefit can be observed as a result. This is especially important to know, because it effectively shows that the molecular target or pathway is not important in the particular disease context under study. Evaluating other members of the class would then be a waste of time. The resources available, which are inevitably finite and limited, can then be transferred to other tumor types or to alternative targets and approaches.

The molecule chaperone Hsp90 is an exciting new molecular target for cancer treatment (for detailed recent reviews, see Refs. 6 and 7). The job of a molecular chaperone in the cell is to assist other, nascent proteins to fold into the correct shape and hence to exert their intrinsic biological activity. Although Hsp90 is a Heat shock protein (hence Hsp) which can help rescue the cell from proteotoxic damage, under normal physiological conditions, the role of Hsp90 is to ensure the correct folding, stability localization, and activity of a relatively small group of proteins, known as “client proteins.” Hsp90 has emerged as a cancer target over the last few years because it is overexpressed in malignant cells (see Refs. 3 and 7) and is involved in the conformational stability of a celebrity A list of client proteins that are mostly all oncogenic in nature. These include Raf-1, Akt/PKB, ErbB2, Cdk4, Polo-1, Met, estrogen and androgen receptors, telomerase hTERT, and mutant p53 (6, 7). Inhibition of Hsp90 leads to the simultaneous degradation of all of these oncogenic client proteins by the ubiquitin proteasome pathway. Because a large number of oncoproteins are degraded at the same time, Hsp90 inhibitors may provide a one-step combinatorial attack on multiple oncogenic pathways (6, 7). This is very important because blocking oncogene addiction in cancer cells (8) may require the simultaneous modulation of several “mission critical” oncogenic pathways (9) to produce an optimal therapeutic effect, especially in the solid epithelial tumors that harbor multiple genetic abnormalities.

Over the last few years, Hsp90 inhibitors have demonstrated proof of principle in cell culture and animal models. Attention has focused most heavily on two classes of natural product Hsp90 inhibitors, namely, the geldanamycin benzoquinone ansamycins and the radicicol class. More recently, a purine-based class of lower molecular weight, fully synthetic Hsp90 inhibitors has also been described (for more details, see Refs. 6 and 7 and for chemical structures see Chiosis et al., Ref. 3). All of these agents work by binding to the ATP/ADP binding site of the NH₂-terminal domain of Hsp90 and thereby blocking the intrinsic ATPase activity that is an essential requirement for its chaperone function (10). The nucleotide-binding pocket in 17AAG is structurally quite distinct from that of kinases and is closer to that of the topoisomerase/gyrase family. In view of the key role of the molecular target in multistep oncogenesis, there is interest in developing a range of Hsp90 inhibitors that act in this way (7). However, much of the current activity is focused on a particular geldanamycin analogue known as 17AAG, especially because this agent is now completing Phase I clinical trials.

17AAG was shown to have a better therapeutic index in animal models than the parent antibiotic geldanamycin, particularly with respect to the ratio of doses required for the inhibition of tumor growth compared with the doses causing hepatotoxicity (11). 17AAG was shown to deplete client protein levels in cancer cell lines in culture and to inhibit the growth of the cells in a way similar to that with geldanamycin (12–14). Furthermore, 17AAG caused similar depletion of client proteins and an impressive cytostatic effect against various human tumor xenografts grown in immunosuppressed mice, including colon, breast, and prostate cancers and melanoma (15–18).

As a result of its action on a novel cancer target and its promising antitumor activity and therapeutic index in xenograft models, 17AAG entered five clinical trials sponsored by the United States National Cancer Institute, one of which was carried out in England under the auspices of Cancer Research UK. Preliminary reports on these trials have appeared in abstract form (19–24).

The majority of drugs are generally less potent when tested against intact cells as compared with evaluation against the molecular target in cell-free systems, for example, using the purified or recombinant protein. This can be for a number of reasons, in particular, factors relating to cellular uptake of drug. In the case of protein kinase inhibitors, it is well known that IC₅₀ values for enzyme inhibition and phenotypic modulation in intact cells are an order or more of magnitude higher then the corresponding values obtained using the recombinant kinase target. One of the major reasons for this is that protein kinase inhibitors are generally competitive with ATP, the concentrations of which are much higher (millimolar) in intact cells than those that are normally used to calculate IC₅₀ values for the target enzyme (micromolar range).

Interestingly, the discrepancy between cellular and target potency for 17AAG goes in the opposite direction. The drug is more potent in cells than against the recombinant Hsp90 target. As Chiosis et al. (3) point out, the affinity of 17AAG, geldanamycin, and related ansamycin benzoquinones for the Hsp90 proteins, is in the low micromolar range (25–27). In contrast, the average IC₅₀ values for 17AAG and geldanamycin against a large panel of cancer cell lines were 220 nm and 47 nm, respectively (15).

The discrepancy was known in the field but had not been explained mechanistically before. Chiosis et al. (3) now demonstrate convincingly that ansamycin benzoquinones accumulate in cells against a concentration gradient. This occurs to such an extent that the exposure of cells to a concentration of 10 nm can lead to an intracellular drug concentration that is in the micromolar range.

In one interesting set of experiments, Chiosis et al. show that growth inhibition against MCF-7 breast cancer cells related more closely to the total molar amount of 17AAG or geldanamycin in the culture medium, rather than to the concentration of added drug. The volume in which the drug was added to the cells was not important. This is very unusual: cellular sensitivity is normally related to the concentration of drug to which the cells are exposed. In contrast to the results obtained with 17AAG and geldanamycin, the unusual discrepancy was not seen with the synthetic purine Hsp90 inhibitors PU3 or PU24FCl (3). It probably also does not apply to the chemically quite different natural product Hsp90 inhibitor radicicol, which is potent against the Hsp90 target and against cells (15).

The uptake of radiolabelled geldanamycin into PC-3M prostate cancer cells was measured in cell culture (3). The results indicated an 80-fold accumulation of the agent within cells compared with the extracellular medium. Uptake was predominantly in the cytosol. In contrast, the purine PU3 was present at 4–5-fold higher concentration in the extracellular medium compared with the cells, as measured by mass spectrometry (3).

In a separate series of experiments, Chiosis et al. (3) noted that there was no difference in the activity of 17AAG or geldanamycin when the growth-inhibitory activity of these agents in cell culture was determined against primary prostate epithelial cells versus prostatic cancer cell lines. The authors propose that this indicates that the accumulation of geldanamycin into normal and malignant prostate cells may be similar, although cellular uptake was not compared in these particular experiments. Interestingly, Chiosis et al. point out that in the previous studies of Egorin et al. (28) accumulation of 17AAG was seen in PC3 human prostate cancer xenografts, to a greater extent than that seen in normal tissues.

The observations of Chiosis et al. (3) are helpful in that they explain why the geldanamycin class of compounds are more potent against cancer cells than their relatively low potency against Hsp90 would predict. Of practical significance, they recommend that full details of in vitro drug exposures be provided, including plate type, number of cells, drug concentration, total amount of drug and the volume of the extracellular medium. This is a good idea, as it will allow investigators in different laboratories to compare their results more effectively.

The mechanism responsible for the cellular accumulation of the benzoquinone ansamycins into cells is not clear and experiments were not carried out to address this issue. Chiosis et al. point out that a similar intracellular accumulation was seen by others with paclitaxel and the epothilones. This is interesting because, like the geldanamycins, paclitaxel and epothilones are also relatively high molecular weight natural products, and they also act on a molecular target that is ubiquitous and relatively abundant in the cell. Chiosis et al. speculate that the cellular accumulation of the ansamycins may relate to the physicochemical properties of the compounds. This notion is not explored in detail, but factors to be taken into account in this regard are molecular characteristics such as solubility, lipophilic/hydrophilic balance, and acidic/basic nature. Further work is required to figure out the mechanisms involved in the uptake of 17AAG and geldanamycin into the cell against a concentration gradient. Investigations with other analogues having greater structural and physicochemical diversity could shed light on the problem.

It is worth pointing out that, because Hsp90 represents 1–2% of intracellular protein, it is possible that binding to the target may contribute to overall uptake to some extent, although probably not sufficiently to account for the results seen. Accumulation via a transport protein or uptake pump is possible. Previous studies suggested that geldanamycin was likely to be a substrate for the P-glycoprotein multidrug resistance efflux transporter, and possibly also the related MRP efflux pump (15). Experiments could be carried out using cell lines transfected with the genes encoding these and other known transporter/pump proteins to identify potential mechanisms involved. Other more simple, now classical, experiments could also be informative, for example, looking at the effects of pH, temperature, metabolic inhibitors, and so on. These could be particularly useful in determining whether an active rather than a passive transport mechanism is involved.

One complicating factor that needs to be highlighted concerns the potential role of the NQO1 gene that encodes the quinone reductase commonly known as DT-diaphorase. In previous work, Kelland et al. (15) have shown that cancer cell lines that overexpress this gene, either naturally or by transfection, are very much more sensitive to the effects of 17AAG. High DT-diaphorase cells retained the Hsp90 mechanism, as measured by markers such as client protein depletion, but the effects were seen at correspondingly lower concentrations, consistent with greater cellular sensitivity. The mechanism of cellular sensitization by DT-diaphorase remains unclear, but the possibility cannot be ruled out that this contributes to cellular accumulation of 17AAG or an active metabolite. On the other hand, DT-diaphorase expression had no effect on sensitivity of cancer cells to geldanamycin. Overall, it seems likely that DT-diaphorase is not responsible for the cellular accumulation reported by Chiosis et al. (3). However, it still represents a significant factor to be taken into account when comparing the activity of 17AAG in different cancer cell lines and is frequently overlooked. On the other hand, the effect of DT-diaphorase is less likely to be important in vivo, because the major 17-amino metabolite of 17AAG, produced by cytochrome P-450, CYP3A4, in the liver (29) is, like geldanamycin, not sensitive to DT-diaphorase expression (15). Nevertheless, the potential involvement of metabolism in the behavior of the geldanamycins in cells and tissues deserves further investigation.

The study by Chiosis et al. (3) highlights a particular aspect of the PK/PD behavior of 17AAG, namely cellular accumulation against a concentration gradient. But given that 17AAG is now in the clinic, it is useful to take a wider view of the pharmacological audit trail for this drug.

Alongside the generic audit trail in Fig. 1, referred to earlier, the right-hand side of the figure identifies some of the factors that could be measured in relation to 17AAG specifically. Some of the salient features of the audit trail are discussed below.

First of all, it will be necessary to understand the role of the level of expression of the members of the Hsp90 gene family on sensitivity to 17AAG. Expression of the molecular target can be an important factor in sensitivity to other drugs. This is not really known for Hsp90 inhibitors. In addition, the role of the expression of the key client proteins needs to be addressed. This may be complex. For example, overexpression of ErbB2 has been shown to increase cellular sensitivity to geldanamycin but not to 17AAG in an ovarian cancer cell line (30). It should be noted that, whereas expression of a particular client protein per se may not necessarily predict for response to 17AAG, the status of a given signaling pathway potentially could do so. Candidate readouts that could be looked at include the phosphorylation of Erk1/2 and Akt/PKB as potential measures of the Ras and PI3-kinase pathways respectively. As mentioned above, expression of metabolizing enzymes, such as DT-diaphorase, and also the level of transporter/pump proteins (15) may also be important.

The drug dose and schedule are clearly important, and it is not yet clear how 17AAG should be given for optimal effect. Chronic administration may have advantages in terms of keeping the Hsp90 inhibited for prolonged periods to ensure sustained depletion of oncogenic client proteins. On the other hand, the ability to do this may be limited by the lack of oral bioavailability and relative insolubility of 17AAG, necessitating the use of a cumbersome i.v. formulation. Once-or-twice-a-week dosing may be the best that can be achieved. Furthermore, it is possible that a more intermittent schedule could be used in the setting of the combinations of 17AAG with cytotoxic agents that are currently being considered.

Clearly, whatever schedule is used will impact on the circulating concentrations of 17AAG and its metabolites in the plasma. The issue of protein bound versus free species also needs consideration as a potential determinant of tissue uptake. The PK of 17AAG in patients have been studied by high-performance liquid chromatography. As Chiosis et al. (3) note, concentrations in the micromolar range are readily achievable and prolonged exposures above 100–200 nm can be obtained in the plasma or serum. Thus, circulating levels above the average IC₅₀ for cancer cells in tissue culture can be achieved in patients. Similarly high levels of the 17-amino metabolite are also present in the circulation, and this metabolite is active as an Hsp90 inhibitor (15). As Chiosis et al. point out, the concentrations of 17AAG achieved in human tumor tissue are not known. Data obtained with human tumor xenografts suggest that accumulation above plasma levels may be seen to a greater extent than in normal tissues (28). Ideally, data would be obtained on tumor uptake in the clinic, but this will not be straightforward. Chiosis et al. argue, based on their data, that micromolar levels must be achieved in the tumor to produce therapeutic effects comparable with those seen in cell culture studies. The clinical PK results suggest that this may well occur, at least as predicted from plasma data and especially if the accumulation of the drug seen in xenograft models also occurs in tumors in human patients. Nevertheless, this represents a gap in the audit trail.

In the absence of such direct measurement of tumor drug levels, Chiosis et al. rightly recommend that the most certain way in which the uptake of 17AAG into tissues will be confirmed will in fact be to assess the PD effects of the drug. A range of preclinical studies have shown that successful inhibition of Hsp90 in cells and tissues can be diagnosed by the presence of a characteristic molecular signature (7). This comprises a decrease in the levels of client proteins that occur at the same time as an increase in the expression of another heat shock protein and molecular chaperone, namely Hsp70. The latter is effected at the transcriptional level by activation through trimerization of the transcription factor heat shock factor 1 or Hsf1, which is normally present in the cell as an inactive monomer bound to Hsp90 (31). Client proteins that are most commonly measured in preclinical and clinical studies include Raf-1, Cdk4, ErbB2, and Lck. The depletion of these Hsp90 clients and the concomitant up-regulation of Hsp70 is seen consistently in cancer cells in culture (12–15) and also in human tumor xenografts (15–18). The specific signature is observed with chemically distinct Hsp90 inhibitors, including geldanamycins, radicicol, and purines, but not with inactive analogues or with other classes of drug such as cisplatin or Taxol. The molecular signature of Hsp90 inhibition was also observed in human peripheral blood lymphocytes treated with 17AAG ex vivo (32). Before use in clinical trials, the biomarker signature was validated in a xenograft model, in which the pattern was seen in both the responding ovarian tumor and the peripheral blood lymphocytes taken from mice treated with 17AAG, with markers returning to normal on cessation of treatment and tumor regrowth (33).

It was, therefore, considered appropriate to measure the depletion of client proteins and elevation of Hsp70 as PD end points in the clinical trials with 17AAG. The characteristic molecular signature of Hsp90 inhibition was indeed observed in both peripheral blood lymphocytes and tumor biopsies from treated patients (22, 23). Thus it can be concluded that the Hsp90 molecular target can be inhibited at doses that are tolerated in patients. Depletion of client proteins was observed at 24 h, and ongoing studies are designed to determine how long the effect persists with the clinically administrable schedules. Overall, these studies should prove invaluable in determining the recommended dose and schedule for Phase 2 efficacy studies.

It is likely that additional PD markers will emerge from basic research and also from systematic approaches to determine the molecular effects of Hsp90 inhibition using the global genomic techniques of gene expression microarray analysis and proteomics, as already seen with 17AAG (32, 34). For example, the newly discovered co-chaperone Aha1, the first co-chaperone known to stimulate the ATPase activity of Hsp90, was found by gene expression microarray and proteomic analysis to be induced by 17AAG at both the mRNA and protein level (34). These powerful techniques could also be applied to the analysis of clinical materials.

Coming next in the pharmacological audit trail is the demonstration that biochemical pathways can be inhibited downstream of client protein effects. Preclinical studies have shown that Erk1/2 phosphorylation is inhibited downstream of Raf-1 depletion and that Akt/PKB phosphorylation is simultaneously decreased, thus demonstrating blockade of both the Raf-to-Erk1/2 and the PI3-kinase pathways. In addition, Rb phosphorylation is also inhibited, potentially as a result of blockade of the above pathways and also resulting from Cdk4 depletion (14, 35). In terms of biological effects, it is clear that 17AAG is able to induce cell cycle arrest, apoptosis, and differentiation in cancer cells (13, 14). It would be very useful indeed, as part of the audit trail, to show that such biochemical and biological outcomes could be achieved in patients. Antibody reagents and other assays are available for this.

One of the issues with measuring such drug effects in clinical studies is that the molecular analysis commonly requires tumor biopsies to be taken before and after treatment. Analysis may involve Western blotting, ELISA, or quantitative reverse transcription-PCR. The need for biopsies raises logistic and ethical issues. Access to ascites tumor cells is more straightforward, and methods are also being developed to measure end points using circulating tumor cells. Of increasing interest is the use of minimally invasive methodologies, such as magnetic resonance spectroscopy/imaging (MRS/MRI) and positron emission tomography (PET), and the development of functional and molecular imaging applications is to be strongly encouraged (1, 5, 36). Recent studies have shown that unusual changes in phosphocholine and phosphoethanolamine, possibly related to changes in membrane turnover or lipid signaling, are induced in human tumor xenografts by 17AAG (37). Other experiments support the use of labeled choline as a tracer to monitor the effects of 17AAG by PET (38). Such noninvasive techniques will greatly facilitate the audit trail for cancer drugs.

A point that is worth stressing in the audit trail is that PD end points are not necessarily the same thing as prognostic biomarkers. Surprisingly, these are often confused, for example, in discussion at scientific and medical meetings. PD end points allow us to determine whether the molecular target has been modulated by the drug. But successful inhibition of the target does not necessarily translate into a response. This will depend on how the cell responds to the target modulation. Definition of the genes that act downstream may provide prognostic markers for predicting response to 17AAG and other Hsp90 inhibitors. Some of these may be genes that regulate cell cycle progression and apoptosis, e.g., p53, Rb, and AKT/PKB (13, 17).

A generic comment to make about all end points, whether we are talking about PK, PD, prognostic biomarkers, or whatever, is that they should be as robust and quantitative as possible. Validation is essential.

A key objective of future studies will be to determine the therapeutic activity and also utility of 17AAG. These studies should be carried out in such a way that the links between efficacy and PK and PD, and also prognostic end points can be established and the pharmacological audit trail can be completed. The relationships between PK, PD, and toxicity should also be established (3). The end points will need to be considered not only in monotherapy trials with 17AAG but also in combination studies with cytotoxic agents, because this approach looks promising based on preclinical data with certain cytotoxics in particular schedules (39).

The studies by Chiosis et al. (3) have emphasized the importance of considering one particular element of pharmacological audit trail, namely the cellular uptake of drugs. They show that the cellular accumulation of 17AAG can increase the effectiveness in cell culture of an agent with relatively modest, micromolar potency on the Hsp90 molecular target. The extent to which this applies in vivo (e.g., in xenograft models and patients) remains to be fully determined. The mechanism underlying the cellular accumulation remains unclear. It is speculated that the physicochemical properties of the geldanamycins may play a role. This needs further analysis, in particular so that any broader generality of the effect can be determined.

The effect does not seem to apply to the chemically simpler, fully synthetic purine class of Hsp90 inhibitors (3, 40, 41). Other novel classes of Hsp90 inhibitor are likely to emerge based on rational design and high throughput screening (42–44). It will be important to study the PK and PD properties of these agents in detail to understand the extent to which they differ from the geldanamycins and also to develop them optimally in their own right.

Broadening the debate beyond Hsp90 inhibitors, it is very well known that preclinical and clinical drug development is a risky, expensive, and sophisticated business. The conduct of mechanistic, hypothesis-testing clinical trials is essential, and rationally based decision making is critical. The inclusion of appropriate PK/PD end points and biomarker measurements is absolutely essential. Although it makes tough demands on the clinic and the laboratory, the completion of the pharmacological audit trail is an essential component of modern drug development. Having your financial accounts audited can be a painful process but the overall gain is worth the pain.

I thank my colleagues and coworkers in the Centre and elsewhere for their collaboration and stimulating discussions.