The procedure can be extended to more than two dimensions; for example, if a plot of n against c suggested that three compounds (1, 2 and 3) were the only biologically active molecules in a larger population, the MSD of the three compounds can be calculated from Equation [5.45] and compared with values for the other combinations of three.
Hansch n values and Hammett c values have a similar range of numbers. When parameters have a significantly different order of magnitude, they must be given equal weight. This can be done by subtracting the mean of each column in the matrix from its individual variables, and dividing by the standard deviation of the column. Such procedures are too protracted for manual calculations, but can be handled by a computer. For example, McFarland and Gans (1986) examined 20 compounds, involving 77 520 clusters of seven compounds.
In recent years neural network (NN) programs have been increasingly used in problem solving, and a number of QSAR applications have been published. NNs model the way that the human brain processes information, and have the potential to be extremely powerful tools. A typical scheme for an NN simulation is shown in Figure 5.10. Input nodes receive information, and pass it to one or more intermediate (hidden) layers of
Figure 5.8 Principal component analysis of skin corrosivity of acids (from Barratt, M.D. (1995) ATLA 23, 111-122) (reproduced
Figure 5.8 Principal component analysis of skin corrosivity of acids (from Barratt, M.D. (1995) ATLA 23, 111-122) (reproduced with permission). The principal components were derived from log P values, molecular volumes, melting points and pKa values.
"neurons" which process the information. The processed data are then passed to the output nodes. In Figure 5.10 all nodes in one layer are shown connected to all nodes in adjacent layers, but this does not have to be the case, so that the NN can be set up in various ways. The network is trained using a training-set of data, so that it can then make predictions about additional data. The level of training is crucial, and it is possible to overtrain a network. Recent QSAR studies using NNs indicate that they are capable of giving better predictions than are Hansch-type regression equations.
Quantitative structure-activity relationships can be said to have brought drug design into the computer age. They have introduced a quantitative element into a subject which had hitherto been entirely qualitative. The medicinal chemists of old used their instinct and experience to predict conformational changes which should bring about increased biological activity, but had little idea of the magnitude of those changes. QSAR requires less instinct and experience, and also estimates the extent of the biological activity. Hansch and Free-Wilson analyses are lead-optimizing techniques. This means that it is necessary to begin with a 'lead' compound, having the basic pharmacological properties; analogues are prepared and tested, and the resultant data used to plan new compounds which will have enhanced biological activity. The technique demands that compounds be selected in a highly rational manner, with a systematic and stepwise progression throughout. Because of this stringent approach, much of the early structure-activity data in the literature are not adaptable to QSAR. These techniques are also self-limiting: as more results become available, the picture becomes clearer, but the number of possible new compounds become more restricted.
Powerful new mathematical techniques in which the computer is made to recognize molecular structures are being extended to drug design, and could move the concept of QSAR from lead-optimizing to lead-generating. For example, if through the use of 3D QSAR the receptor features necessary for activity can be identified, it should be possible
Figure 5.10 A typical neural network scheme.
to design novel compounds that will complement those receptor features, and which should thus possess the requisite activity.
Armstrong, N.A. and James, K.C. (1996) Pharmaceutical Experimental Design and Interpretation. London: Prentice Hall.
Basak, S.C. (1987) Use of molecular complexity indices in predictive pharmacology and toxicology: a QSAR approach. Med. Sci. Res. 15, 605-609.
Buisman, J.A.K. (1977) Biological Activity and Chemical Structure. Amsterdam: Elsevier.
Charton, M. (Ed.) (1996) Advances in Quantitative Structure-Property Relationships, Vol. 1. Greenwich, CT, USA: JAI Press.
Coulson, A.E. (1965) An Introduction to Matrices. London: Longman Group Ltd. Hansch, C. and Leo, A. (1979) Substituent Constants for Correlation Analysis in Chemistry and Biology. New York: Wiley.
Hansch, C. and Leo, A. (1995) Exploring QSAR. Vol. 1. Fundamentals and Applications in Chemistry and Biology. Washington DC: American Chemical Society.
Hansch, C., Leo, A. and Hoekman, D. (1995) Exploring QSAR. Vol. 2. Hydrophobic, Electronic and Steric Constants. Washington DC: American Chemical Society.
James, K.C. (1974) Linear free energy relationships and biological action. In Progress in Medicinal Chemistry, edited by G.P.Ellis and G.B.West, Vol. 10, pp. 205-243. Amsterdam: North Holland Publishing.
James, K.C. (1986) Solubility and Related Properties. New York: Marcel Dekker.
Kendall, M. (1980) Multivariate Analysis. London and High Wycombe: Charles Griffin & Co. Ltd.
Kier, L.B. and Hall, L.H. (a) (1976) Molecular Connectivity in Chemistry and Drug Research. London: Academic Press. (b) (1986) Molecular Connectivity in Structure-Activity Analysis. Letchworth: Research Studies Press.
Kubinyi, H. (1993) QSAR: Hansch Analysis and Related Approaches. Weinheim, Germany: VCH.
Kubinyi, H. (Ed.) (1993) 3-D QSAR in Drug Design—Theory, Methods and Applications. Leiden, Netherlands: ESCOM Science Publishers.
McFarland, J.W. and Gans, D.J. (1986) The significance of clusters in the graphical display of structure-activity relationships. Journal of Medicinal Chemistry 29, 505514.
McGowan, J.C. and Mellors, A. (1986) Molecular Volumes in Chemistry and Biology. Chichester: Ellis Horwood Ltd.
Tute, M.S. (1971) Principles and practice of Hansch analysis: a guide to structure-activity correlation for the medicinal chemist. In Advances in Drug Research, edited by N.J.Harper and A.B.Simmonds, Vol. 6, pp. 1-77. London: Academic Press.
Wells, P.R. (1968) Linear Free Energy Relationships. London: Academic Press.
FROM PROGRAMME SANCTION TO CLINICAL TRIALS: A PARTIAL VIEW OF THE QUEST FOR ARIMIDEX™, A POTENT, SELECTIVE INHIBITOR OF AROMATASE
6.1 INTRODUCTION 208
6.2 BACKGROUND 209
6.3 ICI START ON AROMATASE INHIBITION 213
6.4 AROMATASE RESUMED 217
6.4.1 The legacy from anti-oestrogens 217
6.4.2 The potential importance of uninterrupted drug cover 218
6.4.3 Increasing concerns about timeliness 219
6.4.4 Naphthol-lactones, tight binding and in vitro/in vivo relationships 221
6.4.5 The design principles behind ICI 207658—later named Arimidex™ 222
6.5 BIS- TRIAZOLE (6.17)—A TIMELY AROMATASE COMPOUND IN DEVELOPMENT 227
6.6 THE SEARCH FOR THE IDEAL BACK-UP CANDIDATE 229
6.7 INTO THE CLINIC 232 FURTHER READING 233
Breast cancer is the commonest cancer in women and, despite continuing advances in treatment, each year world-wide an increasing number die from the disease: in Japan the incidence has been increasing by an alarming 10% per annum. In the early stages of the disease, 30-40% of patients respond to hormonal or anti-hormonal therapy. One way to deprive hormone-dependent cancer of its primary mitogens, oestrogens, is to prevent their synthesis—preferably by inhibition of aromatase, the ultimate and biochemically unique enzyme that converts androgens such as testosterone to mitogenic oestradiol.
This account provides some of the background to the author's and ICI's involvement with hormonal modulation, but attempts mainly to cover the cytochrome P450-dependent enzyme, aromatase (oestrogen synthase, P450arom/NADPH cytochrome P450 reductase), its inhibition, and the way in which the ICI Aromatase Team selected a development compound, was forced by long-term toxicity to abandon it, but was more fortunate in its second choice with the compound ICI 207658, numbered D1033 during early development and later given the name anastrozole. During the development phase of D1033, ICI Pharmaceuticals became Zeneca Pharmaceuticals and the designation ZD1033 was used for the drug which is now called Arimidex*.
The medicinal chemistry coverage, in focusing mainly on the author's team contributions to the programme, is a partial account of work that involved several chemistry teams.
An undergraduate course in organic chemistry in the mid 1950s typically made good use of the inspiring work of many groups in the fields of steroid structure determination, conformational analysis, reactivity and synthesis. The potent and multifarious biological properties of such molecules, along with the synthetic challenges presented by, for those times, extremely complex structures, made them synthetic targets for many eminent chemists of the day. The first formal total synthesis of cortisone (6.1), then thought to be a miracle drug, had been briefly reported by Fieser and Woodward in August 1951.
However, the race to achieve the first non-trivial synthesis of cortisone had unexpectedly been won by a group of young chemists in Mexico City who were employed by a small, recently-formed company called Syntex Inc. That synthesis, starting with readily available diosgenin—from Mexican yams—had commercial potential from the sale of intermediates as well as the final product and its dihydroderivative, hydrocortisone or cortisol. Cortisone was in great demand for the treatment of severe inflammatory and immunologically-related conditions, as well as for treating Addison's disease—a previously life-threatening condition caused by a deficiency of cortisol synthesis in the adrenal gland of an afflicted individual.
Clearly, our target aromatase inhibitor would have to avoid in vivo inhibition of the cytochrome P450-dependent enzymes involved in cortisol/cortisone synthesis. Indeed it ideally had to avoid inhibiting any cytochrome P450-dependent enzyme except aromatase. Schenkman and Greim (1993) recently edited a wide-ranging multi-author review of such enzymes.
Carl Djerassi was a leading member of that early Syntex group and he was soon hailed by many in academia as one of the promising synthetic chemists of the day. Such limited accolades were eventually dwarfed when he achieved broadly proclaimed 'immortality' as the 'Inventor of The Pill'. As so often even in those days, the title is largely the result of the mass media requirement for (over)simplification. As Djerassi (1992) emphasises
* Arimidex is a trademark, the property of Zeneca Limited.
in his autobiography, his part in the invention of 'The Pill' was that of main contributor to the search for and discovery of the orally potent progestagen, norethindrone (norethisterone) (6.2); this was achieved during his brief time at Syntex and overlapped the cortisone work. Others had foreseen the potential of such agents and many others were involved in the development and exploitation of this and related compounds—indeed it took more than a decade for such hormonal modulation to gain even limited acceptance as a method of contraception.
It is of interest in the present context that norethindrone and other terminal acetylenes such as ethynyloestradiol may owe some of their improved oral activity to irreversible, mechanism-based, covalent inhibition of drug-metabolising cytochrome P450-dependent enzymes in the liver. This possibility and the nature of such enzymes were unknown at the time of those drugs' discovery, but with the advent of that understanding it is now possible to design inhibitors of P450s based on terminal acetylenes. Potent inhibitors of aromatase have been generated by modifying natural substrates, and close analogues, through the addition of an ethynyl group to C19—the initial site of substrate oxidation by aromatase. Amazingly, it was only in 1982 that norethindrone was shown in vitro to be a rather weak (~2 pM), irreversible inhibitor of aromatase. It is unlikely, however, that this has relevance to its contraceptive use.
By the late 1950s many research groups were involved in hormonal modulation. One of those groups was in ICI Pharmaceuticals and its efforts were rewarded with the discovery and successful development of the oestrogen antagonist tamoxifen, or Nolvadex™ (6.4) as ICI named it (Nolvadex is a trademark, the property of Zeneca Limited). A number of structurally related oestrogen agonists had been discovered in the 1930s, one of which, diethylstilboestrol (6.3), remains, somewhat controversially, in use to this day.
Nolvadex™ (6.4) was found to be an effective treatment for a substantial proportion of post-menopausal patients with oestrogen-dependent breast cancer. It has in recent years been the largest selling chemically-defined anticancer drug of all time.
The author's first year in research, 1957-8, coincided with three events important to this discourse. First, M Klingenberg and later D Garfinkel independently reported the generation of a new absorption peak at 450 nm when homogenised liver supernatant was exposed to carbon monoxide: pigment 450 (P450) was born, but the function, if any, of this pigment was unknown. Second, K J Ryan reported that androgens incubated with human placental microsomes were converted to oestrogens. This amazing process involves the removal, by then unknown chemical steps, of a very hindered, non-activated methyl group from C10 and a non-activated hydrogen atom from C1 of the androgenic precursors testosterone or androstenedione. What agent or agents were at work was again unknown. Third, the Swiss pharmaceutical giant, Ciba, a major force in steroid chemistry, pharmacology and drugs, started clinical trials with aminoglutethimide, (AG) (6.5), as a prospective anticonvulsant drug. Those trials and subsequent use under the name Elipten™ (later Orimeten™: Ciba, and now Cytadren™: Ciba-Geigy) revealed multiple adverse side effects, one of the most serious being adrenal insufficiency. A few years after launch it was withdrawn from sale, but as so often in chemotherapy, one person sees a side effect while another sees an opportunity: a medical adrenalectomy might be useful in various adrenal hormone-dependent diseases—including breast cancer where surgical adrenalectomy was an established hormonal manoeuvre.
Some years later, that possibility became actuality—AG was shown to be useful in several conditions, including advanced breast carcinoma. It originally was assumed that efficacy flowed from suppression of adrenal pregnenolone synthesis. Rather low-potency (~26 pM) inhibition in vitro of an adrenal-derived enzyme, P450scc, that converts cholesterol via side chain cleavage to pregnenolone, had long since been demonstrated and adrenal hypertrophy in various species dosed with AG is ascribed to that gland's attempt to maintain steroidogenic homeostasis. Inhibition of P450scc would in turn limit synthesis of the many other steroids, including oestrogens, which have pregnenolone as a precursor. Scheme 6.1 shows a selection of steroidogenic pathways—unidirectional arrows indicate that one or more of the steps in that pathway involves a cytochrome P450 enzyme. These pathways operate in differing degrees
i according to tissue, species, sex, age and in pregnancy and disease and their products elicit a wide variety of responses depending on the target cell type and its environment (Castagnetta et al, 1990).
Replacement glucocorticoid was administered with AG during breast cancer therapy in part because of the above findings. Later quantitative studies however showed reduced oestrogen levels but normal or even increased levels of androstenedione in AG-treated patients' plasma: androstenedione is the main androgenic precursor of oestrone. These
new findings appeared inconsistent with substantial P450scc involvement. Subsequent in vitro studies—twenty-five years after Ryan's discovery—suggested that efficacy in post-menopausal breast cancer patients resulted mainly from inhibition of the enzyme (or enzymes) that converts androgens to oestrogens. By then the placental enzyme activity was widely known, but, because the enzyme is embedded in microsomal phospholipid membranes, and is functionally dependent on that association, it was not yet well characterised. Crystallisation is likely to be extremely difficult or impossible.
We now know that P450arom is a single enzyme—the product of the CYP19 gene on chromsome 15 in man. Gene expression is subject to complex and multifactorial regulation. The enzyme is widespread in humans, males and females, in the brain and the periphery, but it is much less widespread in most other species. It belongs to the cytochrome P450-dependent class of enzymes and is commonly called aromatase or P450arom, but the latter designation refers specifically to the haem-binding protein of the two component enzyme: the second component, a flavoprotein, is reduced nicotinamide adenine dinucleotide diphosphate(NADPH)-cytochrome P450 reductase; the reductase is common to all P450-dependent enzymes. Use in vitro of oxidants other than air, for example Ph-I=O, or H2O2, allows most P450s to function in the absence of the reductase. Interestingly, and relevant to the mechanism (Scheme 6.2), iodosobenzene is ineffective in the final step of aromatisation while hydrogen peroxide allows all three steps to proceed.
P450arom, or rather the oestrogens it produces, has differing roles according to the sex of the animal and cell type in which it is expressed. Importantly it is expressed and is functional in producing trophic effects in many breast cancer cell lines—a blood-born oestrogen supply is not always necessary for growth.
Testololactone(Teslac™, Squibb) (6.6), used in breast cancer and originally thought to act via androgenicity, may also owe much of its efficacy to aromatase inhibition.
In the last years of the 1970s the Fertility Project Team in ICI was again testing compounds for antifertility potential. Some of that effort was devoted to random screening with the end point being prevention of pregnancy in rats. One of the more potent compounds discovered, the N-'benzyl' imidazole (6.7), was considered to be worthy of further investigation since its structure and overall biological effects did not point to any known mode of action. There was some concern that its fragmentation to a quinone methide might be involved—if that was so, the generation of such a reactive species would make it and any analogues unattractive. The postulated quinone methide is implicated in the lung toxicity of butylated hydroxytoluene (BHT; 3,5-ditertiarybutyl-4-hydroxy-toluene), a widely used antioxidant.
Also at that time, the Team's interest in aromatase had been heightened by the results of clinical and biochemical studies in patients receiving AG. Preclinical results being obtained by Angela Brodie and co-workers with 4-acetoxy-androst-4-ene-3,17-dione were also encouraging. This steroid was active in vivo, especially in the oestrogen-dependent DMBA (dimethylbenzanthracene) rat tumour model, but interest focused later on the 4-hydroxy compound, 4-OHA (6.8), (formestane), a potent, Ki=10 nM, and time-dependent aromatase inhibitor since licensed and named Lentaron™ by Ciba-Geigy. Structure (6.8) is shown with partial van der Waals' radii for some 'atoms' (actually CH2 and CH3 groups); those 'atoms' in the enzyme bound state are postulated to be in contact with the large, extensively-planar protoporphyrin-IX prosthetic group which is depicted, edge-on, as a thick line. Partial van der Waals' radii for atoms in the porphyrin are not shown but extend to contact those shown for the steroid.
The official start of the Aromatase programme—just a few weeks into the new decade, was contemporaneous with the beginning of another team's attempt to find an anti-oestrogen working through inhibition of translocation of the oestrogen receptor from cytosol to nucleus, but more of that later. Chemistry started in two main directions, steroid-based (naturally!) and azole-based: the N-'benzyl' imidazole (6.7) had by now been shown likely to be an aromatase inhibitor. Our compound collection, together with some standard antifungal agents from ICI Plant Protection, generated a structurally diverse set of leads with some remarkably simple azoles, e.g. N-(m-pentanoylbenzyl) imidazole (IC50=2 ng/ml), being very potent inhibitors of human placental microsomal aromatisation—e.g. of testosterone to oestradiol or of androstenedione to oestrone.
The literature evidence at the time was consistent with all the aromatase chemistry being carried out by a single enzyme, but this became certain only much later, during the second part of the programme. The multiple steps involved in this conversion were already broadly established from a vast body of work by many academic groups (Brodie et al, 1993), see Scheme 6.2, but some of the finer mechanistic detail remains controversial (Aktar, Njar and Wright, 1993). The lower part of the Scheme and footnote commentary represent the author's view of the how the final steps might proceed.
At this stage we knew from ICI work on antifungal agents which potently inhibit fungal lanosterol-14-methyl demethylase, a P450-mediated reaction very closely related to that performed by aromatase, as well as from literature reports of azoles inhibiting various P450 enzymes, and already rather extensive studies with the N-'benzyl' imidazole (6.7), that superior selectivity versus AG could be the key to a successful drug. Considering how 'dirty' AG is by modern standards, this seemed at first an easy target. We soon thought otherwise: the in vivo effective azoles then to hand were all clearly deficient in one or more respects. Surprisingly to us because we were not aware of any connection with P450 enzymes, all the in vivo more potent (but
still weakly potent) azoles, including (6.7), caused unacceptable elevation of liver triglycerides at modest multiples of their aromatase-effective doses.
Another frequently observed effect in vivo—adrenal enlargement—indicated unwanted inhibition of non-oestrogenic steroid production. No pattern of selectivity could be discerned. Yet another indicator of potentially inadequate selectivity was the increased sleeping times observed after co-administration to mice of hexobarbital with each of the few azoles tested: such effects are probably due to inhibition of liver P450-mediated oxidative clearance of this xenobiotic sedative. Multiple high doses of all azoles examined caused increases in liver weight to body weight ratios and elevation of some liver mixed function oxidase (P450) enzymes. These elevated P450 levels can cause increased clearance rates and modify metabolite patterns of hormones, drugs and other natural products and xenobiotics. Both these effects and enzyme-inhibitory effects are present in AG-treated patients—but we decided that none of these effects would be acceptable at the therapeutic dose of our target molecule. Increased P450-mediated production of toxic and particularly mutagenic metabolites is one of the consequences of smoking and is implicated in the increased incidence of cancer in smokers. Smoking differs from most drug therapy in causing different P450s to become elevated, but obviously everyone would wish to minimise such risks—even in long-term drug treatment of cancer patients.
So how might one achieve selectivity? There are several possibilities: set up screens and throw everything you have at them; or, ideally with the help of precision models—Dreiding etc. and computer modelling, try to use substrate structures and inhibitor structures as a guide to drug design; or, again using modelling, try to understand the reaction(s) using as much detailed information as exists, make educated guesses about
the three-dimensional interactions needed for recognition and mechanism, then design the drug around as many hopefully unique features as possible.
Mainly in part two of the programme, we did some of each of these and developed other ideas that will be discussed later. Modelling at various levels should be, and was, an on-going process. The amino acid sequence of human aromatase was not known at the time of our work but became so soon thereafter: its sequence of 503 amino acids shows only about 30% homology with other known mammalian P450s; the latter group are highly homologous. This puts P450arom in a unique category. Despite the implications from the foregoing, several groups have published models of the enzyme based on lipid-free, water-soluble, bacterial enzymes, e.g. P450cam, that have been crystallised and the structures determined at high resolution by X-ray diffraction methods. In the author's view none of these models is satisfactory and any model is at present highly speculative. Speculative hydrogen bonds indicated in Scheme 6.3 are those used during our work. The peroxy intermediate (6.9), bound to a partial enzyme model, is shown as a stereoscopic pair in Figure 6.1. This binding mode was the basis of essentially all our modelling—despite the (still) speculative nature of such a species. It is shown essentially as we used it except that the amino acid side-chains on the a-helical protein fragment have been updated: we used those in P450cam. Some inhibitors throughout the chapter are drawn as they would appear in such a model— but with the partial a-helix removed and viewed edge on to the porphyrin multi-ring system: these changes allow easier comparisons of our suggested binding modes.
Steroidal inhibitors might seem intrinsically to hold better prospects for selectivity, but, as shown in part in Scheme 6.2, Nature's wide use of P450 enzymes in chopping, trimming and oxidatively modifying this skeleton argues against overconfidence in this intuitive position. Furthermore, steroids typically have other problems such as rapid clearance and poor efficacy by the oral route, especially in the rat—our preferred test species. Effects through receptor interactions are another concern. At a practical level, synthesis of new compounds can be demanding and slow and several other groups—industrial and academic—were known to have a substantial start on us. Counterbalancing this, we had a steroid expert in the team and we thought it worth a try.
Probably none of these considerations counted for much in the light of the excitement generated by the 'translocation' work yielding some extremely interesting compounds. This new lead, irrespective of mechanism, was anti-oestrogenic in every test that was applied. The observations of apparent translocation of receptors, in response to oestrogens, turned out to be artifactual—they are always predominantly in the nucleus—but this idea nonetheless led to the discovery of the first 'pure' anti-oestrogens (Wakeling 1990).
There is recent evidence to show that these agents are not equivalent to the total absence of oestrogen—the potential outcome of aromatase inhibition. Control of gene transcription is a complex multifactorial process in which the occupied but apparently oestrogenically-inactive receptor still has a role. Ongoing clinical studies may start to tease out some of the therapeutic implications of this complex and still little explored biochemistry.
Unsurprisingly, chemistry effort was switched from aromatase. Soon still more effort was required: it was then that the author came to work for the first time on hormonal modulation.
6.4 AROMATASE RESUMED
A 'pure' anti-oestrogen development candidate was chosen in 1985. ICI 182780 (6.11), is a 7a-(long side chain) substituted oestradiol derivative, but many non-steroidal frameworks were investigated during the programme and most, with appropriate side-chains, yielded potent, 'pure' anti-oestrogens.
Generally these frameworks, linked to azole rings via short side chains, yielded moderate-to high-potency aromatase inhibitors in the ensuing second phase of that programme. Because of its limited conformational freedom and very high inhibitory potency against human placental aromatase, one such azole, the (racaemic) triazole derivative (6.12), is particularly relevant to computer modelling of the enzyme active site.
Several highly potent aromatase inhibitors arose inadvertently during the final stages of the anti-oestrogen work. We were attempting to find a replacement for the potency-enhancing but metabolically-sensitive phenolic hydroxyl group in our pure anti-oestrogens: phenolic compounds bind to the receptor in vitro about 100-fold tighter than non-phenolic analogues. Many alternatives to the phenolic OH group had been tried, but none came even close to matching its 'magical' effect. One possible explanation for these dramatic findings is that strong interactions occur between the receptor protein and both the inplane acidic hydrogen and the in-plane oxygen lone-pair of the aromatic OH group.
With this hypothesis, no benzenoid derivative was likely to match the phenol, but that conclusion need not apply to planar heterocyclic systems: the 4-substituted pyrazole
(6.13) was designed to interact with just such a hypothetical phenol-binding site, as shown, minus double bonds for clarity, in (6.14).
Because there is no positional correspondence of atoms in the two differently-interacting rings, the design of the pyrazole 4-substituent could not be based on the normal steroid structure. Instead it was designed such that, overall, the molecule possesses a similar outline shape to the steroid skeleton and the hydroxyl group could be positioned roughly to correspond to that in testosterone. The design was a miserable failure—inhibition of radiolabelled oestrogen binding to the receptor was undetectable; a substantial volume deficit in the steroid ring B and C regions may contribute to this result. Inhibition of aromatase in contrast was among the best we had then seen: AR1: IC50=2 ng/ml (see below)! Unfortunately, activity in vivo was not detected at the highest dose examined, 20 mg/kg, and none of several pyrazoles was better, even after intra-peritoneal dosing.
These results illustrate a common problem in chemotherapy—good activity in vitro all too often fails to manifest itself in vivo. Sometimes this can be rationalised, in part, in terms of competitive phase effects: the highly lipophilic N-'benzyl' imidazole (6.7) binds to albumin and some other macromolecules and is extracted into fat deposits, phospholipid bilayers and other fatty body components. This drastically reduces the free aqueous concentration of the drug in vivo relative to that in vitro, and binding to the enzyme suffers in proportion. More usually poor bioavailability and rapid clearance are the most relevant parameters. The latter effect is undoubtedly relevant to potency in our OI3 test, but much more so in the OI2 test that we relied on increasingly throughout this second phase. These tests involved oral dosing of compound at 12.00 noon on day 3 (OI3) or 4.00 pm on day 2 (OI2) of the light-synchronised ovarian cycles of female rats; it had been previously determined that suppression of ovarian oestrogen production from mid-afternoon until midnight of day 3 of the 4 day cycles, prevents priming of the hypothalamus for the ovulation-triggering surge of luteinising hormone (LH) on day 4. Ovulation inhibition is the observed end-point.
6.4.2 The potential importance of uninterrupted drug cover
On theoretical and practical grounds (occasional non-compliance) there are good reasons for wanting a cytostatic anticancer drug to have a long half-life (t1/2): even the transient presence each day of growth-promoting levels of oestrogen may reduce response rates, quality or duration of effect. Tumour cells can express aromatase and synthesise oestrogens locally so plasma drug and oestrogen levels might give an incomplete picture of intra-tumour drug effectiveness. On the other hand, too long a half-life may result in serious consequences in the event of a severe adverse reaction. This analysis led us, in this case, to aim for an average t1/2 of our target drug, in patients, (assuming the simplest possible kinetics) of at least 12-16 hours and preferably not greater than two days. Because of competitive pressures, this criterion was made the dominant factor at one stage of the programme, despite the fact that predicting t1/2 values in man from data in other species was known to be little better than guesswork.
By the autumn of 1985, as we restarted the programme, the competitive situation in aromatase was intense. Many analogues of AG had been revealed and even p-cyclohexylaniline had been shown to possess good in vitro potency—equal to AG with respect to human placental aromatase but substantially less so versus rat ovarian enzyme. We had during the initial phase of the programme tested a few compounds in parallel against rat and human enzymes: AG was seven-fold more potent against rat enzyme, while 1-nonylimidazole was three-fold selective in the reverse sense. Almost no other comparative tests were performed—our limited resources were needed elsewhere—so our interpretation of in vitro (human) to in vivo (rat) potency ratios was always potentially flawed by species differences in enzyme binding; we had to hope, we still hope, that we were not seriously misled, but the reader needs to bear this in mind during apposite parts of structure/activity relationship (SAR) discussions.
Some of these new AG analogues showed much improved selectivity for aromatase—sometimes through improvements against the target enzyme, but often through reduced potency against other enzyme(s), typically P450scc. Potency in rats however, where reported, remained disappointing.
Another recently reported analogue of AG had the 4-aminophenyl group changed to 4-pyridyl and it was reported to be more selective opposite P450scc. We decided to make a sample for in-house investigation. While we were doing that, and making a number of analogues, we roughly derived the necessary parameters for imides (at that time they were not available from published lists of Allinger's MM2 force-field parameters) so that we could perform molecular mechanics calculations on such systems: we were thus able to predict that both this new analogue and AG exist very largely with the aromatic rings axially disposed. This interesting prediction led us to perform a slightly modified synthesis aimed at the analogue (6.15), which calculations predicted would exist overwhelmingly in the axial pyridyl conformation, and (6.16) which should have a moderate preference for the equatorial pyridyl conformation. The latter was only 2- to 6-times less potent in AR1 than either the parent or (6.15). No compound in this pyridyl series had sufficient potency in vivo to warrant further interest from us.
By far the largest area of competitive activity concerned steroidal inhibitors, particularly of the time-dependent variety, but similarly disappointing in vivo results generally applied here. Even the Brodie compound, 4-OHA (6.8), had been shown to have unwanted androgenic effects and an intra-muscular depot formulation used for human dosage was not always well tolerated.
Also by this time, we had noted an association between azole-based antifungal activity and aromatase inhibition. And since many drug companies were or had been active in the antifungal area, we needed to give rapid attention to this potential source of leads. Fortunately for us, another team within ICI Pharmaceuticals had recently completed an
antifungal programme based on inhibition of the multi-enzyme mediated biosynthesis of ergosterol—an essential constituent of fungal cell walls which is not synthesised in mammals. The specific target of that programme had been fungal lanosterol-14-methyl demethylase. As part of a frequently-applied ten-day teratology assessment, they had seen placental enlargement and effects on foetal development in pregnant rats—all potentially consistent with aromatase inhibition and a common property of the imidazole/ triazole compounds they had explored. This work had heightened awareness and understanding of selectivity issues in the business, and placental enlargement in rats provided the Aromatase Team with a test (PE9) wherein chronic effects of oestrogen depletion (compounds dosed once daily for 9 days) could be compared with similarly chronic effects on other systems in the same test animal. This overcomes problems of differential handling of compound between individuals, sexes or species, all of which in retrospect can be seen to have misled us at some point of the programme. As with all chronic tests, the accumulation of long half-life compounds (in this context, t1/2>ca. 1 day) can present problems, but may also allow such compounds to be identified at an early stage.
We felt sure that time was not on our side, so we were well pleased when screening of our antifungal agents soon yielded a compound which was potent (IC50=7 ng/ml) in our aromatase screen (AR1: human placental microsomes; substrate, 40 nM [1,2-3H]-androstenedione) and was inconsistently active in OI2/OI3 at 0.25-0.5 mg/kg. Poor aqueous solubility may have led to the inconsistency, but removal from that structure of an ortho-chloro substituent led to the bis-triazole (6.17), which was less active in AR1 (IC50=40 ng/ml), but consistently active in OI2, OI3 and PE9 at 0.2 mg/kg (approx. =ED50). The compound is therefore 20- to 50-times as potent as AG. It was urgently subjected to as detailed an investigation as the perceived time pressure allowed.
That time pressure increased substantially during 1986 as the competitive situation grew still more intense. Schering were claiming long-lived oral effects for atamestane (1-methyl-androsta-1,4-diene-3,17-dione) dosed orally at 1 mg/kg to male volunteers, while Ciba-Geigy disclosed that their racemic bicyclic imidazole derivative, CGS 16949A (6.18), is 1000-times as potent as AG, with ED50 in female rats of 30 pg/kg, and inhibition of aromatase was evident in human male volunteers even at 0.3 mg per man!
At that time, only in its duration of effect in volunteers, 4-10 hours, did the Ciba-Geigy compound seem to present room for improvement. This placed still greater emphasis on the half-life requirement of our target drug.
6.4.4 Naphthol-lactones, tight binding and in vitro/in vivo relationships
While much of the Team's early effort went on antifungal leads, we were also finding widespread activity with azoles attached to stilbenes related to (6.3), cis- and trans-2-aryltetrahydronapthalenes related to (6.12), and to more speculative frameworks based on computer modelling—such as the naphthol-lactone (6.20) whose synthesis and structure are shown in Scheme 6.3. In vivo, lactones and simple phenol esters such as pivalate esters (archetypal prodrugs) are almost always too rapidly hydrolysed, via enzyme-mediated catalysis, to allow the longevity we demanded. But this lactone is a special case: it is impossible for it to be hydrolysed at pH 7.4—the typical value for blood. It even stays ring closed in very dilute sodium hydroxide—due to thermodynamics, not kinetics! The reason lies in the large increase in steric compression strain that accompanies ring opening. In contrast, the negligible problems generated in the ring-opened form of the synthetic intermediate (6.19), Scheme 6.3, leave this compound highly sensitive to hydrolysis—t1/2 in water at room temperature is ~120 seconds at pH 9: this is 5000-times more reactive than ethyl acetate. The more hindered lactone (6.20) is however rapidly reduced by borohydride, but only to a lactol (hemi-acetal); further reduction under the weakly basic conditions would require ring opening, which, like the ester hydrolysis, essentially does not occur.
A lactol ethyl-ether, inadvertently produced in a reaction that had the cyclic ether as its target, was relatively poor in vitro but in vivo it had similar activity to the corresponding lactone. Rather efficient liver cytochrome P450-mediated reoxidation to the lactone seems a likely explanation.
The imidazole (6.20), X=Y=CH, is extremely potent in vitro. In a single AR1 test it inhibited the aromatisation of tritium-labelled androstenedione by 74% at 1.25 ng/ml—the lowest concentration tested, while at higher concentrations the figures were: 2.5 ng/ml—95.5% and 5.0 ng/ml—99%. Such figures, if they can be relied on, are indicative of 'tight binding'—the condition in which, for the simplest case, free drug concentration is significantly depleted from the nominal value by binding to a site, usually the active site, which is present in the test medium at a concentration only somewhat less than or equal to twice the observed 50% inhibitory concentration. Ultimately, half an equivalent of inhibitor—essentially all bound to the target—is the absolute minimum required for 50% inhibition no matter how potent the agent might be (rare, catalytically-active, irreversible inhibitors excepted). In most test situations one cannot rely on 95% inhibition being different from 100% inhibition, but here we are measuring release of tritiated water,
which is easily and completely separable from the precursor, so very small amounts of reaction can quantitatively be measured. Other extremely potent inhibitors show similar responses, while weakly and moderately potent azole-based inhibitors all display classical inhibition curves consistent with the simplest outcome of 1:1 competition between substrate and inhibitor.
The sigmoidal appearance of linear-%inhibition versus log-concentration curves tends to obscure modest deviations from 'normality', but the theoretical curve for the simplest case can be transformed to a linear function, or, more conveniently, the %inhibition axis can be transformed so that experimental points for the simplest case should be linear and lie on a line which passes through 9.09, 50 and 90.9% inhibition at 0.1-, 1- and 10-times the IC50. Graph paper to this design was generated 'in house' some years ago. Data point sets for two tight-binding inhibitors and another set for a borderline case are shown in this format in Figure 6.2, along with three theoretical 'curves' (curved/inclined lines). The thinner central curve should fit observations when an enzyme present at 5 nM, acting on a negligible concentration of substrate, is inhibited by a compound with an equilibrium inhibition constant, Ki, equal to 5 nM: the IC50 of 7.5 nM is only a 1.5-fold underestimate of its true dissociation constant. If a second compound binds 1000-fold tighter, i.e. Ki= 5 pM, the thick curve on the left shows that, under the above conditions, the observed IC50 would be 2.5 nM—the limiting condition corresponding to half-an-equivalent referred to above. Tight binding thus limits the apparent potency advantage of the second compound, over the first, to 3-fold rather than the 1000-fold which would be observed with 'infinitely'-dilute enzyme solutions. The thick 'curve' on the right for a compound with Ki=100 nM differs only minutely from linearity. Experimental data points shown for three compounds, and other observations, pointed to the presence of roughly 3-5 nM binding sites (not necessarily active enzyme) in our typical AR1 test milieu, so compounds with IC50 values less than ~20 nM could, for the best comparisons, be corrected to non-tight binding values, preferably nowadays by computed data-fitting techniques. Assuming a Km for androstenedione of 40 nM, one can estimate pKi (-log Ki) values for the
1 10 10* COMPOUND CONCENTRATION (nM)
O : (6.20; X=Y=H) X : (6.15; S-isomer) + • (6,27)
Was this article helpful?
Learning About 10 Ways Fight Off Cancer Can Have Amazing Benefits For Your Life The Best Tips On How To Keep This Killer At Bay Discovering that you or a loved one has cancer can be utterly terrifying. All the same, once you comprehend the causes of cancer and learn how to reverse those causes, you or your loved one may have more than a fighting chance of beating out cancer.