《天然药物化学》课程参考文献（天然药物研究与开发）The impact of natural products upon modern drug discovery.pdf

Available online at www.sciencedirect.com The impact of natural products upon modern drug discovery A Ganesan In the period 1970–2006, a total of 24 unique natural products were discovered that led to an approved drug. We analyze these successful leads in terms of drug-like properties, and show that they can be divided into two equal subsets. The first falls in the ‘Lipinski universe’ and complies with the Rule of Five. The second is a ‘parallel universe’ that violates the rules. Nevertheless, the latter compounds remain largely compliant in terms of log P and H-bond donors, highlighting the importance of these two metrics in predicting bioavailability. Natural products are often cited as an exception to Lipinski’s rules. We believe this is because nature has learned to maintain low hydrophobicity and intermolecular H-bond donating potential when it needs to make biologically active compounds with high molecular weight and large numbers of rotatable bonds. In addition, natural products are more likely than purely synthetic compounds to resemble biosynthetic intermediates or endogenous metabolites, and hence take advantage of active transport mechanisms. Interestingly, the natural product leads in the Lipinski and parallel universe had an identical success rate (50%) in delivering an oral drug. Address School of Chemistry, University of Southampton, Southampton SO17 1BJ, United Kingdom Corresponding author: Ganesan, A (ganesan@soton.ac.uk) Current Opinion in Chemical Biology 2008, 12:306–317 This review comes from a themed issue on Molecular diversity Edited by Gerry Maggiora and Christopher Hulme 1367-5931/$ – see front matter Published by Elsevier Ltd. DOI 10.1016/j.cbpa.2008.03.016 Introduction Nature has evolved over time to produce a bewildering diversity of secondary metabolites. Based on empirical observations and folklore, natural product extracts were the first, and for a long time, the only medicines available to mankind. Although crude extracts remain the primary healthcare for a majority of the world’s population, they are largely supplanted by active pharmaceutical ingredients in the Western world. Furthermore, the dependence upon natural products is no longer obligatory and many drugs are purely synthetic small molecules or manufactured biologics such as vaccines, antibodies, and recombinant proteins. Given these alternatives, there needs to be a rationale for the continued exploration of natural products as leads, and two major arguments can be put forward: Premise 1: Natural products interrogate a different area of chemical space than synthetic compounds. If this were untrue, it would be more profitable to concentrate on more readily accessible synthetic compounds. However, there are significant differences in the molecular architecture produced by nature when compared to the synthetic molecules of medicinal chemistry [1,2,3 ,4,5]. Although both aim to produce biologically active matter, biosynthesis operates under a different set of constraints and guiding principles than the synthetic organic chemist (Table 1). In nature, a very parsimonious set of building blocks is utilized, whereas we have access to tens of thousands of commercially available chemicals. As a consequence, we achieve numbers by repeating a reliable sequence of reactions over and over again while changing the input. Nature, on the contrary, diversifies by taking its limited building blocks and partitioning them into a multitude of pathways. Further differences occur in the type of synthetic transformation performed. Nature is oxophilic, and has developed enzymes that exquisitely accomplish site-selective C–H activation [6,7 ] to introduce oxygen and discriminate between numerous functional groups at different oxidation levels. Meanwhile, medicinal chemistry concentrates on nitrogen and often includes ancillary atoms such as sulfur and halogens that are relatively rare in nature. Finally, the chiral enzymes of biosynthesis usually yield the product as a single stereoisomer. Although medicinal chemists are themselves chiral and target chiral enzymes or receptors, they prefer to work in ‘flatland’ with molecules low in stereochemical features. Premise 2: Natural products are amenable to further improvement. If this were untrue, the natural product extracts would suffice, or the purified natural product would become the final drug without modification. Although this can often be the case, it runs counter to the drug discovery paradigm where initial leads are subjected to extensive medicinal chemistry campaigns before a candidate is selected. A priori, natural products should undergo the same iterative cycle of improvement, as their evolutionary reason for existence is not for use as a therapeutic agent. Thus, one can expect that the natural product can be further improved, whether in terms of efficacy and selectivity for the target or achieving optimal pharmacokinetic and pharmacodynamic properties. For Current Opinion in Chemical Biology 2008, 12:306–317 www.sciencedirect.com

example, the opium alkaloid morphine is an important drug that is obtained solely from nature and continues to be used in both extract and pure form. At the same time, morphine has encouraged the discovery of many semisynthetic and fully synthetic compounds based on the same pharmacophore that are successful secondgeneration opioid drugs. The molecular architecture of drug-like matter Biological space is modest in size — the human genome is on the order of 3 104 genes of which only a fraction is targeted by current therapeutics [8,9 ]. Meanwhile, chemical space is infinite, and there are an estimated [10] 1060 organic compounds with a molecular weight cutoff of 500. Our imperfect understanding of which areas of chemical space are best suited to interact with biological space is the major bottleneck of drug discovery. In recent years, there were various attempts at narrowing this gap by statistical analyses to define descriptors for small-molecule drug-like space. The most famous, Lipinski’s ‘Rule of Five’ [11] predicts passive oral absorption based on log P, molecular weight, and H-bond donors and acceptors. Subsequently, others have considered the importance of parameters such as the number of rotatable bonds, polar surface area (PSA) [12], and ligand efficiency [13,14]. The success of these rules lies in the ease with which the metrics are calculated, and the careful choice of dataset leading to their derivation. For example, Lipinski restricted his study to compounds reaching Phase II clinical trials, with the assumption that failures because of poor permeability would have dropped out at an earlier stage of the discovery process. In addition, the emphasis was on small molecules, with peptides or nucleotides deliberately excluded. Meanwhile, it is useful to keep in mind that the ‘rules’ are in fact guidelines [15 ] and that 20% of all oral drugs violate at least one rule. Lipinski has noted [16] that many natural products remain bioavailable despite violating the Rule of Five. Computational comparisons (for example [3 ]) of natural product datasets compared to synthetic compounds or drugs have not uncovered a reason for this unexpected behavior. Although the ‘average’ natural product differs from the ‘average’ synthetic drug in terms of elemental composition and stereochemical complexity, both show similar values for Lipinski parameters. In part, this must be because of the smoothing out effect when calculating averages with large datasets. Furthermore, Lipinski’s analysis was based on Phase II candidates, whereas the natural product datasets were not filtered in a similar way. We should really concentrate on the subset of successful natural products that went into development, just as Lipinski did not look at all compounds in medicinal chemistry programs. Surprisingly, this type of analysis has not been previously reported, and is the subject of the next section. The molecular architecture of successful natural products How many marketed drugs have a natural product origin, or are based on a pharmacophore first identified in a natural product? This question is easy to answer, thanks to the excellent and comprehensive surveys by Newman at the National Cancer Institute (NCI). The most recent survey [17], covering the period 1981–June 2006, lists a total of 1184 new chemical entities (NCEs) receiving approval. Of these, 52% have a natural product connection, 18% are biologics, and 30% purely synthetic. The question we would like to ask is the reverse: How many unique natural products led to a successful drug launch? Systematic data on this is lacking, and I have attempted to answer this question by metaanalysis of the 1981–2006 timeslice covered in Newman’s review. In doing so, a number of filters were applied to all the 1184 NCEs reported by Newman, as follows: 1. Drugs that were inspired by a natural product lead discovered pre-1970 were discarded. This is an arbitrary decision to emphasize newer natural products that were the result of modern screening campaigns rather than second or later generation drugs based on classic natural products. 2. Where there are two or more natural product leads of closely related structure, I have selected the first compound disclosed in the literature. For example, the statin class of cholesterol-lowering drugs grew from the natural products mevinolin and compactin, which differ only by a methyl group. 3. One of Newman’s categories is ‘ND’ or natural product derived. This literally refers to the starting material used in the drug’s preparation and is not synonymous with a lead. For instance, semisynthetic steroid hormones are usually manufactured by multistep routes from plant steroids but the latter did not serve as an inspiration for the discovery of the former. Modern drug discovery Ganesan 307 Table 1 Some fundamental differences between biosynthesis and synthesis Biosynthesis Synthesis Building blocks Few Many Strategy Branching of intermediate Alteration of building block Scaffold diversity High Low Functional group tolerance High Low Novel motifs Common Rare C–H activation Common, site-specific Rare Stereocontrol Easy, enantioselective Difficult, caseby-case basis www.sciencedirect.com Current Opinion in Chemical Biology 2008, 12:306–317

4. Drugs that are based on an understanding of human physiology or endogenous ligands are excluded. Thus, the origin of CNS drugs can often be traced back to human neurotransmitters rather than a secondary metabolite isolated from another organism. 5. Drugs that can be rationally predicted by a mechanistic understanding of the target’s action are excluded. In the antiviral area, many drugs are nucleoside analogs or transition state inhibitors of viral proteases. In both cases, drugs that work by such mechanisms did not need the discovery of a natural product lead. Some of the filters may be contentious, but I believe the net result is a reasonable first-pass approximation. It tells us how many unique natural product chemotypes discovered in 1970 or later led to a marketed drug in 1981– 2006. The reader might want to pause and take a stab at the answer before reading on. When I have polled academic and industrial chemists, a pretty broad range of values has ensued! A full list of the natural product linked drugs in Newman’s dataset is provided in the supplementary information, subdivided into categories with light annotation. Applying the above filters yields a set of 24 natural products that were the starting point for marketed drugs in the 25-year period 1981–2006 (Table 2, structures in Figure 1). Of these, 19 were isolated from soil microorganisms, actinomycetes in particular, while the remaining 5 were of plant origin. Marine natural products are conspicuously absent, as their systematic exploration became widespread only recently. The majority of these successful natural products were discovered by the pharmaceutical industry through high-throughput screening methods, with others coming from research institutes specializing in natural product chemistry. Overall, nearly half of these leads were discovered in Japan, a testament to the country’s leading expertise in natural products. The third column from the right illustrates the arduous journey from lead discovery to approved drug, with a lag of over a decade being typical. It also shows that a single compound can be the lead for multiple drugs. The second last column gives the route of administration for the approved drugs, while the final column is an indicator of the success of natural product derived drugs, with 17 entries among the top 500 drugs of 2006. Ten of the 24 leads were antimicrobial agents (3 for intracellular targets, 7 for cell wall or membrane targets) while 14 were against targets in man (11 intracellular, 3 membrane or extracellular). In terms of structure, these 24 leads are predominantly of polyketide, peptide, or terpenoid origin. Alkaloids are absent, and a possible explanation is that this class was among the first to be examined as their basic properties aid isolation, and highly biologically active members were already heavily exploited pre-1970. Although a cursory glance at Figure 1 might indicate a random set of structures, they can be classified into two broad categories (Tables 3 and 4) in terms of chemical space. The tables list values for molecular descriptors used in Lipinski’s rules, as well as Veber’s number of rotatable bonds and PSA, and Hopkins’s ligand efficiency. All values that fall beyond the cutoff are shaded (MW > 500, C log P > 5, Hd > 5, Ha > 10, Rot > 10, PSA > 140, HA > 35). In addition, I have noted the number of stereogenic centers as a measure of architectural complexity and a predictor of tractability for medicinal chemistry efforts. A generous St value >5 has arbitrarily been set as one that would discourage synthesis of analogues. Exactly half of the 24 natural products lie in what can be called the ‘Lipinski universe’ (Table 3). The ‘Rule of Five’ is violated only once (spergualin has too many Hbond donors) in these compounds, and many of them are closer to the tighter constraints placed on lead-like space [18] rather than a drug. The average molecular weight is only 319, although the natural product origin is betrayed in the degree of stereochemical complexity, with an average of 4 stereogenic centres present. The other half (Table 4) displays very different molecular properties. We might describe them as a ‘parallel universe’ to Lipinski space. The existence of these two separate clusters can be visualized graphically, as shown in the molecular weight distribution (Figure 2). With the exception of pseudomonic acid and lipstatin, all of the natural products in the parallel universe have at least two Lipinski ‘alerts’. This is accompanied by high values for rotatable bonds, PSA, heavy atoms, and stereogenic centers. Of the 12 natural product leads in the Lipinski universe, 6 (50%) led to orally administered drugs, while the 12 leads in the parallel universe had the identical outcome of 6 (50%) oral drugs. Thus, for these natural products, compliance or otherwise of Lipinski rules is not a reliable predictor of oral bioavailability. The subset of orally administered drugs, just like the total, shows a bimodal distribution when plotted against molecular weight of the lead (Figure 2). Rules for successful natural products Tens of thousands of biologically active natural products were discovered in the period 1970–2006. Yet, only 24 of these had the ‘right stuff’ that resulted in an approved drug [19 ]. On the basis of the data in the preceding section, we can devise some guiding principles that will help in assessing the worth of natural product leads (or indeed synthetic compounds as well) as potential therapeutic agents. log P is the lord of the rules Although natural products in the ‘parallel universe’ (Table 4) may appear to break all the rules, they are remarkably compliant with regard to log P. This under- 308 Molecular diversity Current Opinion in Chemical Biology 2008, 12:306–317 www.sciencedirect.com

Modern drug discovery Ganesan 309 Table 2 The 24 natural products discovered since 1970 that led to an approved drug in 1981–2006 Lead, year, and structural class Origin Discoverer Drug, year Route Ranking Validamycin, 1970 Actinomycete Takeda (JAP) Acarbose, 1990 po 357 Oligosaccharide Voglibose, 1994 po Midecamycin, 1971 Actinomycete Meiji (JAP) Miocamycin, 1985 po Macrolide Pseudomonic acid, 1971 Bacteria Beecham (UK) Mupirocin, 1995 top 436 Polyketide Taxol, 1971 Plant Res Triangle Inst/NIH (USA) Paclitaxel, 1993 iv 81 Diterpene Docetaxel, 1995 iv 123 Cephamycin C, 1971 Actinomycete Lilly (USA) Moxalactam, 1982 iv b-lactam Cefotetan, 1984 iv Cefbuperazone, 1985 iv Coformycin, 1974 Actinomycete Inst Microbial Chem (JAP) Pentostatin, 1992 iv Nucleoside Echinocandin B, 1974 Fungus Ciba-Geigy (SWI) Caspofungin, 2001 iv 293 Cyclopeptide Micafungin, 2002 iv Anidulafungin, 2006 iv Mizoribine, 1974 Fungus Toyo (JAP) Mizoribine, 1984 po Nucleoside Rapamycin, 1974 Actinomycete Ayerst (CAN) Sirolimus, 1999 po 434 Polyketide Everolimus, 2004 po Zotarolimus, 2005 po Compactin, 1975 Fungus Sankyo (JAP) Lovastatin, 1984 po 264 Polyketide Simvastatin, 1988 po 2 Pravastatin, 1989 po 41 Fluvastatin, 1994 po 195 Atorvastatin, 1997 po 1 Cerivastatin, 1997 po Pitavastatin, 2003 po 71 Rosuvastatin, 2003 po Cyclosporine A, 1975 Fungus Sandoz (SWI) Ciclosporin, 1983 po 122 Cyclopeptide Lipstatin, 1975 Actinomycete Roche (SWI) Orlistat, 1987 po 277 Polyketide Bestatin, 1976 Actinomycete Inst Microbial Chem (JAP) Ubenimex, 1987 po Peptide Thienamycin, 1976 Actinomycete Merck (USA) Imipenem, 1985 iv 247 b-lactam Meropenem, 1994 iv 231 Panipenem, 1994 iv Faropenem, 1997 po Biapenem, 2002 iv Ertapenem, 2002 iv Doripenem, 2005 iv Artemisinin, 1977 Plant Qinghaosu Res Grp (PRC) Artemisinin, 1987 po Sesquiterpene Artemether, 1987 po Artenusate, 1987 po Arteether, 2000 po Forskolin, 1977 Plant Hoechst (IND) Colforsin, 1999 iv Diterpene Plaunotol, 1977 Plant Sankyo (JAP) Plaunotol, 1987 po Diterpene Avermectin B1a, 1979 Actinomycete Kitastato Inst (JAP)/Merck (USA) Ivermectin, 1987 po Polyketide SQ26,180, 1981 Actinomycete Squibb (USA) Aztreonam, 1984 iv b-lactam Carumonam, 1988 iv www.sciencedirect.com Current Opinion in Chemical Biology 2008, 12:306–317

lines the central importance of log P in drug discovery. Although an increase in log P can often yield a higher affinity for the target, it tends to be outweighed [20] by pharmacokinetic liabilities such as solubility, permeability, plasma protein binding, metabolic turnover, and toxicity. The single most important lesson from natural products lies in their ability to maintain low log P regardless of other characteristics. In the Lipinski universe (Table 3), average log P is 0, while in the parallel universe (Table 4) it has only risen to 2.2 despite an average molecular weight of 917. It is thus possible to operate in non-Lipinski space with high molecular weight and large numbers of H-bond acceptors and PSA, provided lipophilicity is not compromised. To do so requires the presence of polar functional groups, and this is compatible with biosynthetic pathways which are extremely chemoselective and regioselective. Making such compounds is a lot more challenging for medicinal chemists, and is likely to involve long routes with protection and deprotection schemes for specific functional groups. Consequently, when compounds with higher molecular weight are made synthetically, log P usually suffers. To quote Lipinski [16], ‘...if you look at companies that are selling compounds they usually quote Rule of Five compliance rates and typically what you find is that the parameter that is most difficult to control combinatorially is lipophilicity’. A study [21 ] by AstraZeneca indicates that lipophilic compounds are the most likely to be discontinued during development. More recent papers from AstraZeneca [22 ,23] track the evolution of molecular properties of oral drugs and drug discovery programs over time at several pharmaceutical companies. These show that the molecular weight of compounds made by medicinal chemists has risen, perhaps because of the complexity of modern targets now that the ‘low-bearing’ fruit in drug discovery has been harvested. At the same time, this increase has not been accompanied by a corresponding increase in log P, showing that medicinal chemists are consciously aware of the importance of avoiding high lipophilicity. If you cannot be passive, be active The rules for assessing druglikeness are based upon passive absorption through the lipid membrane, and are no longer applicable for carrier-mediated or active transport. Such processes may be more common than historically believed [24]. A recent estimate suggests 758 transporters in the human genome, with the substrate tolerances unknown for most of them. Many synthetic orally bioavailable drugs may have a component of active transport, and such mechanisms may account for the anomalously high bioavailability of natural product drugs that violate the rules. Because biosynthesis pathways have common features (Table 1), a foreign natural product is more likely than a foreign synthetic molecule to be similar to endogenous ligands or metabolites and accepted as a substrate by transporters. Of course, once such natural products get into the cell, there is a separate issue in their recognition as a xenobiotic, being susceptible to clearance by active efflux pumps. It is tempting to speculate that the body has evolved two parallel strategies for avoiding high molecular weight xenobiotics. Those with high log P are ‘influx-limited’, because of poor solubility, distribution, and propensity for first-pass metabolism. Thus, the defense mechanism is to avoid such compounds reaching the site of action in the first place. On the contrary, those compounds with low log P are ‘efflux-limited’. They reach target cells upon which they can be absorbed by active transport. The defense mechanism relies on similarly active efflux for clearance. For such compounds, efforts to rationally minimize efflux [25 ] or improve passive transport should be an important stage of the drug development process. Not all H-bonds are equal The energy penalty of a H-bond donor that needs to interact with bulk water is higher than that of a H-bond acceptor that is reacting in a reversible manner. This is implicit in the Lipinski rules, as the cutoff for H-bond donors is half of that for H-bond acceptors. Nature follows the same logic, as only 4/12 compounds in Table 4 exceed 310 Molecular diversity Table 2 (Continued ) Lead, year, and structural class Origin Discoverer Drug, year Route Ranking Spergualin, 1981 Bacteria Inst Microbial Chem (JAP) Gusperimus, 1994 iv Peptide Arglabin, 1982 Plant Inst Phytochem (USSR) Arglabin, 1999 po Sesquiterpene FK506, 1984 Actinomycete Fujisawa (JAP) Tacrolimus, 1993 po 103 Polyketide Daptomycin, 1986 Actinomycete Lilly (USA) Daptomycin, 2003 iv Cyclodepsipeptide Calicheamicin g1, 1988 Actinomycete Lederle (USA) Gemtuzumab, 2000 iv Polyketide The second last column lists the route of administration (po = oral, iv = intravenous, top = topical) and the final column gives the ranking among the global top 500 drugs of 2006, according to IMS Health. Current Opinion in Chemical Biology 2008, 12:306–317 www.sciencedirect.com