Natural Products as a Foundation for Drug Discovery

Abstract

Many natural products have been used as drugs for the treatment of diverse indications. While most U.S. pharmaceutical companies have over the years reduced or eliminated their in-house natural product groups, new approaches for compound screening and chemical synthesis are resurrecting interest in exploring the therapeutic value of natural products. The aim of this report is to review emerging strategies and techniques that have made natural products a viable strategic choice for inclusion in drug discovery programs.

1. History

1.1. Early natural product drugs

Humans have long used naturally occurring substances for medical purposes. In particular, plant products have played a leading medical role in most cultures. With the development of the science of chemistry in the early 19^th century, it became possible to examine plant materials more closely to identify biologically active constituents and to understand their mechanisms of action. In 1804 Sertürner purified morphine from the opium poppy and found that it largely reproduced opium’s analgesic and sedative effects (Lockemann, 1951). His success led others to seek “active principles” of medicinal plants. During the 19^th century bioactive natural products were purified from cinchona (quinine) (Pelletier and Caventou, 1820, Borchardt, 1996), coca (cocaine) (Gaedcke, 1855, Gay et al., 1975) and many other plants. The ability to determine chemical structures developed later, with the planar structure of cocaine first reported in 1898 (Willstätter and Müller, 1898) of quinine in 1908 (Rabe, 1908) and morphine in 1923 (Gulland and Robinson, 1923). The technology needed to synthesize these evolved mainly in the 20^th century, with morphine being synthesized in 1956 (Gates and Tschudi, 1956). While the active principle hypothesis has not provided an explanation for all biological activities of natural substances, it has been the most fruitful approach for addressing this issue.

1.2. The antibiotic era

The identification by Fleming (Fleming, 1929) of the antibacterial properties of penicillin and its subsequent isolation by Chain and Florey (Chain et al., 1940) revolutionized the treatment of bacterial infections. This led to extensive screening of microbes, particularly soil actinomycetes and fungi, in an attempt to identify other antibiotics. Using simple bioassays, microbes from soil samples were cultured and their constituents extracted. This led to the isolation and identification of dozens of new classes of antibiotics, many of which were commercialized. Large numbers of these original agents, as well as chemical derivatives, are still used for the treatment of infectious disease (da Cunha et al., 2019). While the development of drug resistance has limited the continued use of many naturally occurring antibiotics, their discovery was the cornerstone for the development of the modern pharmaceutical industry after World War II. Semisynthetic analogues of microbial antibacterials are still being approved as new molecular entities (NME). As detailed in Table 1, a large proportion of natural product NMEs (7/26) were approved for antibacterial use in the period from 2008-2018.

Table 1.

Recent natural product drug introductions, 2008-2018.

Year	Drug name	Company	Indication	NP template
2008	Biolimus stent	Biosensors	Stenosis	Rapamycin
	Ceftobiprole medocaril	Roche	Antibacterial	Cephalosporin
	Methyl naltrexone bromide	Wyeth/Progenis	Constipation	Morphinan
2009	Nalfurafine HCl	Toray	Pruritis	Morphinan
	Tebipenem pivoxil	Spero	Antibacterial	Thienamycin
	Televancin	Theravance	Antibacterial	Vancomycin
2010	Cabazitaxel	Sanofi-Aventis	Cancer	Taxane
	Ceftaroline fosamil	Takeda	Antibacterial	Cephalosporin
	Eribulin mesylate	Eisai	Cancer	Halichondrin
	Fingolimod	Novartis	Multiple sclerosis	Myricocin
	Mifamurtide	Takeda	Cancer	Muramyl dipeptide
	Romidepsin	Fujisawa/Astellas	Cancer	Cyclopeptide
	Vinflunine ditartate	Pierre Fabre	Cancer	Vinca alkaloid
	Zucapsaicin	E Merck AG	Topical analgesic	Capsaicin
2011	Fidaxomicin	Optimer	Antibacterial	Macrolide
	Spinosad	ParaPro	Antiparasitic	Spinosyn A
2012	Arterolane	Ranbaxy	Antiparasitic	Artemisinin
	Carfilzomib	Proteolix.Onyx	Cancer	Peptide
	Crofelemer	Salix/Shaman	Diarrhea	Oligomeric proanthocyanin
	Dapagliflozin	Astra Zeneca	Diabetes	Phlorizin
	Ingenol mebutate	Peplin/Leo	Cancer	Phorbol ester
	Novolimus	Elixir	Stenosis	Rapamycin
	Omacetaxine mepsuccinate	ChemGenex/ Teva	Cancer	Homoharringtonine
	Pixanthone dimaleate	Cell Therapeutics	Cancer	Anthracycline
2013	Canagliflozin	Jannsen	Diabetes	Phlorizin
2014	---
2015	Ceftolozame	Astellas/Cubist	Antibacterial	Cephalosporin
	Dalbavancin	Vicuron/Durata	Antibacterial	Vancomycin
	Empagliflozin	Boehringer/Lilly	Diabetes	Phlorizin
	Nalogexol	Nektar/Astra Zeneca	Consitpation	Morphinan
	Oritavancin	Lilly/Medicines Co.	Antibacterial	Vancomycin
	Tofoglozin	Chugai/Sanofi/Kowa	Diabetes	Phlorizin
2016	---
2017	---
2018	Midostaurin	Novartis	Cancer	Staurosporine
	Naldimedine	Shionagi	Constipation	Morphinan

1.3. Taxol

To stimulate interest in the development of cancer therapeutics, beginning in 1960 the US National Cancer Institute supported an extensive academic network examining plant sources for potential anti-cancer agents. Taxol (Wani et al., 1971) and camptothecin analogs (Wani and Wall, 1969, Lerchen, 2002) were the most prominent products that resulted from that program, although neither was commercialized until the early 1990s. Difficulty in obtaining the quantities of taxol needed for commercialization slowed its advancement, while the poor solubility of camptothecin required modifications to its chemical structure for it to have the physicochemical properties needed for clinical use. Once launched, taxol revolutionized the treatment of certain cancers and continues to play a major role in the management of these conditions.

2. Why has pharmaceutical company interest in natural products waned?

Over the past two decades research on natural products has declined in the pharmaceutical industry. Companies such as Merck (Mullin, 2008), Pfizer, and Bristol Myers Squibb have during this time closed in-house programs in natural products. This trend has been most visible in the United States, with a few European and Japanese companies continuing to support natural products groups. Several reasons have been given for this trend:

2.1. Discovery and development of natural products is perceived as a slow process.

This is a valid concern. Current high throughput screening (HTS) campaigns are designed to identify and prioritize chemical leads from compound libraries within weeks or months. If unpurified natural product extracts are tested initially, the time needed to isolate and identify active constituents make it impossible for natural product isolation to match the speed achieved by testing libraries of pure compounds. As detailed below, strategies have been developed to address this issue.

Natural product samples are most often tested as whole fermentation broths, or as crude extracts of plants and marine organisms. Once a hit is identified in a biological screen the extract must be fractionated to isolate the active compounds. This process typically requires that bioassays be conducted at each level of purification. Thus, the length of time required to conduct the bioassay and report the results, and the number of separation cycles needed to obtain pure compounds, are factors which dictate the time required to process a natural product hit. Even when cycles are performed on a weekly basis using a rapid bioassay, it is unusual for a natural product extract hit to yield a pure compound in less than a month. Other factors that may affect the speed and efficiency of compound identification are chemical instability, challenging separations, and between-run variation in bioassays.

2.2. All of the easy natural product drug discoveries have been made.

This perception is often reinforced by the expression “That pond’s all fished out.” However, while the number of plant and animal species is finite, only a very small fraction of all species has been chemically investigated, let alone broadly examined in a panel of bioassays. The number of higher plant species is estimated to be between 300,000 and 400,000. The largest plant screening program in pharma was conducted by Smith Kline & French, with about 19,000 species screened for alkaloid content using a simple color test (Raffauf, 1996). Over the past 20 years the U.S. National Cancer Institute has collected higher plants for screening, with the collection currently containing about 30,000 species, or roughly 10 percent of the total.

There is no simple way to tally the number of microbial samples that have been screened for biological activity, as the typical protocol in microbial screening is to perform only minimal species identification before undertaking tests for biological activity. Although the number of microbial samples screened is enormous, the taxonomic diversity of these samples is limited by the bias for collecting soil samples and the difficulty in growing all but a small fraction of microbes in culture. Recent advances in environmental microbiology have revealed that there is an enormous amount of unsampled microbiota (Epstein and López-Garcia, 2008, Tedersoo et al., 2014). Also, genomics studies have shown there is a correspondingly enormous amount of biosynthetic “dark matter” lurking in the uncultivated microbial world (Katz and Baltz, 2016). Thus, rather than the pond having been fished out, it appears that new types of bait or new fishing strategies are required to exploit fully the therapeutic potential of natural products. Heterologous expression of biosynthetic modules in host microbes amenable to manipulation is one approach developed for this purpose (Harvey et al., 2018), with activation of silent gene clusters being another (Ravindran, 2017).

Marine invertebrates were extensively studied over the past three decades yielding a great deal of information on the chemistry and biology of their bioactive constituents (Carroll et al., 2019). While the extent of biodiversity among marine invertebrates is unknown it is most likely quite large, given that life evolved first in the marine environment. The true diversity of marine life will not soon be known, at least by using classical methods as there are too few taxonomists to identify and classify new species, and only the easily SCUBA-accessible, shallow, warm marine waters have been extensively explored. Metagenomic methods are now being applied to marine invertebrates to address this issue (Lopez et al., 2019). Symbiosis between microbes and marine invertebrates is an important process that is often mediated by small molecules produced by the microbial partner(s) to defend the macro-organism partner (Florez et al., 2015).

The argument that there is little more to be discovered in natural products is reminiscent of the claim by some 19^th century physicists that their field was nearing completion. While this was perhaps true for Newtonian physics, events of the last century have clearly demonstrated how misguided was their perception. Even if new developments in natural products consist of humble improvements in techniques and understanding rather than revolutionary advances in drug development, it is clear that many “fish” still remain in the pond.

2.3. The complex structures of natural products are too difficult to synthesize.

The chemical structures of natural products range from very simple (e.g., pinene, Figure 1a) to extremely complex. Improvements in structure elucidation techniques have made it possible to determine the complete stereostructure of natural compounds as complex as maitotoxin (Sasaki et al., 1996, Zheng et al., 1996), a compound with a molecular weight of 3425 Da incorporating 98 chiral centers (Figure 1b). Although such agents are likely to never be candidates for commercial scale total synthesis, the molecular weights of most of the natural products isolated and elucidated to date are <1000 Da, and would be reasonable candidates for commercial synthesis. In addition, many commercial drug products have been developed by synthetic modification of a naturally produced precursor. In that case total chemical synthesis is not required. Alternatively, structure-activity studies in conjunction with total synthesis may identify fragments of the parent structure with biological activity, thereby allowing for a drastic reduction in the size and chirality of a bioactive natural product. As discussed below, two examples where this approach has succeeded are bryostatin (Wender et al., 2017) and halichondrin (Yu et al., 2011).

Figure 1.

Extremes of complexity in natural-product structures: (A ) Pinene and (B ) Maitotoxin.

2.4. Resupply of the source organism is difficult.

Obtaining the large amount of a natural compound needed for preclinical studies can be challenging. If derived from a plant which grows in a remote tropical location, physical access for recollection may be difficult, or permission to collect and ship the material difficult to obtain. Another risk is that the plant may only produce quantities of the desired compound under certain environmental or ecological conditions. A marine organism may require an expensive expedition, especially if the animal inhabits deep water or regions with strong or unpredictable currents. Even with a constant supply of material, the factors that induce production of the target substance may be poorly understood. This can result in a high degree of variability in the amount of active material harvested among batches. Understandably, pharmaceutical companies prefer predictable, controllable sources of starting material. Also, for commercial viability, solutions must be found to overcome the vagaries of natural product production. Approaches being taken to addressing these issues are covered in section 5, below.

2.6. High throughput chemistry is viewed as better than natural products.

Parallel synthesis techniques make possible the production of synthetic libraries containing thousands of distinct compounds. However, such rapid synthetic techniques have not significantly increased the rate of identification of drug candidates. Early combinatorial libraries were composed of compounds with poor solubility, with few useful hits found. In some cases, the quantities of compound produced were quite small and the purity highly variable. In recent years, smaller, more focused libraries have increased the yield of potential drug leads. Experience suggests that parallel synthesis is most useful for expanding an existing lead, rather than in creating screening libraries that yield NCEs of with therapeutic potential.

2.7. Natural product compositions of matter may not be patentable in the US.

Two US Supreme Court decisions have raised potential barriers to the patenting of natural products in the United States: Mayo v. Prometheus (Prometheus, 2011) and Association for Molecular Pathology v. Myriad (Myriad, 2013). Mayo concerned patenting a method of administering thiopurine drugs to treat autoimmune diseases which entailed the measurement of thiopurine metabolites to guide dosage of the parent drug to reduce toxicity. The Court struck down the patent because it felt the claims were made for underlying laws of nature, i.e., the variations in drug metabolism. In Myriad, the patents concerned the BRCA1 and BRCA2 tumor suppressor genes. Because mutations of these genes increase the risk of breast and ovarian cancer, it was argued these mutations would be useful diagnostic targets. The claims covered the naturally occurring gene and the DNA sequence, with the Court ruling the patents were similarly invalid. Both these decisions had immediate negative impacts on the patenting of diagnostics (Aceto, 2013, Dorn, 2013, Gordon, 2014).

With respect to natural products, in 2014 the US Patent and Trademark Office issued guidance to its patent examiners that interpreted the Mayo and Myriad cases as blocking composition of matter claims on isolated natural products (Anonymous, 2015, Lee, 2015). The unpopularity of the policy prompted revisions (Servick, 2015, Bahr, 2016, Lee, 2016) but these did not substantively change the position of natural products. A proposal for a legislative remedy (or a test case) was made (Tallmadge, 2017), and a draft bill was recently submitted in the US Senate (Servick, 2019). It remains to be seen how this issue will be resolved. No cases of rejection of patent applications or invalidation of existing composition of matter claims for natural product structures have yet occurred.

3. Why natural products are attractive for drug discovery

3.1. Secondary metabolites have evolved to be bioactive.

The metabolic energy and the genetic cost of the biosynthetic machinery to synthesize a small molecule requires that the compound provide some benefit to the organism, whether by defending it against predators, communicating with other members of its species, or interfering with competing organisms. While most functions of natural products in their producing organism are currently unknown, opinion has shifted markedly since the days when natural products were viewed as mere waste products (Mothes, 1969). Whatever the precise role, it appears now that, like drugs and hormones, many natural products interact with receptor sites on or within cells. The large number of pure natural products found to interact with particular mammalian receptors testifies to the inherent bioactivity of natural products. For example, natural product ligands known to interact with the mammalian GABA receptor include muscimol (Brehm et al., 1972), bicucculine (Johnston et al., 1972), securinine (Beutler et al., 1985), and picrotoxin (Akaike et al., 1985).

3.2. Structures are not limited by the chemist’s imagination.

While chemists may be as creative as natural systems, the natural systems have been at it for a much longer time. The most obvious value of natural products chemistry is the introduction of novel molecular skeletons and functionalities that have not previously been conceived of by humans. Some examples include the cephalostatins (Pettit et al., 1988), mitomycins (Stevens et al., 1965), and esperamicins (Golik et al., 1987), (Figures 2a, b, and c, respectively).

Figure 2.

Structures of (A ) cephalostatin 1, (B ) mitomycin C, and (C ) esperamicin A1.

3.3. The Lipinski rules of five do not apply to natural products.

These rules were developed to drive synthetic chemists towards creating compounds that have the biophysical properties required of an orally active drug candidate. The rules require that compounds should have a molecular weight < 500 Da, posses <5 hydrogen bond donors, <10 hydrogen bond acceptors, and have log P<5 (Lipinski et al., 1997). Lipinski explicitly excluded natural products from these rules, primarily for the reasons set forth above (see Secondary metabolites have evolved to be bioactive), and because they often enter cells by way of membrane transporters rather than passive diffusion (Lipinski, Lombardo, Dominy and Feeney, 1997).

4. High throughput screening and natural products

4.1. Miniaturization and reductionism.

High throughput chemical screening to identify drug candidates evolved in the 1980s from automated clinical analyzer technology and miniaturization aimed at increasing the pace of testing and lowering the cost of testing per sample. As a result, robotic methods of sample manipulation and specialized detectors capable of reading 96-well microtiter plates were developed. At the same time the testing endpoint shifted from empirical measures of cell growth or function to specific interactions with protein molecular targets.

4.2. Cell-free or cellular assays?

In its most extreme reductionist forms, targeted screening involves detection of the interaction of test compounds with a purified, naked protein. Hits from that experimental model are then tested in a functional assay before progression to a cellular, and then tissue level of complexity. Because the highest level of reductionism provides the lowest barrier to successfully identifying hits, the large number of compounds that test positive must be winnowed using secondary, tertiary, and even quaternary assays. Abundant and common natural products such as tannins (7.2.1) overwhelmed reductionist assay strategies with very high hit rates.

There has been a substantial shift in the last decade the use of screening assays that measure biological function directly in cells. As these can typically be tuned to higher stringency, they yield lower hit rates, with positive samples more likely displaying the desired biological properties than is the case with less discriminating tests.

4.3. Change the assay or change the sample?

Natural products do not always behave as expected even in cellular assays, or in functional cell-free assays. The question arises as to whether it is better to adapt the assay to the sample or vice versa. Both approaches have had some success, with the path chosen depending on the availability of resources in chemistry and biology.

A common problem with screening natural product extracts is that many fluoresce in the fluorescein wavelength range (emission maximum 521 nm). This results in a high false positive rate in a screen with a fluorescent endpoint. If the fluorophore endpoint is changed to a label that emits at >560 nm (Cy3B, for example), much less sample autofluorescence is seen, and the false positive rate declines. Alternatively, use of a time-resolved fluorescence label also substantially decreases sample interference. Most sample autofluorescence has a short half-life (i.e., 10 ns), while europium fluorescence labels, for example, have a much longer half-life (ca. 700 ms). Thus, by gating the photodetector to record the signal after a 1 ms delay, the majority of the sample autofluorescence is filtered out, while the label is sensitively detected (Hemmilä and Webb, 1997).

4.4. Prefractionation of extracts

One approach to sample modification that has attracted significant interest is “prefractionating” the crude extract. In its most complex form, this means isolating pure compounds and partially characterizing them before they are tested in bioassays. Several companies have taken this approach, with mixed success (Bindseil et al., 2001, Eldridge et al., 2002). Simpler, lower cost strategies that separate the crude extract into 5-15 samples based on a single chromatography step, followed by solvent evaporation, may provide much of the benefit at a reduced cost (Bugni et al., 2008, Wagenaar, 2008). These approaches all require an investment in automation. Automated weighing capability, flexible programmable liquid handling, and low cost separation media are required to produce the samples and conduct the assays.

There are several benefits with this type of approach: 1) cytotoxic compounds which might mask the activity of another compound in a cellular assay may be separated; 2) minor constituents are concentrated and can be tested at higher concentrations; 3) very polar or lipophilic constituents of an extract can be ignored or discarded entirely. The initial testing results from using prefractionation strategies instead of crude extracts have demonstrated higher hit rates in cellular screening assays (Henrich et al., 2006, Ruocco et al., 2007, Booth et al., 2009, Blees et al., 2010), and higher reconfirmation rates in biochemical screens (Bermingham et al., 2017). The US National Cancer Institute recently instituted a program to prefractionate its complete natural product extract collection for distribution to external screening groups. A library of 150,000 samples is currently available, and the number of samples is projected to reach 1 million (Thornburg et al., 2018).

5. Sourcing

5.1. Natural product sources for drug discovery

As noted above, plants have historically played the leading role in providing drugs or templates for drugs, with microbes following as illustrated in the antibiotic era. Screeners began examining marine sources in earnest once SCUBA facilitated the collection of algae and marine invertebrates. While only a few marine natural products have yielded commercial drug products, many marine compounds have demonstrated activity in screens, with a number of them undergoing preclinical evaluation. Obtaining an adequate supply of compound has been a major hurdle in the study of marine invertebrate agents. For example, bryostatin 1 was initially obtained for clinical studies from its marine source organism, Bugula neritina, using Good Manufacturing Practices (Schaufelberger et al., 1991). However, a mere 18 g of material was purified from 14,000 kg of the producing bryozoan. Mariculture of the same animal was subsequently accomplished, leading to the successful production of large quantities of bryostatin 1 (Mendola, 2003).

A few programs have used insects as a screening source, such as in a collaboration between the Merck and InBio in Costa Rica (Sittenfeld et al., 1999), and in the Eisner lab at Cornell University (Schröder et al., 1998). Also notable is the work of John Daly using amphibians as a rich source of bioactive compounds (Daly et al., 2005). Epibatidine, a frog alkaloid (Badio and Daly, 1994), served as the stimulus for the design of the analgesic drug candidate ABT-594 (Arneric et al., 2007).

5.2. Microbial or dietary origin of marine and plant metabolites

Natural product investigators often encounter difficulties in obtaining reliable production of desired compounds from the producing organism. For example, it is common in microbial screening to confirm bioactivity by regrowing the microbe in the same conditions under which the initial screening sample was produced. In these cases, a success rate of only 50 percent is common. Similarly, when a plant is collected for isolation of larger amounts of constituents, it is not unusual to find low amounts of the desired metabolite, or no compound at all. This is even more common with marine invertebrates.

The reasons for these problems are not fully appreciated, but undoubtedly manifold. With microbes, obtaining good production of a desired metabolite is often a matter of studying the culture conditions (growth media, time, temperature, oxygenation, etc.) and defining the best conditions for reliable production. With plants, the chief problem may be a poor understanding of taxonomy. That is, careful botanical field studies may reveal several closely related species, only one of which produces the compound in question (McKee et al., 1998a). Dependence of metabolite production on environmental factors (climate, season, herbivore pressure) often plays an important role for plants. In marine invertebrates a poor understanding of taxonomy contributes to lack of reproducibility. Additionally, vectoring of metabolites from one organism to another and sequestration in the second organism is known to contribute to variability in production among samples (Thoms et al., 2006, Puglisi et al., 2019, Zan et al., 2019). Dietary sources of bioactive compounds have also been identified in amphibians that consume arthropods and other small leaf litter animals, along with the mechanisms of sequestration (Daly, Spande and Garraffo, 2005, Saporito et al., 2007, Caty et al., 2019) .

Another reason for erratic production may be that the higher organism is not the source of the compound which is instead produced by a microbial symbiont. Support of this comes from the fact that in many cases marine invertebrates have been found to contain compounds which look suspiciously like microbial metabolites (Simmons et al., 2008). Occasionally , similar compounds have been isolated from both a marine invertebrate and a microbe (Suzumura et al., 1997, McKee et al., 1998b). If the microbe is an obligate symbiont, it may be difficult to obtain proof of the relationship. A case has been made by Haygood’s group that bryostatins are produced by a symbiont, with the details of the symbiosis gradually being defined (Hildebrand et al., 2004, Miller et al., 2016). Similar reports for endophyte-plant derived compounds are also intriguing (Smith et al., 2008). An example is the isolation of taxol from an endophytic fungal associate of the Pacific yew and its antifungal utility to the plant host (Stierle et al., 1993, Soliman et al., 2015).

5.3. Synthesis of natural products versus biological production

Organic chemists have made great strides in being able to synthesize complex, chiral molecules such as natural products. While difficulty and cost are directly related to the number of chiral centers and molecular weight, total synthetic approaches for the production of natural products are becoming more common. Given sufficient resources, it is possible to reduce the number of synthetic steps required to reach the target molecule and improve the yield at each step, while using inexpensive starting materials. Wender’s group has developed synthetic routes to both bryostatin 1 (Figure 3a), and to “bryologs” (Figure 3b), compounds that display a potency similar to bryostatin 1 but with more simplified chemical structures. One recent, highly active bryolog, with differential potency between two PKC isoforms, was prepared in 10 steps in an overall yield of 30 percent (Wender, Hardman, Ho, Jeffreys, Maclaren, Quiroz, Ryckbosch, Shimizu, Sloane and Stevens, 2017).

Figure 3.

Structures of (A ) bryostatin 1 and (B ) a bryolog.

Another example of synthetic success with a complex natural product is halichondrin B (Figure 4b). Wild collection of the producing sponge gave poor yields of the target compound (Uemura et al., 1985, Pettit et al., 1991). Mariculture in New Zealand yielded similar levels of halichondrin B (Munro et al., 1999). When a total synthesis was accomplished by the Kishi group (Aicher et al., 1992), it was discovered that fragments half the size of the natural product possess all of the bioactivity (Wang et al., 2000, Dabydeen et al., 2006). In 2010 Eisai gained FDA approval of eribulin (Figure 4b) for the clinical management of some types of cancer (Huyck et al., 2011).

Figure 4.

Structures of (A ) halichondrin B and (B ) its simplified analog eribulin.

An example from my own group involves englerin A, whose structure was published in January 2009 (Ratnayake et al., 2009). This small sesquiterpene diester contains seven adjacent chiral centers, which constitutes a substantial challenge for total synthesis. Nonetheless, within a year, with no prior disclosure, the Christmann group reported its total synthesis (Willot et al., 2009). Many other groups followed with various efficient synthetic approaches, notably those of the Chain (Li et al., 2011) and Echavarren (Molawai et al., 2010) laboratories. Together all of the many studies have provided access to over 100 analogues in a relatively short time. Analogues of englerin A are currently in advanced preclinical development as antineoplastic agents (Wu et al., 2017).

5.4. Biosynthesis in heterologous organisms

Improvement in DNA sequencing tools have facilitated investigations of the biosynthetic pathways that lead to secondary natural products (Walsh, 2015). The biosynthesis of polyketide natural products has attracted a great deal of attention because many commercial antibiotics are derived from this pathway. However, non-ribosomal peptide synthesis, terpenoid biosynthesis and flavonoid pathways have also been elucidated in many organisms. A key observation is that many such pathways consist of modular gene clusters that can be manipulated as an entire unit, particularly in bacteria (Donadio et al., 1991). Polyketide synthase modules share enough homology that they can be identified in relatively distantly related organisms by lowering the stringency of hybridization reactions. While most biosynthetic gene clusters are not expressed under typical cultivation conditions, cryptic clusters can be turned on by a variety of elicitors to stimulate production (Xu et al., 2017).

In fact, such modules are detected in uncultivatable microbes (Piel, 2002), where elicitation is not an option. Metagenomic studies of environmental samples have become quite popular, with analysis of different environments yielding detailed information on the capabilities of organisms that cannot be cultivated by conventional techniques. For example, a recent study of northern California grassland soil reconstructed hundreds of nearly complete bacterial genomes containing a large repertoire of biosynthetic genes (Crits-Christoph et al., 2018).

This opens up the possibility of expressing the module in a convenient heterologous organism to obtain the desired secondary metabolite, assuming the needed precursors are available and other cellular machinery is compatible with the production of the metabolite (Zhang et al., 2008). This involves “refactoring” to adjust the suitability of the conditions and host organism to allow the desired gene cluster to be expressed in a functional state. One such platform has been described for fungal natural products (Harvey, Tang, Schlecht, Horecka, Fischer, Lin, Li, Naughton, Cherry, Miranda, Li, Chu, Hennessy, Vandova, Inglis, Aiyar, Steinmetz, Davis, Medema, Sattely, Khosla, St Onge, Tang and Hillenmeyer, 2018).

In addition, by altering the module, altered analogous metabolites may also be produced (Xu et al., 2009). It has even become common to predict the biosynthetic product from the sequence of a polyketide module (Banskota et al., 2006).

Progress with plant biosynthetic pathways has been slower than for microbes, primarily due to the scattered distribution of biosynthetic genes in the organism. The Keasling group recently succeeded in expressing the complete biosynthetic pathway for cannabinoids in yeast, (Luo et al., 2019), as well as a partial pathway to artemisinin in yeast, completed by chemical synthesis (Paddon et al., 2013). The Møller laboratory has made good progress on selected terpenoid plant systems inserted in yeast, including the concept of combinatorial biosynthesis (Møller and Ratcliffe, 2014, Andersen-Ranberg et al., 2016). These strategies were reviewed recently (Cravens et al., 2019).

5.5. Ethnobotany

Knowledge of the medical effects of plants is not limited to European science and cultural traditions. Botanists trained in anthropology have studied many non-western cultures to inventory their use of plants and other natural substances for medical and other purposes (Buenz et al., 2018). Chemical and pharmacologic investigation of ethnobotanical information is a viable alternate pathway to high throughput screening for drug discovery, although it has its own limitations.

First, cultural concepts of disease can vary. While most cultures readily recognize a superficial fungal infection or diabetes in a similar manner, most cancers are not diagnosed and interpreted in the same manner across cultures (Hartwell, 1967). Nonetheless, some have claimed that plants used traditionally for medicinal purposes yield a higher fraction of anti-cancer activity than unselected plants (Spjut, 2005). In addition, the medical effects of many plants used in traditional medicine may be less specific than what is desired by western pharmacologists. For example, while tannins are often found in herbal preparations and may contribute to their biological activity, tannins are usually not well-suited to drug development. Nonetheless, SP-303, developed by Shaman Pharmaceuticals, (Holodniy et al., 1999) was a carefully defined tannin preparation from Croton lechleri, a Peruvian ethnobotanical (Williams, 2001) that was tested for several antiviral indications before being sold an over-the-counter anti-diarrheal agent. In 2013 it entered the prescription market as Crofelemer (Chordia and MacArthur, 2013).

Another issue is exemplified by both Chinese traditional medicine and the Indian Ayurvedic system. Both of these ancient traditions utilize polyherbal preparations for the majority of prescriptions. Each component is thought to play a particular role, in some cases by modulating the toxicity of another component. This complexity makes active principle analysis difficult. Reductionist approaches to Chinese and Ayurvedic preparations have been largely unsuccessful in validating their traditional uses, although many bioactive molecules have been isolated from both (Tang and Eisenbrand, 1992, Deocaris et al., 2008). The use of microarrays to study the in vivo effects of complex preparations may hold some promise for better understanding of these products and there future clinical use (Yin et al., 2004).

5.6. Chemical ecology

While natural products are now known to not be waste products of the producing organism, (Mothes, 1969) the purpose they serve for the producer is rarely similar to their potential use in human medicine. Most drugs act through interaction with protein receptors. Domains of proteins, though not their precise function, are widely conserved (Rompler et al., 2007). Thus, ligands targeted to a particular domain in nature may also have activity in an orthologous or paralogous receptor in humans. C. elegans, for example, has been proposed as a model organism for anti-Parkinson drug screening; many of the compounds which affect dopaminergic systems in humans also have more or less parallel effects in worms (Nass et al., 2008).

Investigation of the ecological function of natural products is a field unto itself, and elucidation of the role a compound plays can be experimentally challenging. Those that have been identified include antifeedant effects (Lidert et al., 1987), allelopathy (interference with growth of competitors) (Tseng et al., 2001), and endocrine disruption (Dinan and Lafont, 2006).

5.7. NCI Letter of Collection

In the late 1980s, contracts were developed by the U.S. National Cancer Institute for the collection of large numbers of plant, microbial and marine samples worldwide. The collectors required permits to collect in many different countries and those countries needed assurance that their rights would be respected in the drug development process. To that end, the NCI developed a standard Letter of Collection which is signed by both parties (Cragg and Newman, 2005). This letter states the willingness of the NCI to collaborate with source country scientists, to deposit voucher specimens in source country repositories, and to develop benefit-sharing arrangements when patents were filed. In addition, Memoranda of Understanding could also be developed to frame direct collaborations.

5.8. Convention on biodiversity

The NCI agreements predated and presaged the 1992 Rio Convention on Biological Diversity (CBD). While the U.S., alone among United Nations member states, has ratified but not signed the treaty, U.S. Department of State policy requires following its principles. The CBD calls for preservation of biological diversity, for protection from exploitation of source country genetic resources, for equitable sharing of the benefits of technology, and for technology transfer to the source country. The 2010 Nagoya Protocol focused on implementation of equitable benefit sharing under the CBD (Cragg et al., 2012, Joppa et al., 2013).

While it is generally perceived that the CBD made access to natural products resources more difficult, it has reduced the worst abuses of source countries by the developed world. It has not resolved the political issue of how benefits should be distributed within the source country, however. See for example the case of Hoodia, a weight loss product from the San people in South Africa where active constituents were patented by government scientists at the South African CSIR and licensed to Phytopharm plc and Unilever (Wynberg, 2004, Anonymous, 2006, Bladt and Wagner, 2007). Currently Hoodia products are only available as dietary supplements.

6. Techniques in natural products drug discovery

6.1. Extraction

Before tissues of an organism can be tested, they must undergo an initial extraction to separate the desired small molecules from the biopolymers (proteins, cellulose, chitin, nucleic acids) that make up the bulk of the material. For plants, before extraction it is common to dry plant parts thoroughly in the field at the point of collection, so the material does not decompose en route to the laboratory. To accelerate extraction, the dry tissue is ground using any of several mills (e.g., a Wiley mill, or a hammer mill). Alternatively, tissues may be frozen, although this is often expensive and cumbersome. Frozen material may be lyophilized. If DNA or mRNA is sought for cloning of proteins, flash freezing the freshly collected tissue in liquid nitrogen is required to obtain useful material.

There are very few standard techniques for extraction, because choice of solvent and conditions depends on the spectrum of small molecules desired. For extraction of drug-like molecules of intermediate polarity, the NCI has found percolation at room temperature with a 1:1 v/v mixture of dichloromethane and methanol to be useful (McCloud, 2010). Extraction techniques that involve heating the solvent and extracted compounds, as in a Soxhlet apparatus, are generally avoided unless the desired compounds are known to be heat stable. Otherwise, heating should be avoided in preparing samples of unknown composition for biological screening.

Marine invertebrate tissues present unique extraction problems because of high water and salt content. An approach adopted at the NCI has proven generally applicable for a wide variety of marine specimens. Frozen samples are broken into pieces small enough to be fed into a commercial hamburger grinder with CO₂ pellets. The resulting powdered material is stored frozen long enough for the CO₂ to sublime, then thawed briefly and stirred with water as a slurry. Filtration through paper in a low-speed centrifuge removes the mucilaginous tissue, and the resulting aqueous extract is freeze dried. The marc (remaining solid residue) is also lyophilized and then extracted with the methylene chloride-methanol mixture (McCloud, 2010).

The solvent must be removed from the solutions which result from any of these extraction procedures. This is done to make it possible to obtain a weight for the extracted material and to avoid chemical reactions in solution which may alter the constituents. Aqueous solutions are lyophilized, while organic solvent mixtures are dried using rotary evaporators. A final finishing under high vacuum removes most traces of the solvent. Materials should be stored in borosilicate glass bottles or vials at −20° C to ensure stability.

For high throughput screening applications, it is common to store libraries in dimethyl sulfoxide (DMSO). DMSO is an extraordinarily good solvent for most natural product samples, including extracts. Organic extracts can often be entirely dissolved in DMSO at concentrations of 10-100 mg/ml, while 50% DMSO solutions of aqueous extracts are possible. It should be noted that DMSO concentrations >25% generally suppress bacterial growth in solutions of aqueous extracts. The bulk extract material should not be stored in DMSO, however, since this solvent can facilitate a number of oxidation reactions. In addition, because of its hygroscopic nature DMSO causes moisture absorption even in nominally sealed microplates in the freezer (Ellson et al., 2005). Such extract plates should be reconstituted from bulk stocks on a regular basis to avoid sample deterioration.

Each bioassay will have a limit on the amount of DMSO it can tolerate. Typically this is 0.5-1% of assay volume for cellular assays. For biochemical assays, it is often as high as 5-10% of assay volume. The limit should be established in advance for a particular test and DMSO controls run in every experiment.

6.2. Separations

Once an extract has been confirmed as a hit in a biological assay, the active compounds in the extract must be identified. This is accomplished using an iterative process termed bioassay-guided fractionation. To this end, an extract is separated into several fractions and the parent extract and fractions tested in the assay. Several outcomes are possible. One is that all activity may be lost in the daughter fractions, in which case the separation method is deemed unsuitable. Loss of biological activity may be due to irreversible binding to the separation media, or to instability of the active compound. A second possibility is for all or most daughter fractions to display some low level of activity. This too is undesirable and is taken to indicate that the separation mode is unsuitable. The third and desired outcome is where one or several daughter fractions contain substantial amounts of bioactivity, with the mass of active fractions reduced from the parent with a corresponding increase in potency. A useful technique in monitoring separations is to calculate both mass and activity recoveries. Thus, if a 5 g sample of a parent extract is separated and yields a summed fraction mass of 4.5 g, the mass recovery is 90 percent. Using concentration-response curve data from the assay, bioactivity recovery is calculated by the equation

$$\frac{\sum ({\mathrm{M}}_{\mathrm{i}}∕{\mathrm{I}}_{\mathrm{i}})}{{\mathrm{M}}_{\mathrm{p}}∕{\mathrm{I}}_{\mathrm{p}}}$$ Equation 1

where M_i are the masses of the fractions, I_i are the IC₅₀ values for each fraction, and M_p and I_p the respective values for the parent extract. If a fraction displays no activity, the term can be ignored. This calculation is limited by the precision of the bioassay, but can be useful in judging the success of a trial separation.

The invocation of synergism to explain loss of activity on fractionation has only rarely been proven experimentally. If activity is lost, most commonly it is attributed to compound instability or irreversible binding to chromatography media. Given a suitably precise assay, calculation of mass and activity recovery will often provide clues to the source of the problem.

A single separation step is rarely sufficient to obtain purified active compounds. While use of high performance chromatography can often yield a superb separation of complex materials, it is more cost-effective to save the high performance step for last because crude extracts can damage expensive preparative HPLC columns. The most useful first separation process is one based on polarity. For example, the so-called Kupchan partition employs a series of two-phase mixtures in a separatory funnel to sort components by partition coefficient. While simple, the technique suffers from a tendency to form emulsions, and from difficulty in evaporating to dryness the water-saturated organic layers. A more convenient approach for organic plant extracts involves solid phase extraction with diol bonded phase media, with increasingly polar solvents used to elute successive fractions (Beutler et al., 1990). This procedure can be scaled over a wide range of volumes, and introduces no water into the samples. For marine samples, a wide-pore C₄ bonded phase scheme can be employed with methanol-water mixtures to separate the large amount of salts and other polar material from the more drug-like intermediate polarity fractions (Cardellina et al., 1993).

Intermediate resolution techniques such as flash chromatography, or gel permeation chromatography are useful once the polarity cuts are made. Open column systems using Sephadex LH-20 with a variety of solvents separate based on both size exclusion and adsorption mechanisms, can be very useful.

Final purification is most often accomplished by preparative HPLC. A wide variety of bonded phases are available (e.g., cyano, C18, phenyl, diol, amino,) which can be operated in reversed-phase or normal phase modes, as well as by ion exchange or hydrophilic interaction chromatography. Pilot thin layer chromatography experiments can provide useful hints on the best choice of column packing and elution conditions. Ultimately, analytical scale HPLC is used to define precise flow and solvent strength parameters. Even with relatively purified fractions, it is often useful to employ gradient elution to obtain an optimum separation. While C18 bonded phases dominate the analytical chemistry market, they are only one of the tools in the HPLC column drawer of a natural products isolation laboratory.

It is also important to carefully consider peak detection. It is common analytical practice to use UV detection at 254 nm, which is useful for compounds with suitable chromophores. However, many constituents of natural materials lack absorbance in this range. The most effective strategy is to use lower wavelengths for detection. For acetonitrile-water systems it is possible to use wavelengths as low as 200 nm to observe compounds with poor UV absorbance. Alternative methods are evaporative light scattering detection or refractive index detection. However, neither of these approaches are very well suited for larger scale separations.

Next, the separation must be scaled up to semi-preparative or preparative scale using larger diameter HPLC columns with the same length, column chemistry, particle size, and porosity used for the smaller samples. Loading studies with increasing injections of material establish the amount of mass than can be effectively separated in a single run. The high cost of larger columns is offset by the shorter time required to run the separation, with columns as large as 41 mm diameter being used with laboratory scale pumping systems capable of delivering to the column 50-100 ml/min of solvent. If flow rates and injection volumes are scaled proportionately, preparative separations can be obtained with the same reproducibility and resolution as analytical separations. The sample injected on an expensive preparative scale column must be carefully filtered and the solvent conditions chosen to elute virtually all of the applied sample. Otherwise, particles and other uneluted material will rapidly degrade column performance.

An excellent overview of preparative chromatographic techniques applied to natural product isolation is available in book form (Hostettmann et al., 1998) and in a recent review (Bucar et al., 2013).

6.4. Structure elucidation

Once the active compounds are obtained in pure form, they are subjected to structure elucidation. The key technique for this is NMR, specifically a series of two-dimensional experiments (COSY, HSQC, HMBC, NOESY) that make it possible to establish the connectivity of all hydrogen and carbon atoms. Serving a very important complementary role is high resolution mass spectrometry (MS), which is capable of providing precise mass measurements that identify the molecular formula of the compound. It is often possible to identify the structure of an unknown molecule using only NMR and MS. Other spectroscopic techniques, such as UV, IR, circular dichroism (ECD) and optical rotation, serve ancillary roles, although they may be critical in specific cases. As the number of atoms in a molecule increases, structure elucidation becomes more difficult because of the exponential increase in possible structures for a given formula. It is routine to determine structures of compounds under a molecular weight of 500 Da, while those over 2,000 Da nearly always require extensive chemical degradation or transformations to establish their structures. Exceptions are smaller biopolymers such as peptides which can be easily sequenced if all of the constituent repeating components are known.

The ability of NMR and MS to provide useful information from smaller amounts of compound has increased substantially in recent years. In particular, advances in NMR probe design, especially gradient probes, flow probes and cryoprobes, have greatly increased sensitivity (Reynolds and Enriquez, 2002). Higher field strength magnets have increased NMR spectral dispersion so that more peaks can be resolved in a spectrum. Improved NMR pulse sequences have reduced experiment time and increased resolution. One advancement is the construction of pulse “supersequences” which chain multiple 2D sequences in a single measurement, dramatically reducing acquisition time (Kupče and Claridge, 2017). Another is the use of residual chemical shift anisotropy in proton-poor molecules (Liu et al., 2018, Milanowski et al., 2018).

Similar improvements have been made in MS, with electrospray ionization and matrix assisted laser desorption being two ionization techniques that have proven to be highly valuable in natural product characterization. Some cutting edge techniques, such as Fourier transform cyclotron resonance mass spectrometry (FTICR-MS), are used mainly in industrial settings because of costs (Feng and Siegel, 2007).

Computational prediction of spectroscopic properties such as ¹³C-NMR shifts and ECD curves, has become more common in structure elucidation, particularly in cases where stereochemical features are difficult to define (El-Elimat et al., 2015, Tran et al., 2019).

An alternative technique for structure elucidation is x-ray crystallography, which has for decades been employed in natural product structure elucidation. It remains an important tool, especially for determining the absolute configuration of complex chiral molecules. The obvious limitation of this approach is that the compound of interest must be prepared in a crystalline form. If crystallization using conventional techniques is not possible, the compound can be derivatized with a variety of modifiers in an attempt to improve its ability to form crystals. Application of robotics to automatically generate many small-scale crystallization experiments has increased the likelihood of identifying crystallization conditions.

6.5. Crystalline sponges

A new development in crystallography is the “crystal sponge method” developed by the Fujita group at the University of Tokyo (Inokuma et al., 2013). As noted above, many natural products fail to crystallize using standard protocols. With host-guest technology the analyte is soaked into a porous metal-organic complex which is itself highly crystalline. If the analyte binds to the metal complex in a consistent orientation, x-ray diffraction studies can then be conducted on a very small scale. The Clardy group has explored the process to determine its scope and transferability and generated some guidelines for success (Ramadhar et al., 2015). Moreover, Fujita has updated the methodology (Hoshino et al., 2016). To date, most of the applications of this method for characterizing natural product structures were performed by the Fujita laboratory (Yoshioka et al., 2015, Matsuda et al., 2016, Urban et al., 2016, Brkljaca et al., 2017, Lee et al., 2017, Kai et al., 2018), although a Chinese group has published the absolute configuration determination of a plant metabolite, asarinin, using crystalline sponge methods (Li et al., 2019). It has been reported that crystalline sponges could be used as an affinity matrix to trap sesquiterpenes from a crude red algal extract for analysis (Wada et al., 2018).

6.6. Microelectron diffraction

A newer technique derived from cryoelectron microscopy has been developed that utilizes electron diffraction rather than x-rays to determine the structure of small molecules. Cryogenic cooling is important in reducing damage to the sample due to high energy electrons. Two groups published this procedure in 2018 describing their success in imaging very small crystals (~100 nm) of small molecules with a cryoelectron microscope (Gruene et al., 2018, Jones et al., 2018). Tiny crystals have the advantage of forming much more readily and with fewer defects than macroscopic crystals. While cryoelectron microscopes are expensive (~$6 million) and heavily booked for single particle studies of proteins and protein complexes, microED can be efficiently performed on equipment no more expensive than an NMR. The scope and difficulty of microED remain to be determined (Brown and Clardy, 2018, Nannenga and Gonen, 2019).

While the ability to determine novel structures with spectroscopic and crystallographic methods with smaller samples are important advances, animal testing has not been miniaturized. Therefore, it is always necessary to conduct preparative separations to obtain sufficient material for in vivo analysis.

6.7. Hyphenated techniques

Hyphenated techniques such as HPLC-MS, HPLC-UV, and HPLC-NMR are useful analytical platforms for detection, identification, and quantification of compounds in extracts. These serve as important tools for determining the compounds in a sample and may inform preparative separation methods. They also are important techniques for chemical dereplication (see below). In addition to coupling several different detection methods, HTS bioassays may be conducted on individual fractions to complement the physicochemical data. One of the drawbacks of using hyphenated techniques is the large data sets generated from each run. Managing, analyzing, and interpreting the results can be daunting.

7. Dereplication

7.1. Biological and chemical

With over 300,000 known small molecules characterized from natural sources, it is no surprise that previously known natural products are often re-isolated during bioassay-guided fractionation. While this may be a welcome discovery if the biological activity is new, a great deal of time and effort can be spent in de novo structure elucidation of known compounds. This problem first emerged during the search for new antibiotics, where the taxonomy of microbial cultures was generally not identified prior to screening. Efforts to avoid investing resources in the elucidation of known compounds are identified by the term dereplication (Corley and Durley, 1994). In all its forms, dereplication is based on attempts to shift the identification of known compounds to an earlier point in the discovery process, either before a pure active substance is isolated, or before a complete NMR data set is acquired and analyzed. Direct physical comparison with standard compounds is a very effective tactic. However, amassing a library of known compounds is an unattainable task for most laboratories.

Most effective is a combination of biological and chemical methods. If the source organism has been identified, reference to databases of known compounds such as the CRC Press Dictionary of Natural Products (Buckingham and Thompson, 1997), can suggest candidate structures. Physicochemical data, in particular ultraviolet spectra and mass spectra, can rapidly limit the scope of possible compounds, especially when combined with analytical HPLC (Lang et al., 2008). Efforts to construct a large public database of mass spectral data are beginning to bear fruit (Allard et al., 2016, Wang et al., 2016, Covington et al., 2017, Olivon et al., 2017, Nothias et al., 2018, Olivon et al., 2019). While NMR database tools have also advanced they are less comprehensive than the mass spectral findings (Bakiri et al., 2017, Zhang et al., 2017, Buedenbender et al., 2018). An effort to build a comprehensive NMR dataset of natural products has been proposed (McAlpine et al., 2019). A system capable of merging mass spectral and NMR data would be ideal (Zani and Carroll, 2017).

7.2. Nuisance compounds

Not all compounds contained in a natural product extract are viable as drug candidates. Several classes of such undesirables are described below.

7.2.1. Tannins

Tannins are polyphenolic plant metabolites that were initially discovered as the compounds responsible for tanning leather. Oak bark and many other plant materials contain substantial quantities of tannins, complex molecular structures which incorporate gallate esters (hydrolysable tannins, e.g. Fig. 5a) or flavanol polymers (condensed tannins, e.g. Fig. 5b) (Khanbabaee and Ree, 2001). Phlorotannins are a third class found in brown algae. Tannins play an important ecological role in deterring feeding by herbivores, and may be produced in response to tissue injury. Many tannins are reported to be antinutritional in that they reduce the digestible protein in foods (Butler, 1992). The mechanism for both the tanning effect and the antifeedant/antinutritional roles is related to noncovalent binding to proteins. Because this effect is a relatively nonspecific effect, with a given tannin being capable of binding to many different proteins, tannins are considered poor candidates for drug development. Much effort has been expended over the years in removing tannins from natural product screening samples because they can be active in a wide variety of cell-free and cell-based assays (Cardellina, Munro, Fuller, Manfredi, McKee, Tischler, Bokesch, Gustafson, Beutler and Boyd, 1993, Wall et al., 1996).

Figure 5.

Structures of some common nuisance compounds. (A ) A common condensed tannin, proanthocyanidin C1. (B ) A hydrolyzable tannin. (C ) A phorbol ester, phorbol 12-tigliate 13-decanoate. (D ) A saponin, ginsenoside Rb2. (E ) General repeating structure of a marine anionic polysaccharide. (F ) General repeating structure of a cationic polymeric alkylpyridine, halitoxin.

7.2.2. Phorbol esters

Phorbol esters are diterpenes produced exclusively by plants in the Euphorbiaceae and Thymelaeaceae families (e.g. Fig. 5c). Many compounds of the class are skin irritants and tumor promoters. Their mechanism of action is activation of intracellular protein kinase C (PKC) (Nishizuka, 1984). Because many cellular functions are PKC-dependant, phorbol esters are pleiotropic agents that affect many cellular pathways. Hence, although phorbol esters appear as hits in a number of cellular screens, they are not good drug candidates because of their potential toxicities and tumor-promoting properties. The general distribution of phorbol esters in different species has been described (Beutler et al., 1989, Beutler et al., 1990, Beutler et al., 1995, Beutler et al., 1996).

7.2.3. Saponins

Saponins are glycosides of triterpenes or sterols produced by many plants (Hostettmann and Marston, 1995). The number of sugar residue varies from one to a dozen, and other chemical functionalities may be appended in various ways (e.g. Fig. 5d). Their ability to act as detergents and form foams in water solution is related to their use as soaps and for killing fish. In the context of biomedical screening assays these properties lead to cell lysis, which can be yield either a false positive response or to interference, depending on the nature of the assay endpoint. In addition, some saponins cause hemolysis, an undesirable property for a drug candidate. A diagnostic feature of saponins in a cell growth assay is that cell lysis is an extremely rapid process, occurring over several minutes, whereas other cell-killing mechanisms generally require numbers of hours. Thus, time-course studies are used to help to distinguish saponins from other types of hits. It is important to note that because not all saponins are detergents or hemolytic some may be viable drug leads (Bento et al., 2003, Tang et al., 2007).

7.2.4. Anionic polysaccharides.

Cellulose, a neutral polysaccharide, is the primary structural material of plant tissues. For animals, cartilage, which is composed of collagen and proteoglycan, plays a similar role in animal tissues. The carbohydrate portion of proteoglycan is composed of heavily sulfated N-acetylglucosamine and hexuronic acid units (e.g. Fig. 5e). These materials, which are often found in marine invertebrate aqueous extracts, are of high molecular weight and carry a substantial negative charge (Beutler et al., 1993). Although anionic polysaccharides are highly active in cellular HIV assays (Beutler, McKee, Fuller, Tischler, Cardellina, Snader, McCloud and Boyd, 1993), their high molecular weight and heterogeneity make them unattractive as drug candidates. Sulfated cyclodextrins, which have been studied as potential antiviral agents, have substantially the same antiviral activity as anionic polysaccharides without some of their liabilities (Moriya et al., 1993). Sulfated polysaccharides are encountered as hits in a variety of cellular screens. They may be removed from extracts by precipitation from ethanol solution at low temperatures. Plants also produce anionic polysaccharides, although they generally display weaker activity than those found in animal tissues.

7.2.5. Cationic polymeric alkylpyridines.

Less often, cationic polymers are found as nonselective hits from natural product extracts. Marine sponges are the usual source of these compounds (e.g. Fig. 4f) (Schmitz et al., 1978, Davies-Coleman et al., 1993).

7.3. Pattern matching

Another approach to identifying or eliminating known natural products without investing resources in their re-isolation and characterization is to compare biological and chemical “fingerprints” to standards. By using the results of multiple biological and chromatographic experiments in which the standard compounds have previously been tested, it is possible group similar samples together, and to propose a dereplication hypothesis for the samples whose results match those of a known compound.

The most data-rich environment in which this has been done is for the NCI 60-cell results. Because thousands of natural products have been tested, these can be used as reference points in data analysis in comparison with results from fractions of crude extracts. A variety of mathematical approaches have been used for the analysis, including calculation of Pearson correlation coefficients (Paull et al., 1989), neural networks (Weinstein et al., 1992), and self-organizing maps (Keskin et al., 2000). If the mechanism of action of the reference compound is known, the correlated test samples can be rapidly screened to determine whether they display similar mechanisms (Paull et al., 1992, Weinstein et al., 1997). This approach has been utilized for agents affecting tubulin (Paull, Lin, Malspeis and Hamel, 1992), epidermal growth factor pathways (Wosikowski et al., 1997), and vacuolar ATPase inhibitors (Boyd et al., 2001), among others.

There is no a priori reason why pattern matching must be limited to cell growth inhibition data. In fact, any type of data can, in principle, be mixed, even chromatographic, spectroscopic and taxonomic information. The utility of pattern matching depends mainly on the number of dimensions present in the data matrix. While redundant dimensions (i.e., cells which respond identically) do not contribute, scattered missing data are only a minor issue if the appropriate analytical techniques are applied.

8. Recent natural product drug introductions

Natural products and their derivatives continue to be approved as new drugs (Table 1). The information on Table 1 is not comprehensive as it excludes peptide drugs, antibody-drug conjugates and other agents that could arguably be considered natural product derivatives. For more comprehensive discussions of approved natural product drugs or those undergoing clinical testing, see the reviews by Cragg and Newman, and Butler (Butler et al., 2014, Newman and Cragg, 2016).

9. Which companies are still conducting natural products discovery ?

While natural products groups have been eliminated in most large pharmaceutical companies in the U.S., they remain active in Europe and Japan. Of the companies listed in Table 1, Bristol Myers Squibb, Merck, Johnson & Johnson, Pfizer, Glaxo Smith Kline, and Lilly no longer maintain in-house natural product discovery groups. Up until its merger with Pfizer, Wyeth had such a group at its Pearl River facility, which has since been closed. In Europe, Novartis continues its efforts in natural product drug discovery.

A corresponding trend is development of smaller, boutique companies that specialize in natural product drug discovery with the aim of licensing their leads to larger entities for clinical development and marketing (Gullo and Hughes, 2005). Pharmamar is an example of a small company that has had success in bringing a natural product (Yondelis) to market in collaboration with a major pharmaceutical firm. Nereus Pharmaceuticals advanced a marine microbial proteasome inhibitor (NP-0052, Marizomib) into phase Ib combination trials in cancer. Triphase purchased the rights to this agent and is conducting phase III trials in partnership with Celgene (ClinicalTrials.gov). Kosan Biosciences, which developed epothilone analogs using biosynthetic technology, was acquired in 2008 by Bristol Myers Squibb on the strength of its development pipeline, with ixabepilone being approved for cancer therapy (Cianfrocca, 2008). Small companies can also serve as screening contractors or provide the natural product libraries and expertise for pharma screening (e.g., Albany Molecular Research).

The public sector has taken on a larger role in drug discovery generally, and natural products drug discovery in particular. For example, an NIH-organized consortium has crowd-sourced 346 pure natural product screening samples from 45 primarily academic groups and screened the pilot set in 50 HTS assays (Kearney et al., 2018).

It is clear that the landscape of natural product research and drug development is rapidly changing. It is a major challenge to maintain the natural product knowledge base and resources developed by large companies. These resources are often lost with corporate mergers, acquisitions and restructuring. Exceptions include the preservation of the Merck natural products libraries at two locations (the Fundación Medina, in Spain, and the Natural Products Discovery Institute, in the US), and the Pfizer microbial culture collection donated to the Scripps Research Institute.

10. Diversity-oriented synthesis

The objective of diversity-oriented synthesis is to take its structural cues from nature. As a daughter of combinatorial chemistry, it seeks to meld parallel synthesis with chiral synthesis technologies. Thus, natural product scaffolds are designated as privileged structures and then functionalized by parallel synthesis (Balthaser et al., 2011, Smith and Kim, 2011, O’ Connor et al., 2012, Garcia et al., 2016). While attractive in concept, for the same reasons that natural products are desirable for drug leads (Section 3, above), the efficiency of this strategy remains to be determined. After all, functionalization choices are still made by the chemist.

11. Specialist journals in natural product science

Shown on Table 2 is a listing of specialist journals popular among those engaged in natural products research.

Table 2.

Specialist journals in natural products research.

Journal Name	Publisher
Economic Botany	NY Botanical Garden Press
Fitoterapia	Elsevier
Journal of Antibiotics	Japan Antibiotics Research Association/Nature
Journal of Chemical Ecology	Springer
Journal of Ethnopharmacology	Elsevier
Journal of Natural Products	American Society of Pharmacognosy/American Chemical Society
Marine Drugs	Molecular Diversity Preservation International
Molecules	Molecular Diversity Preservation International
Natural Product Reports	Royal Society of Chemistry
Natural Product Research	Taylor & Francis
Pharmaceutical Biology	Taylor & Francis
Phytochemical Analysis	Wiley Interscience
Phytochemistry	Elsevier
Phytochemistry Reviews	Springer
Phytomedicine	Elsevier
Phytotherapy Research	Wiley Interscience
Planta Medica	Thieme
Toxicon	Elsevier