How are drugs invented? Part II — Modern drug discovery

Source: FDA

In the first part, I talked about the origins of drug discovery and some of the methods used in today’s process. This second part is devoted to an overview of the modern techniques of drug discovery and why the process is so expensive. I will use the terms drug discovery and drug invention more or less interchangeably because the process has elements of both.

Drug discovery overview

You might recall from Part I of this article that drug screening and structure modification played key roles in the early scientific drug discovery. That is still the case today. But before we get into the actual process, why does a pharmaceutical company decide to work on a particular disease? There are many reasons that need to be carefully balanced together since drug development is a very risky business. It takes up to about 15 years to develop a new drug with a success rate of ~ 10% and a cost of up to more than $ 2.5 billion per approved drug(1). Some of those reasons are:

• Unmet medical need: this is perhaps the stronger motivator since there are many diseases that still do not have an adequate treatment like Alzheimer’s, heart failure, pulmonary hypertension, some autoimmune disorders, chronic pain, etc. just to name a few. Most rare diseases fall in this category too.
• New basic research: mainly coming from Academia, new research can suddenly open the possibility to treat a condition that was previously considered untreatable.
• Existing therapeutic areas: for established companies, it is usually preferred to work on a small number of therapeutic areas (like oncology, cardiovascular, immunology, etc.) since they have the experts in all areas, from research all the way to marketing and sales.
• Balanced portfolio: while this does not have any scientific or medical basis, a balanced portfolio is important for the survival of the company. Following the philosophy of not putting all the eggs in the same basket, companies will generally try to mix and match several projects: from risky novel targets but with potential for a high impact to low risk better versions of existing drugs (more potent, safer, less side effects, more compliance).
• Drug repurposing: using an old existing drug for new indications.

Let’s go back to the drug discovery process in the pharmaceutical industry. It consists of two broad steps: Research which produces a compound that should be suitable for human testing, and Development that refines the compound into a state that can be administered to humans, is actually tested in humans and is hopefully approved for sale. We can include a previous step of basic research where relationships about targets and diseases are usually discovered and that tends to happen in the Academic research world.

Targets and diseases

Preclinical strategies that are used to identify potential drug candidates include: target-based screening, phenotypic screening, modification of natural substances, and biologic-based approaches(2). Target-based and phenotypic are the most common. In the context of clinical medicine, phenotype is any observable characteristic or trait of a disease (morphology, development, biochemical or physiological properties, behavior, etc.) A biological target is “anything within a living organism to which some other entity (like an endogenous molecule or a drug) is directed and/or binds, resulting in a change in its behavior or function.” If there is experimental evidence that modulation of a target has an effect on the disease of interest (see for example Open Targets as a source of data), then the project will be the development of a molecule (small molecule or biologic molecule) that can produce such modulation. If there is no known target, then a phenotypic approach will be followed looking to develop a molecule that can correct the diseased phenotype. As we have seen in Part I, the concept of a target originated in part on Paul Ehrlich’s “magic bullet” idea, but there was no way to identify the target back then. That is why Ehrlich’s studies were all phenotypic based and they would continue to be for decades.

Drug Discovery workflow

Drug Discovery follows a more or less fixed workflow shown here for a target-based approach; the approximated times for completion are listed. In the case of a phenotypic approach, the first step is skipped.

1. Target discovery and validation: 1 - 2 years
2. Compound screening, hit identification, lead selection: 0.5 - 1 year
3. Lead optimization: 1 - 3 years
4. Preclinical drug development: 1 - 2 years
5. Clinical trials: 3 - 7 years (Phase I: <1 year, Phase II: 1 - 2 years, Phase III: 1 - 4 years)

Source: © Omar Stradella 2019

Target discovery and validation

Most targets or receptors are proteins and proteins are the products (expression) of genes. If someone can figure out what a gene does, then there is a good chance that they can figure out the role in a particular disease. A common method to do that is a gene knockout, which is a genetic technique that renders one (or more) of an organism genes inoperative. If a knockout eliminates or improves the disease, then it is a good candidate for a drug target. A full knockout sometimes can be too drastic and lead to bad consequences for the test organism, including death. An alternative is a knockdown where the expression of the gene is reduced instead of eliminated. The reduction can occur either through genetic modification or by treatment with a reagent such as a short DNA or RNA oligonucleotide that has a sequence complementary to either gene or an mRNA transcript. There are also knockin techniques that can be used to introduce specific mutations that can provide clues about the contribution of the gene to the disease(3).

One common classification of targets is by protein families(4):

• GPCRs (G protein-coupled receptors): a large protein family of receptors that detect molecules outside the cell and activate internal signal transduction pathways and, ultimately, cellular responses.
• Kinases: enzymes that catalyze the transfer of phosphate groups from high-energy, phosphate-donating molecules to specific substrates.
• Ion Channels: pore-forming cell membrane proteins that allow ions to pass through the channel pore.
• Nuclear Receptors: responsible for sensing hormones and certain other molecules and then regulate the expression of specific genes, thereby controlling the development, homeostasis, and metabolism of the organism.

Two properties characterize the interaction between a drug and a target: affinity and intrinsic efficacy. Affinity is the property of a drug that describes its ability to bind to a target and it is a constant unique for each drug-target pair, as it is dependent on both the structures of the drug and the target. Intrinsic efficacy is the drug property that describes the effect a drug has on the target activity that can lead to a change in cellular activity. Like affinity, intrinsic efficacy is a constant that is dependent on both the structures of the drug and the target and thus is unique for each drug-target pair. In general, drugs can behave as agonists or antagonists. Agonists are drugs with both affinity (they bind to the target) and intrinsic efficacy (they change target activity to produce a response). Antagonists have affinity but zero intrinsic efficacy; therefore they bind to the target but do not produce a response. By virtue of occupying a fraction of the target population, an antagonist reduces the probability of occupancy by an agonist which leads to a corresponding reduction in response. There are also inverse agonists that decrease the activity of an otherwise normally active target(5).

1536 well screening plate (Source: Fluotics) and Primary screening vs Secondary (confirmatory) assay results

Dose response curve (Source: Deranged Physiology)

Compound screening

Once a target or suitable phenotype has been identified, and experiment (called assay) is created to find chemical compounds that would have an effect on them. Until the 1980s, the number of compounds that could be screened by a single facility in a week was between 10 and 100. In the 1990s technology had progressed to the point where tens of thousands of compounds could be screened a week in what would be called High Throughput Screening (HTS) and, by 2005, 1 million compounds a day could be tested in ultra High Throughput Screening (uHTS). Testing tens of millions of compounds was now feasible and pharmaceutical companies build very large collections of chemical compounds mostly purchased from specialized chemical companies but also containing representative compounds synthesized in-house over the years. The idea is to test all these compounds for activity. Why test so many? The so-called “drug-like” chemical space has been estimated to be in the order of 10^60 structures(6), which is about the number of atoms in 1000 of our Sun, so even 10 million compounds represent an infinitesimal fraction of that number. The assay used in HTS and uHTS has to be simple and easy to run at such a scale. Each compound is dissolved to create a solution of a fixed concentration (typically 10 μM) and a tiny volume (few μL) is deposited in a well of a special plate (containing 1000s of wells) together with the components and reagents needed to generate some signal of activity that can be automatically read (like fluorescence). After the assay is run, an analysis of the data is necessary. The assay readout is scaled between 0% (no activity) and 100% (full activity) using data from control compounds if possible, sorted in descending order of activity, and an arbitrary cutoff is set to select a reasonable amount (few thousands) of active compounds or hits that can undergo further tests. The HTS/uHTS assays are by nature not very accurate and prone to false positives (compounds that seem active but they are not) and false negatives (compounds that do not seem active but they are) that require the use of confirmatory assays to figure out if the hits selected are really active. Confirmatory assays are run at several concentrations (2 to 10) of the compound with the idea that, if a compound is active, then increasing the concentration should increase the activity too, these are known as dose-response assays. The compounds that pass these tests are considered to be validated hits. We might still have hundreds or thousands of validated hits, but we can not pursue all of them and a small number (less than 5) has to be selected to become what are know as lead series heads. Several criteria are used to select lead series compounds:

• Potency: choose the most active compounds.
• Scaffold: choose compounds with a good scaffold. Useful compounds contain a central core or scaffold, usually made of one ring or a small number of fused rings, to which other chemical groups are attached. Thus, it is easy to make new compounds by adding and/or replacing groups attached to the scaffold.

Structure of naproxen showing the scaffold or core

Some features of a good scaffold:

• IP space: choose scaffolds that are not in existing patents.
• Structural diversity: choose scaffolds that “look” as different as possible from other scaffolds in the set, so that there will be some alternative options in case some of the other series fail.
• Synthetic feasibility: choose scaffolds that are relatively easy to make.
• Liabilities: exclude scaffolds that can be potentially toxic or quickly metabolized.
• Clusters: if there are several similar structures or compounds that share a common scaffold in the screen and that are all active, then choosing one from this group is a good option since it hints at a workable SAR (Structure Activity Relationship).
• Size: compounds are only going to get bigger, so start with a reasonably small scaffold.
• Selectivity: try to exclude compounds that are known to be active on other assays, unless the targets on those assays are related to the screen target.

Lead optimization

After the lead series have been chosen, the lead optimization can start. The idea is to systematically make changes to the lead compounds to optimize or improve several parameters at the same time. The most important is potency or activity. But potency alone is useless if the compound cannot reach its target, for example. The most important parameters fall under five categories known as ADMET, and there are standard assays to test them all:

· Absorption: How much of the compound is absorbed and how quickly? (bioavailability)

· Distribution: Where is the compound distributed within the body? What is the rate and extent of the distribution?

· Metabolism: How fast is the compound metabolized? What is the mechanism of action? What metabolite is formed and is it active or toxic?

· Elimination: How is the compound excreted and how quickly?

· Toxicity: Does this compound have a toxic effect on body systems or organs?

As mentioned above, compounds consist typically of a central core or scaffold surrounded by chemical groups (“decoration”). During lead optimization, the chemical groups around the scaffold are systematically changed one by one and the activity of the resultant compound is recorded. With this data, we can build a Structure-Activity Relationship (SAR) that allows us to make rough predictions of new compounds in the same series. The data can also be used in computer simulations (CADD: Computer Aided-Drug Design) that can provide more accurate estimates of activity, particularly if the 3D structure of the target is known. Potency is tested in in vitro assays that consist of more or less the target alone. But, since compounds usually have to get into cells to interact with the target, cellular potency assays are typically used as well. Ultimately, cellular potency matters more than in vitro potency. Eventually, when cellular potency and all ADMET parameters reach reasonable levels, it is time to start testing in animal models. Healthy animals (rats, mice, etc.) are used in pharmacokinetic (PK) assays to be able to predict how the drug is going the be absorbed, distributed, and excreted in humans when the time for clinical trials comes and also to monitor potential toxicity. Animals that can model the disease are used in pharmacodynamics (PD) assays to predict how the drug is going to affect the disease in humans.

Preclinical drug development

The goal in drug development is FDA approval of a new drug application (NDA) and prescribed use in the clinic. Once the lead optimization goals for in vivo potency, toxicity, PK, and PD are met by a few compounds, one is chosen as the development candidate. A typical preclinical development program consists of six major efforts(7):

• Synthesis scale-up: the chemical reactions that worked to produce the small amounts used in the lead optimization assays might not be the best, in terms of yields for example, to synthesize the larger quantities needed for clinical trials and later for large-scale manufacturing
• Formulation: the mixing of the candidate compound with other chemical ingredients to create the drug product (DP). The goal is to optimize the delivery of the compound according to the delivery method: oral, topical, injectable, etc.
• Analytical and bioanalytical methods development and validation: necessary to quantify both the amount of compound given to the patient and the amount of compound and metabolites present in biological samples.
• Metabolism and pharmacokinetics: PK parameters are used to estimate the dose and frequency of administration. Metabolism studies are useful to evaluate the potential for drug-drug interactions, possible inhibition of enzymes involved in the metabolism of drugs, and to generate drug metabolite profiles.
• Toxicology: both safety and genetic toxicology and possibly safety pharmacology.
• Good manufacturing practice (GMP): manufacture and documentation of drug product for use in clinical trials.

Clinical trials

The boundary between preclinical development and clinical trials is sharply defined by the filing of an Investigational New Drug (IND) application, which is required prior to the initiation of the clinical trials and summarizes the activities and findings during development to present to the US Food and Drug Administration (FDA). When preclinical studies show that the drug is basically safe, work moves to experiments in human volunteers through a phased series of clinical trials(8):

• Phase I clinical trials provide initial safety data to support further testing with larger samples. As the focus of these studies is primarily safety rather than efficacy, the study subjects are frequently a small number of healthy volunteers (tens of participants).
• Phase II clinical trials seek further safety data and preliminary evidence in support of biological effect. A slightly larger sample of subjects (hundreds) is administered the treatment at a dose or doses that were preliminary judged safe in the Phase I studies.
• Phase III clinical trials are large (hundreds to thousands of participants) confirmatory studies meant to establish an acceptable benefit/safety profile in order to gain regulatory approval for a precisely defined indication. Evidence from these studies that strongly supports the proposed indication will generally lead to the approval of the drug.
• Phase IV clinical trials are postmarketing trials that are meant to evaluate rare but serious effects that cannot be assessed in the smaller Phase III studies

The design of every clinical trial starts with a primary clinical research question. It is possible to ask more than one question, but the results of a trial cannot be used to answer a question that was included in the design. Studies are designed in a way that allows for the comparison of two (or more) groups of patients, including a treatment group and a group receiving a placebo or treatment with a reference drug. Randomization and blinding are essential to get definitive answers to the clinical research question. Randomization means that patients are allocated to the treatment and control groups at random, thus reducing possible bias. The efficacy of treatments can depend on whether or not the patient knows that he belongs to the treatment or control group. To eliminate this major difficulty and maintain comparable groups throughout the study, it is necessary to implement a blind procedure: either simple blind (generally the patient does not know) or double-blind (the patient and the physician do not know). Finally, the correct application of statistical methods is key to obtain valid answers.

After the Phase III trial, the efficacy and the safety data of the new drug are known. The IND application file will be prepared and submitted to the registration authorities of the different countries. This file should establish the major criteria: pharmaceutical quality, efficacy and safety. It should also provide answers to all possible questions concerning the choice of patients to be treated and the modalities of prescription.

A risky business

Drug discovery is a risky and expensive process that can fail at many points along it. These are some of the reasons:

• The number of theoretically possible “drug-like” is enormous, about 10^60, in comparison to the number of available compounds of about 107 million (registered in Chemical Abstract Services as of January 2020) or 52 orders of magnitude less.
• The high-throughput screening hit rate is low and at most about 1%, but typically lower. Of millions of compounds tested we can expect a few tens of thousands hits at most. But, it happens that sometimes no hits can be found. Costs vary wildly but a ballpark of $ 0.25/compound would put the price of a 5 million compound HTS at about $ 1.25 million.
• Even with tens of thousands of hits, we can only “gamble” on a few (no more than about five) to be selected as lead series, which eventually will have to be reduced to only one. The reason is that in a typical lead series, by the time that a development candidate is selected, a few thousand compounds would have had to be synthesized and tested which is time-consuming and expensive. The probability of a project to end up producing a development candidate is low (few percentage points), the reasons could range from difficulties to improve potency to failure to achieve any of the ADMET goals, to lack of in vivo efficacy. There is a tendency to park or cancel projects if there is doubt of success for any of the reasons stated above. At this point, before clinical trials, not too much money has been spent (about 25% of the total for a successful approval).
• Once a compound enters clinical trials, the probability of approval is only about 12%. It can fail in Phase I (40%) because of toxicity in humans that was not detected in animals. It can fail in Phase II (64%) usually for lack of efficacy. It can fail in Phase III (38%) also usually for lack of efficacy even though it seemed efficacious in Phase II, or because of toxicity that was not seen in the smaller Phase II trial. Finally, it can fail in Phase IV when the drug is already in the market, again for toxicity reasons in most cases.

Estimated phase transition probability and overall clinical approval success rates. Source: Journal of Health Economics (9)

• The capitalized cost from screening to approval is on average about $ 1.8 billion (10) with more than half (63%) spent in clinical trials, and the whole process can take up to 15 years. The later a project is killed, the more money is wasted.

Source: Nature Reviews Drug Discovery (10)

• Only novel compounds must be synthesized during the lead optimization phase, otherwise no patents would be granted. At some point, those compounds will have to be submitted to the patent offices. The current policy is that the first to file is the one that gets the patent, which means that after a compound is synthesized/tested and before submission, there is a window of opportunity for another company to submit the same compound and get the patent. One would think that submitting as soon as possible would be a good strategy, but since it can take up to 12 years from the beginning of lead optimization to approval and the patents last only for 20 years from filing, that can reduce the usable patent life to less than half.

References

1. Plenge RM. Disciplined approach to drug discovery and early development. Sci Transl Med. 2016;8(349):349ps15–349ps15. doi:10.1126/scitranslmed.aaf2608
2. Swinney DC, Anthony J. How were new medicines discovered? Nat Rev Drug Discov. 2011;10(7):507–519. doi:10.1038/nrd3480
3. Doyle A, McGarry MP, Lee NA, Lee JJ. The Construction of Transgenic and Gene Knockout/Knockin Mouse Models of Human Disease. Transgenic Res. 2012;21(2):327–349. doi:10.1007/s11248–011–9537–3
4. Lin Y, Mehta S, Küçük-McGinty H, et al. Drug target ontology to classify and integrate drug discovery data. J Biomed Semant. 2017;8(1):50. doi:10.1186/s13326–017–0161-x
5. Berg KA, Clarke WP. Making Sense of Pharmacology: Inverse Agonism and Functional Selectivity. Int J Neuropsychopharmacol. 2018;21(10):962–977. doi:10.1093/ijnp/pyy071
6. Reymond J-L. The Chemical Space Project. Acc Chem Res. 2015;48(3):722–730. doi:10.1021/ar500432k
7. Steinmetz KL, Spack EG. The basics of preclinical drug development for neurodegenerative disease indications. BMC Neurol. 2009;9(1):S2. doi:10.1186/1471–2377–9-S1-S2
8. National Research Council (U.S.), National Research Council (U.S.), National Academies Press (U.S.), eds. The Prevention and Treatment of Missing Data in Clinical Trials. Washington, D.C: National Academies Press; 2010.
9. DiMasi JA, Grabowski HG, Hansen RW. Innovation in the pharmaceutical industry: New estimates of R&D costs. J Health Econ. 2016;47:20–33. doi:10.1016/j.jhealeco.2016.01.012
10. Paul SM, Mytelka DS, Dunwiddie CT, et al. How to improve R&D productivity: the pharmaceutical industry’s grand challenge. Nat Rev Drug Discov. 2010;9(3):203–214. doi:10.1038/nrd3078