Empirical Analysis of Drug Targets for Nervous System Disorders

Abstract

The discovery and development of drugs to treat diseases of the nervous system remains challenging. There is a higher attrition rate in the clinical stage for nervous system experimental drugs compared to other disease areas. In the preclinical stage, additional challenges arise from the considerable effort required to find molecules that penetrate the blood–brain barrier (BBB) coupled with the poor predictive value of many preclinical models of nervous system diseases. In the era of target-based drug discovery, the critical first step of drug discovery projects is the selection of a therapeutic target which is largely driven by its presumed pathogenic involvement. For nervous system diseases, however, the feasibility of identifying potent molecules within the stringent range of molecular properties necessary for BBB penetration should represent another important factor in target selection. To address the latter, the present review analyzes the distribution of human protein targets of FDA-approved drugs for nervous system disorders and compares it with drugs for other disease areas. We observed a substantial difference in the distribution of therapeutic targets across the two clusters. We expanded on this finding by analyzing the physicochemical properties of nervous and non-nervous system drugs in each target class by using the central nervous system multiparameter optimization (CNS MPO) algorithm. These data may serve as useful guidance in making more informed decisions when selecting therapeutic targets for nervous system disorders.

Graphical Abstract

INTRODUCTION

Drug discovery and development is a cost-intensive and time-consuming process with a high failure rate. The challenges are particularly evident in the area of nervous system disorders. The clinical success rate of neurological drugs is significantly lower than other disease areas when comparing the probability of launch (FDA approval) from entry into phase I clinical trials with a success rate of only 3% over the 2010–2017 time period.¹ The high attrition rate can be attributed to several factors including the complexity and heterogeneity of nervous system disorders, as well as the scarcity of validated and easily accessible biomarkers to monitor therapeutic response.² The difficulty in predicting clinical efficacy with nervous system disease preclinical models has also contributed to the low success rate in phase II clinical trials, where efficacy is first evaluated in patients. These challenges prompted some pharmaceutical companies to deprioritize and/or restructure efforts on developing new therapeutic agents treating nervous system disorders.³

Although no statistics are available to verify, it is plausible that the preclinical phase of neurotherapeutic development also suffers from higher failure rates than other therapeutic areas. One major challenge unique to the nervous system disorder space is identifying compounds that can be distributed to the central nervous system (CNS), such as the brain, which is protected by the BBB. Under these circumstances, the probability of success is largely influenced by the druggability of the chosen therapeutic target. In other words, how likely is it that potent molecules can be found for a target within the narrow range of physicochemical properties required for optimal distribution to the CNS?

To address this question, we assessed the distribution of human protein targets of FDA-approved drugs treating nervous system disorders and non-nervous system disorders, respectively. Such an empirical analysis could serve as guidance in making an informed decision on nervous system therapeutic target selection.

METHODS

Small molecule drugs (<900 g/mol) approved by the FDA up until 2022 were subjected to analysis. The analysis of molecular targets provided by Santos et al. served as the basis for target class assignment.⁴ Since their analysis covered FDA-approved drugs up until June 2015, targets for drugs approved between June 2015 and December 2022 were sourced from DrugBank (https://go.drugbank.com/). Antibiotics, antivirals, and antifungals that target pathogen biological molecules were excluded from the analysis since the primary focus of this review article is to decipher the druggability of human therapeutic targets. Drugs targeting non-proteins (e.g., DNA, RNA, etc.) were also excluded because of the current inadequate understanding of these targets from a druggability standpoint. Pharmaceutical mixtures were included with separate entries for their individual components if those components were not previously approved as monotherapies and act on targets directly associated with the disease mechanism. For example, an oral combination of decitabine/cedazuridine was approved in 2020.⁵ Decitabine was, however, previously approved as a monotherapy in 2006, so it was not duplicated on the list with a 2020 entry. Furthermore, cedazuridine was not included in the analysis because it acts on cytidine deaminase, which is unrelated to the disease mechanism, and rather serves to enhance decitabine exposure by inhibiting its metabolism.⁶ Prodrugs were included in our analysis and assigned to the target of their active form.

Drugs were classified by their targets into the major classes of GPCR, enzyme, nuclear receptor, ion channel, or transporter. The five major classes collectively cover 96% of the analyzed FDA-approved drugs. Subclasses were assigned within the major classes of enzyme, (oxidoreductases, kinases, proteases, transferases, hydrolases, phosphodiesterases, lyases, isomerases, ntpases, and other), ion channel (LGIC and VGIC), and GPCR (7TM1, 7TM2, and 7TM3). Drugs that do not target one of the five major classes were categorized as “other”. The online database UniProt (https://www.uniprot.org/) aided in target classification.

Some drugs have multiple therapeutic targets. If the targets were in the same major class and subclass, the drug was only counted once in our analysis. For example, amisulpride, a drug used to treat schizophrenia, was only counted once even though it binds to both 5-HT7 and D2 receptors both of which belong to the major class of GPCR and the subclass of 7TM1. Drugs with multiple therapeutic targets from different major classes were counted more than once in our analysis. For example, vortioxetine was counted twice because it targets both 5-HT1A receptor (classed as a GPCR) and 5-HT3A receptor (classed as an ion channel).

When a single target could be classified into two or more different classes, we focused on the specific function being modulated by the drug. For example, proton pumps, also known as H⁺/K⁺ ATPases possess both enzymatic and transporter activities. FDA-approved proton pump inhibitors covalently inhibit the ATPase activity⁷ and, therefore, we assigned their target as nucleoside-triphosphatase which falls under the major class of enzyme.

The primary objective of the present work is to analyze the differences between nervous system drug targets and non-nervous system drug targets which necessitated an objective strategy to sort FDA-approved drugs into the two clusters. We chose to use Anatomical Therapeutic Chemical codes (ATC codes) assigned to clinically used drugs by the WHO Collaborating Centre for Drug Statistics Methodology (https://www.whocc.no/). The first letter of the ATC code indicates the anatomical main group. There are 14 groups in total. For example, C represents the cardiovascular system, while N represents the nervous system. We categorized FDA-approved drugs with an ATC code that begins with N (e.g., N05CM19 coding suvorexant) as drugs for nervous system disorders. Conversely, those with an ATC code that begins with one of the 13 remaining letters (e.g., C03XA02 coding conivaptan) were classified as drugs for non-nervous system disorders. ATC code assignments for drugs approved by the FDA up until June 2015 were incorporated from the previous work reported by Santos et al.⁴ while codes for the additional drugs approved through 2022 were obtained from the WHO Collaborating Centre for Drug Statistics Methodology ATC index. Drugs often have multiple ATC codes. If a single drug had multiple ATC codes, with at least one corresponding to the nervous system and another corresponding to the non-nervous system, it was included in both clusters. It should be noted that not all drugs with a nervous system ATC code act on targets in the nervous system. Similarly, some drugs with a non-nervous system ATC code may be distributed to the nervous system. This analysis, however, did not consider these cases to maintain a clearly defined and objective benchmark for drug categorization.

For the majority of FDA-approved small molecule drugs, the six physicochemical parameters required to calculate CNS Multiparameter Optimization (MPO) scores⁸ were obtained from ChEMBL (https://www.ebi.ac.uk/chembl/). Missing values were supplemented using ChemAxon. For compounds devoid of a basic atom (basic pK_a not applicable, N/A), T0 values of 1 were given for pK_a.

The resulting list was assembled into an Excel spreadsheet (provided as supporting information) for the subsequent analysis.

RESULTS AND DISCUSSION

We first analyzed the distribution of FDA-approved drugs (up to December 2022) by target class (Figure 1A). GPCR- and enzyme-targeting drugs are the two most common classes, each representing about 30% of FDA-approved drugs. Ion channel-targeting and nuclear receptor-targeting drugs account for 15% and 13%, respectively. Transporter-targeting drugs represent the smallest share (8%) among the major classes that we defined. The distribution is in good agreement with a previous analysis for drugs approved up until June 2015.⁴ However, when drugs approved by the FDA between July 2015 and December 2022 were extracted for the same analysis (Figure 1B), enzyme-targeting drugs represented a substantially higher percentage (48%) as compared to 29% of all FDA-approved drugs up to December 2022. Upon closer examination of enzyme-targeting drugs approved between July 2015 and December 2022, we found that two-thirds of them were kinase inhibitors, representing 32% of all drugs approved during this period. In contrast, kinase-targeting drugs represented only 3% of small molecule drugs approved up until June 2015.⁴ Therefore, the growing number of kinase inhibitors approved by the FDA between July 2015 and December 2022 appears to be the main reason for the increased share of enzyme-targeting drugs during this period.

Figure 1.

Distribution of all FDA-approved small molecule drugs by major target class. (A) All drugs approved through December 2022. (B) Drugs approved from July 2015 to December 2022.

Figure 2 presents the distribution of FDA-approved drugs by target class for the treatment of non-nervous system disorders (Figure 2A) and nervous system disorders (Figure 2B), respectively. Some notable differences were observed. First, while nuclear receptor-targeting drugs account for 17% of drugs treating non-nervous system disorders, none of them were approved for the treatment of nervous system disorders. Second, enzyme-targeting drugs account for a substantially lower proportion of drugs treating nervous system disorders compared to that of drugs treating non-nervous system disorders. Third, the proportions of ion channel- and transporter-targeting drugs are substantially larger in the cluster of drugs treating nervous system disorders. Each of these findings is discussed in more depth below.

Figure 2.

Distribution of FDA-approved drugs by target class for the treatment of (A) non-nervous system disorders and (B) nervous system disorders.

Despite the broad therapeutic impact of nuclear receptor-targeting drugs in non-nervous system disorders, none have been approved for nervous system disorders. This, by no means, reflects the insignificance of nuclear receptors in the nervous system. Indeed, nuclear receptors are considered highly attractive therapeutic targets in neurodegenerative diseases.^9–10 The failure to translate efficacy in animal models to patients can be cited as one reason for the lack of success in nuclear receptor-targeting drugs treating nervous system disorders.¹¹ In addition, many nuclear receptors are expressed in both the nervous and non-nervous systems, which raises concerns over side effects elicited by modulation of peripheral receptors. These challenges are, however, relevant to many therapeutic targets in the nervous system. Another possibility is our inability to identify potent and BBB-penetrant nuclear receptor ligands. To further investigate this possibility, we analyzed the physicochemical properties of FDA-approved drugs to calculate CNS MPO score,⁸ a well-recognized algorithm to predict the probability of success in the CNS therapeutic area using a scale from 0 to 6 (the greater number, the more desirable). The CNS MPO algorithm could be used to identify compounds with not only desirable CNS permeability but also low P-glycoprotein (P-gp) efflux, good metabolic stability, and favorable safety profile.⁸ Since its implementation, the CNS MPO algorithm has contributed to a higher percentage of compounds nominated for clinical development at Pfizer.¹² Table 1 lists the median CNS MPO scores for FDA-approved drugs clustered into different major target classes and further separated into those treating nervous system and non-nervous system disorders. Nuclear receptor-targeting drugs have a median CNS MPO score of 4.1 on a scale of 6, which is barely above the minimum target score of 4.0 recommended for increased probability of success in the clinic. Indeed, the median CNS MPO score of 4.1 is the lowest among drugs targeting the five major classes, underscoring the ability of the CNS MPO algorithm to capture molecular features undesirable for CNS therapeutic agents. In an attempt to identify specific molecular properties contributing to the low median CNS MPO scores for nuclear receptor-targeting drugs, we analyzed median transformed values (T0) for the six properties, namely, ClogP, ClogD (at pH 7.4), molecular weight (MW), topological polar surface area (TPSA), hydrogen bond donor count (HBD), and pK_a. Each of these six T0 values is expressed on a scale from 0 to 1 (T0 = 1.0, most desirable) and accounts for one-sixth of the CNS MPO scores. As shown in Figure 3, the lower median CNS MPO score of nuclear receptor-targeting drugs is mainly attributed to their higher lipophilicity (ClogD, ClogP) and molecular weight. It should be noted that T0 values for these physicochemical properties were lower than not only those of drugs treating nervous system disorders but also those of drugs treating non-nervous system disorders. Since these three parameters tend to shift in concert, whether simultaneous reduction of molecular weight and lipophilicity can be achieved without sacrificing the potency is the key question in targeting nuclear receptors in the nervous system.

Table 1.

Median CNS MPO Scores of FDA-Approved Drugs by Target Class

target class	all disease areas	nervous system disorders	non-nervous system disorders
all FDA-approved drugs	4.4	4.8	4.2
GPCR	4.3	4.4	4.3
enzyme	4.4	5.1	4.2
ion channel	5.0	5.1	4.9
nuclear receptor	4.1		4.1
transporter	4.2	4.4	4.2
other	3.8	4.2	3.4

Figure 3.

(A). Median T0 values for the six physicochemical parameters that constitute CNS MPO scores of all nervous system (blue), all non-nervous system (yellow), nuclear receptor-targeting (gray), and kinase-targeting (orange) drugs. The labels above the bars represent median physicochemical property values (not applicable to pK_a of nuclear receptor-targeting drugs because more than 50% of them have no basic centers). (B) Median CNS MPO scores for the four drug clusters.

The substantially smaller proportion of drugs targeting enzymes in the area of nervous system disorders compared to that of non-nervous system disorders does not suggest a scarcity of enzymes as promising nervous system therapeutic targets. In fact, enzymes play a critical role in nervous system pathology, and the development of enzyme-targeting drugs remains an important therapeutic approach to many neurological disease areas.^13–16 For example, kinases in the nervous system are largely untapped, despite the evidence of their pathogenic roles in various neurological conditions.¹⁷ To date, no kinase-targeting drugs have been approved for nervous system disorders while over 70 kinase inhibitors have been approved for non-nervous system disorders from 2001 to 2022. These findings prompted us to analyze the physicochemical properties of FDA-approved kinase inhibitors (Figure 3A). As seen with nuclear-receptor targeting drugs, kinase inhibitors show low median T0 values for ClogP, ClogD, and molecular weight. Furthermore, the median number of hydrogen bond donors is 2, serving as another negative factor in the CNS MPO algorithm. Collectively, FDA-approved kinase inhibitors have a median CNS MPO score of 3.5, the lowest among the four clusters analyzed (Figure 3B). These findings are in agreement with the prior analyses reported by Shi et al.¹⁸ The trend in physicochemical properties of kinase inhibitors highlights general challenges inherent to the design of enzyme-targeting drugs for the treatment of nervous system disorders. Enzyme inhibitors tend to have higher molecular weight and more hydrogen bond donors in order to bind to a large active site pocket often surrounded by polar and/or charged residues essential for the catalytic process, thereby departing from desirable molecular properties for CNS drugs.

Despite these challenges, medicinal chemists were not discouraged from pursuing kinase inhibitors for the treatment of nervous system disorders. For example, DNL-201 is a first-in-class CNS-penetrant leucine-rich repeat kinase 2 (LRRK2) inhibitor that advanced into clinical trials in patients with Parkinson’s disease.¹⁹ DNL-201 has a CNS MPO score of 5.5, substantially higher than the median score of FDA-approved kinase inhibitors, demonstrating the feasibility of designing potent kinase inhibitors within a narrow range of physicochemical properties allowed for neurotherapeutic agents. Outside the area of kinase inhibitors, the evolution of β-site amyloid-precursor-protein-cleaving enzyme (BACE-1) inhibitors represents another example of sustained and strategic efforts leading to successful discovery of brain-penetrant enzyme inhibitors.^20–21

In contrast to nuclear receptor- and enzyme-targeting drugs, substantially larger percentages of ion channel-targeting (34%) and transporter-targeting (15%) drugs were found in the nervous system disorder space compared to the non-nervous system disorder space (9% and 5%, respectively). We propose two possible explanations. First, both target classes play a more prominent role in the nervous system, where the process of neurotransmission relies heavily on the coordinated action of ion channels and transporters. Second, ion channels and transporters have been known to be effectively modulated by molecules with restricted physicochemical properties,²² making them more feasible nervous system targets. With that being said, many drugs in these clusters share targets. For instance, more than 40% of ion channel-targeting drugs act on the GABA_A receptor, and nearly 90% of transporter-targeting drugs inhibit dopamine, serotonin and/or norepinephrine transporters. Thus, the greater percentages represented by these drugs in the nervous system disorder space do not categorically reflect the greater diversity of therapeutic targets within the ion channel and transporter clusters. On a related note, the origin of benzodiazepines, which account for the majority of GABA_A receptor-targeting drugs, can be traced back to the serendipitous discovery of chlordiazepoxide in 1955 without the knowledge of its target at the time.²³ Interestingly, many antipsychotics discovered around the same time were later found to act on monoamine GPCRs in the nervous system.²⁴ The fact that nearly two-thirds of GPCR-targeting drugs in the nervous system act on monoamine receptors signifies the important impact of phenotypic drug discovery approaches to uncovering druggable neurotherapeutic targets.²⁵

As shown in Table 1, the median CNS MPO score of 4.1 for nuclear receptor-targeting drugs is only slightly below the median scores of 4.4 for GPCR- and transporter-targeting drugs treating nervous system disorders, questioning the broader utility of the CNS MPO algorithm. This prompted us to analyze the six median transformed values (T0) constituting the CNS MPO score for these clusters. As shown in Figure 4A, drugs targeting GPCRs and transporters in the nervous system had higher T0 values for ClogP, ClogD, and molecular weight, which are the three parameters that nuclear receptor-targeting drugs performed poorly as discussed earlier. On the other hand, nuclear receptor-targeting drugs scored better on TPSA (compared to transporter-targeting drugs) and pK_a (compared to both), resulting in similar CNS MPO scores for the three clusters (Figure 4B). These findings indicate that the first three T0 values may have more impact on CNS druglikeness than the 50% of CNS MPO score that they collectively represent. Thus, close attention should be paid to not only the CNS MPO scores but also their compositions when using the CNS MPO algorithm to guide molecular design.

Figure 4.

(A). Median T0 values for the six physicochemical parameters that constitute CNS MPO scores of nervous system GPCR-targeting drugs (purple), nervous system transporter-targeting drugs (green), and non-nervous system nuclear receptor-targeting drugs (gray). The labels above the bars represent median physicochemical property values (not applicable to pK_a of nuclear receptor-targeting drugs because more than 50% of them have no basic centers). (B) Median CNS MPO scores for the three drug clusters.

CONCLUSION

Comparison of human protein target distributions between FDA-approved drugs for the treatment of nervous system disorders and non-nervous system disorders highlights that some target classes (ion channels and transporters) are more amenable to neurotherapeutic development than other target classes (enzymes and nuclear receptors). Analysis of physicochemical properties in the context of CNS MPO suggests that the observed trend can be, at least partially, attributed to the different degrees of CNS druggability across various target classes. To this end, the feasibility of identifying potent molecules within the stringent range of molecular properties necessary for CNS penetration should serve as another important factor in the process of therapeutic target selection, which has been historically driven by the pathological significance of the targets.

Our analytical results, however, should not be prematurely reduced into general rules, such as avoiding enzymes and nuclear receptors as neurotherapeutic targets. As exemplified by the development of LRRK2 inhibitors, brain-penetrant drugs can be found against seemingly undruggable CNS therapeutic targets with the right molecular design strategy. As eloquently stated by Dr. Eric Lander, druggability should not be considered as a definition of nature but rather as a description of the current state of our abilities.²⁶ Nevertheless, the key is to involve both biologists and medicinal chemists at an early stage of the target selection process to factor in both biological significance and druggability so that more informed decisions can be made toward selecting and/or prioritizing nervous system therapeutic targets.