THE NATURE OF DRUG TARGET
A prerequisite for counting the number of targets is defining what a target is. Indeed, this is the crucial, most difficult and also most arbitrary part of the present approach. For the purpose of this paper, we consider a target to be a molecular structure (chemically definable by at least a molecular mass) that will undergo a specific interaction with chemicals that we call drugs because they are administered to treat or diagnose a disease. The interaction has a connection with the clinical effect(s).
This definition implies several constraints. First, the medicinal goal excludes pharmacological and biochemical tools from the present approach. Second, a major constraint is a lack of technique. Life, including disease, is dynamic, but as we do not yet directly observe the interactions of drugs and targets, and only partly notice the subsequent biochemical 'ripples' they produce; we are generally limited to 'still life' (for example, X-ray crystal structures) and to treating targets as static objects. In the case of G-protein-coupled receptors (GPCRs), the pharmaceutically most useful class of receptors, a re-organization of the protein after drug binding was derived from biochemical data, but such approaches are still in their infancy.
For most drugs, several if not many targets were identified. Consequently, we had to decide for every drug substance or drug class which target(s) to include in our list. For this, we relied on the existence of literature data that showed some connection between the interaction of the drug with the biochemical structure of the target and the clinical effect(s) (not side effects). A chemical with a certain reactivity or binding property is used as a drug because of its clinical effects, but it should be stressed that it can be challenging to prove that a certain molecular interaction is indeed the one triggering the effect(s). In this respect, knockout mice are proving increasingly useful. For example, a lack of effect of a drug in mice lacking a particular target can provide strong support that the effects of the drug are mediated by that target .We therefore considered the construction of knockout animals that lack the target, with pertinent observation of effects, strong proof or disproof for a certain mechanism of action. In the case of receptors, we regarded the availability and testing of both agonists and antagonists (and/or inverse agonists) proof for a mechanism. In the case of enzyme inhibitors (for example, cyclooxygenase inhibitors), molecular interactions and effects of structurally unrelated substances that are largely identical were considered proof of the mechanism. In cases where a drug inter-action on the biochemical level was found, but the biochemical pathway was not yet known to be connected with the observed drug effect, the target was not counted. For antipsychotic drugs in particular, a plethora of target receptors and receptor subtypes are known .
The dynamics of drug effects. It would ultimately be desirable to move away from a static target definition, but this is hindered mainly by our inability to gauge the inter-action of the aforementioned 'ripples' — in other words, the actual pharmacodynamics of drugs. All drugs somehow interfere with signal transduction, receptor signalling and biochemical equilibria. For many drugs we know, and for most we suspect, that they interact with more than one target. So, there will be simultaneous changes in several biochemical signals, and there will be feedback reactions of the pathways disturbed. In most cases, the net result will not be linearly deducible from single effects. For drug combinations, this is even more complicated. A mechanism-based simulation of pharmacodynamic drug–drug interactions was published recently, highlighting the complexity of interaction analyses for biological systems. Awareness is also increasing of the nonlinear correlation of molecular interactions and clinical effects. For example, the importance of receptor–receptor interactions (receptor mosaics) was recently summarized for GPCRs, resulting in the hypothesis that cooperativity is important for the decoding of signals, including drug signals Another paper reported dopamine fluctuations after administration of cocaine, followed by a gradual increase in steady-state dopamine concentration.Indeed, the dynamics of the response are what really matters, but are difficult to assess experimentally. Further examples of dynamic (process) mechanisms of drug action include non-covalent modifications of the active centre (for example, acetylation of bacterial transpeptidases by -amino butyric acid) receptors); drugs that require the receptor to be in a certain state for binding and inhibition (for example, 'trapping' of K+ channels by methanesulphoanilide anti-arrhythmic agents); drugs that exert their effect indirectly and require a functional background (for example, the catechol-O-methyl transferase inhibitor entacapone, the effect of which is due to the accumulation of non-metabolized dopamine); anti-infectives that require the target organism to be in an active, growing state (for example
The macro- and micro-world of targets. So, for estimations of the total number of targets, a clinically relevant 'target' might consist not of a single biochemical entity, but the simultaneous interference of a number of receptors (pathways, enzymes and so on). Only this will give a net clinical effect that might be considered beneficial. As yet, we are unable to count 'targets' in this sense ('macro-targets'), and it is only by chance that most of the current in vitro screening techniques will identify drugs that work through such targets.
Greater knowledge of how drugs interact with the body (mechanisms of action, drug–target interactions) has led to a reduction of established drug doses and inspired the development of newer, highly specific drug substances with a known mechanism of action. However, a preoccupation with the molecular details has sometimes resulted in a tendency to focus only on this one aspect of the drug effects. For example, cumulative evidence now suggests that the proven influence of certain psychopharmaceuticals on neuro-transmitter metabolism has little to do with the treatment of schizophrenia or the effectiveness of the drug for this indication
With diseases such as type 1 diabetes, for example, the molecule insulin is indeed all that is needed to produce a cure, although we cannot imitate its regulated secretion. With diseases such as psychoses, for example, antipsychotic drugs might not correct nor even interfere with the aspect of the human constitution that is actually deranged, and with such drugs molecular determinism might be counterproductive to the use and development of therapeutic approaches. It is thought that rather than chemically providing a 'cure', these drugs make the patient more responsive to a therapy that acts at a different level. Reflections on molecular targets are very important because drugs are molecules, but it is important not to be too simplistic.
Returning to the key question, what do we count as a target? In the search for molecular reaction partners of drug substances, we will have to be content with losing sight of some of the net biochemical and especially clinical effects of the drug's action. A target definition derived from the net effect rather than the direct chemical interaction will require input from systems biology, a nascent research field that promises to significantly affect the drug discovery process11. At the other end of the scale of precision, we can define some targets very precisely on the molecular level: for example, we can say that dihydropyridines block the CaV1.2a splicing variant in heart muscle cells of L-type high-voltage activated calcium channels. This is an example of a 'micro-target'. It does make sense to define it because a subtype or even splicing variant selectivity could alter the effectiveness of calcium channel blockers. We could further differentiate between genetic, transcriptional, post-transcriptional or age differences between individuals, and again this will make sense in some cases. But for a target count, a line needs to be drawn somewhere, otherwise the number of individual patients that receive a drug could be counted and equated with the number of known targets. In summary, we will count neither macro- nor micro-targets, but something in between — admittedly a somewhat arbitrary distinction.
Classification of current drugs
There are a number of possible ways to classify drug substances (active pharmaceutical ingredients). From the end of the nine-teenth century until the 1970s, drug substances were classified in the same way as other chemical entities: by the nature of their primary elements, functional moieties or organic substance class. Recently, the idea of classifying drug substances strictly according to their chemical constitution or structure has been revived. Numerous databases now attempt to gather and organize information on existing or potential drug substances according to their chemical structure and diversity. The objective is to create substance 'libraries' that contain pertinent information about possible ligands for new targets (for example, an enzyme or receptor) of clinical interest and, more importantly, to understand the systematics of molecular recognition (ligand–receptor).
At present, the most commonly used classification system for drug substances is the ATC system16 (see WHO Collaborating Centre for Drug Statistics Methodology, Further information). It categorizes drug substances at different levels: anatomy, therapeutic properties and chemical properties. We recently proposed an alternative classification system.
Classification of drug substances according to targets. The term 'mechanism of action' itself implies a classification according to the dynamics of drug substance effects at the molecular level, the dynamics of these interactions are only speculative models at present, and so mechanism of action can currently only be used to describe static (micro)targets,
The actual depth of detail used to define the target is primarily dependent on the amount of knowledge available about the target and its interactions with a drug. If the target structure has already been determined, it could still be that the molecular effect of the drug cannot be fully described by the interactions with one target protein alone. For example, antibacterial oxazolidinones interact with 23S-rRNA, tRNA and two polypeptides, ultimately leading to inhibition of protein synthesis. In this case, a description of the mechanism of action that only includes interactions with the 23S-rRNA target would be too narrowly defined. In particular, in situations in which the dynamic actions of the drug substance stimulate, or inhibit, a biological process, it is necessary to move away from the descriptions of single proteins, receptors and so on and to view the entire signal chain as the target. Indeed, it has been pointed out by Swinney in an article on this topic that "two components are important to the mechanism of action... The first component is the initial mass-action-dependent interaction... The second component requires a coupled biochemical event to create a transition away from mass-action equilibrium" and "drug mechanisms that create transitions to a non-equilibrium state will be more efficient". This consideration again stresses that dynamics are essential for effective drug action and, as discussed above, indicates that an effective drug target comprises a biochemical system rather than a single molecule.
A further criterion needed for the full categorization of drug substances according to their target is the anatomical localization of the target. This is essential for a differentiation between substances with the same biochemical target, but a different organ specificity (for example, nifedipine and verapamil are both L-type calcium channel inhibitors; the former interacts primarily with vascular calcium channels and the latter with cardiac calcium channels). However, in the tables, we chose not to include this criterion as it would have made the list more cumbersome.
The number of drug targets
The most prominent target families included hydrolases in the enzyme family, GPCRs in the receptor family and voltage-gated Ca2+ channels in the ion-channel family. The usefulness of a target family in this count is probably a consequence of its commonness, the format of assays (with recent binding-affinity based assays having contributed little as yet), and the nature of the diseases that affect the developed world.
Many successful drugs have emerged from the simplistic 'one drug, one target, one disease' approach that continues to dominate pharmaceutical thinking, and we have generally used this approach when counting targets here. The recent progress made in our understanding of biochemical pathways and their interaction with drugs is impressive. However, it may be that 'the more you know, the harder it gets'. It is not the final number of targets we counted that is the most important aspect of this Perspective; rather, we stress how considerations about what to count can help us gauge the scope and limitations of our understanding of the molecular reaction partners of active pharmaceutical ingredients. Targets are highly sophisticated, delicate regulatory pathways and feedback loops but, at present, we are still mainly designing drugs that can single out and, as we tellingly say, 'hit' certain biochemical units — the simple definable, identifiable targets as described here. This is not as much as we might have hoped for, but in keeping with the saying of one of the earliest medical practitioners, Hippocrates: "Life is short, and art long; the crisis fleeting; experience perilous, and decision difficult." Humility remains important in medical and pharmaceutical sciences and practice.
