Considerations on Pharmaceutical Research and Biological Complexity: Toward System‐Based Drug Discovery

The search for remedies capable of relieving the suffering caused by disease, trauma, or more complex syndromes is an activity that has always accompanied the history of humankind. The enormous progress and the successes achieved in the therapeutic field especially over the course of the twentieth century are undoubtedly important, but they should not make us forget how many diseases still lack a cure. This highlights the full current relevance of drug discovery, a challenging undertaking, whose first key goal is the identification of new bioactive ingredients (usually small molecules, technically New Chemical Entities: NCEs) potentially capable of counteracting disease states. While the pharmacological approach is only one of the available ways of treating disease, and even taking into account the various new therapeutic modalities identified in recent years, it should be stressed that the ability to maintain an adequate flow of natural or synthetic NCEs entering the preclinical and clinical development pipeline represents a worldwide health necessity. Fortunately, considerable progress has been made in the design and synthesis of new chemical compounds, such that today's computational and synthetic chemistry technologies make it possible to obtain practically any molecule targeted at a given therapeutic goal. At this point, it may also be useful to broaden our view of the foundational horizon of drug research as a scientific discipline, attempting to place it within a conceptual framework that takes into account some current theoretical and experimental orientations in biology and beyond. Precisely, I am referring to the category of complexity, a salient feature of biological systems whose significance and scope are increasingly being explored also in medicine. Indeed, this same complexity-oriented view is already shaping new avenues of drug discovery, in the literature as well as in actual practice, which makes it worthwhile to reflect on how we arrived here and on what theoretical grounds this shift may rest. In this article, I will trace the evolution of the drug discovery paradigm, starting from a brief account of its development over the course of the twentieth century, and arriving at an updated definition consistent with certain current directions in scientific research.

To simplify as much as possible the historical development of drug research, and to limit ourselves to the last century, we might say that the strategies for identifying new agents with therapeutic potential have essentially been two: a) the testing in vivo or in vitro of compounds without hypothesizing a mechanism of action a priori (a top-down approach), and b) the selection or production, based on a mechanistic hypothesis, of molecules to be tested in in vitro systems such as enzymes in solution, cell cultures, or isolated organs (a bottom-up approach). Following the top-down approach (often called "phenotypic drug discovery"), in the past, the discovery of new drugs was based mainly on clinical investigations or chance observations almost always followed by chemical modifications of prototypes (see, for example, the penicillins), whereas more recently so-called compound libraries have been widely exploited. These are more or less extensive collections of molecules systematically screened at high speed ("high throughput screening") until a promising compound is found. In the bottom-up strategy, by contrast, drug candidates are designed starting from hypotheses derived from knowledge of the disease to be treated, in particular of its pathogenesis at the molecular level. In this sense, it is at least from the 1960s onward that we speak of "rational drug design". This approach was strongly aided at first by advances in molecular and structural biology, and later by the application of statistical and/or molecular-modeling computational methods. In the early period, when the most important and frequent advances in biology consisted in the ability to identify and express in cell lines the genes associated with particular aspects of disease, the goal of the research was identified as the ability to specifically target the expression products of these genes with suitable molecules, so as to activate or inhibit a cellular response believed to play a crucial role in the progression or manifestations of the disease. This gave rise to the paradigm, inspired by Ehrlich's "magic bullet"¹, referred to as "one disease – one target – one drug": to treat a disease it is necessary to hit a molecular target (generally a protein) with a compound that binds specifically to it. This way of conducting drug research became widespread and made it possible—and still does—to identify a great many NCEs with extremely high specificity for individual targets, although not many of these molecules, when studied in the clinical setting, turned out to be capable of acting in the expected specific way. In the late 1990s, however, just as this truly reductionist approach reached its peak of diffusion and acceptance, a somewhat different path emerged, less exclusive in identifying the target to be hit, and people began to look for "multi-target" compounds: molecules capable of binding to more than one target, with the aim of striking the biological processes associated with the disease not just at a single point but at several steps along the pathogenetic way. And so, from the 2000s onward, the paradigm became "one disease – multiple targets – one drug", supported by a great deal of experimental evidence contradicting the reductionist hypothesis of the single target.

Even from the broad outline sketched above, one can glimpse an evolution of the general drug discovery strategy toward increasing complexity. And for a schematic inspection, the two paradigms stated above can be depicted in a graph-based representation as follows:

where D, t, and d stand for Disease, target, and drug, respectively. In the A) case, a single molecular element is hypothesized to link drug and disease, namely the target itself, be it an enzyme that is inhibited, a receptor that is activated or blocked, an ion channel that is opened or closed, and so on. In the B) case, the drug's action is assumed to be mediated by its interaction with more than one of the elements listed above, a representation that, even if only slightly, appears more realistic than the previous one. In practice, this shift from single-target pharmacology to “polypharmacology” has been supported by increasingly in-depth studies of drug mechanisms of action, which have made it possible to obtain a level of detail unimaginable just a few years earlier. Indeed, drugs designed and initially found to act through interaction with a single target later turned out to be multi-target molecules: a case in point is the well-known anticancer drug imatinib, a compound conceived and developed to selectively inhibit the BCR-ABL kinase enzyme responsible for uncontrolled cell proliferation in chronic myeloid leukemia, which was later found to be a poly-kinase inhibitor.

If we consider the two graphs shown above, we can say that B) seems to represent a drug–disease relationship more consistent with experimental knowledge than that represented by A), owing to the fact that B) takes into account several—ideally all—of the possible protein targets associated with the disease and capable of being hit by the drug. To take the reasoning a step further, we might recall a well-established finding of modern proteomic studies, namely that within a cell, proteins do not exist or act in isolation but interact with one another, possibly in various ways, the most frequent of which is direct contact. At the cellular level, then, proteins give rise to a network in which each protein is connected to one or more others, forming the so-called "protein–protein interactome" that can be represented as a graph in which individual proteins are the nodes and the connections between them are the edges. Using dedicated computational tools, it is possible to model and visualize this interactome, which, depending on the starting data and the construction criteria, can contain even more than 20k proteins and 1M connections. Remarkably, a recent observation made by some researchers is that the proteins involved in the pathogenesis and manifestations of a disease are almost always directly connected, or at least very close to one another within the interactome. This has led to the concept of the “disease module”, which is the sub-network of the interactome formed by the proteins associated with the disease together with their neighboring and connecting proteins, as tentatively shown by the scheme below:

The disease module is a central concept in the innovative approach to medicine known as "network medicine"² and can be used to continue the discussion of the paradigms introduced above with graphs A) and B), which, recall, represent the drug–disease relationship mediated by a single target (A) or several targets (B). We can now postulate that a drug does not act on the disease simply through a single or several target(s), but plausibly through some proteins of the disease module, and we can represent this version of the paradigm as follows:

where Dm denotes the Disease module. This hypothesis, among other things, has given rise to numerous drug-repositioning studies, following the discovery that the modules of different diseases can partially overlap within the interactome; consequently, a drug originally developed to interact with some protein targets associated with a disease can be used to treat another condition that shares targets in common with the first one³.

Although paradigm C) may serve as a valid starting point for outlining the theoretical context of the search for new drugs, we note that it is based entirely on what we might consider a protein-level representation of a living system. It would be like saying that the protein–protein interactome stands for the whole individual (the patient) and the disease module for the part of the organism affected by the disease: the drug acts on the disease by hitting targets included in the disease module. But in this description, reductionism reappears. Indeed, the action of drugs on a living organism does not involve proteins alone. For example, proteins are the product of the expression of the genes that encode them, that is, "below" the protein level there is a genetic level with which a drug can directly interfere. On the other hand, proteins may take part in various biochemical pathways that enable the processes necessary for life and for cell replication, and drugs affect these pathways as well; here, then, is another level of description of the cellular system, "above" the protein level, at which we can study drug action. Continuing along this line of reasoning, we might further consider that cells interact with one another and may organize themselves into different types of aggregates depending on their characteristics (think of neuronal cells or, in a pathological setting, cancer cells) to give rise to tissues, and once again we are faced with another viewpoint on pharmacological action: drugs act on cells. And going further still, we come to the organs of our body, which carry out the various vital functions for the whole organism and with respect to which drug action can be considered ("drugs for the heart" or "drugs that act on the central nervous system"). What has been briefly introduced in these last lines does nothing more than apply to pharmacology the approach to describing living organisms known as "systems biology", which aims to study such organisms comprehensively (holistically⁴), thereby overcoming both the reductionism typical of traditional biology and the discretization of biological domains that leads to their being investigated separately. According to this way of viewing things, now increasingly accepted within the scientific community, the organization of living beings can be conceptualized as a sequence of intra- and inter-connected hierarchical levels, from molecules to cells and up to tissues and organs, up to individuals, possibly considered in their relationships with one another and with the environment: these levels, taken together and in their sequence, provide a global representation of the complexity of living systems. The scheme below gives an illustration of this concept:

De Domenico, M. (2020, November 17). Multilayer Networks Illustrated. https://doi.org/10.17605/OSF.IO/GY53K

It should be noted that considering the complexity of living systems makes it possible to avoid a purely deterministic interpretation of biological phenomena, giving weight instead to the so-called "emergent properties" that arise from the horizontal and vertical interactions among the various entities making up the whole system. In this sense, diseases can be investigated not as simple consequences of single physiological aberrations (e.g., “a gene mutation causes the disease”), but as emergent properties, that is, as phenomena resulting from possible evolutions of a biological system along a path different from the one considered "physiological", yet equally compatible with the complexity of the system itself. For example, the mutation of a gene changes the network of relationships around it, and this, in connection with other organic and environmental factors, induces a different (“pathological”) configuration of the system⁵.

On the basis of the above, returning now to the development of the paradigms, we can broaden the range of pharmaceutical targets by considering the individual as a whole, across all levels of organization. It will therefore be a matter of rewriting graph C) in a form that takes the systems approach into account, as follows:

where S denotes the living organism (System) made up of the various entities that characterize its nature, interconnected according to the hierarchy of levels mentioned above. The paradigm can thus be redefined as "one disease – one system – one drug". This rewriting of the drug discovery paradigm (“System-Based Drug Discovery”) ultimately places this activity within the broader, more general context of systems biology, situating it in a multidisciplinary domain grounded in Complexity Science⁶, the current discipline that studies complex systems and their behavior.

Having introduced this new way of considering the drug–disease relationship, it remains to be seen whether and how paradigm D) can be useful in assisting and guiding the identification of new molecular entities⁷ in drug research. In this regard, it should first be noted that the overall system described by the simple graph D) lends itself very well to being modeled in silico through the use of network science methods. Indeed, the real system denoted by S can be depicted as a network, in which the nodes represent the various biological entities that make it up (genes, proteins, cells, organs, etc.) and the connections make explicit the known relationships among them. To construct the network, one can use data collected in the numerous publicly accessible databases, which are in turn assembled on the basis of experimental studies in genomics, transcriptomics, proteomics, metabolomics, and microbiomics, as well as, moving across all levels, in molecular, cellular, and functional biology, anatomy, and physiology. Furthermore, to complete the overall network, one can also include the entities D and d of D), considering data concerning diseases (D), such as pathological manifestations and symptoms (phenotypes), definitions, and even environmental effects related to the onset and progression of diseases following exposure to external factors. Finally, the network can also comprise known drugs (d) and all information regarding their use, like their known targets, mechanisms of action, side effects and toxicities, as well as interactions with other drugs. With this data it is possible to build multiscale networks that provide an overall picture of the disease–organism–drug system, encompassing hundreds of thousands of nodes and millions of relationships among them. Various types of these network models exist, for example "multilayer networks" or "knowledge graphs", and network science provides us with the tools to extract from these representations information that cannot be detected through investigations based on less comprehensive and detailed models, such as those often obtained in vitro or even in vivo. To mention an example in this context, one challenge that has begun to be tackled is that of building "digital twins" of patients that are not merely an in silico replica of a person's pathophysiological characteristics, but that integrate the network models described above into a personalized predictive tool⁸. This should make it possible to develop a therapeutic intervention tailored to the patient, e.g., by identifying the specific targets to be hit by drugs and predicting the effects of those drugs on the entire organism.

It also follows from what has been said above that, in order to be useful for drug discovery, such network models must be analyzed, and simulations and predictions must be carried out on their basis. This field of activity is well known to people involved in drug design, since the use of simulative and predictive computational techniques has long been part of the skills required for this type of research. In particular, in the case of network models of complex systems, one can draw on applications from graph theory both to analyze their topology and dynamics and to predict the consequences of changes in the system's configuration, such as the silencing of a gene within a gene-regulatory network or the evolution over time of an intracellular metabolic network.

Recalling the question posed at the beginning of this section, that is, the applicability of the system-based paradigm in drug discovery, we may ask how the more or less complex and elaborate models can assist in the search for new molecular entities to be considered for therapeutic use. Undoubtedly, whatever type of model is used, it will provide a knowledge context on the basis of which the requirements that new therapeutic molecules must possess can be defined; then, virtual or real molecules possessing such required characteristics can be generated⁹ or extracted from the available databases. This way of operating can be applied to the search for small or large molecules (biomolecules), of natural or synthetic origin, and can be regarded as an extension of the classic approaches to the identification of new compounds, one that includes high-level modeling of the biological context, and possibly of preclinical and clinical issues, like pharmacokinetics, patentability, patient stratification, clinical outcome prediction, and so on. It is worth noting that all the drug discovery process so far described can be—and actually has been—partly or even completely automated through AI-informed platforms that carry out the whole sequence of activities, from target identification to simulated clinical trials of the proposed drug candidates¹⁰.

Needless to say, however, we must remember that the new molecular entities identified by computational models will then have to be produced, tested in the laboratory, and evaluated clinically before finally reaching, if all goes well, the patient.

I hope that what has been described above can convey an idea of the contemporary conceptual and theoretical landscape within which to situate the search for new drugs, understood as molecular entities acting on a complex system such as the human organism. Recalling the traditional approaches mentioned above, I would like to stress that the current perspective does not exclude or replace the design methodologies and techniques developed to date; rather, it includes and combines them within a more advanced paradigm. The two top-down and bottom-up strategies are no longer viewed as mutually exclusive, but are instead integrated into a scenario that considers the organism as a system in its entirety, while at the same time allowing for the specificity of interventions on targeted objectives. What makes the difference compared with the recent past is, first and foremost, network thinking, the conceptual shift that lets us virtualize and investigate the complex system represented by our organism as a web of relationships rather than a collection of isolated parts. This way of thinking finds its most concrete expression in network modeling: an extremely powerful and flexible tool, open to the continuous acquisition of new information and capable of revealing global system characteristics not evident in disaggregated data. Finally, from a conceptual standpoint, this paradigm rests on the recognition of the complex nature of living organisms and on the now well-established awareness that the functioning of these systems is based on the relationships established among their various components, at various levels. Today we have the tools needed to investigate this complexity, and the time has come to harness this knowledge to improve our ability to combat the diseases that still afflict humankind.