Precision medicine review: rare driver mutations and their biophysical classification

How can biophysical principles help precision medicine identify rare driver mutations? A major tenet of pragmatic approaches to precision oncology and pharmacology is that driver mutations are very frequent. However, frequency is a statistical attribute, not a mechanistic one. Rare mutations can also act through the same mechanism, and as we discuss below, “latent driver” mutations may also follow the same route, with “helper” mutations. Here, we review how biophysics provides mechanistic guidelines that extend precision medicine. We outline principles and strategies, especially focusing on mutations that drive cancer. Biophysics has contributed profoundly to deciphering biological processes. However, driven by data science, precision medicine has skirted some of its major tenets. Data science embodies genomics, tissue- and cell-specific expression levels, making it capable of defining genome- and systems-wide molecular disease signatures. It classifies cancer driver genes/mutations and affected pathways, and its associated protein structural data guide drug discovery. Biophysics complements data science. It considers structures and their heterogeneous ensembles, explains how mutational variants can signal through distinct pathways, and how allo-network drugs can be harnessed. Biophysics clarifies how one mutation—frequent or rare—can affect multiple phenotypic traits by populating conformations that favor interactions with other network modules. It also suggests how to identify such mutations and their signaling consequences. Biophysics offers principles and strategies that can help precision medicine push the boundaries to transform our insight into biological processes and the practice of personalized medicine. By contrast, “phenotypic drug discovery,” which capitalizes on physiological cellular conditions and first-in-class drug discovery, may not capture the proper molecular variant. This is because variants of the same protein can express more than one phenotype, and a phenotype can be encoded by several variants.


Introduction: precision pharmacology faces challenges
Drug development has been facing challenges. Recent statistics quoted 44% of the discontinued phase III drugs as being due to inadequate efficacy, and 24% safety issues (Mills 2016). Several reasons were proposed as drivers of failure (Grignolo and Pretorius 2016): inadequate basic science, suboptimal dose selection, insufficient assessment of the standard of care and of disease area landscape, flawed study design, flawed data collection/analysis, and study operations problems. Recent publications argued that even though improvements in medicinal chemistry should have led to higher reliability and reproducibility, that is not the case due to irreproducibility of the results (Begley and Ellis 2012;Ioannidis 2005;Perrin 2014;Prinz et al. 2011) and higher failure rates in clinical trials (DiMasi 1995;Hay et al. 2014) as well as higher R&D costs (Booth and Zemmel 2004;Munos 2009;Scannell et al. 2012;Scannell and Bosley 2016).
The advent of precision medicine aims to reset the pharmacology button (Cheng et al. 2019). The NIH-funded thrust Bto catalogue and discover major cancer-causing genome alterations in large cohorts… through large-scale genome sequencing and integrated multi-dimensional analyses^ (Tomczak et al. 2015), coupled with individual scientist-inspired research and pharmaceutical companies' efforts, has propelled the drug discovery frontiers with innovative, data-and concept-driven approaches. The vast amount of generated data along with advanced, more powerful computational strategies for its analysis and interpretation promise to transform diagnosis and cure. Precision medicine abstracts genome signatures and considers cell, tissue, and clinical observations, recently made even more powerful by high-throughput timeseries analyses that permit following cancer cell evolution (Cheng et al. 2019). The recent TCGA (The Cancer Genome Atlas) collection tells what appears to be a success story (Welcome to the Pan-Cancer Atlas 2016). Its three categories, embracing patterns in cells-of-origin (Hoadley et al. 2018), oncogenic processes (Ding et al. 2018), and signaling pathways (Sanchez-Vega et al. 2018), illustrate depth and breadth. Among the key emerging questions looms one of the most important to human health: how to leverage this information in drug discovery. This challenge has led the TCGA teams to focus on patterns of cancer vulnerabilities that will help in new combination therapies. Among these are individual and cooccurring actionable alterations in prominent cancer pathways including those involved in MYC, RAS, ubiquitin, DNA damage repair, splicing, and metabolism (Ge et al. 2018;Knijnenburg et al. 2018;Peng et al. 2018;Schaub et al. 2018;Way et al. 2018). The teams are large and multidisciplinary. They include experimental and computational biologists, basic sciences, and the clinics. The methods are diverse and comprehensive and span disciplines. They include analysis of genomes, transcriptomes, and proteomes; they also encompass analysis of protein structures to more reliably identify driver mutations, commonly defined as statistically frequent mutations that drive cancer.
Nonetheless, biophysical principles have largely been skirted (Nussinov and Wolynes 2014). Biophysics has been called Bthe bridging science.^Biophysicists seek to explain complex biological phenomena. Modern technologies produce huge amounts of data. To arrive at conclusions and make predictions, biophysics considers the fundamental underpinnings of the constituents of cells and tissues, their dynamic environments, and the communication between the molecular entities. Even though to date biophysics has been applied increasingly to the biological sciences, this has not been the case in the emergent branch of precision medicine, despite its ability to help decipher enigmatic questions, such as how changes in the DNA of healthy cells can trigger their transformation into cancer cells and how exactly pathogens can trigger cancer and neurodegenerative diseases, such as Parkinson .
Here, we offer some principles and guidelines for extending precision medicine through biophysics. Key among these principles is that protein structures exist as conformational ensembles which can be expressed as phenotypic traits Alhadeff et al. 2018;Cukier 2018;Frauenfelder et al. 1991;Gunasekaran et al. 2004;Jang et al. 2016a, b;Jenkins et al. 2018;Kumar et al. 2000;Mickert and Gorris 2018;Naganathan 2018;Nguemaha et al. 2018;Nussinov 2016;Nussinov et al. 2017;Nussinov et al. 2016Nussinov et al. , 2018aQiao et al. 2018;Tsai and Nussinov 2018). We consider how a mutation can affect phenotypic traits by linking to different network modules and, especially, how such conformational effects may extend and empower innovative drug discovery concepts. Notably, this applies to frequent and rare mutations.
Successful identification of drug candidates with either single cell or several cell types, as compared with targeted discovery, has reinvigorated phenotypic drug discovery (Giuliani et al. 2018;Heilker et al. 2018;Vaidya et al. 2018). At first glance, the re-emergence of the phenotypic drug discovery discipline is anathema to genome-based precision drug discovery, where a panoply of mutations, or genomic aberrations, can be expressed by similar disease phenotypes at the cellular or organism level. However, phenotypic drug discovery can produce first-in-class drugs and assess a target in its physiological cell environment. As such, it can complement precision medicine. Nonetheless, it is unable to reliably target the underlying mutational origin while minimizing toxicity.

Nomenclature/definitions
Here, we clarify some of the terminology used below: Orthosteric site The functional site, e.g., active sites for enzymes, or protein-protein binding sites.
Allosteric site A site away from the orthosteric site but whose perturbation by binding will affect the conformation at the orthosteric site.

Conformational distributions
The distributions of distinct conformations of the ensemble in the free energy landscape, which display their relative populations (i.e., corresponding free energy levels).
BOR,^incremental Bgraded,^BAND^all-or-none logic gate operations For simplicity, we explain these logic gates with kinase examples (Bradshaw 2010). In an BOR^logic gate switch, one event is sufficient for complete enzyme activation (e.g., Syk kinase is an BOR-gate switch^since either phosphorylation OR ITAM binding is sufficient to fully activate it). In a Bgraded^switch, there is an incremental increase in kinase activity with an increasing number of activating stimuli. Unlike a Bgraded^switch, an Ball-or-none^switch has only two possible responses: either inactive or fully active (Bradshaw et al. 2003;Lisman and Zhabotinsky 2001;Ninfa and Mayo 2004). Thus, an Ball-or-none^switch can be either an BOR^gate or an BAND^gate. In an BOR^gate, a single stimulus is sufficient for full activation; however, an BANDĝ ate requires multiple stimuli (Prehoda and Lim 2002); Tec kinase activation is an BAND^gate switch (Bradshaw 2010). The type of gate-switching mechanism is decided by the kinase features. In Tec kinases, both phosphorylation of the activation loop and the interaction with the SH2 linker are required.

How biophysics can empower concepts and approaches in precision medicine
With the same phenomena, biophysicists and biologists ask different questions (Bialek 2011). To biophysicists, mechanisms can only be fully understood at the detailed structural level. Pathway diagrams are informative and important, but insufficient. Biophysicists venture into microscopic scales. They seek to understand why evolution has selected a certain presumably optimal solution to a problem that is essential in the life of the organism, and how it is reached and tunedwith the adage that if they are able to understand, they may be able to predict the observable behavior and exploit it in design. To biophysicists, understanding implies comprehending the molecular behavior on the conformational level.
Precision, or personalized, medicine rests primarily on individual genomic sequences (Broes et al. 2018;Caskey 2018;Hampel et al. 2017;Hyman et al. 2017;Martin et al. 2015;Nakagawa and Fujita 2018;Poulos and Wong 2018;Prasad 2016;Senft et al. 2017;Tannock and Hickman 2016;Vargas and Harris 2016;Voest and Bernards 2016;Yu et al. 2017). The variability in human DNA is large, and the paramount question becomes how to identify among the many variations in the genomic sequences those which cause-or can causesusceptibility to the disease. Standard approaches involve a priori identification of driver mutations, i.e., those whose association with the disease is statistically significant (Fig. 1 provides an example for KRAS mutations). If the statistical significance of a mutation is high, its correlation with the disease is indicated. These driver mutations are then compiled and stored in databases. Subsequent observations of these mutations in an individual can testify to the origin of the illness (Raphael et al. 2014;Tsang et al. 2017).
However, statistics cannot explain how these mutations elicit the observed phenotype (Barone et al. 2017;Bilal et al. 2013;Blucher et al. 2017;Hogeweg 2011;Mathew et al. 2007;Payne 2012;Winter et al. 2012;Yakhini and Jurisica 2011). It also falls short when disease-promoting mutations are rare. Structural knowledge can help in both identifying mutations with low statistical significance and revealing the mechanism behind those statistically significant (Raphael 2012). The first step would involve mapping the mutations onto their corresponding macromolecular structures to identify their location and structural environment, and uncovering the interactions the protein is involved in, and which the mutations may affect. The mutations would be analyzed with respect to how they may affect these interactions and thereby rewire the cellular network. If located in highly flexible, or disordered, regions, the chances of their driver character are higher. Residue type may also help in providing clues: chances are higher if the mutated residue is aromatic or charged. Such regions are more likely to play a role in conformational changes with the altered interactions shifting the relative stabilities, typically of the inactive in favor of the active conformations. Mutations involving charged or aromatic residues are also likely to lead to a similar outcome. Eventually, the networks of protein-protein and protein-DNA (or RNA) interactions determine cell behavior. Importantly, for membrane-interacting proteins, the mutations should be considered within this framework as well. Hydrophobic or charged residues may be particularly involved in altered membrane-interaction tendencies. Interactions with signaling lipids such as eicosanoids, phosphoinositides, sphingolipids, and fatty acids control critical cellular processes, including cell proliferation, apoptosis, metabolism, and migration (Wymann and Schneiter 2008). Mechanistic consequences may also involve mutations of residues which can be post-translationally modified (PTMs). PTM effects can be either via an allosteric mechanism (Nussinov et al. 2012), or by abolishing or altering direct molecular interactions, thereby shifting an inactive protein state to an active one, or vice versa.
Altogether, mechanistic underpinnings that decide the mutational effects-whether statistically frequent or rare, at the tail of the distribution-can involve orthosteric or allosteric effects (Cheng and Nussinov 2018;Cheng et al. 2016;Collier and Ortiz 2013;Feher et al. 2014;Fetics et al. 2015;Liu and Nussinov 2008;Lu et al. 2015Lu et al. , 2016aMarcus and Mattos 2015;Morra et al. 2009;Munro et al. 2018;Nikolaev et al. 2018;Park et al. 2018;Risques and Kennedy 2018;Shen et al. 2017;Stout and Campbell 2018;Tehver et al. 2009;Nussinov 2014, 2017;Tuncbag et al. 2017;Waters and Der 2018;Xu et al. 2017;Zhan et al. 2016). Orthosteric mutations are at binding or active sites and can directly abolish (most frequent) or strengthen (much less frequent) interactions. The interactions can be between proteins, between the protein and nucleic acids, and between the protein and the membrane. Allosteric effects can rewire the cell network via short-or longer-range effects. Notably, all involve a shift in the conformational ensemble, including orthosteric mutations, since they too disturb the structure and elicit changes in the atomic interactions. Network rewiring reflects the changes in the conformational distributions whose outcome may favor different partners at shared binding sites (Boehr et al. 2009;del Sol et al. 2009). The shift of the ensemble can also explain the propagation of signaling, i.e., Ballo-networks^ (Nussinov et al. 2011;Szilagyi et al. 2013). As we have postulated earlier, Ballosteric signal propagation does not stop at the 'end' of a protein; but may be dynamically transmitted across the cell.T he signaling can be transmitted within a macromolecule, such as protein, DNA, or RNA, or across interacting molecules, enhancing or inhibiting specific interactions as the Ballosteric wave^propagates.
The challenge in the identification of mutations and the complexity involved in their mechanisms, which is expected to involve combinations of mutations within the same or different proteins whose conjoint effects may alter the network, can be strikingly observed from recent statistics (Alexov 2014). The 1000 Genomes pilot project (Clarke et al. 2012) indicated that most individuals carry 250 to 300 loss-offunction mutations in annotated genes with 50-100 mutations identified earlier as likely to be involved in inherited disorders (Genomes Project et al. 2010). Critically, the mutational load and the cell/tissue environment affect the severity of a disease. These may reflect the level of expression, protein and RNA concentrations, metabolites, presence/absence of microbiota that can mitigate or aggravate disease phenotypes, and more. The emerging complexity and ramifications pose an enormous challenge. In the absence of good biophysical models, exactly how to incorporate thousands of genetic variations of proteins is, however, unclear. While feasible, combining individual gene variability may lead to poor predictors of disease.
BLatent driver^mutations, statistically rare mutations, and the cellular network Although many mutations are inherited, most appear during normal cell growth Vogelstein et al. 2013;Welch et al. 2012). Most of these are considered Bpassengers,^as are most of those emerging during cancer evolution. By contrast, Bdriver mutations,^whether inherited or gained, acquire certain hallmarks of cancer, thereby conferring cancer cell advantage, including drug resistance ( Fig. 2) (Egeblad et al. 2010;Weinberg 2000, 2011;Merlo et al. 2006;Nussinov and Tsai 2013;Thiagalingam et al. 1996). While mutation rates vary (Bignell et al. 2010;Stratton et al. 2009;Youn and Simon 2011), a puzzling question has been why a primary cancer can develop over many years whereas drug resistance develops rapidly . Even though resistance mechanisms are diverse (e.g., Knievel et al. 2014;Nakamura et al. 2013;Poulikakos et al. 2010;Roskoski Jr. 2014), Blatent driver^mutations may be among them. Like a passenger mutation, on its own a latent driver mutation does not confer a cancer hallmark; however, together with a newly evolved mutation, it can express a cancer cell phenotype and cause drug resistance. A latent driver mutation can rewire the cellular network via BAND^all-ornone or incremental Bgraded^logic gate operations . Since, to date, Bactionable mutations^have only been those identified by their statistically significant frequencies (Cheng et al. 2019), latent driver mutations have been overlooked.
The Blatent driver^concept builds on a conformational view of macromolecules. According to this view, a macromolecule has a certain distribution of its ensemble, which reflects its environment. An environment can include concentration; sequence; post-translation modifications; interactors, such as proteins, nucleic acids, lipids, and small molecules; pH; temperature; ions; the crowded cellular environment; etc. A mutation may stabilize or destabilize the native state. The extent of the stabilization/destabilization with respect to the wild type may be minor, in which case the phenotypic change may not be observed; however, integrated with an emerging mutation whose contribution may also be minor, it may promote cell transformation or drug resistance. Together, the mutations can incrementally shift the protein ensembles toward populating a constitutively active (or inactive for a repressor protein) state , eliciting a cancer hallmark. Identifying such mutational partners is highly challenging.
Rare mutations can drive cancer development even if they are at the tail of the distribution. Even though not identified as drivers, the mechanisms of these mutations resemble those of mutations which are frequent and thus identified as drivers. Like statistically significant mutations, rare drivers can occur at the binding site, thus directly abolishing binding or allosterically shifting the protein ensemble from an inactive to an active state (or vice versa). Here, too, the challenge is their identification, pulling them out of the background passenger mutational clutter.
A protein (node) in a signaling pathway is typically modulated by upstream activation (or inhibition), with the effects (transmission of the signals or blocking them) propagating via interactors to pathways downstream, thereby affecting the network (Fig. 2). At the basic, conformational level, the modulation switches the population of the ensemble. In the absence of an upstream signal, passenger mutations cause no change in the relative (active versus inactive) populations, thus no downstream signaling shift. Even if the effects of several passenger mutations are integrated, there is still no change toward the distinct active conformation. This contrasts with driver and latent driver mutations. Both lead to alterations in the populations of the active vs inactive state-however to different extents. Latent drivers would shift the distributions in favor of the active (or inactive) state; however, the extent is insufficient for an observable phenotypic change. On the other hand, by definition, a driver mutation would result in a clear observable change. A rare driver is a driver, except that its statistics is lower. Heterogeneous protein ensembles can explain how mutations can elicit the observed phenotype; thus, the definitions of driver, latent driver, passenger, and rare driver hinge on their change of relative conformational populations between the active and inactive states with respect to that of the wild type and their statistics.
Taken together, we can view protein molecules in terms of their conformational ensembles; this contrasts with a network view which perceives proteins as nodes in networks. Considering research disciplines, chemists aim to obtain drugs that target actionable mutants; mathematicians and physicists consider the consequences of the inhibition on the cellular networks which they view as nodes and edges (the interactions connecting them). However, allosteric propagation across the protein, and via its interactors to the pathways, merges the two views and thereby can extend the database of drug targets Szilagyi et al. 2013). The propagation of allosteric signals can dynamically transcend their interfaces. However, we would still not expect the signal to travel a long way, through multiple pathway-associated proteins. For this to take place, there should be additional perturbations downstream. Binding to a protein may elicit distance-limited, specific, and network-neighboring changes (Antal et al. 2009;Ma and Nussinov 2009;Nussinov et al. 2011;Tsai et al. 2009).
Protein ensembles link physicochemical principles with the cellular network and protein behavior in vitro and in vivo. As we discuss below, they are also critical components in pharmacogenomics and pharmacogenetics.

Driver mutations, interactions, and cell signaling
Driver mutations can destabilize or stabilize proteins, shifting their states between active (or ON) or inactive (OFF) states. They can also rewire cell signaling. Identifying those mutations that shift states can help identify rare driver mutations. Key proteins, especially hubs, are epitomized by shared binding sites (Fig. 3a) (Gursoy et al. 2008;Tuncbag et al. 2009). Mutations that abolish an interaction of a partner protein binding at a shared site can favor binding of another protein at that site, thereby rewiring the network, which may now favor, e.g., pro-inflammatory signaling rather than apoptosis (Fig. 3b) (Guven-Maiorov et al. 2015). One example is the case of C27* mutation in the Fas-associated protein with death domain (FADD) protein in lung squamous carcinoma (0.21%, TCGA), which our data (unpublished) suggest can abolish its interaction with myeloid differentiation factor 88 (MyD88), permitting interleukin-1 receptor-associated kinase 4 (IRAK4, the NF-κB pathway) to occupy that MyD88 site (Fig. 3c). Similarly, TNF receptor-associated factor 6 (TRAF6) and TRAF3 occupy overlapping binding sites on Fig. 2 Mutations in the cellular network. In diseases, driver mutations affect the cellular network to alter phenotypic traits. A latent driver mutation, like a passenger mutation, does not confer a cancer cell phenotype. However, together with a newly evolved mutation, it can express a cancer cell phenotype Fig. 3 The protein interaction network and its mutational effects. a A classical representation of a cellular pathway (top left). In the cellular network, proteins (nodes) connect with each other through proteinprotein interactions (edges). Node traits can be highlighted by structural representation in the cellular pathway (lower left panel). Hub proteins provide shared binding sites for many binding partners. Edge traits can be characterized by specific combinations of the protein-protein interactions (right panel). The cartoons were inspired by a previous publication (Tuncbag et al. 2009). b An example of the hub protein MyD88 (PDB code: 3MOP) and its binding partners. FADD (PDB code: 2GF5) and IRAK4 (PDB code: 3MOP) partially overlap at the same binding site of MyD88, and TRAF6 (PDB code: 1LB5) and TRAF3 (PDB code: 1FLL) share the same binding site of MyD88 and completely overlap at the interface. These proteins compete to bind to MyD88, causing distinct downstream signaling pathways. c Mutations can alter edge traits. Examples are shown for R34H missense and C27* nonsense mutants of FADD. These mutations abolish the interaction of FADD with MyD88, rewiring cellular network. Protein complex constructions were inspired by a previous publication (Guven-Maiorov et al. 2015) MyD88; thus, MyD88 induces either pro-or antiinflammatory cytokines. A R505* TRAF3 mutation in head and neck tumors (4.7% frequency) abolishes its binding to MyD88, which promotes TRAF6 binding, possibly explaining how it contributes to cancer.
Mutations can shift not only the preferred binary interactions but also interactions in multimolecular assemblies (or Bsignalosomes^), and in this way alter cell signaling. This can alter the intramolecular crosstalk (Rowland et al. 2017) and influence drug activity. Structural data, coupled with efficient structural comparison algorithms, appropriate datasets, and filters can assist in getting an insight into such shared sites, and in obtaining concrete predictions of interactions consistent with experiment. Mutations that favor such a switch, even if rare, are likely to be drivers. Driver mutations can alter the regulation of the signalosome assembly/disassembly. Notably, when inhibited, the impacts on different tissues will vary, indicating that cellular context is crucial for targeting signaling networks.

Conformational heterogeneity can help bridge pleiotropy and pharmacogenetics
As an example for how biophysics can help link pharmacogenomics with precision medicine, consider the emerging principle that variation in any one pharmacogene may impact the clinical outcome for more than one drug (Oberg et al. 2016), since one gene often impacts more than one clinical outcome. This can be straightforwardly understood in terms of a shift in the ensemble, leading to an altered favored interacting partner. It can also reflect an altered pattern of posttranslational modifications in distinct tissues. Notably, according to the standard definition, unlike pharmacogenomics, pharmacogenetics considers how variation in a single gene influences the response to a single drug.
Thus, even though classically not described in this light, pleiotropy can reflect the presence of multiple states (Fig. 4). While definitions vary, a pleiotropic gene (or genetic variant) is often conceived as affecting multiple traits (Hodgkin 1998;Houle et al. 2010;Solovieff et al. 2013;Stearns 2010;Tyler et al. 2016;Wagner and Zhang 2011;Wang et al. 2010), as for example in the case of the small heat shock family α-crystallin B chain R120G mutation in mice which causes cataracts and cardiomyopathy (Andley et al. 2011). Pleiotropic genes that affect related traits cluster into relatively independent modules, with a mutation more likely to have smaller effects on unrelated traits in a different module than on related traits (Pendergrass and Ritchie 2015;Wagner and Zhang 2011). Comprehensive phenome-wide association study (PheWAS) of the National Health and Nutrition Examination Surveys (NHANES) observed potential pleiotropic genes with 13 SNPs associated with more than one phenotype (Hall et al. 2014), and a broad analysis of disease-causing proteins in UniProt revealed that 12% are pleiotropic, i.e., variants in the same protein cause more than one disease (Ittisoponpisan et al. 2017). Pleiotropy was also suggested as a reason why a drug that targets only one phenotype may fail, and thus, developing drugs that target only one phenotypic effect of a pleiotropic gene could be futile (Wagner and Zhang 2011). Classically, pleiotropy was captured in cases of clear, observable linkage between disease and a distinct mutation. Biophysics can extend such observations, offer their underlying mechanisms, and bridge the disciplines. A major tenet of biophysics is that a single mutation-frequent or rare-or another allosteric event can induce different phenotypic traits by populating conformations within the same or different network modules, and favor certain signaling pathways.

Phenotypic drug discovery
Phenotypic drug discovery has been dubbed Bclassical pharmacology,^which is the historical basis of drug discovery-in contrast to Breverse pharmacology,^which is considered as target-based drug discovery (Lage et al. 2018). The classical concept of Bphenotypic drug discovery^is thus not new. However, questions about the sustainability of the current target-based drug discovery process have triggered its renaissance which capitalizes on physiological cellular conditions and first-in-class drug discovery, and a recent perspective attempted to merge the phenotypic drug discovery concept with a target-based view (Heilker et al. 2018). The root of the phenotypic drug discovery concept is an anathema to the structure-based biophysics view, as well as to precision medicine whose basis is specific genomic sequences and mutations. Phenotypic drug discovery is now considered in databases, with assays that it employs grouped by animal disease model or phenotypic endpoint, accounting for assay data on protein targets or cell-or tissue-based systems (Hunter et al. 2018). Accounting for phenotypic consideration was also proposed to help computationally guided drug repurposing (Giuliani et al. 2018). Phenotypic screens were also proposed to be useful in chemogenomics (Jacoby and Brown 2018), as well as bridging the gap between phenotypic and biochemical assays (Denny 2018). Exploiting image-based profiling assays for assessing single-cell phenotypes has also been explored, including by machine learning (Scheeder et al. 2018). Powerful cell-based assay technologies that permit tighter linkage between in vitro and physio-pathological conditions and environments, including induced pluripotent stem cells, three-dimensional models, co-culture, and organ-on-achip systems, as well as advances in gene-editing technologies (Dorval et al. 2018), were put forward as the reasons for this renaissance and its promising future. Methods have also been developed to capture it (Joslin et al. 2018). These have triggered excitement for phenotypic drug discovery approaches, which can complement large-scale drug discovery efforts. It was further suggested that some diseases would benefit from a focus on a relevant condition or phenotype, rather than a specific target, via coupling network pharmacology with phenotypic screening (Sidders et al. 2018). A recent review provides an excellent overview of these and other small-molecule discovery technologies with examples (Markossian et al. 2018), and comparative analyses have been carried out as well (Riddy et al. 2018). A search of the literature yields a plethora of phenotypic drug discovery approaches, developments, and applications (e.g., Ayotte et al. 2018;Chatelain and Ioset 2018;Lagunin et al. 2018;Lane et al. 2018;Orellana et al. 2018;Ortiz et al. 2017), and finally, a cautionary note was also put forward (Copeland and Boriack-Sjodin 2018).
From the biophysical standpoint, protein (i.e., target) variability, even that induced by a single mutation or a change in the environment, may elicit altered phenotypes. The converse also holds: an observed phenotype, such as a disease, may reflect not only different proteins for example in the same signaling pathway but also variants of the same proteins. A phenotype may have diverse origins, and thus, even though the phenotypic approach may be helpful under certain circumstances in speeding up discovery, it may or may not capture the proper variant and accurately predict its source. The heated controversy between the supporters of phenotypic and targetfocused screening, which is at the core of precision medicine, as to which provides a more reliable path to successful drug development is expected. Even though attempts to reconcile the two have been advocated, the biophysics axiom argues that no matter how efficient phenotypic methods are, a phenotype may not equate to a single molecular shape. Phenotype-based discovery does not identify the pathway or the protein target, and since no structure is involved, it also does not permit ab initio design, or design optimization.

How biophysics applications can identify candidate cancer driver mutations
Examples of how biophysics applications can identify cancer driver mutations include molecular dynamics simulations of the native state and the mutant state: if the mutation promotes destabilization of the inactive state, it is a candidate for a driver mutation. Such destabilization can elicit a release of an autoinhibited state, as for example in PI3K (Nussinov et al. 2018b; unpublished data), Raf, and EGFR kinases. Simulations can also reveal mutations that can deactivate an essential enzymatic action which makes the enzyme switch back into the inactive state, as in the case of Ras oncogenic mutations, e.g., Q61L and G12 mutations (Lu et al. 2016a, b;Pantsar et al. 2018). Trapping different states experimentally, as well as simulations in the case of EGFR (Ruan and Kannan 2018;Verma et al. 2018), Raf (Rukhlenko et al. 2018), and other proteins, and integrating these with activity either directly or indirectly by observing their resulting actions downstream (e.g., on phosphorylation or transcription/translation) can provide first-hand confirmation, with the detailed structural data showing exactly how the mutation works. In another recent modeling strategy, exploration of structure, energetics, and dynamics of p53, PTEN, and SMAD4 tumor suppressor Fig. 4 Genotype and phenotype. a Monogenic phenotype occurs when a single gene expresses each trait, b while polygenic phenotype arises when multiple genes influence a single trait. c Pleiotropy occurs when seemingly unrelated phenotypic traits are expressed by a single gene. d In a new genotypephenotype paradigm, a pleiotropic gene can encode a distinct conformational ensemble in all states with certain populations of the specific phenotype traits that link to genotype. See text for further details proteins has also revealed that driver mutations in these proteins inactivate structurally stable residues that play a fundamental role in global propagation of dynamic fluctuations and mediating allosteric interaction network (Verkhivker 2019). p53 mutations were also shown to alter the interaction of the protein with the genome, promoting carcinogenesis, with a biophysical model for a gain-of-function mechanism that consolidates many literature observations (Stiewe and Haran 2018). Crystallography provides the key building block in such studies (Gomes et al. 2018), as also shown by TrkA (tropomyosin receptor kinase A, a nerve growth factor receptor) and broadly in activating mutations in tyrosine kinase domains in cancer (Artim et al. 2012).

Biophysics and precision medicine: principles and strategies
A fundamental biophysical principle is that molecules are not rigid rocks and should not be considered as single entities (Nussinov and Wolynes 2014). Statistics based on such a description may be inaccurate, especially when related to disease and treatment decisions. Molecules exist as dynamic ensembles, which integrate into and enhance the dynamic nature of interactome (Alexov 2014;Bohnenberger et al. 2011;Foerster et al. 2013). Molecules fluctuate, and these fluctuations are required for cell life . Fluctuation involves sampling different states, some of which have distinct functions. Fluctuations are critical for enzymes to work, for a receptor to switch between states, and for the chromatin to express the right protein at the right time (Rychkov et al. 2017). Overlooking these fluctuations does not necessarily affect the identification of targetable states, although it is required in the optimization of the drug design process. However, overlooking protein conformations and their dynamics may result in omissions in identification of protein mutations, and their consequences on the dynamic cell states, as well as accurate identification of the origin of an observed phenotypic trait. Upon mutational events, altered environment, covalent or non-covalent linkages or associations, and conformational heterogeneity may lead to different favored partners at shared binding sites and explain propagation of signaling. Biophysics can explain how variants can signal through distinct pathways, and how allo-network drugs can be harnessed. It clarifies how one mutation-frequent or rare-can affect multiple phenotypic traits by populating conformations that favor other network modules, and that such conformational effects may extend and empower innovative drug discovery concepts.
How to identify rare driver and latent driver mutations and how to construct the cellular allo-network? Mapping the mutations on the protein structure will indicate whether they adjoin functional or allosteric sites, and whether they exist in clusters, in which case they have a higher likelihood of being rare or latent drivers. Mutations that elicit a switch in partners at a shared binding site are also likely driver mutations, even if such mutations are rare, as those in highly flexible regions. Mutations that prevent execution of essential regulatory actions, such as abolishing ubiquitination of the Ras hypervariable region, thereby leading to extended membrane anchorage and mitogen-activated protein kinase (MAPK) signaling (Bigenzahn et al. 2018;Steklov et al. 2018), are also highly likely candidates. Molecular dynamics simulations might also observe changes in the distributions of the ensemble of the native vs the mutational state.
Paraphrasing the Biophysical Society (Biophysical Society 2018), one may ask what do the laws of physics, like those that define forces or energy, have to do with biology? As the Society befittingly explains, Bthese laws and concepts are essential to unraveling complex biological questions like how plants extract energy from sunlight and how changes in a protein's shape affect its function… Biophysics is critical to integrating systems biology, genomics, and proteomics data into information that can guide diagnostics and medical treatment.^This is the essence of precision medicine.
Thus, while precision medicine can be viewed as driven by systems biology, and the phenotypes of diseases are complex, biophysics can help in addressing the challenge (Ali and Aittokallio 2018;Filipp 2017;Schurdak et al. 2018). Still the question looms: driver mutations are not always frequent; how to identify them? Can biophysics suggest strategies that would broaden the pharmacological targets to include not only rare driver mutations (Ciriello et al. 2013) but also latent drivers, which by definition are rare as well? distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.