The RAS/ERK pathway has been intensely studied for about three decades, not least because of its role in human pathologies. ERK activation is observed in the majority of human cancers; in about one-third of them, it is driven by mutational activation of pathway components. The pathway is arguably one of the best targets for molecule-based pharmacological intervention, and several small-molecule inhibitors are in clinical use. Genetically engineered mouse models have greatly contributed to our understanding of signaling pathways in development, tissue homeostasis, and disease. In the specific case of the RAS/ERK pathway, they have revealed unique biological roles of structurally and functionally similar proteins, new kinase-independent effectors, and unsuspected relationships with other cascades. This short review summarizes the contribution of mouse models to our current understanding of the pathway.
Signal strength versus signal diversification — the role of RAS/ERK pathway component in tissue development and homeostasis
The extracellular signal-regulated kinase (ERK) pathway is a highly conserved signaling cascade critically regulating cell morphology, migration, metabolism, proliferation, differentiation, and survival. In response to extracellular stimuli relayed by tyrosine kinase or other receptors, the small G-proteins RAS are activated at the membrane and recruit >10 downstream effectors that can affect cell proliferation and survival. These include PI3K, GAP, RalGDS, PLCɛ and AF6 as well as the RAF kinases, which are the activators of the ERK pathway . Once recruited by RAS, RAF kinases are activated by a complex mechanism involving phosphorylation/dephosphorylation events and the formation of RAF dimers, which phosphorylate the dual-specificity kinases MEK (MAPK/ERK kinase). These, in turn, activate ERK, which diversifies and amplifies the signal by phosphorylating a variety of cytosolic or nuclear targets mediating the complex and context-dependent biological outcomes of pathway activation . Over the years, cell-based studies have led to the identification of many other proteins that can regulate the pathway at the level of RAF and MEK, signal downstream from these proteins, or tether the complexes to specific subcellular localizations. These results, reviewed elsewhere [3–5], emphasize the complexity of a cascade originally thought to be a linear array of kinases and its high level of integration with other pathways within the cell.
The ‘core’ ERK pathway was discovered in the 1980s, and fundamental knowledge of its architecture, regulation, and function was accrued using cellular systems and model systems expressing only one paralog, notably Drosophila melanogaster. It was in this system, and more specifically in the specification of the R7 cell of the composite Drosophila eye, that a linear cascade starting with RTKs and flowing through RAS, RAF, MEK, and ERK was discovered. This concept was later extended to the Tor RTK system, specifying the anterior and posterior terminal structures, and the DER RTK pathway, mediating the determination of the follicular epithelium and ultimately the specification of ventral structures. Loss of function D-raf and Dsor-1 (DMEK-1) are lethal at the larval/pupal stage, and their contribution to signaling was mainly analyzed in germline mosaics, in which they exhibited developmental disorders similar to Tor and DER loss of function [6,7]. An involvement of the Drosophila ERK paralog rolled in the differentiation of terminal embryonic structures downstream from Tor was demonstrated using shRNA-mediated silencing . Similarly, a hypomorphic Ras1 allele has been generated that affects dorsoventral patterning downstream from DER . It was not until 1997, however, when the first knockout (KO) model of Kras was generated , that we started understanding the physiological function of the individual components of the RAS/ERK pathway in the more complex context of mammalian development, where each tier of the pathway is occupied by multiple enzymes that add both complexity and plasticity to the system (Figure 1). In mammals, RAS is represented by four proteins: HRAS, NRAS, and KRAS4a and b (coded by three genes, one of which, KRAS, is alternatively spliced), and three RAF proteins: ARAF, BRAF, and RAF1 (also called CRAF). The same regulatory layer also contains two pseudokinases, KSR1 and KSR2 (kinase suppressors of Ras), which promote MEK/ERK activation. These proteins work as scaffolds promoting the interaction of RAF, MEK, and ERK at the membrane; in addition, KSR and RAF can allosterically interact with each other, resulting in MEK phosphorylation . Each of the next two tiers is occupied by two enzymes, MEK1 and MEK2 and ERK1 and ERK2. Indeed, components genetic ablation in mice has revealed that they have both redundant and non-redundant functions in development (Figure 1). Knocking out K-ras, B-Raf, Mek1, or Erk2 in mice results in embryonic lethality accompanied by fetal liver defects, cardiac anomalies, and neuronal cell death (KRas−/− [10,12]) or by abnormal placental development (B-Raf −/− , Mek1−/− , and Erk2−/− ). Many of these anomalies are due to the decrease in ERK pathway signal strength. In support of this, a compound Ras KO has been shown to exacerbate the K-ras phenotype, whereas the expression of an H-ras transgene  or the knockin of H-ras in the K-ras locus could rescue embryonic lethality. Similarly, knocking in Mek2 in the Mek1 locus rescues embryonic lethality , as does transgenic expression of ERK1 in Erk2−/− embryos . The common denominator of these studies is that the strength, rather than the quality of the signal, is responsible for the phenotype.
The RAS/ERK pathway in development and disease.
The situation is slightly different in the case of RAF. Here, it is clear that BRAF is essential for full-fledged ERK activation in the placenta  and during the postnatal development of the central nervous system (CNS) where B-Raf depletion leads to neuronal defects, severe dysmyelination and defective oligodendrocyte differentiation . The neuronal phenotype is exacerbated by compound B-Raf/C-Raf-1 ablation , confirming that the CNS defects are due to the role of BRAF as a MEK activator; the Mek KO displays a closely related phenotype, with anomalies in the oligodendrocyte lineage [22,23]. Ablation of KSR1, which is expressed primarily in the CNS, does not cause gross anomalies, but affects PKC-dependent ERK activation, associative learning, and long-term potentiation , a phenotype similar to that induced by Erk2 ablation in neuronal progenitors . Thus, full-fledged ERK signaling is necessary for the development and functioning of the CNS. A similar situation is observed in the epidermis; in this tissue, ablation of B-Raf  or Ksr-1  reduces ERK activation but does not cause any overt phenotypes, whereas combined depletion of MEK1 and MEK2 severely affects epidermal integrity [28,29]. Finally, reduced ERK signaling as a result of Ksr-1 ablation also affects T-cell proliferation  and contributes to the increased cytokine-induced apoptosis observed in the intestinal epithelium during inflammation . Raf-1 ablation has similar consequences in intestinal inflammation. In this case, however, the phenotype is for the most part MEK-independent, and it is caused instead by reduced prosurvival signaling through NF-κB . Taken together, these loss-of-functions studies underscore the importance of attaining appropriate ERK signal strength in organismal development and homeostasis.
On the other hand, mouse models have also allowed us to discover unique, non-redundant functions of RAS/ERK pathway components. While this is probably the most exciting application of genetically engineered mouse models (GEMMs), it is important to note here that when comparing the results of GEMM experiments ablating one family member, a critical parameter to consider is tissue-specific expression, which could confound the analysis of gene-specific phenotypes. For example, the three RAS genes are expressed at different levels in mouse embryonic and adult tissues ; if only one RAS paralog is expressed in a given tissue, its ablation would be equivalent to the ablation of all family members. The analysis of paralog expression is, therefore, mandatory to draw the right conclusion and avoid to mistakenly interpret a phenotype as paralog-specific.
In the case of RAS, the knockin of H-ras in the K-ras locus rescues embryonic lethality, but these animals still develop cardiomyopathy ; and K-ras, but not N-ras, ablation alters hematopoietic stem cell renewal , although paralogs are expressed in these tissues and the KO of one paralog does not affect the expression of the other. This function of KRAS can therefore be considered non-redundant and is still, at least in part, linked to its ability to activate ERK signaling in hematopoietic stem cells. In contrast, the study of both germline and tissue-restricted Raf-1 KO has led to the discovery of unique, kinase-independent functions of this protein. Raf-1 embryos die due to FASL-induced fetal liver apoptosis [19,36]; this antiapoptotic function of RAF1 is independent of its MEK kinase activity  and is connected instead to the ability of RAF1 to interact with the kinases MST2 , implicated in controlling apoptosis downstream from the tumor suppressor RASSF1A , and ROKα . ROKα is a Rho effector that controls cytoskeletal rearrangements. In Raf-1 KO cells, hyperactive ROKα causes changes in cell shapes, restricts motility , and slows down FAS internalization, leading to sustained proapoptotic signaling . In addition, RAF1 antagonizes apoptosis during postnatal cardiomyocyte development; in this case, the dilative cardiomyopathy caused by RAF1 ablation in the heart muscle can be rescued by concomitant ablation of another RAF1 interactor, the proapoptotic kinase ASK1 . Importantly, RAF paralogs are ubiquitously expressed in the mouse, albeit at different levels [42,43]; and in all cases where this was addressed, compensatory up-regulation was not observed.
In the same tier, an intriguing, MEK-independent function has been reported for KSR2 in fatty acid metabolism. Mice lacking Ksr-2 are hypophagic and hyperactive, but instead of being leaner than controls, they are obese and insulin-resistant . This counterintuitive phenotype is due to the reduced activity of AMP-activated protein kinase (AMPK), an enzyme that promotes catabolic activity and inhibits anabolic activity. In MEFs, the activity of AMPK is enhanced by physical interaction with KSR1 and -2. However, Ksr-1 KO mice are not obese, nor do they show other metabolic phenotypes characteristic of Ksr-2 KOs . As for Raf-1, differential tissue expression cannot be invoked to explain the different phenotypes of Ksr-1 and -2 mice; in fact, if anything, KSR2 is expressed at lower levels than KSR1 in all the relevant tissues (adipose tissue, liver, and skeletal muscle). The only exception is the brain, and this has led to the hypothesis that the KSR2–AMPK interaction would play a pivotal role in the CNS .
In the MEK tier, both paralogs are expressed ubiquitously albeit at different levels; for instance, MEK2 is expressed more strongly than MEK1 in embryonic tissues . However, Mek2-deficient mice are essentially normal, and this is not due to compensatory MEK1 overexpression [14,47]. In contrast, Mek1 KO animals and cells have highlighted two unexpected roles of this kinase. Both functions are connected to MEK1's ability to form complexes with other enzymes. In the context of a MEK1/MEK2 dimer, MEK1 is the acceptor of a negative feedback phosphorylation that restricts the extent and duration of MEK2/ERK signaling . Strikingly, the same phosphorylation site is responsible for the formation of a ternary complex consisting of MEK1, the lipid and protein phosphatase PTEN, and a specific isoform of the adaptor protein MAGI. This ternary complex is required for the recruitment of PTEN to the membrane, where it turns over the second messenger PIP3. In Mek1 KO animals, this feedback is disabled; the consequence is a complex autoimmune disease accompanied by myeloproliferation and by the hyperactivation of AKT in the affected cells and organs .
Finally, unique functions have also been described for Erk genes; Erk1 KO mice are viable, show subtle defects in thymocyte differentiation , and enhanced long-term memory potentiation ; in sharp contrast with the latter phenotype, loss of Erk2 in neural progenitor cells decreases cell proliferation and induces a defect in long-term potentiation .
Pathway activation in ‘RASopathies’ and cancers
All of the above demonstrates that the RAS/ERK pathway plays a pivotal role in organism development and homeostasis. Thus, it is not surprising that mutations in pathway components are involved in disease. Germline mutations in KRAS, HRAS, RAF1, BRAF, and MEK1 and 2 cause a set of human pathologies aptly termed ‘RASopathies’ . RASopathies share similar features, including craniofacial dysmorphia, neurocognitive defects, cardiac malformations, ocular and cutaneous anomalies, and increased cancer risk [52,53]. Animal models have helped in establishing the activation of ERK by RAS and RAF mutations as causative of the diseases. Knockin mouse models expressing mutations observed in the Costello syndrome (H-RasG12V, observed at low frequency in the human disease [54,55]), in the Noonan syndrome (K-RasV14I  or Raf1L613V ), and in the cardiofacial cutaneous syndrome (BrafQ241R/ ) recapitulate the features of these diseases; in the Noonan and cardiofacial cutaneous syndrome models, the developmental defects can be rescued, to varying degrees, by chemical MEK inhibition [56,57]. In addition, RAF mutants associated with RASopathies require dimerization for ERK activation, and a peptide modeled on the RAF dimer interface can interrupt RAF signaling in cultured cells . Knockin models of RASopathies will be useful for the preclinical testing of these and other new therapies.
The deregulation of the RAS/ERK pathway frequently observed in a wide variety of cancers has been an area of intense research for almost three decades. Activating mutations are most commonly observed within the RAS and BRAF genes, and represent a critical event in the onset of neoplasia (Figure 1). RAS were among the first oncogenes identified and are mutationally activated in 30% of all human cancer types. Mutations affecting amino acids 12, 13, or 61 are the most frequent and lock the proteins in an active state. KRAS mutations are found predominantly in pancreatic, lung, and colorectal carcinoma, and account for up to 85% of all RAS-mutated cancers, followed by NRAS (11%, mainly in melanoma and leukemia) and HRAS mutations (4%; mainly squamous cell carcinoma and urothelial carcinoma) . GEMM models most faithfully recapitulating tumorigenesis rely on the sporadic oncogenic activation of KRAS . In this system, a synthetic ‘stop’ element flanked by loxP sites is inserted in front of oncogenic K-ras and can be excised by the Cre recombinase delivered to the appropriate cell types through adenoviral or lentiviral vectors  or expressed in a tissue-specific manner in transgenic mouse models, allowing the development of GEMMs useful for the study of KRAS-driven lung and pancreatic cancer that have been recently reviewed elsewhere [63,64].
GEMMs expressing oncogenic RAS and RAF have greatly contributed to the understanding of the mechanisms involved in tumor development and maintenance. Interestingly, the signaling pathway downstream of K-rasG12D or K-rasG12V in lung tumorigenesis was found to be selectively dependent on RAF1 [65,66]. Although Mek/Erk ablation also prevented KRAS-driven lung tumor development, Raf-1 ablation did not affect ERK activation, suggesting an ERK-independent function of RAF1 in this context. Such an ERK-independent function had been demonstrated earlier in the skin, where Raf-1 ablation prevented the development of Ras-driven tumors and caused the regression of established lesions . In this case, the role of RAF1 was elucidated and found to rely on the inhibition of ROKα, which, when activated, strongly promotes keratinocyte differentiation . In contrast, epidermis-restricted B-Raf ablation stops tumor progression by limiting MEK/ERK activation and cell proliferation, slowing down tumor growth. Thus, RAF1 and BRAF have essential but unique roles in RAS-driven skin carcinogenesis . In contrast, MEK1 and MEK2 are functionally redundant in this context . The relative contribution of the two Ksr paralogs to tumorigenesis has not been investigated systematically; however, ablation of the more ubiquitously expressed KSR1 induces a phenotype similar to BRAF, in that it reduces both ERK activation and tumorigenesis in models of carcinogenesis driven by activated RAS [27,70,71], but also MYC .
NRAS and BRAF mutations are frequent in human melanoma, and their effects have been modeled in the mouse. The most common BRAF mutation, V600E, causes a 500-fold increase in kinase activity compared with wild-type BRAF and acts as a monomer [59,73]. The high levels of RAS/ERK pathway activation induced by BRAFV600E cause senescence, which must be bypassed through the loss of tumor suppressor genes to drive melanomagenesis . In mice, the expression of BRAFV600E in melanocytes promotes melanoma development even without other mutational events , but melanomagenesis is accelerated by TP53 mutations . The Danio rerio (zebrafish) model is also currently in use to study melanomagenesis; indeed, co-operation with BRAFV600E and p53 have been originally described in this model . More recently, a study in Danio has identified the re-emergence of neural crest progenitors as the key event of melanomagenesis, significantly contributing to our understanding of melanoma .
BRAFV600E inhibitors have shown impressive responses in patients with metastatic melanoma; however, these responses are transient and resistance through reactivation of the RAS/ERK pathway almost inevitably develops, precluding their use in single-agent therapy . Thus, the race is on for the most efficient combinatorial therapy, whether vertical (inhibition of RAF and MEK) or horizontal (e.g. inhibition of the RAS/ERK and PI3-K pathway), and NRAS and BRAF models of melanoma are extremely useful to investigate the effect of these combinations on disease development. Cutaneous toxicities are prominent among the side effects of RAF inhibitors; about half of them consist in abnormal proliferation, from hyperkeratosis to the development of drug-related cutaneous tumors . These effects are mechanism-based: RAF inhibitors stabilize a conformation of the RAF kinase domain that promotes RAS-driven RAF dimerization [80–82]. Unless the inhibitors are present in saturating concentrations, these dimers induce, rather than inhibit, RAS/ERK pathway activation; accordingly, RAF inhibitor-induced cutaneous toxicities are alleviated by co-treatment with MEK inhibitors . The relevance of RAS-driven RAF dimerization for the emergence of inhibitor resistance and for the cutaneous side effects has been demonstrated in mouse models of melanoma  and of epidermal tumors [83,84] and, more recently, in patients . In a model of RAS + RAF inhibitor-driven tumorigenesis, Raf-1 ablation and the ensuing activation of ROKα were shown to be extremely effective in preventing tumorigenesis, suggesting that combined therapies targeting kinase and non-kinase functions of RAF could be highly efficacious in skin tumor treatment . Animals lacking both BRAF and RAF1 in the epidermis have been generated to mimic this situation; surprisingly, these mice develop an allergic disease similar to human atopic dermatitis, due to a transient weakening of the epidermal barrier combined with the increased production of inflammatory cytokines by keratinocytes. The phenotype results from a combination of decreased ERK signaling, caused by B-raf ablation, and increased JNK activity, due to the loss of RAF1 . These experiments highlight both the pivotal role and the non-redundant function of RAF in the epidermis.
The studies summarized above focused on the cell-autonomous effects of RAS/ERK pathway components in tumorigenesis. The pathway, however, also plays an important role in establishing the tumor niche, remodeling the environment to evade immune surveillance and to generate the tumor-associated vasculature providing nutrients and oxygen to the growing lesion. In melanoma, for instance, it has been shown that both BRAF and MEK inhibitors are capable of up-regulating melanosomal antigen expression, and this is accompanied by increased infiltration of T-lymphocytes . This has raised the hope that combining ERK pathway inhibition with immune checkpoint blockade inhibitors could overcome immune evasion by the tumor and increase the response rate. This strategy has recently been validated in a mouse model of BRAFV600E-driven melanoma, in which the combination of RAF and MEK inhibitors with checkpoint blockers yielded optimal antitumor response .
The role of the RAS/ERK pathway in angiogenesis has been demonstrated in xenograft models. In lung or melanoma tumor cells, constitutively active RAS [88–90] or BRAF [91,92] induces VEGF expression, which in most cases depends on MEK/ERK activation; in agreement with this, BRAF is required for VEGF production in a model of pancreatic insulinoma which depends on the angiogenic switch for progression , as well as in the placenta [19,93]. In the endothelium itself, RAS signaling has been shown to be important for angiogenic sprouting and permeability, functions depending on the RAS/ERK and the PI3K pathways, respectively [94,95]; while the RAF1/ROKα complex acts downstream from another small GTPase, RAP1, to stabilize endothelial cell–cell junctions during sprouting angiogenesis .
At the level of MEK, MEK1 is essential for labyrinthine angiogenesis during placental development, a phenotype that correlates with reduced ERK activation in cells migrating toward extracellular matrix components [14,48], whereas both Erk1 and Erk2 must be ablated to reveal the crucial role of ERK in embryonic angiogenesis .
Mouse models have greatly advanced our understanding of the RAS/ERK pathway in physiological and pathological contexts, revealing both redundant and unique functions of the proteins involved. We now need to dig deeper and start probing the effect of mutations affecting kinase activity, preventing the phosphophorylation of specific sites, and/or the interaction between pathway components. Examples of these strategies are still rather rare; whenever they have been applied, however, the results have been always informative and sometimes surprising. Among these are the demonstration of the importance of PI3K signaling downstream from RAS in lung tumorigenesis [98,99] and tumor-induced angiogenesis ; the validation of the role of kinase-inactive BRAF mutants in melanomagenesis and of their RAF dimer-promoting function when combined with active RAS ; and finally, the demonstration that point mutations in the BRAF activation loop at least partially phenocopy the defects of BRAF KO animals. A further unexpected outcome of the latter study was the selective impact of the mutation on the interaction between BRAF and KSR, rather than with RAF1 [100,101]. The advent of the CRISPR-Cas9 genome-editing tool offers excellent opportunities for the faster development of such mutants, alone or in combination with the ablation or point mutation of other pathway components. CRISPR-Cas9 will also allow the generation of human disease models in which multiple genes are mutated at the same time, facilitating the investigation of genetic networks and pathway cross-talk  in GEMMs but also in patient-derived xenografts (PDX). These human-in-mouse models, particularly when the xenografts are orthotopic, are gaining relevance as a promising translational platform to develop and test new pharmacological agents , with the remaining caveat that the appreciation of the cross-talk between tumor cells and stroma may be limited by species-specific effects . While all models are by nature an approximation, their constant evolution continues to raise new opportunities for the analysis of the RAS/ERK and other pathways in development and disease; in parallel, however, we need to continue developing tools to analyze the models properly. These include protocols designed to preserve phosphorylation and protein–protein interactions during isolation of the proteins of interest from tissue as well as validated, histology-grade reagents for the detection of phosphorylation sites in situ, needed to clarify their relevance as well as their potential use as biomarkers or druggable targets.
We thank all the staff of Baccarini laboratory for support and helpful discussions. We apologize to authors whose work we could not include because of space limitations.
C.D. is the recipient of a long-term EMBO fellowship [ALTF 191-2015]. Work in the Baccarini laboratory is/was supported by grants of the Austrian Research Fund [FWF W-1220, SFB-021, P-19530, P-26874, P-26303, and P-22831], EU-project INFLACARE, Bridge project 827500 (Austrian Research Promotion Agency, FFG), and the Johanna Mahlke geb. Obermann-Stiftung of the University of Vienna.
The Authors declare that there are no competing interests associated with the manuscript.