Document resource
Introduction The rat sarcoma viral oncogene homolog ( RAS) gene family—comprising HRAS, NRAS and KRAS—encodes small GTPases that function as molecular switches, toggling between active (guanosine triphosphate (GTP)-bound) and inactive (guanosine diphosphate (GDP)-bound) states to regulate cell proliferation, survival and differentiation. The name ‘RAS’ originates from RaS viruses, where viral oncogenes were first identified in the 1960s and 1970s.1 2 A major breakthrough came in 1982, when three independent groups used NIH 3T3 mouse embryonic fibroblast transfection assays to show that human tumours—most notably bladder carcinomas—carry transforming genes homologous to the viral ras oncogenes of the Harvey and Kirsten sarcoma viruses.3–5 These findings were soon extended to additional malignancies, and mutated HRAS and KRAS were confirmed as oncogenic drivers in human bladder, lung and colorectal cancers (CRC).6–9 This provided the first definitive evidence that non-viral human cancers can result from activating point mutations in host-intrinsic proto-oncogenes. These seminal discoveries established a new paradigm in cancer research and laid the foundation for the field of molecular oncology.Among the RAS isoforms, KRAS is the most frequently mutated in human cancers and is particularly prevalent in epithelial malignancies, such as pancreatic, colorectal and non-small cell lung cancers (NSCLCs).10 Historically, KRAS was considered undruggable due to its high affinity for GTP/GDP and lack of suitable binding pockets.11 However, recent breakthroughs—most notably the development of KRAS-G12C protein inhibitors (G12Ci) such as sotorasib and adagrasib—have overcome this barrier, resulting in US Food and Drug Administration (FDA) approval for the treatment of KRAS-G12C-mutant NSCLC.12 13 These clinical advances have fuelled intensive efforts to develop inhibitors against additional KRAS mutants, as well as pan-KRAS and pan-RAS strategies, targeted protein degradation and RNA-based therapies for cancer.Despite this progress, resistance to KRAS-targeted therapy—both intrinsic and acquired—remains a major challenge. Tumour adaptation through secondary mutations, bypass signalling, metabolic reprogramming and immunosuppressive tumour microenvironment (TME) limits treatment durability.10 Addressing these barriers will require rational combination strategies and precision medicine approaches.14In this review, we aim to provide a concise yet integrative overview of the rapidly evolving field of KRAS-targeted cancer therapy. Rather than offering an exhaustive catalogue of studies, our focus is to distil key mechanistic insights and translational advances that define current therapeutic paradigms. We begin with the fundamental biology and oncogenic signalling of KRAS, followed by an overview of approved and emerging therapeutic strategies. We then discuss major resistance mechanisms and their implications for clinical management. Finally, we highlight the remaining challenges—such as tumour heterogeneity, adaptive resistance and the complexity of the TME—and outline future directions for developing next-generation KRAS inhibitors (KRASi) and rational combination therapies.KRAS biology and oncogenic signalling KRAS functions as a molecular switch, cycling between an inactive GDP-bound state and an active GTP-bound state ( figure 1).15 This cycle is tightly regulated by guanine nucleotide exchange factors (GEFs), such as SOS Ras/Rac guanine nucleotide exchange factor 1 (SOS1), which promote the exchange of GDP for GTP,16 17 and GTPase-activating proteins (GAPs), such as neurofibromin 1 (NF1),18 19 which enhance the intrinsic GTPase activity of KRAS, facilitating GTP hydrolysis and returning KRAS to its inactive state. Protein tyrosine phosphatase non-receptor type 11 (PTPN11, also known as SHP2), a protein tyrosine phosphatase, facilitates SOS1-mediated loading of GTP onto RAS, thereby promoting the active RAS ‘on’ state. In normal cells, KRAS activation is transient and tightly regulated by upstream growth factors, receptor tyrosine kinases (RTKs) and intrinsic negative feedback mechanisms. However, oncogenic KRAS mutations frequently impair GTP hydrolysis by reducing intrinsic GTPase activity or conferring resistance to GAP-mediated inactivation, resulting in constitutive activation of KRAS and sustained downstream signalling.Figure 1Regulation and downstream signalling of KRAS in normal and oncogenic contexts. Growth factor binding to receptor tyrosine kinases (RTKs) promotes receptor phosphorylation and recruitment of adaptor proteins, including protein tyrosine phosphatase non-receptor type 11 (PTPN11) (SHP2) and SOS1, which act as guanine nucleotide exchange factors (GEFs) to catalyse guanosine diphosphate (GDP)–guanosine triphosphate (GTP) exchange on KRAS. Activated GTP-bound KRAS transiently signals to downstream effectors, including the rapidly accelerated fibrosarcoma (RAF)–mitogen-activated protein kinase (MEK)–extracellular signal-regulated kinase (ERK) cascade that controls proliferation, transcription and differentiation, and the phosphatidylinositol 3-kinase (PI3K)–protein kinase B (AKT)–mechanistic target of rapamycin (mTOR) pathway that regulates growth, survival and metabolism. These signals are counterbalanced by GTPase-activating proteins (GAPs) such as neurofibromin 1, which enhance GTP hydrolysis and return KRAS to its inactive GDP-bound state. The tumour suppressor PTEN antagonises PI3K signalling by converting phosphatidylinositol-3,4,5-trisphosphate back to phosphatidylinositol-4,5-bisphosphate, thereby limiting AKT activation. Oncogenic KRAS mutations (eg, G12C, G12D) impair GAP-mediated inactivation, resulting in constitutive KRAS activation and persistent downstream signalling that drives tumourigenesis. PTEN, phosphatase and tensin homolog; WT, wild type; MU: mutant.Once activated, KRAS engages several key effector pathways that coordinate diverse cellular functions (figure 1). The rapidly accelerated fibrosarcoma (RAF)–mitogen-activated protein kinase (MEK)–extracellular signal-regulated kinase (ERK) cascade is a major signalling route that regulates gene transcription, cell cycle progression and differentiation. The phosphatidylinositol 3-kinase (PI3K)–protein kinase B (AKT, also known as PKB)–mechanistic target of rapamycin (mTOR) pathway also plays a central role, promoting cell growth, survival, protein synthesis and metabolic adaptation. Additional KRAS effectors, such as ral guanine nucleotide dissociation stimulator, ras association domain family member 1 and TIAM rac1 associated GEF 1, link KRAS to processes such as vesicle trafficking, cytoskeletal remodelling and cell migration, thereby contributing to tumourigenesis in various tissues including the pancreas and intestine.20–22 The relative contribution and interplay of these signalling cascades depend on both the specific KRAS mutation and the cellular context, contributing to functional heterogeneity across tumour types.KRAS mutations are among the most prevalent oncogenic alterations in human cancer and most commonly affect codons 12, 13 and 61.23 In pancreatic ductal adenocarcinoma (PDAC), more than 90% of tumours harbour activating KRAS mutations, with G12D and G12V being the most frequent. In NSCLC, the KRAS-G12C mutation is enriched, occurring in approximately 11–13% of cases, whereas about 40% of CRC harbour G12D, G13D or Q61H/L substitutions. These mutations are not functionally equivalent; they differ in intrinsic GTP hydrolysis rates, affinity for downstream effectors and dependence on upstream signalling. For example, KRAS-G12C harbours a unique cysteine residue that enables covalent binding by allele-specific inhibitors,24 a feature absent in G12D and G12V, which limits their druggability and necessitates alternative targeting strategies. Although mutations in exon 2 are the most common, KRAS produces two splice variants—KRAS4A and KRAS4B—through alternative usage of exon 4. Thus, effective KRAS-targeted therapies must account not only for the biochemical consequences of specific point mutations but also for isoform-specific structural and functional differences.The functional diversity among KRAS mutants has implications for therapeutic response. Allele-specific features influence effector pathway bias, oncogenic potency and dependence on upstream signals epidermal growth factor receptor (EGFR), PTPN11 or fibroblast growth factor receptors (FGFRs). Moreover, co-occurring genetic alterations, including loss-of-function mutations in EGFR, tumour protein p53 (TP53, also known as p53), serine/threonine kinase 11 (STK11, also known as LKB1) and kelch-like ECH-associated protein 1 (KEAP1), further modulate KRAS-driven signalling.25 26 These comutations can alter tumour metabolism, immune cell infiltration, oxidative stress responses and resistance to therapy. For instance, STK11 mutations are associated with reduced T-cell infiltration and poor response to immune checkpoint blockade (ICB) in KRAS-mutant NSCLC,27 whereas KEAP1 mutation or loss enhances NFE2-like BZIP transcription factor 2 (NFE2L2) activity, promoting resistance to oxidative damage and targeted therapies.28 In addition, KRAS and TP53 comutations in advanced lung adenocarcinomas predict improved survival benefit from ICB and are associated with distinct molecular features, including elevated tumour mutational burden, CD274 (also known as PD-L1) expression and a unique gene expression signature.26 Thus, a comprehensive understanding of KRAS mutant biology, including structural differences, effector engagement and interaction with the TME, is essential for developing rational therapeutic strategies.Current therapeutic strategies targeting KRAS Current approaches to suppress KRAS-driven cancers can be broadly classified into direct KRASi and indirect targeting strategies.KRAS-G12C inhibitors KRAS-G12C is a common oncogenic mutation in NSCLC and occurs less frequently in CRC and PDAC. The discovery in 2013 by Ostrem et al of a druggable pocket within the switch II region (S-IIP) of KRAS-G12C protein enabled the development of covalent inhibitors that selectively target the mutant cysteine residue in its GDP-bound state.24 This breakthrough led to the development of sotorasib (AMG 510) and adagrasib (MRTX849),13 29–31 the first FDA-approved G12Ci for locally advanced or metastatic NSCLC.Both agents have demonstrated reproducible efficacy in clinical trials. In the CodeBreaK 100 trial, sotorasib achieved an objective response rate (ORR) of ~32–37%, disease control in >80% of patients with NSCLC, median progression-free survival (PFS) of 6–7 months and overall survival (OS) of ~12.5 months.12 32 The Phase 3 CodeBreaK 200 study confirmed modest benefit over docetaxel in NSCLC (PFS 5.6 vs 4.5 months) but without OS improvement.33 As a result, the drug retained only accelerated—not full—FDA approval, leaving its long-term benefit uncertain and subject to further confirmatory evidence. Adagrasib, evaluated in KRYSTAL-1, showed comparable efficacy in NSCLC (ORR ~43%, PFS 6.5 months, OS 12.6 months) and similarly received accelerated FDA approval in 2022.34 35Efficacy in CRC has been markedly lower with monotherapy (ORR<10–20%, PFS~4 months),32 largely due to adaptive resistance driven by EGFR reactivation.36 This has been overcome by rational combination strategies: sotorasib plus panitumumab (CodeBreaK 300) improved ORR to ~30% and PFS to 5.6 months,37 gaining FDA approval in 2025, while adagrasib plus cetuximab (KRYSTAL-1) achieved an ORR of 46%, PFS of 6.9 months and OS of 13.4 months,38 39 leading to full FDA approval in 2024. Adagrasib has also shown modest activity in PDAC (ORR 33%, PFS 5.4 months).35Collectively, these results establish first-generation G12Ci as a therapeutic standard in NSCLC and as part of combination regimens in CRC. However, their clinical benefit is limited by short duration of response (~4–6 months) and the emergence of resistance, highlighting the need for next-generation G12Ci and rational combinatorial approaches, several of which are summarised in online supplemental table S1.SP110.1136/bmjonc-2025-000946.supp1Supplementary dataAmong the secondary generation of G12Ci, divarasib (GDC-6036) showed durable activity in advanced NSCLC with an ORR of 55.6%, median response duration of 18 months and PFS up to 15.3 months, suggesting superiority over first-generation agents.40 41 In a phase I study, D3S-001 with faster target engagement compared with first-generation G12Ci was well tolerated with no dose-limiting toxicities and demonstrated robust antitumour activity, achieving an ORR of 73.5% in G12Ci-naïve patients (including 66.7% in NSCLC, 88.9% in CRC and 75.0% in PDAC) and 30.0% in G12Ci-pretreated NSCLC.42 43Although most second-generation G12Ci remain OFF-state agents, novel approaches are being developed. BBO-8520 targets both GDP-bound and GTP-bound KRAS-G12C,44 whereas RMC-6291 functions as an ON-state inhibitor by forming a tricomplex with peptidylprolyl isomerase A (PPIA, also known as cyclophilin A), a cellular chaperone protein.45 This tricomplex stabilises drug binding to the active, GTP-loaded conformation of KRAS, thereby maintaining inhibition even when the protein cycles into its ON state. Together, these compounds illustrate alternative strategies for KRAS-G12C inhibition, though head-to-head comparisons will be required to establish their relative efficacy and safety.Non-G12C mutant inhibitors Non-G12C KRAS mutations, particularly G12D and G12V, are highly prevalent in PDAC and CRC and lack a cysteine residue for covalent targeting, necessitating alternative therapeutic strategies. 46 MRTX1133 was the first non-covalent KRAS-G12D inhibitor (G12Di) discovered through structure-based design, binding both GDP-bound and GTP-bound states with ~700-fold selectivity over wild-type KRAS protein.47 In PDAC mouse models, it induced tumour regression and shifted tumours towards a classical phenotype, enhancing chemosensitivity.48 Although MRTX1133 advanced to a phase 1/2 trial (NCT05737706), the study was terminated in 2025 due to variable and suboptimal pharmacokinetics despite an acceptable safety profile.Several other G12Di are in early clinical testing, including GFH375/VS-7375, HRS-4642, INCB161734, LY3962673, TSN1611, QLC1101 and QTX3046 (summarised in online supplemental table S2). Among these, HRS-4642 is a highly selective non-covalent inhibitor that targets both ON and OFF states of KRAS-G12D protein, demonstrating potent preclinical activity, durable tumour retention and early clinical responses with a favourable safety profile.49An alternative ON-state approach is represented by RMC-9805, which forms a tricomplex with PPIA to engage GTP-bound KRAS-G12D. It induced robust responses in mice bearing PDAC and NSCLC xenografts, and early clinical data (NCT06040541) reported a 30% ORR and 80% disease control, with manageable gastrointestinal toxicity and rash.50Beyond small-molecule inhibitors, targeted protein degradation strategies are emerging. Proteolysis-targeting chimaeras degraders such as ACBI3 (a pan-KRAS degrader active against KRAS-G12D and KRAS-G12V in xenografts),51 ASP3082 (a KRAS-G12D-selective, in phase 1 trials (NCT05382559)),52 RP03707 (a cereblon (CRBN)-based degrader with durable in vitro and in vivo activity)53 and CH091138/compound 6 (a Von Hippel-Lindau tumour suppressor (VHL)-based degrader effective in cell lines, organoids and mouse xenograft models)54 highlight the potential of degradation-based approaches to overcome limitations of classical inhibitors.RNA-based therapies provide another avenue to silence KRAS expressions. The biodegradable implant siG12D-LODER, delivering siRNA against KRAS-G12D, has shown safety and tolerability in phase I/II trials for locally advanced PDAC when combined with chemotherapy (NCT01676259).55 Building on this foundation, ligand-conjugated RNAi approaches have been developed to improve mutation and tissue specificity. For instance, EFTX-G12V, an EGFR-targeted RNAi molecule, selectively silences KRAS-G12V transcripts while sparing wild-type KRAS, illustrating the potential of receptor-mediated delivery to enhance precision and reduce off-target toxicity.56 More recently, a dual pan-KRAS+MYC RNAi modality has been engineered using an ‘inverted’ design, in which the passenger strand of a MYC-targeting siRNA is fused to the guide strand of a KRAS-targeting siRNA, enabling concurrent suppression of two key oncogenic drivers.57 In parallel, engineered exosomes loaded with KRAS-G12D siRNA have entered phase I clinical trials in PDAC patients,58 representing an innovative strategy for improving biocompatibility and tumour-selective delivery. Other transcript-silencing modalities include antisense oligonucleotides, such as AZD4785, which achieved potent and selective KRAS knockdown with corresponding tumour growth inhibition in preclinical models,59 and CRISPR–CasRx (Cas13d) systems that precisely silenced KRAS-G12D transcripts and suppressed PDAC progression in vitro and in vivo.60 Despite these advances, the clinical translation of RNA-based therapeutics remains challenging. Major limitations include inefficient intracellular delivery, restricted tissue penetration, rapid nuclease degradation and undesired immune activation. Addressing these barriers—through next-generation lipid nanoparticles, tumour-specific ligands or exosome-based vehicles—will be essential to realise the full therapeutic potential of RNA-based KRAS silencing.Macrocyclic peptide inhibitors have recently emerged as a promising strategy to target non-covalent KRAS mutants such as KRAS-G12D and KRAS-G12V, which lack the reactive cysteine exploited by covalent inhibitors. Through mRNA display and structure-guided screening, several macrocyclic scaffolds have been identified that bind the switch I/II pocket of inactive, GDP-bound KRAS-G12D with nanomolar affinity. The prototypic compound KRpep-2d blocks the interaction between KRAS and SOS1, thereby preventing nucleotide exchange and downstream RAF–MEK–ERK activation.61 Subsequent chemical optimisation, including hydrocarbon stapling and arginine-rich linkers, has improved membrane permeability and metabolic stability, leading to measurable suppression of pERK signalling in KRAS-G12D-mutant cell lines.62 Further derivatives, such as cyclorasin and LYRAS analogues, extend activity across KRAS-G12D and KRAS-G12V mutants by disrupting RAS–effector interactions and inducing tumour regression in preclinical xenografts.63 Although challenges remain in enhancing intracellular delivery and systemic stability, these macrocyclic peptides demonstrate that even ‘undruggable’ KRAS alleles possess tractable binding surfaces, offering a conceptual bridge between small-molecule inhibitors and biologic therapeutics.Pan-KRASi Because KRAS mutations extend beyond G12C and G12D, pan-KRASi are being developed to target multiple alleles as well as KRAS wild type. Early tool compounds such as BI-2852 and its analogues (eg, BI-2865) bind non-covalently to the GDP-bound state and suppress growth across KRAS-G12C, G12D, G12V and WT models.64 These agents also inhibited tumour growth in patient-derived xenografts (PDXs), and BI-2865 further reversed ATP binding cassette subfamily B member 1-mediated drug resistance.64 More recently, BI-2493, an allele-agnostic pan-KRAS inhibitor and structural analogue of BI-2865 with improved rigidity, metabolic stability and permeability, was shown to suppress tumour growth in diverse preclinical PDAC models, including xenografts and genetically engineered mice.65 In addition to blocking canonical KRAS effector pathways, BI-2493 reprograms tumour cell states by reducing epithelial-to-mesenchymal transition and activating STK11/AMP-activated protein kinase–microtubule affinity regulating kinase 2 signalling, consistent with a cytostatic shift towards oxidative phosphorylation.65 BI-2493 also remodels the tumour immune microenvironment by enhancing CD8+ and CD4+ T-cell infiltration while reducing suppressive myeloid populations, thereby increasing responsiveness to ICB.65Further advances include MCB-294, a potent pan-KRAS inhibitor that targets multiple mutants (G12D/C/V/S, G13D, Q61H) by binding dual KRAS states through a water-mediated hydrogen bond network.66 In preclinical models, MCB-294 achieved strong antitumour efficacy, overcame G12Ci resistance and enhanced immune cell infiltration.66 Building on this scaffold, MCB-36 was designed as a VHL-recruiting pan-KRAS degrader that induces sustained KRAS degradation.66 Several additional compounds have now entered the clinic, including BI-3706674, which is in phase I trials (NCT06056024) for KRAS WT amplified and KRAS-mutant cancers. Other phase I candidates of pan-KRASi include PF-07934040 (Pfizer), LY-4066434 (Eli Lilly) and QTX3034 (QuantaTx). Collectively, these agents can be expected to extend therapeutic coverage across the heterogeneous spectrum of KRAS alterations. The key challenges for pan-KRASi are achieving broad efficacy across KRAS alleles while maintaining sufficient selectivity to minimise toxicity and overcoming adaptive resistance mechanisms that tumours rapidly deploy.Pan-RAS inhibitors Pan-RAS inhibitors were developed to address a key limitation of allele-selective KRASi—the reactivation of WT KRAS and other RAS isoforms via RTK signalling feedback. Early concerns that pan-RAS inhibition would result in prohibitive toxicity have been alleviated, as several candidates have demonstrated manageable safety profiles in preclinical and early clinical studies. Tool compounds such as RMC-7977 and clinical candidates like RMC-6236 (a PPIA-based tri-complex RAS(ON) inhibitor) have shown potent activity across PDAC, NSCLC and CRC models, with tumour-selective apoptosis and minimal normal tissue damage. 67 68 Clinically, RMC-6236 has produced encouraging results: in 127 patients with RAS-mutant tumours, it achieved an ORR of 29% and disease control in 91%, with two complete responses (one in KRAS-G12D PDAC and one in KRAS-G12V NSCLC).69 Most adverse events were low-grade rashes, though 35% of patients required dose modification, and no treatment discontinuations were reported.69 The advantage of using pan-RAS inhibitors lies in the ability to suppress a broad spectrum of RAS mutants and overcome bypass reactivation; the disadvantage includes weaker responses in CRC, where combination therapy may be required, and relatively short PFS in heavily pretreated PDAC (8.1 months second line, 4.2 months third line). Moreover, as with KRAS-selective inhibitors, resistance to pan-RAS inhibitors emerges rapidly, raising concerns about the durability of benefit. The ongoing RASolute-302 phase 3 trial (NCT06625320) will determine whether pan-RAS inhibition can achieve sustained efficacy across RAS-mutant cancers. Ultimately, the clinical success of pan-RAS inhibitors will depend not only on their single-agent activity but also on rational combination strategies that extend durability and overcome adaptive resistance.Indirect targeting strategies Several approaches have been developed to indirectly inhibit KRAS. One strategy targets KRAS processing, as membrane localisation is essential for its activation. KRAS prenylation is mediated by two enzymes, farnesyltransferase (FT) and geranylgeranyltransferase I (GGT1), which facilitate its anchoring to the plasma membrane. FT inhibitors initially showed promise but were rendered ineffective by compensatory geranylgeranylation. Dual FT and GGT1 inhibitors such as FGTI-2734, 70 and agents targeting downstream processing enzymes such as isoprenylcysteine carboxyl methyltransferase inhibitors,71 may help overcome this limitation. Synthetic proteins like PTPN11 can also block KRAS self-association at the membrane, although potential immunogenicity remains a concern.72Because KRAS activity depends on upstream inputs, inhibition of RTKs has also been explored. Although EGFR inhibitors and antibodies such as cetuximab benefit selected patient subsets, their efficacy in KRAS-mutant cancers is generally limited. A potential exception is KRAS-G13D colorectal cancer, where some studies have reported modest responses, although this remains controversial and is not widely adopted in clinical practice.73–75 Targeting mediators of RAS activation, including SOS1 and PTPN11, represents another promising avenue. The SOS1 inhibitor BI-1701963 and the PTPN11 inhibitor TNO155 are currently in clinical evaluation, often in combination with MEK or KRAS-G12C inhibitors.76 77Downstream blockade of KRAS effectors has also been tested extensively. RAF (eg, LXH-254), MEK (eg, trametinib, cobimetinib and binimetinib) and ERK (eg, ulixertinib) inhibitors demonstrate modest single-agent activity but may provide improved efficacy when combined, whereas PI3K pathway inhibitors remain limited by toxicity and insufficient selectivity. In addition, synthetic lethal strategies and disruption of KRAS-driven metabolic adaptations, including autophagy and macropinocytosis, represent emerging therapeutic approaches. Importantly, reactivation of these signalling pathways contributes to resistance against KRASi (discussed later). Thus, combining direct and indirect strategies may represent the most effective clinical approach to achieve durable responses in KRAS-driven cancers.Clinical safety and adverse reactions of KRASi The safety profiles of covalent G12Ci have been systematically characterised across multiple phase 1–3 clinical trials. In the pivotal CodeBreaK 100 and CodeBreaK 200 studies, 12 33 sotorasib demonstrated a manageable toxicity profile, with treatment-related adverse events (TRAEs) reported in 88–98% of patients and grade ≥3 TRAEs in 20–33%. The most frequent toxicities were diarrhoea (42%), nausea (26%), fatigue (26%) and transaminase elevations (20%). Hepatic enzyme elevations—alanine aminotransferase (ALT, 8%) and aspartate aminotransferase (AST, 6%)—were the most common grade ≥3 laboratory abnormalities, typically reversible with dose interruption. Serious TRAEs such as pneumonitis were rare (<2%), and no treatment-related deaths were observed. In CodeBreaK 200, hepatotoxicity occurred more frequently when sotorasib was initiated within 2.6 months after prior immune-checkpoint blockade, suggesting immune-primed hepatic sensitivity.Combination regimens have yielded comparable or improved tolerability. In CodeBreaK 300,37 sotorasib+panitumumab for KRAS-G12C-mutant metastatic colorectal cancer resulted in grade ≥3 TRAEs in 30–36% of patients, mainly rash, pruritus, dermatitis and hypomagnesaemia due to EGFR blockade, with low-grade gastrointestinal toxicity and few discontinuations. Similarly, KRYSTAL-139 reported adagrasib±cetuximab to cause TRAEs in nearly all patients, but grade 3–4 events in ~28%, most commonly nausea (61%), vomiting (52%), diarrhoea (49%) and acneiform dermatitis (48%), with no treatment-related fatalities and limited hepatotoxicity.Next-generation inhibitors have shown promising safety improvements. In the phase 1 trial of D3S-001,42 grade ≥3 TRAEs occurred in 16.7% of patients in the KRAS-G12C-naïve cohort and 10% in the pretreated cohort, with no grade 4–5 events. The most common adverse events were nausea (45%), diarrhoea (31%), vomiting (19%), hypertriglyceridaemia (19%), amylase/lipase elevations (16–19%) and mild hepatotoxicity (ALT/AST or GGT increases, mostly grade 1). No interstitial lung disease or pneumonitis was reported, and no treatment discontinuations occurred. These results underscore a favourable tolerability profile for D3S-001, with no dose-limiting toxicities up to 900 mg and pharmacokinetics supporting a 600 mg phase 2 dose.In a long-term follow-up of a phase I study in 65 patients with KRAS-G12C-mutant NSCLC, divarasib (GDC-6036) demonstrated a manageable and durable safety profile consistent across treatment durations.40 Overall, 94% of patients experienced TRAEs, the majority being grade 1–2 gastrointestinal toxicities, including nausea (78%), vomiting (66%) and diarrhoea (62%). Grade≥3 TRAEs occurred in 17%, primarily hepatic enzyme elevations (ALT 6.2%, AST 4.6%) and rare lipase increases (3.1%). No grade 5 events or dose-limiting toxicities were reported. Treatment modification was required in 39% of patients, with only 5% discontinuing due to adverse events. Among patients treated beyond 1 year, 55% developed new-onset mild TRAEs such as amylase or lipase elevation, but none of grade ≥3 severity—supporting the long-term tolerability of divarasib and its potential compatibility with ICB.Similarly, glecirasib, evaluated in a global phase 2 study of 172 patients with KRAS-G12C-mutant NSCLC,78 displayed grade ≥3 TRAEs in 12.8%, most commonly ALT/AST elevations (4–5%) and diarrhoea (3%), with no treatment-related deaths and only 1.7% discontinuations due to toxicity. The overall incidence of any-grade TRAEs was 69%, dominated by mild gastrointestinal and hepatic events, reinforcing the safety consistency among GDP-bound KRASi.Collectively, KRASi demonstrate a predictable and clinically manageable toxicity spectrum, primarily involving gastrointestinal (diarrhoea, nausea, vomiting ≈40–60%), hepatic (ALT/AST ≈5–10%) and dermatologic (rash ≈30–50%) effects. Hematologic and systemic toxicities are markedly less frequent than with cytotoxic chemotherapy, and fatal events remain rare (< 1%). Most grade ≥3 toxicities resolve after dose modification, supporting the use of routine liver-function monitoring and pharmacodynamic-guided dose adjustment to optimise safety in KRAS-targeted therapy.Mechanisms of resistance to KRAS-targeted therapies Despite recent advances in KRASi development, clinical responses remain transient and heterogeneous. KRAS-mutant tumours exhibit intrinsic resistance or acquire adaptive resistance through diverse molecular and cellular mechanisms, as summarised below (figure 2).Figure 2Mechanisms of primary, acquired and adaptive resistance to KRAS-targeted therapies. Primary resistance arises from comutations (eg, STK11 driving immune-cold tumour microenvironments, KEAP1 promoting NFE2-like BZIP transcription factor 2 (NFE2L2)-dependent antioxidant defenses, TP53 impairing apoptosis) and tumour microenvironmental barriers (regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSCs), tumour-associated macrophages (TAMs), cancer-associated fibroblast (CAF)-driven fibrosis) that suppress immunity and hinder drug delivery. Acquired and adaptive resistance develops through diverse mechanisms, including (1) KRAS alterations such as secondary pocket mutations (Y96, H95, R68) and gene amplification; (2) bypass activation via NRAS/HRAS signalling, NF1 loss or reactivation of epidermal growth factor receptor (EGFR) and mesenchymal–epithelial transition factor (MET); (3) downstream reactivation through BRAF, MEK and PI3K mutations or SRC–RAF1 signalling; (4) phenotypic plasticity, including epithelial–mesenchymal transition (EMT), Yes1 associated transcriptional regulator (YAP1)/WW domain containing transcription regulator 1 (TAZ)–TEA domain transcription factor (TEAD) signalling, and lineage transitions; and (5) metabolic rewiring, such as advanced glycosylation end-product specific receptor (AGER)-driven macropinocytosis and induction of autophagy. TME, tumour microenvironment.Primary resistance Primary resistance refers to the failure of tumours to respond to KRASi at treatment initiation, occurring in approximately 9–24% of patients. 32 This resistance arises from both tumour-intrinsic genomic alterations and extrinsic influences from the TME.Co-occurring mutations in tumour suppressor genes strongly shape therapeutic response. In lung adenocarcinoma, STK11 and KEAP1 mutations frequently accompany oncogenic KRAS.79 STK11 loss is associated with an immunologically ‘cold’ TME characterised by poor T-cell infiltration, reducing responsiveness to G12Ci and other targeted therapies.80 81 In contrast, KEAP1 inactivation drives constitutive NFE2L2 activation, promoting antioxidant defenses and metabolic rewiring that enable tumour cells to withstand oxidative stress and evade cytotoxicity induced by KRAS blockade.81 A meta-analysis of 1224 patients confirmed KEAP1 comutation as a negative predictor of response to adagrasib or sotorasib in KRAS-G12C-mutant NSCLC, showing reduced efficacy compared with KEAP1 WT (OR 0.35).82 A second analysis of 1132 patients validated G12Ci monotherapy efficacy, particularly in elderly and female patients, but highlighted poor outcomes in those with liver or brain metastases, with KEAP1 mutation emerging as the dominant negative prognostic factor.83 Moreover, STK11 mutations and high CD274 expression trended towards resistance but did not reach statistical significance.83 Recent clinical data further confirmed that KEAP1–NFE2L2 pathway activation and low transcription termination factor 1 expression are major determinants of poor sotorasib efficacy, while the KL (KRAS–LKB1/STK11) transcriptional subtype—characterised by STK11 loss, KEAP1 comutation and a non-inflamed ‘cold’ TME—exhibited only modest benefit compared with docetaxel, underscoring the interplay between transcriptional subtype, redox state and immune contexture in shaping KRAS-targeted therapy outcomes.84TP53 mutations, common across cancers, contribute further by promoting heterogeneity and impairing apoptotic priming, thereby enabling survival despite KRAS pathway suppression. In KRAS-G12C-mutant NSCLC, TP53 comutations limit inhibitor efficacy; however, CRISPR-Cas9 and drug screening identified WEE1 G2 checkpoint kinase inhibition as a strategy to restore apoptosis.85 WEE1 blockade enforces premature G2/M transition, impairs DNA damage response signalling and induces mitotic catastrophe, thereby synergising with G12Ci to enhance tumour control in NSCLC mouse models.85 Caution is warranted, however, as WEE1 inhibition carries the risk of dose-limiting toxicities, including myelosuppression and genomic instability in normal proliferating tissues, which may restrict the therapeutic window of this combination.Beyond genomic drivers, the TME may establish an immunosuppressive niche that fosters therapeutic resistance. KRAS-mutant tumours frequently accumulate regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSCs) and tumour-associated macrophages, which collectively dampen cytotoxic T-cell activity through the secretion of inhibitory cytokines (eg, interleukin 10 (IL-10), transforming growth factor beta 1) and the expression of immune checkpoint ligands.86 In parallel, dense stromal fibrosis, driven by activated cancer-associated fibroblasts, creates a physical barrier to immune infiltration and drug delivery, while also releasing growth factors and extracellular matrix components that sustain tumour survival.87Collectively, these intrinsic mutations and extrinsic microenvironmental factors blunt KRASi activity, underscoring the need for rational combination strategies to overcome primary resistance.Acquired and adaptive resistance Although KRASi can induce initial responses, most tumours develop adaptive or acquired resistance within 4–6 months through genetic, epigenetic and metabolic adaptations that restore oncogenic signalling or activate bypass survival pathways. 88 In the KRYSTAL-1 trial, 45% of patients who achieved stable disease for at least 12 weeks eventually progressed on adagrasib.88Resistance can arise through secondary KRAS alterations or activation of alternative RAS isoforms. Mutations within the switch-II pocket—including Y96C/D/H/N, H95D/G/N/R and R68S—directly disrupt drug binding and reduce inhibitor affinity, restoring KRAS activity despite covalent blockade.88 89 Gene amplification of KRAS-G12C increases the total pool of KRAS protein available, saturating inhibitor binding capacity. Parallel activation of RAS family members, such as NRAS-Q61K/L/R or HRAS-G13V, can similarly bypass KRAS inhibition by re-engaging downstream MAPK and PI3K signalling. In addition, loss of the tumour suppressor NF1, a GAP (RAS-GAP), prolongs RAS-GTP loading and sustains downstream proliferative signalling, further undermining KRASi efficacy.88Upstream bypass activation is another major contributor. Amplification or overexpression of RTKs, such as EGFR or mesenchymal–epithelial transition factor (MET), can reactivate MAPK and PI3K pathways independent of KRAS inhibition.90 MET (also known as hepatocyte growth factor receptor) is a potent RTK that drives PI3K–AKT, RAS–MAPK and signal transducer and activator of transcription 3 signalling, promoting cell proliferation, survival and motility. Aberrant MET activity can result from exon 14 skipping mutations, gene amplification or receptor overexpression, and selective inhibitors such as capmatinib and tepotinib are clinically approved for MET-altered tumours.91 In KRAS-G12C-mutant NSCLC models, depletion of MET with siRNA or pharmacologic inhibition using crizotinib restored sensitivity to sotorasib, both in vitro and in MET-amplified xenografts.92 In CRC, EGFR reactivation represents a dominant bypass mechanism, driving resistance to G12Ci. This has led to the clinical development of dual inhibition strategies: in the phase III CodeBreaK 300 trial, sotorasib (960 mg) combined with the EGFR antibody panitumumab significantly prolonged PFS and improved ORRs (30.2%) compared with investigator’s choice chemotherapy, supporting this regimen as a new treatment standard in chemorefractory KRAS-G12C-mutant CRC.37 93 94Downstream reactivation of KRAS effectors also contributes to therapy resistance. Mutations in BRAF (V600E) or MAP2K1/MEK1 (K57N/T or in-frame deletions) restore RAF–MEK–ERK signalling despite upstream KRAS inhibition.88 In resistant cells, SRC–RAF1 interactions drive JUN activation and transcription of ATP binding cassette subfamily C member 1 and cyclin D1, conferring multidrug resistance; this can be reversed by SRC inhibitors such as dasatinib, bosutinib or DGY-06–116.95 PI3K–AKT–mTOR signalling may also be re-engaged by gain-of-function mutations such as PIK3CA-H1047R, promoting survival and therapeutic escape.96 Other rare drivers include RET mutations and fusion events (eg, EML4–ALK, CCDC6–RET, NRF1–BRAF), though their impact requires further clarification.88Epithelial–mesenchymal transition (EMT) is increasingly recognised as a resistance mechanism to G12Ci. Mesenchymal-like cells maintain PI3K signalling and activate RTKs such as FGFR and AXL, which converge on the Hippo signalling pathway.97 98 Yes1 associated transcriptional regulator (YAP1, also known as YAP)-WW domain containing transcription regulator 1 (TAZ)–TEA domain transcription factor (TEAD) transcriptional activity reinforces EMT programmes, as confirmed by genome-wide CRISPR screens in KRAS-G12C-mutant NSCLC.99 Pharmacologic TEAD inhibition (eg, VT3989) has been shown to restore sensitivity to adagrasib in preclinical models.100 Recent studies further highlight that combining KRASi with pan-TEAD inhibitors (eg, IAG933) can produce synergistic antitumour effects and mitigate adaptive resistance, as evidenced by dual blockade of the KRAS–MAPK and YAP/TAZ–TEAD pathways.101 102 Similarly, resistance to BI-2493 is characterised by YAP1/TAZ activation and an immunosuppressive phenotype, suggesting that rational combination strategies integrating KRAS inhibition with pan-TEAD or ICB may effectively overcome Hippo pathway-mediated resistance.Epigenetic rewiring and phenotypic plasticity enable lineage transitions that reduce KRAS dependence. scRNA-seq studies of KRAS-G12C-mutant lung cancer cells exposed to ARS1620 revealed divergent states: one with suppressed growth and reduced KRAS signalling, and another with restored proliferation via RTK-driven nucleotide exchange.103 Integration with CRISPR-based screens identified HB-EGF and aurora kinase A (AURKA) as mediators of adaptive resistance to G12Ci ARS-1620.103 Mechanistically, AURKA binds nascent KRAS-G12C, stabilising KRAS–CRAF interaction and sustaining ERK-driven cell-cycle progression.103 These findings illustrate how transcriptional plasticity and signalling crosstalk converge to bypass KRAS inhibition, emphasising the need for combination strategies that simultaneously target oncogenic drivers and epigenetic rewiring pathways.Many tumours, particularly those with oncogenic KRAS mutations, depend on macropinocytosis to sustain anabolic metabolism in nutrient-poor microenvironments such as PDAC.104 105 In MRTX1133-sensitive PDAC cells, G12Di suppresses macropinocytosis via the indirect inhibition of Rac family small GTPase 1 (RAC1), which operates downstream of KRAS. However, G12Di-resistant cells re-establish the process of macropinocytosis through the activation of RAC1 that occurs via β-catenin and advanced glycosylation end-product specific receptor (AGER, also known as RAGE)–diaphanous related formin 1.106 Macropinocytosis promotes amino acid uptake and glutathione synthesis, thereby restricting cell death, particularly apoptosis.106 Similar AGER-dependent mechanisms contribute to adaptive resistance to G12Ci in NSCLC.106 These findings suggest that targeting AGER-driven macropinocytosis may overcome metabolic resistance and improve the efficacy of KRASi in PDAC and NSCLC.Autophagy is a conserved lysosomal degradation pathway that recycles damaged organelles, misfolded proteins and other cellular components, thereby contributing to drug resistance.107 Early-phase clinical trials combining gemcitabine/nab-paclitaxel with hydroxychloroquine in PDAC demonstrated the feasibility of pharmacologic autophagy inhibition and showed improvements in pathological response and serum CA19-9 reduction.108 109 However, these benefits did not translate into a significant OS advantage, underscoring the limited efficacy of nonspecific autophagy blockade and the need for more potent, selective autophagy inhibitors or mechanistically informed combination regimens with KRAS-targeted therapies. Mechanistic studies in KRAS-G12D-mutant PDAC have revealed that MRTX1133 induces compensatory autophagy through mTOR suppression, leading to enhanced glutathione synthesis, decreased reactive oxygen species accumulation and inhibition of cytochrome c-mediated apoptosis.110 Genetic deletion of autophagy-related 5 or beclin 1, as well as pharmacologic autophagy inhibition with chloroquine, enhances MRTX1133 efficacy in vitro and in mouse models, including PDXs.110 These findings highlight the therapeutic promise of integrating KRAS inhibition with precise autophagy modulation to overcome metabolic adaptation and improve treatment durability in PDAC.111Together, these diverse mechanisms—secondary mutations, pathway reactivation, EMT, lineage plasticity, metabolic rewiring, macropinocytosis and autophagy—underscore the complexity of resistance in KRAS-driven tumours. Overcoming resistance will require integrative strategies that combine KRASi with agents targeting comutations, parallel pathways, metabolic dependencies or the immune microenvironment to achieve durable clinical benefit.Strategies to overcome KRASi resistance As resistance to KRAS-targeted therapies inevitably emerges in most patients, there is an urgent need to develop strategies that sustain or restore therapeutic responses. Current approaches focus on rational combination therapies, targeting bypass signalling and tumour adaptation, and implementing precision medicine guided by real-time molecular profiling. These strategies aim to prevent the outgrowth of resistant clones, resensitise tumours and personalise therapy based on individual tumour biology.Combination therapies One of the most effective approaches to overcome resistance is the rational combination of KRASi with agents that target complementary nodes within the signalling network or immune microenvironment ( online supplemental table S3).KRASi+RTKi/PTPN11i/MEKi/PI3Ki A landmark example is the recent FDA approval (January 2025) of sotorasib in combination with panitumumab, a fully human IgG2 monoclonal antibody against EGFR. Panitumumab blocks receptor dimerisation, autophosphorylation and downstream activation of the RAS–RAF–MEK–ERK and PI3K–AKT pathways. In the CodeBreaK 300 trial, this combination significantly improved PFS compared with standard therapy (5.6 vs 2.0 months; HR 0.48, p=0.005) and achieved a 26% response rate vs 0% with standard of care. 37Additional targets include PTPN11 and SOS1, which function as common nodes for RTK-mediated signalling. PTPN11 inhibitors (eg, TNO155, SHP099, RMC-4550) and SOS1 inhibitors (eg, BAY-293) enhance G12Ci activity or reverse adaptive resistance. These hypotheses have been validated in preclinical models,76 77 98 103 112–116 and such agents are now advancing into clinical trials (eg, NCT04330664).Downstream blockade of the RAF–MEK–ERK or PI3K–AKT–mTOR cascades provides another layer of suppression. For example, selumetinib or trametinib combined with KRASi limits compensatory MAPK rebound, whereas buparlisib or alpelisib counteracts PI3K-driven metabolic resilience.13 117 118 These approaches exemplify ‘vertical’ inhibition strategies that neutralise both the driver mutation and adaptive signalling circuits.Despite encouraging results, challenges remain. Broad pathway inhibition often induces overlapping toxicities, such as dermatologic and gastrointestinal adverse events with MEK-ERK inhibitors or metabolic disturbances with PI3K–AKT–mTOR blockade. Moreover, signalling plasticity enables cancer cells to bypass single-node inhibition. Thus, the future of KRAS-targeted therapy relies on rationally designed combinations and biomarker-driven patient stratification to maximise efficacy while minimising toxicity.KRASi+immunotherapy KRAS-mutant tumours are frequently embedded within an immunosuppressive microenvironment enriched in Tregs, MDSCs and exhausted cytotoxic lymphocytes.119 KRASi can partially remodel this milieu by restoring antigen presentation and enhancing effector T-cell infiltration,29 87 120–122 providing a rationale for combining KRASi with ICB.123 Preclinical studies demonstrate that G12Ci, G12Di, as well as pan-RAS inhibitors, enhance tumour immunogenicity and synergise with PDCD1/PD-1, CD274/PD-L1 or CTLA4 blockade to produce more durable tumour regressions.29 87 120 122 124 125 Conversely, oncogenic KRAS expression reduces T-cell infiltration and impairs ICB efficacy in otherwise immunotherapy-sensitive models such as MC38 colorectal cancer.126Beyond ICB, adoptive T-cell therapies are emerging. A novel TCR (051), isolated from a PDAC patient with KRAS-G12V and restricted by HLA-A11:01, exhibited high specificity and potent antitumour activity when engineered into T cells.127 The resulting TCR-T product (IX001) demonstrated robust efficacy, persistence and safety in preclinical models without detectable off-target toxicity,127 supporting further clinical development for HLA-A11:01 patients with KRAS-G12V-mutant cancers. Similarly, IL-36γ, a proinflammatory IL-1 family cytokine, can activate both innate and adaptive immunity. When delivered via oncolytic viruses, IL-36γ enhances T-cell-mediated responses and synergises with KRASi and ICB.121 G12Ci suppresses IL-36γ expression,103 but combining MRTX1257 with an IL-36γ-armed oncolytic virus restored superior tumour control, further enhanced by PDCD1 blockade.121 The efficacy of these combinations depended primarily on CD8+ T cells, with additional contributions from CD4+ T cells and natural killer cells.121Another innovative approach involves bispecific T-cell engagers (BiTEs). G12Ci generates covalent drug–peptide neoantigens that can be presented by major histocompatibility complex class I molecules. These complexes can be recognised by antibodies and engineered into BiTEs, redirecting cytotoxic T-cell activity specifically against KRAS-G12C-mutant cancer cells.128 129In addition, combining KRASi with metabolic pathway blockade—such as targeting AGER-dependent macropinocytosis with RAGE229 or autophagy with chloroquine—can further enhance KRASi-induced cell death by increasing immunogenicity and activating damage-associated molecular pattern-mediated pathways (eg, high mobility group box 1-driven immunogenic cell death).106 110Clinical translation of KRASi–ICB combinations has produced mixed outcomes. In CodeBreaK 100/101, concurrent sotorasib plus ICB caused high rates of grade 3–4 hepatotoxicity, leading to frequent discontinuations, though lead-in dosing modestly reduced toxicity. By contrast, adagrasib plus pembrolizumab (KRYSTAL-1/7) was associated with mainly low-grade liver events, a more manageable safety profile and encouraging efficacy in CD274–high NSCLC (ORR 63%). Translational studies suggest comutations (eg, KEAP1) may negatively influence outcomes, underscoring the need for biomarker-driven stratification. In PDAC, G12Di such as RMC-9805 have shown preclinical synergy with ICB, although clinical benefit remains uncertain given the profoundly immunosuppressive PDAC microenvironment.Ongoing clinical efforts include the phase 3 CodeBreaK 202 trial (NCT05920356), which is comparing sotorasib versus pembrolizumab, each in combination with platinum doublet chemotherapy, in ~750 patients with KRAS-G12C-mutant, CD274-negative NSCLC. In parallel, Moderna is testing mRNA-5671, an mRNA vaccine targeting KRAS-G12D, G12V, G13D and G12C mutations, in combination with pembrolizumab (NCT03948763). Although initially evaluated with ICB, such vaccines may also be explored alongside KRASi to eliminate drug-resistant clones.A central challenge for all these strategies is optimising dosing and sequencing to minimise immune-related toxicities while maximising efficacy. Predictive biomarkers will be essential to guide patient selection and determine when immune potentiation is most effective.KRASi+radiotherapy Preclinical data support combining KRASi with radiotherapy to deepen tumour control. In KRAS-G12C models, the inhibitor MRTX1257 acted as a radiosensitiser—enhancing radiotherapy-induced tumour control in vitro and in vivo, with the most durable effects observed in immunocompetent settings, consistent with an immune-mediated contribution to efficacy. 130 Mechanistically related work shows that blocking the KRAS pathway downstream (eg, MEK inhibition) augments radiotherapy by boosting antitumor immunity, reinforcing the biological rationale for pairing KRAS-pathway blockade with radiation.131Early clinical translation is underway. The single-arm K-SAB study (NCT06127940) is evaluating induction sotorasib followed by stereotactic body radiotherapy to limited lesions in KRAS-G12C-mutant NSCLC, with feasibility and safety/efficacy readouts pending.Together, these data suggest that radiotherapy can convert the primarily cytostatic effects of KRASi into more durable tumour control by engaging antitumour immunity; ongoing trials will clarify the clinical magnitude and optimal scheduling of this strategy.Precision medicine and biomarker-guided approaches A key component of resistance management is the application of real-time tumour profiling to guide personalised treatment strategies.Multiomics for biomarker discovery Resistance to KRASi is highly heterogeneous, reflecting diverse tumour-intrinsic and microenvironmental adaptations. Integrated multiomics approaches—spanning genomics, transcriptomics, epigenomics and proteomics—are increasingly applied to identify biomarkers that forecast therapeutic response and resistance. For instance, co-occurring mutations in STK11 or KEAP1 are recurrent in KRAS-mutant NSCLC and have emerged as strong negative prognostic factors: STK11 loss is associated with an immunologically ‘cold’ microenvironment with poor T-cell infiltration, while KEAP1 inactivation leads to constitutive NFE2L2 activation, enhancing antioxidant defenses and reducing sensitivity to G12Ci.83 84 132 Moreover, transcriptomic profiling has revealed immune gene expression signatures that correlate with enhanced response to immunotherapy, suggesting that subsets of KRAS-mutant patients may derive greater benefit from KRASi–ICB combinations.133 These insights highlight the importance of molecular stratification: rather than applying KRASi in a ‘one-size-fits-all’ manner, tailoring therapy based on comutation status, immune signatures and other omics-derived biomarkers may improve efficacy and durability of clinical benefit.Real-time monitoring of resistance evolution Because resistance mutations often emerge before radiographic progression, longitudinal tracking through circulating tumour DNA (ctDNA) and other liquid biopsy platforms provides an early-warning system. For example, secondary KRAS mutations (such as Y96D, H95Q/R and R68S) have been identified in patients treated with sotorasib or adagrasib and are associated with impaired drug binding. 88 Similarly, bypass pathway alterations such as MET amplification or EGFR amplification have been detected in ctDNA as mechanisms of resistance, leading to reactivation of upstream RTK signalling.91 Downstream rewiring is also captured: alterations in MAP2K1/MEK1 or BRAF V600E can restore MAPK signalling despite KRAS blockade.88 These events can be detected months before clinical or radiographic relapse, offering a therapeutic window for proactive intervention. Integration of ctDNA profiling into clinical trial designs—such as adaptive treatment arms triggered by emerging resistance mutations—is beginning to guide real-time therapeutic modifications, enabling earlier incorporation of PTPN11 inhibitors, MET inhibitors or next-generation KRASi.Adaptive treatment strategies based on clonal dynamics A deeper understanding of tumour clonal evolution is opening the door to adaptive therapy paradigms. Instead of continuous high-intensity dosing, dynamic schedules that alternate or cycle therapies are being investigated to shape clonal competition. Mathematical and evolutionary models suggest that resistant subclones often carry a fitness cost in the absence of drug, meaning that drug holidays or reduced-intensity intervals can allow sensitive populations to re-expand and suppress resistant lineages, if they compete for the same niche. For example, alternating KRASi with agents that exploit collateral sensitivities—such as DNA damage response inhibitors or metabolic pathway blockers—may exploit vulnerabilities that emerge under selective pressure. Adaptive therapy trials in other cancers (eg, BRAF-mutant melanoma) have already shown delayed resistance with intermittent dosing, providing a framework that is now being extended to KRAS-driven tumours.134 Such strategies emphasise the principle of managing resistance evolution, rather than attempting eradication with continuous monotherapy.Conclusion and outlook The KRAS field has moved from ‘undruggable’ to clinically actionable, yet the durability of the therapeutic effect remains the central obstacle. Across alleles and tumour contexts, therapeutic escape reflects a moving target shaped by tumour-intrinsic plasticity, parallel pathway rewiring, metabolic compensation and an immunosuppressive microenvironment. The next phase should therefore treat KRAS not as a single node but as a networked vulnerability—to be intercepted at multiple layers with schedules and partners chosen by biology, not convention.Three priorities can accelerate progress. First, prospectively multiomics and real-time ctDNA should be embedded into trials to anticipate resistance, trigger on-study adaptations (eg, RTK/PTPN11 or MET add-ons, allele-switching, degraders) and discontinue futile regimens early. Second, rational combinations must be biomarker-driven, matching liabilities (eg, KEAP1/STK11 comutation, EMT/YAP1 programmes, macropinocytosis/autophagy dependence) with pathway-specific partners and immunologic cotherapies; vertical inhibition should be balanced against toxicity with thoughtful sequencing and intermittent or pulsed dosing when supported by pharmacodynamics. Third, it appears advisable to diversify modalities, such as allele-selective ON-state and OFF-state inhibitors, pan-KRAS/RAS agents, targeted degraders, RNA-based silencers, TCR-T, BiTEs, cytokine-armed oncolytic platforms and radiotherapy, and combine them not redundantly but orthogonally to block escape routes.Importantly, recent findings revealed that systemic KRAS ablation in adult mice disrupts haematopoietic homeostasis, leading to myelodysplastic syndromes and chronic myelomonocytic leukaemia, and that treatment with the pan-RAS inhibitor daraxonrasib (RMC-6236) alters myeloid lineage differentiation.135 These results underscore that KRAS signalling is essential not only for tumour maintenance but also for normal myeloid cell function, emphasising the need for careful haematologic monitoring and selective target engagement in future clinical development of pan-KRAS and pan-RAS inhibitors.Trial design will matter as much as drug design. Adaptive platforms that allow response-guided intensification, de-escalation or partner-swapping can align therapy with clonal dynamics. Integration of immune contexture metrics (eg, T-cell inflamed signatures, myeloid programmes) and metabolic readouts (eg, macropinocytosis or redox dependence) should move from correlative endpoints to eligibility and stratification criteria. For immunotherapy combinations, schedule optimisation (lead-in, staggered or maintenance) and organ-specific risk management (eg, hepatotoxicity) are essential to widen the therapeutic window.Finally, success will hinge on interdisciplinary playbooks: structural biology to outpace on-target mutations; systems pharmacology to choreograph dosing and timing; imaging and liquid biopsy to track evolving clones; and rigorous data sharing to harmonise biomarkers across tumour types. If the field embraces this precision-plus-adaptation mindset, KRAS-targeted strategies can shift from short-lived responses to durable disease control, converting a hard-won foothold into sustained clinical benefit.