Drug resistance in cancer: molecular mechanisms and emerging treatment strategies

Abstract

Therapeutic resistance remains a defining challenge in oncology, limiting the durability of current therapies and contributing to disease relapse and poor patient outcomes. This review systematically integrates recent progress in understanding the molecular, cellular, and ecological foundations of drug resistance across chemotherapy, targeted therapy, and immunotherapy. We delineate how genetic alterations, epigenetic reprogramming, post-translational modifications, and non-coding RNA networks cooperate with metabolic reprogramming and tumor microenvironment remodeling to sustain resistant phenotypes. The influence of the microbiome is highlighted as an emerging determinant of therapeutic response through immune modulation and metabolic cross-talk. By summarizing key regulatory circuits, We establishe a unified framework linking clonal evolution, metabolic adaptability, and tumor ecological dynamics. We further synthesizes novel therapeutic strategies that convert resistance mechanisms into therapeutic vulnerabilities, including synthetic lethality approaches, metabolic targeting, and disruption of stem cell and stromal niches. Advances in single-cell and spatial omics, liquid biopsy, and artificial intelligence are emphasized as transformative tools for early detection and real-time prediction of resistance evolution. This review also identifies major translational gaps in preclinical modeling and proposes precision oncology frameworks guided by evolutionary principles. By bridging mechanistic understanding with adaptive clinical design, this work provides an integrated roadmap for overcoming therapeutic resistance and achieving sustained, long-term cancer control.

Introduction

Cancer has become one of the leading causes of mortality worldwide. In 2022, an estimated 20 million new cancer cases were reported globally, with approximately 9.7 million deaths attributed to cancer and its complications [1]. This imposes not only a significant public health burden but also a major threat to human life expectancy [2]. For the vast majority of solid tumors, curative resection remains the primary treatment option [3]. However, for certain early-stage tumors, particularly those with no obvious clinical symptoms and high invasiveness, it is often challenging to achieve early detection and timely intervention, resulting in the loss of the optimal treatment window [4].

To address this challenge, comprehensive treatment strategies centered on drug therapy have become the cornerstone of management for most tumors. From traditional chemotherapy to targeted therapy and immunotherapy, the spectrum of available treatments has continued to expand, leading to significant reductions in mortality rates and improvements in overall survival [5–7]. However, drug resistance, the most fundamental challenge in cancer treatment, is directly and fatally linked to tumors. It is not an isolated complication but a core biological process that causes treatment failure and triggers tumor recurrence and metastasis. The complexity of this issue stems from the extreme diversity of its mechanisms [8]. Tumor cells can evade killing through multiple pathways, including activating drug efflux pumps, inducing target mutations, and activating alternative signaling pathways [9–11]. Moreover, these mechanisms exhibit spatial heterogeneity within tumors and dynamically evolve over the course of treatment.

More alarmingly, the prevalence of drug resistance extends across all mainstream therapies. From the perspective of drug classes, whether chemotherapy agents, targeted therapies, immunotherapeutic agents, or novel modalities such as (ADCs, resistance remains a major concern [12–15]. From the perspective of treatment timelines, resistance may present as either primary (intrinsic) insensitivity or as acquired resistance developing during the course of therapy [16, 17]. Regardless of type, all forms of resistance ultimately lead to tumor recurrence or metastasis, culminating in treatment failure. Therefore, the early identification of resistance types and elucidation of their molecular mechanisms are crucial for optimizing therapeutic strategies and improving clinical outcomes.

This review begins with the current clinical landscape of cancer treatment and examines the various forms of therapeutic resistance, including both intrinsic and acquired types. It further analyzes resistance across different treatment modalities. Based on this framework, we provide a detailed synthesis of resistance mechanisms at multiple biological levels, including gene expression and mutation, epigenetic modifications, protein expression and post-translational modifications, and tumor immune microenvironmental factors. More importantly, we systematically summarize emerging strategies proposed in recent years to overcome drug resistance, with a particular focus on novel therapies. Finally, we highlight limitations in current research and propose future directions that may guide subsequent studies. Through this synthesis and analysis, we aim to provide a valuable reference for elucidating the molecular basis of tumor drug resistance and for developing more effective treatment strategies.

Clinical burden and evolving resistance paradigms

Tumor drug resistance fundamentally limits the clinical benefits of cancer therapy. Currently, approximately 90% of chemotherapy failures and more than 50% of targeted or immunotherapy failures are directly attributable to resistance, which not only impedes improvements in survival rates but also results in substantial waste of medical resources [18]. As precision medicine guided by molecular subtyping advances, the dual paradigm of resistance, intrinsic versus acquired, along with tumor-specific barriers, has become a critical consideration in clinical decision-making. The unique resistance mechanisms associated with targeted therapies, immune checkpoint inhibitors, and ADCs further underscore the dynamic evolutionary capacity of tumors under therapeutic pressure. Therefore, by assessing the hierarchical clinical burden imposed by resistance, deconstructing its taxonomic framework, and elucidating the molecular logic underlying resistance in these three leading therapeutic modalities, a mechanistic roadmap can be established to help overcome current treatment bottlenecks.

Therapeutic failure in practice

In clinical practice, tumor resistance has become a core challenge in cancer management. The emergence of resistance is the principal cause of most tumor recurrences and a major factor contributing to cancer-related mortality [19]. Importantly, this challenge is not confined to any single treatment type but is observed across virtually all malignancies [20]. Resistance is thus a universal obstacle faced by clinicians and researchers alike.

Among available therapeutic options, resistance has been documented in both conventional chemotherapy and newer modalities such as targeted therapy and immunotherapy. Reports indicate that up to 90% of chemotherapy failures are attributable to resistance, a finding consistently observed across most malignant tumors [18]. In breast cancer, resistance to paclitaxel [21], 5-fluorouracil [22], doxorubicin [23], cyclophosphamide [24], and carboplatin [25] not only results in tumor recurrence but also predicts poor patient prognosis. Similarly, in colorectal and gastric cancers, resistance to agents such as 5-fluorouracil [26, 27], and oxaliplatin [28, 29] poses a serious threat to patient survival. Targeted agents and immune checkpoint inhibitors, including small molecules and biologics, have demonstrated significant efficacy but remain subject to resistance. For example, imatinib mesylate, widely recognized for its success in chronic myeloid leukemia and gastrointestinal stromal tumors, eventually encounters resistance [30]. Targeted therapies against human epidermal growth factor receptor 2 (HER2) can develop resistance within one year [31]. In melanoma, immunotherapy resistance can occur within five years, and in non–small cell lung cancer (NSCLC), disease progression occurs in approximately 56% of patients within four years [32].

To address resistance to monotherapy, combination strategies have been proposed. Theoretically, due to differing mechanisms of action and resistance, combination therapy should maximize efficacy while minimizing the likelihood of resistance. However, in clinical reality, combination regimens often fail, particularly in advanced-stage tumors, because of emerging resistance. Moreover, while response rates may improve, multidrug chemotherapy significantly increases treatment-related toxicities [33]. This challenge extends beyond chemotherapy to combinations involving targeted therapies and immunotherapies. For instance, in metastatic melanoma, dual immune checkpoint blockade with anti–PD-1 and anti–CTLA-4 antibodies achieved high initial response rates, but long-term survival benefits remain uncertain, with nearly half of patients experiencing severe treatment-related adverse events [34]. Given the existence of cross-resistance and multidrug resistance mechanisms, combination therapy does not represent a definitive solution to the problem [35].

Clinical categorization of drug resistance

Broadly, drug resistance can be classified into two paradigms: intrinsic resistance (primary resistance) and acquired resistance (secondary resistance) (Fig. 1). Intrinsic resistance refers to a lack of response to initial treatment, indicating that resistance mechanisms pre-exist before therapy begins [16]. Acquired resistance develops during or after treatment, implying that there was an initial therapeutic response followed by the emergence of resistance [36]. This therapeutic escape severely compromises the achievement of complete remission.

Fig. 1.

The evolution of intrinsic and acquired drug resistance

Tumor cells possess remarkable phenotypic plasticity, enabling survival through continuous adaptation under immune surveillance and anticancer therapy. During phenotypic conversion, certain tumor cells undergo Darwinian passive selection and enrichment, ensuring the survival of the fittest, while others actively respond to diverse internal and external stimuli [37]. Under the combined pressures of tumor heterogeneity and drug selection, a minority of tumor cells divide and form distinct subpopulations [38, 39]. These resistant subclones often exhibit dormancy and stem cell–like properties, proliferate slowly, and ultimately result in persistent resistance leading to disease progression. In some cases, they may regain drug sensitivity after treatment cessation [40].

In addition to these general categories, tumor-specific resistance mechanisms pose unique challenges. For example, molecular heterogeneity often drives resistance. In NSCLC, the epidermal growth factor receptor (EGFR) is a well-established therapeutic target. Despite mutation rates of up to 30%, three generations of EGFR tyrosine kinase inhibitors (EGFR-TKIs) have been developed for this patient population, yet resistance inevitably emerges [41]. First-generation EGFR-TKIs such as gefitinib and second-generation agents such as afatinib are effective in patients with and L858R mutations. However, the emergence of the T790M mutation leads to acquired resistance in a substantial proportion of patients within 9–14 months of treatment [42, 43]. Third-generation EGFR-TKIs, exemplified by osimertinib, display improved efficacy against T790M-positive tumors but have driven the appearance of the C797S mutation. Consequently, resistance can arise via both EGFR-dependent and EGFR-independent pathways [44–46].

Microenvironmental characteristics also play pivotal roles in resistance. In pancreatic ductal adenocarcinoma (PDAC), the acellular matrix can constitute up to 90% of tumor volume, displaying extensive fibrosis [47]. Excess extracellular matrix (ECM) deposition elevates interstitial fluid pressure, impairs vascularization, and, when driven by activated cancer-associated fibroblasts (CAFs), creates a physical barrier to drug delivery [48, 49]. Such features significantly limit the penetration of agents like gemcitabine and are associated with poor prognosis. In glioblastoma, vascular abnormalities may disrupt the blood–brain barrier (BBB) and increase permeability; however, this disruption is uneven, and many regions maintain an intact BBB [50]. Moreover, overexpression of efflux pumps further reduces drug concentrations, resulting in diminished therapeutic efficacy [51].

Hematological malignancies, unlike solid tumors, are not impeded by physical barriers or dense ECM but depend on specialized mechanisms such as stem cell dormancy and bone marrow niche dynamics. In chronic myeloid leukemia (CML), the BCR–ABL fusion protein is a classical target for TKI therapy with imatinib [52]. Nonetheless, mutations in SH3 and SH2 kinase domains, alterations in the P-loop via the T315I mutation, and single amino acid substitutions can substantially impair TKI efficacy [53]. In multiple myeloma (MM), which arises from plasma cells, the bone marrow microenvironment (BMME) plays a central role [54]. Interactions between immune cells and MM cells, through both direct contact and soluble mediators, activate multiple signaling pathways, enhancing tumor survival, reducing drug sensitivity, and fostering resistance [55].

Addressing the spatial heterogeneity of tumors, barriers to microenvironmental remodeling, and tumor type–specific resistance mechanisms is essential for establishing a multidimensional understanding of drug resistance and advancing therapeutic strategies.

Resistance in targeted therapy, immunotherapy, and antibody–drug conjugates

Targeted therapy is currently one of the main comprehensive treatment approaches for diverse malignancies. To date, numerous molecular targets across multiple signaling pathways have been identified and applied in clinical practice, yet drug resistance remains an inevitable obstacle. Broadly, the evolution of resistance to targeted therapy involves both target-dependent and target-independent mechanisms (Table 1). The former primarily includes target modifications (mutations, amplifications, fusions) and bypass activation (upregulation of compensatory signaling pathways), whereas the latter encompasses phenotypic conversion (lineage plasticity) and microenvironmental protection (physical barriers, immune suppression), among others.

Table 1.

The main targets of targeted therapy and immunotherapy and their mechanisms of resistance

Signaling pathway	Target	Representative drugs	Main mechanisms of resistance	References
RTKs pathway	EGFR	Osimertinib	Secondary mutations (T790M/C797S) MET/HER2 amplification phenotypic transformation	[56]
	ALK	Lorlatinib	ALK kinase-domain mutations Activation of bypass signalling pathways	[57]
	ROS1	Crizotinib	Secondary mutations (G2032R/L2026M) ROS1 fusion	[58]
	MET	Capmatinib	Secondary mutations (D1228/Y1230) EGFR compensatory activation	[59]
	HER2	Trastuzumab	Intracellular domain truncation (p95-HER2) PI3K/AKT sustained activation	[60]
MAPK pathway	KRAS G12C	Sotorasib	Acquired Y96D/H95Q mutation NRAS/AKT feedback activation	[61]
	MEK	Trametinib	Acquired mutations in MEK1/2 ERK reactivation	[62]
Angiogenesis pathway	VEGFR	Bevacizumab	Alternative vascular growth pathway (FGF) Hypoxia-induced VEGF expression	[63]
	HIF-2α	Belzutifan	Compensatory activation of HIF-1α Acquired mutations in EPAS1	[64]
DDR pathway	PARP	Olaparib	Restoration of BRCA1/2 function 53BP1 deletion Overexpression of drug efflux pump (ABCB1)	[65]
	ATM/ATR	AZD1390	Upregulation of replication fork protection protein	[66]
DDR pathway	Polθ	RP-6685	53BP1 deletion	[66]
Epigenetics pathway	IDH1/2	Enasidenib Ivosidenib	Acquired IDH2 mutation (Q316E) 2-HG metabolic bypass activation	[67]
	BET	RVX208	BRD4 hyperphosphorylation Enhanced Wnt/β-catenin signaling	[68]
Cell cycle pathway	CDK4/6	Palbociclib	RB1 deletion/mutation CDK2/cyclin E amplification FGFR1 signal compensation	[69]
immune checkpoint	PD1/PDL1	Nivolumab Avelumab	T Cell Exhaustion Disfunction of tumor antigen Compensatory Inhibitory Signaling	[70]

RTKs Receptor tyrosine kinases, MAPK Mitogen-activated protein kinase, DDR DNA damage repair

For example, the therapeutic efficacy of trastuzumab, which targets HER2, is influenced not only by expression levels of the p95-HER2 isoform but also by mutations in phosphatidylinositide 3-kinase (PI3K) [71]. Alterations in PI3K catalytic subunit alpha (PIK3CA) or loss of phosphatase and tensin homolog (PTEN) can hyperactivate the PI3K–AKT pathway, thereby promoting resistance to anti-HER2 therapy [72]. Due to the complexity of intracellular signaling and regulatory networks, inhibition of a single target often triggers feedback loops that establish new compensatory pathways, fostering resistance. For instance, capmatinib, a MET inhibitor used in NSCLC and gastric cancer, can induce activation of the EGFR pathway, leading to resistance [73–75].

Immune suppression can also undermine targeted therapy. In tumors resistant to osimertinib, increased macrophage infiltration, heightened M2 polarization, reduced T cell infiltration, and diminished activation collectively create an immunosuppressive microenvironment that compromises therapeutic efficacy [76]. Thus, resistance to targeted therapy reflects a multifaceted process involving molecular alterations, signaling pathway reprogramming, and tumor microenvironmental modulation, underscoring the need for integrative perspectives to decipher its underlying logic.

Immunotherapy likewise demonstrates both primary and secondary resistance. Primary resistance frequently occurs in so-called “cold tumors”, characterized by low antigenicity, minimal or absent T cell infiltration, and an immunosuppressive microenvironment, resulting in poor responsiveness [77]. For example, prostate cancer is considered an “immune desert” with low T cell infiltration and weak immunogenicity, severely limiting the efficacy of immune checkpoint inhibitors [78, 79]. Even “hot tumors” are not uniformly responsive. In renal cell carcinoma, for example, although immunotherapy is a first-line option, immune exhaustion and “cold” subtypes persist, remaining insensitive to immune checkpoint blockade but potentially responsive to targeted agents [80].

Tumor heterogeneity, both inter- and intra-tumoral, significantly impacts immunotherapy outcomes. Even among initial responders, more than 60% of NSCLC patients and 30% of those with metastatic melanoma eventually develop resistant lesions, leading to recurrence [81, 82]. Mechanistically, resistance, whether primary or secondary, ultimately converges on T cell dysfunction and the accumulation of immunosuppressive cells, with immune checkpoint inhibitors failing to trigger effective immune responses. Current research therefore focuses on restoring effector cell function and overcoming immune evasion, including the development of molecular classification systems based on biomarkers to guide patient stratification and precision treatment, thereby mitigating resistance development [83]. Additionally, breaking away from traditional pathological classification concepts and establishing tumor molecular classification characteristics based on molecular markers will aid in patient stratification, thereby establishing more targeted precision treatment models, which will help mitigate the development of immunotherapy resistance.

ADCs combine targeted delivery with potent cytotoxic payloads, offering high efficacy with reduced systemic toxicity. However, resistance to ADCs has emerged as a substantial barrier to their clinical benefit. Target antigen–dependent resistance, characterized by reduced antigen expression, is common. For example, high HER2 expression correlates with better responses to breast cancer ADCs [84], whereas reduced expression or heterogeneity adversely affects outcomes [85]. Defects in endocytosis and lysosomal processing also contribute. ADCs require internalization and lysosomal degradation to release their cytotoxic payload. Impaired expression of endophilin A2 (Endo II) reduces HER2 internalization in breast cancer, diminishing response to T-DM1[86]. Abnormal caveolin-1 (CAV1) mediated endocytosis has been observed in resistant cell lines, lowering uptake efficiency [87]. Non-cleavable ADCs transported into cells are degraded in the acidic environment of lysosomes and by active lysosomal enzymes. However, in T-DM1-resistant tumor cell lines, not only were lysosomal alkalization and impaired lysosomal protease activity observed, but also abnormal V-ATPase (a proton pump regulating lysosomal acidification) activity was detected [88, 89].

Resistance can also arise from modifications in the payload target or from drug efflux. For instance, in non-Hodgkin lymphoma resistant to microtubule inhibitor monomethyl auristatin E (MMAE)-containing ADCs such as pinatuzumab vedotin and polatuzumab vedotin, multidrug resistance 1 (MDR1) upregulation mediates enhanced drug efflux [90]. Resistance to sacituzumab govitecan (SG) has been associated with topoisomerase-1 (TOP1) mutations (E418K and frameshift) and a novel tumor-associated calcium signal transducer 2 (TACSTD2) and its encoded protein trophoblast cell surface antigen-2 (TROP2) T256R missense mutation, both of which impair drug activity [91]. Overall, ADC resistance is a multifactorial process involving alterations at every stage of drug targeting, internalization, processing, and payload action, necessitating comprehensive mechanistic understanding for the development of next-generation ADC strategies.

Molecular mechanisms underpinning drug resistance

Drug resistance in cancer is a dynamic, multifactorial adaptive process driven by complex interactions among host genetic background, intrinsic tumor cell plasticity, immune system dynamics, and the tumor microenvironment (TME) [92, 93]. As treatment paradigms have evolved from conventional chemotherapy to targeted therapy, immunotherapy, and combination regimens, the spectrum of resistance mechanisms has become increasingly diverse and complex. Across cancer types, tumor genetic and phenotypic heterogeneity, clonal evolution, and therapeutic pressure collectively shape the temporal and spatial dynamics of resistance, posing persistent clinical challenges [94–97].

This section provides a systematic overview of the core molecular mechanisms underlying drug resistance, encompassing genetic alterations, epigenetic and transcriptional reprogramming, proteomic and post-translational adaptations, and TME-mediated resistance (Fig. 2). By summarizing classical mechanisms alongside recent research advances, we aim to deepen the understanding of resistance diversity across cancer types and therapeutic modalities, thereby informing prediction and intervention strategies.

Fig. 2.

Molecular mechanisms underpinning drug resistance. Multiple mechanisms collectively mediate tumor resistance, including genetic mechanisms, epigenetic mechanisms, post-translational modifications, and the tumor microenvironment. TAM: Tumor-associated macrophages; MDSC: Myeloid-derived suppressor cells; Treg: Regulatory T cells; TAN: Tumor-associated neutrophils; CAF: Cancer-associated fibroblasts

Genetic mechanisms

Genetic alterations represent some of the earliest and most fundamental drivers of drug resistance. By modifying drug target structures, activating alternative signaling pathways, or enhancing cellular survival and repair capacity, these alterations confer selective advantages to tumor cells under therapeutic pressure (Fig. 2). Common genetic changes include acquired driver mutations, gene amplifications, fusions, and deletions, all of which have been implicated in resistance across multiple tumor types and treatment modalities [98–100].

In chemotherapy, genetic alterations can impair drug efficacy through various mechanisms [101]. TP53 mutations are particularly prevalent, reducing apoptosis in response to DNA damage and conferring broad resistance to chemotherapeutic agents [102, 103]. Upregulation of MDR1 leads to P-glycoprotein–mediated drug efflux, a well-established multidrug resistance mechanism [104, 105]. Resistance may also stem from mutations in direct drug targets. For example, TOP1 mutations (p.L617I, p.R621H, p.E710G) reduce the formation of TOP1–DNA cleavage complexes in response to camptothecin derivatives, leading to resistance to topoisomerase I inhibitors [106, 107]. β-tubulin gene mutations identified in paclitaxel-resistant cell lines reduce drug binding affinity to microtubules, lowering sensitivity [108, 109].

In targeted therapy, resistance often arises from mutations within the drug target or activation of bypass and downstream signaling pathways [110]. Steric hindrance mutations introduce bulkier amino acids at the drug-binding interface, obstructing inhibitor access. For example, the BCR–ABL1 T315I gatekeeper mutation blocks imatinib binding [111, 112]. Altered affinity mutations, such as EGFR T790M mutation increases ATP binding, reducing first- and second-generation EGFR inhibitor efficacy [113, 114]. Conformational change mutations disrupt the inactive conformation of kinases or stabilize their active state, sustaining kinase activation. For example, ALK mutations F1174L and L1196M stabilize the active conformation of ALK, reducing sensitivity to first-generation ALK inhibitors like crizotinib [115, 116]. Bypass signaling activation, such as PIK3CA mutations or PTEN loss activating PI3K/AKT, KRAS/NRAS/BRAF mutations reactivating ras proto-oncogene (RAS)/mitogen-activated protein kinases (MAPK), or janus kinase 2 (JAK2)/signal transducer and activator of transcription 3 (STAT3) activation, can restore proliferation despite target inhibition [117–122].

Immunotherapy resistance is frequently driven by genetic defects in antigen presentation and interferon-gamma (IFN-γ) signaling [110]. The former disrupts the stability of the MHC-I complex, leading to loss of MHC-I expression on tumor cells and preventing T cell-mediated recognition and killing, with B2M mutations or deletions being a key example [123, 124], while JAK1/JAK2 loss-of-function mutations impair IFN-γ responsiveness, reducing PD-L1 and MHC-I induction [125–127]. Additional alterations include ZNF689 deficiency, which promotes genomic instability and impairs antigen presentation [128], and recombination activating gene 21 (RAD21) amplification, which cooperates with yes-associated protein 1 (YAP)/TEA domain family member 4 (TEAD4) to suppress interferon signaling genes such as STAT1 and interferon regulatory factor 9 (IRF9) [129].

Epigenetic and transcriptional reprogramming

Epigenetic regulation and transcriptional reprogramming are central to the development and modulation of drug resistance in cancer [130, 131]. Epigenetic regulation refers to heritable changes in gene expression that occur without alterations in the DNA sequence, mediated by mechanisms such as DNA methylation, histone modifications, chromatin remodeling, and non-coding RNAs (Fig. 2). These processes shape chromatin states and determine transcriptional potential [132]. Transcriptional reprogramming describes the dynamic adjustment of transcription factor activity, chromatin accessibility, and regulatory networks in response to external pressures, enabling tumor cells to establish gene expression programs that support adaptation and survival [133, 134]. Epigenetic alterations provide the permissive chromatin landscape that facilitates such transcriptional plasticity under stress [135, 136]. Dysregulation of these processes can silence tumor suppressor genes, remodel signaling pathways, alter the immune microenvironment, and drive metabolic adaptation, ultimately undermining therapeutic efficacy [137–139]. Importantly, epigenetic interventions, such as DNA methyltransferase and histone deacetylase inhibitors, have shown promise in partially reversing resistance [140, 141].

DNA methylation is a classical epigenetic mechanism involving the addition of a methyl group to cytosine residues, typically forming 5-methylcytosine, which alters chromatin conformation and transcriptional activity [142]. The mechanisms by which DNA methylation contributes to drug resistance vary across different therapeutic modalities [143]. In chemotherapy, hypermethylation-mediated silencing of tumor suppressor genes and regulation of DNA damage repair pathways are common. For instance, MutL homolog 1 (MLH1) promoter hypermethylation impairs mismatch repair, reducing sensitivity to platinum-induced DNA damage and promoting chemoresistance [144]. In targeted therapy, DNA methylation can modulate oncogenic signaling, alter target expression, or affect drug transport and metabolism. In clear cell renal cell carcinoma, QPCT promoter hypomethylation enhances nuclear factor kappa-B (NF-κB)-mediated transcription, stabilizes HRAS, and activates the extracellular regulated protein kinases (ERK) pathway, promoting sunitinib resistance [145]. In immunotherapy, DNA methylation can suppress antigen presentation, inhibit T cell–attracting chemokines, and induce immune evasion. For example, in hepatocellular carcinoma, Riplet promoter hypermethylation increases fatty acid synthase stability, enhances palmitic acid–dependent STAT3 activation, and induces CD8⁺ T cell exhaustion, reducing PD-1 blockade efficacy [146].

Histone modifications regulate gene expression by altering chromatin accessibility via covalent modification of histone tails [147]. Key modifications include acetylation, methylation, phosphorylation, and ubiquitination [148]. For example, trimethylation of histone H3 lysine 9 (H3K9me3) catalyzed by SUV39H1 promotes heterochromatin at the androgen receptor variant 7 (AR-V7) promoter in prostate cancer, sustaining AR-V7 expression and conferring resistance to enzalutamide [149]. Recently identified modifications such as lactylation, succinylation, citrullination, and crotonylation also affect transcription and resistance [150–152]. In colorectal cancer, hypoxia-induced lactate accumulation increases H3K18 lactylation (H3K18la), upregulating RUBCNL and activating autophagy, thereby mediating bevacizumab resistance [153]. In glioblastoma, lysine catabolism increases crotonyl-CoA and histone H4 crotonylation, suppressing interferon signaling and enabling immunotherapy resistance [154]. Furthermore, alterations in histone modification pathways also represent critical genetic events in tumor progression. For example, mutations in lysine demethylase 6A (KDM6A), exert gender-dual effects in urothelial carcinoma: enhancing immunotherapy response in males but promoting angiogenesis and poor outcomes in females, revealing mechanisms underlying gender differences in tumor immunotherapy response [155].Chromatin remodeling, the ATP-dependent repositioning of nucleosomes, regulates transcription factor access and gene expression [156, 157]. In EGFR-mutant NSCLC, SWI/SNF complexes sustain accessibility at oxidative stress–responsive loci, supporting NRF2 pathway activation and osimertinib resistance [158]. AP-1 transcription factors, particularly FOS-like antigen 1 (FOSL1) and JUN, cooperate with chromatin remodeling to drive resistance-specific transcriptional programs [159]. The presence of mutations in chromatin remodeling genes like ARID1A, coupled with high CD8 + T cell infiltration, is linked to improved prognosis in urothelial carcinoma patients receiving adjuvant chemotherapy and immunotherapy [160]. These findings underscore how chromatin remodeling facilitates resistance not merely through individual gene regulation but by orchestrating global, adaptable transcriptional networks [161].

Non-coding RNAs (ncRNAs), including microRNA (miRNA), long non-coding RNAs (lncRNA), and circular RNA (circRNA), modulate gene expression by influencing chromatin structure, mRNA stability, translation, and signaling pathways [162, 163]. In the context of cancer therapy resistance, ncRNAs contribute to adaptive survival under treatment stress by targeting critical signaling pathways, regulating drug transport and metabolism, and modulating apoptosis and immune evasion, thereby constituting a key epigenetic mechanism across multiple treatment modalities [162, 164, 165]. In breast cancer, miR-1275 downregulation derepresses MDK, activates PI3K/AKT, and increases stemness, promoting epirubicin resistance [166]. In clear cell renal cell carcinoma, lncARSR functions as a competing endogenous RNA (ceRNA) for miR-34 and miR-449, upregulating AXL and c-MET and activating downstream AKT and STAT3 signaling to drive sunitinib resistance [167]. In colorectal cancer, circPHLPP2 enhances interleukin-36 gamma (IL36γ) transcription via interleukin enhancer binding factor 3 (ILF3), suppressing natural killer (NK) cell infiltration and effector function, promoting anti-PD-1 resistance [168]. Collectively, these examples highlight the diverse roles of ncRNAs in driving therapy resistance. Beyond these, additional classes of ncRNAs, such as PIWI-interacting RNA (piRNA) and tRNA-derived fragments (tRFs), have also been implicated in drug resistance, pointing to an increasingly complex and still-evolving regulatory network [169–171].

RNA modifications represent a crucial layer of post-transcriptional gene regulation, and a wide array of chemical modifications have been identified across various RNA species [172]. Among the most well-characterized are N6-methyladenosine (m6A), N1-methyladenosine (m1A), 5-methylcytosine (m5C), pseudouridine (Ψ), and adenosine-to-inosine (A-to-I) editing [173]. These modifications affect splicing, stability, localization, translation, and degradation of mRNA, tRNA, rRNA, and lncRNA, thereby participating in fundamental cellular processes [174]. In the context of tumor progression, dysregulation of RNA modification regulators, including writers, erasers, and readers, has been widely implicated [175]. Through modulating oncogenic signaling pathways, enhancing stemness, promoting immune evasion, or enabling metabolic reprogramming, RNA modifications play central roles in maintaining tumor phenotypes and facilitating therapeutic resistance [176]. Among various RNA modifications, m6A is frequently highlighted due to its well-defined regulatory roles [177, 178]. Methyltransferase like 3 (METTL3), an m6A “writer,” is upregulated in lenvatinib-resistant hepatocellular carcinoma, enhancing EGFR mRNA translation and activating ERK signaling [179]. METTL3 also promotes PD-L1 mRNA stability via insulin-like growth factor 2 mRNA binding protein 3 (IGF2BP3), thereby impairing T cell-mediated cytotoxicity and reducing anti–PD-L1 therapy efficacy in breast cancer [180]. In parallel, m6A “erasers” also contribute to drug resistance. The m6A demethylase AlkB homolog 5 (ALKBH5) is upregulated in cisplatin-resistant ovarian cancer, where it forms a positive feedback loop with the transcription factor homeobox A10 (HOXA10), and promotes resistance by demethylating JAK2 mRNA, preventing its degradation by YTHDF2 and activating the JAK2/STAT3 signaling pathway [181]. Other modifications such as m1A and m7G, mediated by tRNA methyltransferase 6 (TRMT6)/TRMT61A and METTL1/WD repeat domain 4 (WDR4), have also been increasingly associated with therapy resistance, further highlighting the multifaceted roles of RNA modifications in shaping drug response [182–185].

During therapy, drug-induced selective pressure can trigger transcriptional reprogramming that remodels gene networks to sustain resistance [186]. Transcription factors such as cellular myelocytomatosis oncogene (c-MYC), forkhead box protein M1 (FOXM1), and hypoxia-inducible factor 1 alpha (HIF-1α) promote stemness, epithelial–mesenchymal transition (EMT), and immune evasion [187–189]. For instance, in chemotherapy-resistant breast cancer, MYC cooperates with myeloid cell leukemia sequence 1 (MCL1) to enhance oxidative phosphorylation and maintain cancer stem cell properties [190]. Enhancers, especially super-enhancers (SEs), can increase chromatin accessibility and recruit transcriptional coactivators, thereby amplifying the transcription of resistance-related genes and facilitating tumor cell adaptation under therapeutic pressure [191, 192]. In poly ADP-ribose polymerase (PARP) inhibitor (PARPi)–resistant ovarian cancer, lactate accumulation induced by enhanced glycolysis promotes H4K12la histone modification at the RAD23A promoter and its associated super-enhancer (Nira-SE), which facilitates MYC recruitment and drives RAD23A overexpression, ultimately enhancing DNA repair capacity and conferring niraparib resistance [193].

Post-translational adaptations and modifications

Post-translational modifications (PTMs) of proteins profoundly influence their stability, localization, and activity, thereby regulating essential cellular processes and contributing to drug resistance in cancer [194, 195] (Fig. 2).

The dynamic regulation of ubiquitination and de-ubiquitination is a central mechanism in resistance to chemotherapy, targeted therapy, and immunotherapy by precisely regulating key proteins in tumor cells [196]. Alterations in these processes can stabilize drug transporters such as MDR1 and breast cancer resistance protein (BCRP) or DNA damage repair (DDR) proteins, enhancing drug efflux, promoting DNA repair, and facilitating cell survival [197–199]. Regulation of target protein stability also influences the sensitivity of targeted therapy and immunotherapy. For instance, ubiquitin-specific peptidase 28 (USP28) enhances SRY-box 9 (SOX9) stability and DNA damage repair, mediating resistance to PARP inhibitors [200]. Inhibiting USP8 increases PD-L1 and MHC-I expression, thereby improving responsiveness to PD-1/PD-L1 blockade [201].

Phosphorylation, typically regulated by kinase activity, modulates downstream signaling cascades and contributes to resistance. In renal cell carcinoma, reciprocal activation between ERK2 and T-LAK cell–derived protein kinase (TOPK) promotes sorafenib resistance in a phosphorylation-dependent manner [202]. Phosphorylation can also interact with ubiquitination to regulate protein stability. For example, in prostate cancer, methionine 1-linked ubiquitination (M1-Ubi) of PTEN at K144 and K197 inhibits its phosphatase activity, accelerating tumor progression and reducing sensitivity to enzalutamide [203]. Additionally, NDR1 phosphorylates the E3 ligase F-box protein 11 (FBXO11), enhancing β-catenin ubiquitination, promoting its degradation, and inhibiting nuclear translocation, thereby suppressing metastasis [204].

Acetylation, especially lysine acetylation, is another key PTM in resistance. On one hand, histone acetyltransferases (HATs) such as CBP and p300 can upregulate resistance-related genes, promoting tumor progression and drug resistance [205]. On the other hand, histone deacetylases (HDACs) can induce chromatin remodeling and epigenetic changes that counteract therapy [206]. In renal cell carcinoma, HDAC1-mediated deacetylation of p53 promotes lipid accumulation and sorafenib resistance [207]. Targeting acetylation, particularly through HDAC inhibitors, is therefore a promising strategy.

Emerging PTMs such as lactylation and SUMOylation have gained attention. Lactylation, catalyzed by lactate, modifies lysine residues on histones and non-histone proteins, regulating resistance pathways [208]. For example, H3 lysine 18 lactylation (H3K18la) mediates cisplatin resistance in bladder cancer [209], while H3K9la accumulates in the LUC7L2 promoter, promoting its transcriptional expression and temozolomide resistance in glioblastoma [210]. Non-histones such as X-ray repair cross-complementing 1 (XRCC1) undergo lactylation at the lysine 247 site (K247), which enhances transport to the cell nucleus and strengthens DNA repair capacity, thereby mediating glioblastoma resistance to therapy [211]. SUMOylation, the covalent attachment of small ubiquitin-like modifiers (SUMO) to proteins, can modulate their function, localization, and stability [212]. Upregulation of the SUMO protein E2 enzyme, UBC9, correlates with increased resistance in breast and liver cancers, and targeting UBC9 can reverse resistance [213]. SUMO2/3 modification of Ku80 at K307 inhibits oxaliplatin-induced apoptosis in colorectal cancer [214].

Importantly, PTM pathways often interact rather than operate independently. For example, SUMOylation can crosstalk with phosphorylation, acetylation, and ubiquitination to regulate drug sensitivity [215]. Such interplay highlights the complexity of PTM networks in resistance and the necessity of integrative strategies when developing interventions.

Tumor microenvironment-mediated resistance

TME plays a pivotal role in mediating drug resistance by providing physical barriers, biochemical protection, and a dynamic ecosystem that enables tumor cells to evade immune clearance, adapt to therapeutic pressure, and activate survival pathways [216] (Fig. 2).

Immune cells, as the primary components of the TME, comprise both anti-tumor and immunosuppressive populations. Tumor-associated macrophages (TAMs) can suppress T cell activity by secreting cytokines such as transforming growth factor-β‌ (TGF-β) and by releasing exosomes, as well as by upregulating PD-L1 expression, thereby mediating resistance to immune checkpoint inhibitors (ICIs) [217]. Other immunosuppressive cells, including myeloid-derived suppressor cells (MDSCs), tumor-associated neutrophils (TANs), regulatory T cells (Tregs), and tumor-associated dendritic cells (tDCs), also contribute to immune evasion and tumor progression [218]. Tumor-associated fibroblasts (CAFs) produce dense extracellular matrix (ECM) components such as collagen, fibronectin, and hyaluronic acid, while secreting pro-survival factors like hepatocyte growth factor (HGF) [219]. These actions not only create physical and biochemical barriers that hinder drug penetration but also directly promote tumor survival and suppress T cell and NK cell activity. Additionally, CAFs can directly suppress T cell and NK cell infiltration and activity, and by inducing Tregs and TAMs, they shape an inhibitory immune microenvironment [220].

The tumor vasculature also contributes to resistance. Newly formed microvessels are often in the TME tortuous and leaky, reducing drug delivery efficiency and fostering hypoxia [221]. HIFs induced under these conditions promote resistance to targeted therapies such as TKIs and suppress anti-tumor immunity by increasing the production of metabolic byproducts like lactate [222]. Hypoxia also drives nutrient competition between tumor and immune cells, particularly for glucose, thereby impairing T cell and NK cell cytotoxicity and pushing the TME toward an immunosuppressive state [223, 224]. Furthermore, hypoxia-mediated acidification lowers the efficacy of weakly basic chemotherapeutics like adriamycin, whereas pH modulation toward alkalinity can mitigate resistance [225]. Endothelial cells within the vasculature can sense oxidative stress and secrete cytokines such as fibroblast growth factor 2 (FGF2) and chemokine (C-X-C motif) ligand 12 (CXCL12), directly inducing resistance and promoting immune escape by upregulating PD-L1, recruiting MDSCs, and inhibiting T cell function [226–229].

The ECM plays multiple critical roles in tumorigenesis and progression by providing mechanical support, regulating the microenvironment, and serving as a source of signaling molecules [230]. ECM–integrin interactions, particularly involving β1-integrin, αvβ3, and αvβ5, promote cancer stem cell (CSC) phenotypes and activate survival pathways such as PI3K/AKT and focal adhesion kinase (FAK), conferring drug resistance [231–233]. ECM components like TGF-β and type I collagen enhance CSC marker expression, reinforcing stemness [234, 235], while hyaluronic acid (HA) supports CSC maintenance via Twist- and TGFβ–Snail-dependent signaling [236]. ECM stiffness also influences therapy response. Increased stiffness enhances YAP/TAZ activity, promoting resistance to targeted agents in breast cancer and melanoma [237, 238]. Stiffness-associated resistance can involve augmented DNA repair. For instance, MAP4K4/6/7-mediated ubiquitin phosphorylation is reduced in tumor cells adjacent to stiff ECM, enhancing double-strand break repair and decreasing drug sensitivity [239]. Moreover, dense ECM deposition forms a physical barrier that impedes immune cell infiltration, while collagen fibers can directly damage T cells and impair cytotoxic function [240–242]. ECM-mediated YAP/TAZ nuclear translocation can also upregulate PD-L1 expression in tumor cells, further suppressing immune responses [243].

Emerging strategies to overcome or prevent resistance

Tumor drug resistance remains a formidable challenge in modern oncology, often likened to the “Sword of Damocles” hanging over current treatment paradigms. Building on an in-depth understanding of the underlying mechanisms, the next step is to develop strategies that can overcome resistance and restore anti-tumor efficacy. Contemporary research addresses this problem from molecular, cellular, and tumor ecosystem perspectives, with the ultimate goal of achieving truly precise cancer therapy.

Targeting resistance mechanisms directly

One of the most direct strategies involves designing specific inhibitors to target known drug-resistant mutations and their associated signaling pathways (Fig. 3 and Table 2). This approach relies on precise molecular characterization of resistance mechanisms. For example, fourth-generation EGFR-TKIs have been developed to address the T790M mutation, which confers resistance to first- and second-generation EGFR-TKIs, as well as the C797S mutation, which emerges after treatment with third-generation agents like osimertinib. These novel inhibitors not only retain activity against sensitive mutations such as L858R but also inhibit EGFR +/T790M and EGFR +/T790M/C797S variants, thereby overcoming resistance [244]. In chronic lymphocytic leukemia (CLL), covalent Bruton tyrosine kinase (BTK) inhibitors exert their activity by binding to cysteine residue 481 of the BTK protein [245]. When this residue mutates, covalent inhibitors lose binding ability and efficacy. In such cases, non-covalent BTK inhibitors such as pirtobrutinib can act independently of BTK mutation status, showing clinical activity in patients resistant to ibrutinib [246].

Fig. 3.

Emerging strategies to overcome or prevent resistance. The current main strategies for treating tumor resistance include directly targeting resistance mechanisms, targeting the vulnerabilities of resistant cells, repairing the tumor ecosystem, and using precision and adaptive treatment methods. TAM: Tumor-associated macrophages; MDSC: Myeloid-derived suppressor cells; Treg: Regulatory T cells; CAF: Cancer-associated fibroblasts

Table 2.

Representative clinical studies currently conducted on strategies to overcome tumor drug resistance (2023–2025)

Agents	Purpose/Result	Status	Register ID
177Lu-PSMA-I&T/ Abiraterone/ Enzalutamide	Comparing the safety and efficacy of 177Lu-PSMA-I&T versus hormone therapy in patients with mCRPC	Active, not recruiting	NCT05204927
KN046/ Regorafenib/ Apatinib	Evaluating the efficacy and safety of KN046 in combination with regorafenib or apatinib for digestive system cancers resistant to PD-1/PD-L1 blockade	Recruiting	NCT06099821
B013/ Paclitaxel	Evaluating the efficacy and safety of B013 in patients with platinum-resistant recurrent ovarian cancer	Recruiting	NCT06434610
ILDR/ Anti-PD-1/PD-L1	Evaluating the efficacy and safety of combining intestinal low dose radiotherapy and PD-1/PD-L1 inhibitors for metastatic malignant solid tumors after acquired resistance to anti-PD1/PD-L1 treatment	Recruiting	NCT07071103
LTC004/ Toripalimab	Evaluating the efficacy and safety of LTC004 in combination with toripalimab in patients with advanced solid tumors resistant to first-line immunotherapy	Recruiting	NCT06490068
Cadonilimab/ Pemetrexed/ Anlotinib	Evaluating the efficacy and safety of cadonilimabin combination with pemetrexed and anlotinib for treatment of elderly patients with T790M-negative advanced non-squamous NSCLC following resistance to EGFR-TKI	Recruiting	NCT06277674
TVB-2640/ Enzalutamide	Evaluating the TVB-2640 administered in combination with enzalutamide in Men with mCRPC	Recruiting	NCT05743621
AB-1015	Evaluating the safety and efficacy of AB-1015 in patients with resistant/refractory epithelial ovarian cancer	Active, not recruiting	NCT05617755
Enhertu	Evaluating the efficacy and safety of Fam-Trastuzumab Deruxtecan-Nxki (T-DXd) as a subsequent line of therapy in HER2-positive metastatic castration-resistant prostate adenocarcinoma	Recruiting	NCT06610825
HRS-5041	Evaluating the safety, tolerability, pharmacokinetics and efficacy of 5041–103 in subjects with mCRPC	Recruiting	NCT06830850
Decitabine/ Carboplatin/ Paclitaxel/ Selinexor	Evaluating the efficacy and safety of combination of the hypomethylating agent decitabine and the nuclear export receptor XPO-1 inhibitor selinexor to reverse platinum resistance in relapsed/refractory epithelial ovarian cancer	Recruiting	NCT05983276
Inosine/Chemotherapy agents	Exploring the Efficacy of Inosine Reversing Chemo Resistance in Triple Negative Breast Cancer	Completed	NCT06355024
Adebrelimab/ Fuzuloparib	Evaluating the efficacy and safety of adebrelimab combined with fuzuloparib in the treatment of patients with recurrent platinum-resistant ovarian cancer	Recruiting	NCT05753826
Pembrolizumab/ Encorafenib/ Binimetinib	Evaluating the efficacy of transient addition BRAF and MEK Inhibitors to overcome primary resistance to immunotherapy in metastatic melanoma patients	Completed	NCT05304546
CUSP06	Evaluating the safety, tolerability, pharmacokinetics, and efficacy of CUSP06 in patients with platinum-refractory/resistant ovarian cancer and other advanced solid tumors	Recruiting	NCT06234423
SON-1010	Assessing the safety, tolerability, and PK/PD of SON-1010 in combination with atezolizumab administered to patients with advanced solid tumors and patients with platinum-resistant ovarian cancer	Recruiting	NCT05756907
Lenvatinib/ VIC1911	Testing the safety and efficacy of lenvatinib in combination with Aurora kinase A inhibitor VIC-1911 in participate with lenvatinib-unresponsive or lenvatinib-resistant hepatocellular carcinoma	Recruiting	NCT05718882
Nivolumab/ Ipilimumab	Evaluating the nivolumab and ipilimumab in patients with MSI/dMMR mCRC resistant to anti-PD1 monotherapy	Recruiting	NCT05310643
JANX007/ Darolutamide	Assessing the safety, tolerability, pharmacokinetic, pharmacodynamic, and the preliminary efficacy of JANX007 in adults with mCRPC	Recruiting	NCT05519449
HS-20093	Study to evaluate the efficacy, safety, tolerability and pharmacokinetic of HS-20093 as a monotherapy in subjects with mCRPC and other solid tumors	Recruiting	NCT06001255
LG002	Investigating the safety and efficacy of Neo-DCVac combined with ICIs in the treatment of advanced lung cancer resistant to ICIs	Recruiting	NCT06329908
Cabozantinib/ Nivolumab	Evaluating the combination of cabozantinib and nivolumab in subjects with advanced castration-resistant prostate cancer	Recruiting	NCT05502315
SX-682/ Enzalutamide	Studing the combination of SX-682 plus enzalutamide in men with mCRPC who have failed abiraterone	Recruiting	NCT06228053
BAY3546828	Evaluating the safety, tolerability, pharmacokinetics, and antitumor activity of actinium-225-macropa-pelgifatamab (BAY 3546828) in participants with advanced mCRPC	Recruiting	NCT06052306
SV-102	Evaluating the safety, tolerability, and efficacy of SYNC-T Therapy SV-102 and to identify the maximum tolerated dose (MTD) and/or selected dose for phase 2b study	Recruiting	NCT06533644
Gedatolisib/ Darolutamide	Evaluating the safety, preliminary efficacy, and pharmacokinetics of gedatolisib in combination with darolutamide in subjects with mCRPC	Recruiting	NCT06190899
TORL-1–23	Evaluating the safety and efficacy of TORL-1–23 in patients with advanced platinum-resistant epithelial ovarian cancer	Recruiting	NCT06690775
ZA-001	Evaluating the safety, whole-body distribution and radiation dosimetry of ZA-001 in mCRPC	Completed	NCT06359821
FPI-2265	Evaluating the safety and efficacy of FPI-2265 in patients with PSMA-positive mCRPC	Recruiting	NCT06402331
Apalutamide/ Carotuximab	Study of apalutamide with carotuximab in patients with mCRPC	Recruiting	NCT05534646
L-TIL/tislelizumab	Investigating the safety and efficacy of Liquid Tumor Infiltrating Lymphocytes (L-TIL) plus tislelizumab as second line therapy for PD-1 inhibitor resistant advanced NSCLC Patients	Recruiting	NCT05878028
OPB-101	Evaluating the safety and efficacy of OPB-101 in platinum-resistant ovarian cancer	Recruiting	NCT07030907
Carboplatin/ 177Lu-PSMA-617	Study of 177Lu-PSMA-617 plus carboplatin in mCRPC	Recruiting	NCT06303713
Docetaxel/ Carboplatin	Investigating platinum and taxane chemotherapy in mCRPC patients with alterations in DNA damage response genes	Recruiting	NCT06439225
JSB462/ AAA617	Study of JSB462 (Luxdegalutamide) in combination with lutetium (177Lu) vipivotide tetraxetan in adult male patients with PSMA-positive mCRPC	Recruiting	NCT07047118
225Ac-LNC1011/ 68 Ga-PSMA-11	Exploring the safety and efficacy of 225Ac-labeled LNC1011 for treating patients with PSMA-positive mCRPC	Recruiting	NCT07117760
HLD-0915	Assessing of the safety and efficacy of HLD-0915 as monotherapy in patients with mCRPC	Recruiting	NCT06800313
Dalpicilib/ Cetuximab	Evaluating the efficacy and safety of cetuximab combined with dalpicilib compared to cetuximab monotherapy in patients with HPV-negative, anti-PD-1-resistant recurrent or metastatic head and neck squamous cell carcinoma	Recruiting	NCT06935188
Bicalutamide/ Sunitinib	Evaluating the safety and efficacy of bicalutamide in combination with sunitinib in patients with receptor tyrosine kinase inhibitor resistant renal cell carcinoma	Recruiting	NCT06222593
Nimotuzumab/ Capecitabine	Exploring the efficacy and safety of a combination regimen of Anti-PD1 monoclonal antibody, nimotuzumab, and capecitabin in treating recurrent or metastatic nasopharyngeal carcinoma patients who have failed first-line platinum-based chemotherapy	Recruiting	NCT06259721
CD19/CD22 -CAR T cells	Investigating the safety, tolerability, and pharmacokinetic properties of human CD19-CD22 targeted T cells infusion for refractory/relapsed leukemia/lymphoma patients with or without central nervous system involvement	Recruiting	NCT06213636

mCRPC Metastatic Castration-Resistant Prostate Cancer, ILDR Low-dose radiotherapy to the intestine, MSI Microsatellite Instability, dMMR Deficient mismatch Repair, NSCLC Non-small cell lung cancer

Data sources: clinical registration website (https://clinicaltrials.gov)

The core of the strategy targeting resistance mechanisms directly lies in developing highly selective targeted drugs that address the direct molecular causes of resistance. Thus, by converting the mechanism of resistance from therapeutic endpoint into a starting point for drug design, this strategy marks a shift from passive adaptation to active elimination, enabling more precise targeting of tumor evolutionary escape routes.

Exploiting vulnerabilities in resistant cells

Drug resistance is an outcome of evolutionary selection. While resistant cells survive by activating compensatory pathways or undergoing phenotypic changes, these adaptations may create new dependencies that can be exploited therapeutically. This concept aligns with synthetic lethality, targeting vulnerabilities such as DNA repair, metabolic reprogramming, or lineage plasticity, thereby transforming cancer cells' adaptive traits into fatal weaknesses (Fig. 3 and Table 2).

In platinum-resistant ovarian cancer, resistance acquisition is accompanied by increased ataxia-telangiectasia-mutated-and-Rad3-related kinase (ATR)-checkpoint kinase 1 (CHK1) activity. When PARP inhibitor–resistant cells are treated with a combination of PARPi and ATR inhibitors, sensitivity is restored through synergistic replication fork stalling, increased double-strand breaks, and apoptosis [247]. In KRAS ^G12C-mutant lung and colorectal cancers, combining KRAS ^G12C inhibitors adagrasib with sevenless homologue 1 (SOS1) inhibitors (BI-3406) or src homology region 2-containing protein tyrosine phosphatase 2 (SHP2) inhibitors disrupts MRAS-driven receptor tyrosine kinase (RTK) feedback activation, enhancing anti-tumor efficacy and delaying resistance onset [248].

Metabolic targeting is another approach. In ibrutinib-resistant mantle cell lymphoma (MCL), glutamine synthase (GLS) overexpression correlates with glutamine dependency and metabolic levels. Inhibiting GLS with telaglenastat in combination with ibrutinib has shown synergistic activity [249]. In refractory melanoma, simultaneous inhibition of anti-apoptotic proteins MCL1 and BCLXL has achieved greater tumor suppression compared to single-agent inhibition [250].

Therefore, targeting the vulnerability of drug-resistant cells shifts cancer treatment from simply blocking signaling pathways to applying evolutionary principles, aiming to achieve long-term disease control by exploiting resistance-induced weaknesses.

Modulating the tumor ecosystem

Drug resistance is rarely an isolated cellular event. It is shaped by the tumor’s broader ecological niche, including the physical barrier of the stroma, the immunosuppressive cellular infiltrate, and the metabolic environment. Targeting this ecosystem seeks to dismantle the protective niche of resistant clones (Fig. 3).

Degrading tumor stroma and normalizing vasculature can enhance drug delivery. HA is one of the main components of the extracellular matrix. By regulating the gel fluid pressure of the tumor interstitium, it can reduce the delivery of anticancer drugs [251]. In PDAC, pretreatment with pegylated recombinant human hyaluronidase (PEGPH20) reduces HA levels, lowers interstitial fluid pressure, and improves paclitaxel distribution, thereby enhancing efficacy [252]. Novel anti-angiogenic agents such as vanucizumab, which simultaneously inhibits VEGF-A and angiopoietin-2 (Ang-2), thereby significantly reducing abnormal tumor angiogenesis [253]. More importantly, this treatment approach provides a transient vascular normalization window that significantly enhances the efficacy of radiotherapy and chemotherapy [254].

Reprogramming the immune microenvironment is also promising. Macrophages, particularly M2 macrophages, have a unique inhibitory role in the tumor microenvironment, and the colony-stimulating factor 1 (CSF-1) receptor (CSF-1R) plays a crucial role in the recruitment and differentiation of monocytes into pro-tumor M2 macrophages and their survival [255]. On the other hand, inhibiting CSF-1R promotes the infiltration and activation of CD3 + CD8 + T cells in the tumor microenvironment, suppresses tumor immune escape, and enhances the antitumor activity of PD-1/PD-L1 inhibitors, thereby overcoming resistance to PD-1/PD-L1 axis blockade [256]. Enhancing antigen presentation processes has similar immunotherapy-sensitizing effects. Wu et al. found that higher levels of antigen-presenting mast cells (apMCs) in breast tissue are associated with enhanced anti-PD-1 therapy efficacy, while the allergy medication, cromolyn, can activate apMC-mediated T cell immunity and enhance tumor sensitivity to anti-PD-1 therapy, providing a new therapeutic strategy to overcome anti-PD-1 resistance [257].

Advanced drug delivery systems can overcome physical barriers. Nanocarriers engineered for receptor-mediated uptake, transporter exploitation, or lipophilic transcellular passage can cross the BBB in glioblastoma, enabling multifunctional delivery of agents that regulate metabolism, induce ferroptosis, or modulate immunity [258–260]. Light-responsive nanoplatforms targeting the TME can induce reactive oxygen species (ROS) production, enable multimodal imaging, and trigger immunogenic cell death in PDAC [261]. Co-delivery of gemcitabine and MnFe₂O₄ exhibit dual effects of chemotherapy and induction of ferroptosis, with this synergistic effect significantly enhancing treatment sensitivity [262]. These approaches all open new avenues for improving PDAC drug resistance and enhancing treatment sensitivity.

Precision medicine and adaptive therapeutic approaches

The key to achieving precise treatment for tumor drug resistance lies in the early and dynamic identification of the evolution of drug resistance (Fig. 3). Therefore, establishing an early drug resistance warning model is an important means of achieving drug resistance monitoring. Jones et al. detected mutations in the mismatch repair (MMR) genes MSH2 and MSH6 in the plasma of glioma patients [263]. These mutations are not only the most common circulating genetic alterations following temozolomide treatment but are also detected earlier than recurrence, thereby achieving the goal of early resistance warning through circulating tumor DNA (ctDNA). Similarly, in colorectal cancer, ctDNA has been used for early identification of tumor resistance, monitoring of treatment response, and recurrence warning [264]. Some of the latest technologies are also being used for resistance monitoring. For example, single-cell transcriptomics and spatial transcriptomics technologies can not only identify resistant tumor cell subpopulations but also map tumor evolution at single-cell resolution [265]. This not only deepens our understanding of resistance mechanisms but also provides a more efficient way to discover new intervention targets [209, 266–268].

Building on these, large models established by deeply integrating multi-omics data such as epigenomics, genomics, transcriptomics, proteomics, metabolomics, and pharmacogenomics using artificial intelligence (AI) technology can not only further explore the underlying logic of tumor resistance but also enable early prediction of tumor resistance [269–271]. More importantly, by classifying and stratifying tumors with different gene expression characteristics, it can strongly support clinical decision-making and significantly enhance the precision of tumor treatment [272].

Additionally, some adaptive treatment modalities beyond traditional standard therapies have made important contributions to improving tumor treatment sensitivity. For example, in prostate cancer with PTEN deletion, Qi et al. found that intermittent administration of PI3K inhibitors, compared to daily dosing, more effectively activates CD8 + T cell-dependent antitumor immune responses, promoting sustained responses to anti-PD-1 therapy [273]. Based on the dynamic changes in the androgen receptor (AR) gene and AR-V7 in circulating tumor cells, researchers have established a method to predict sensitivity to chemotherapy and endocrine therapy in metastatic castration-resistant prostate cancer (mCRPC) [274]. Following docetaxel treatment for mCRPC, patients with liquid biopsy-negative AR and AR-V7 status benefited from combination therapy with enzalutamide, while those with positive status did not.

These measures effectively support the entire process of tumor drug resistance monitoring, prediction, and intervention, thereby promoting a paradigm shift in drug resistance control from empirical treatment to data-driven precision treatment.

Innovative therapeutic approaches

Microbiome modulation

The human microbiome, particularly the gut microbiota, has emerged as a critical regulator of therapeutic response in oncology (Fig. 4). Clinical evidence increasingly shows that microbial dysbiosis contributes to both primary and acquired resistance across multiple tumor types through three interconnected mechanisms: immunomodulation via metabolite-mediated T cell polarization, enzymatic drug metabolism altering chemotherapeutic bioavailability, and barrier disruption that promotes pro-tumorigenic inflammation.

Fig. 4.

Clinical research progress regarding drug resistance. This schematic illustrates three innovative approaches targeting therapeutic resistance: microbiome-targeted therapies, PROTAC technology, AI-driven clinical decision support

Modulating the microbiome, through approaches such as fecal microbiota transplantation (FMT), precision probiotics, and phage-targeted decolonization, represents a paradigm-shifting strategy for resensitizing refractory malignancies. This section synthesizes recent advances in microbiome engineering for overcoming resistance, highlighting biomarkers and intervention frameworks with translational potential in precision oncology.

Antibiotic treatment has been shown to diminish the efficacy of PD-1 blockade in cancer patients, although the underlying mechanisms remain incompletely understood. Fidelle et al. demonstrated that antibiotic-induced downregulation of mucosal addressin cell adhesion molecule-1 (MAdCAM-1) in the ileum facilitates gut recolonization by Enterobacteriaceae [275]. This, in turn, reprograms gut-tropic α4β7⁺CD4⁺ regulatory T17 cells toward tumor dissemination. Moreover, low serum levels of soluble MAdCAM-1 were correlated with poor prognosis in independent cohorts of lung, kidney, and bladder cancer patients, suggesting that the MAdCAM-1–α4β7 axis could be a targetable gut immune checkpoint in cancer immunosurveillance. Joachim et al. developed an in vivo model in which C57BL/6J mice received oral supplementation with the bacterial metabolite deaminated tyrosine (DAT) under the regulation of type I interferon (IFN-I) [276]. DAT delayed tumor growth and enhanced the effects of anti-CTLA-4 or anti-PD-1 immune checkpoint inhibitors, effects that were dependent on host IFN-I signaling. Oral DAT supplementation altered gut microbiota composition, increasing bacterial taxa associated with favorable responses to immunotherapy.

Research into traditional Chinese medicine has also provided insights. Huang et al. investigated whether combining ginseng polysaccharides (GPs) with an anti-PD-1 monoclonal antibody could enhance therapeutic efficacy via modulation of the gut microbiota [277]. They found that the combination upregulated the microbial metabolite valeric acid, downregulated L-tryptophan, and reduced the kynurenine/tryptophan ratio, thereby decreasing regulatory T cell abundance and increasing effector T cell activity. This suggests that GPs in combination with anti-PD-1 may improve immunotherapy sensitivity in NSCLC.

Rodríguez-García et al. examined the role of urolithin A (UroA), a metabolite produced by the gut microbiota, in multiple myeloma (MM). Analysis of a retrospective cohort of 45 patients revealed that UroA synergized with bortezomib in vitro, indicating its potential as an adjunctive therapy to overcome resistance in MM [278]. Wu et al. investigated the inhibitory effect of flaxseed lignans (FL) on breast cancer biological behavior and evaluated the role of FL in enhancing the anticancer efficacy of PD-1/PD-L1 inhibitors (PDi) [279]. Following conversion of FL to enterolactone (ENL) by the gut microbiota, FL administration suppressed BC progression. ENL inhibited malignant BC behavior by downregulating CD38—a key gene associated with immunosuppression and resistance to PD-1/PD-L1 blockade. Consequently, FL enhances the anticancer efficacy of PDi through modulation of gut microbiota and host immunity.

Prior to the study by Xu et al., the potential role of the gut microbiota in modulating gastric cancer (GC) sensitivity to oxaliplatin remained uninvestigated [280]. Antibiotic treatment diminished the therapeutic efficacy of oxaliplatin in GC mouse models, an effect transferable to germ-free mice via fecal microbiota transplantation (FMT), suggesting gut microbiome involvement in oxaliplatin response. Further metabolomic data revealed that metabolically active Akkermansia muciniphila potentiated oxaliplatin efficacy. The researchers demonstrated that A. muciniphila-derived phenethylamine (PEA), acting as a glycolysis inhibitor, enhanced oxaliplatin response in gastric cancer cells by directly counteracting FUBP1 activity.

Current studies report inconsistent findings regarding the gut microbiome as a biomarker for ICIs response. Kim et al. identified TANB77, a previously uncultured and unique bacterial taxon, as exhibiting the highest enrichment in responders through a meta-analysis of ten independent ICIs-treated cohorts [281]. Murine models with higher gut TANB77 abundance showed improved responses to anti-PD-1 therapy. Furthermore, mice administered intraperitoneal injections of TANB77-derived proteins demonstrated enhanced anti-PD-1 responses, providing in vivo evidence for the therapeutic benefit of pirin-like proteins. These findings suggest that pirins from the TANB77 clade may potentiate ICIs responses across diverse cancer patient populations.

PROTACs in early-phase trials

PROTAC (Proteolysis-Targeting Chimera), a new biomedical tech, is used for selective and targeted protein degradation. It works by using the cell's natural ubiquitin–proteasome pathway. The PROTAC molecule has three parts: a targeting ligand that identifies and binds to the target protein; a linker that connects the parts; and an E3 ubiquitin ligase ligand that recruits the E3 ubiquitin ligase. When the PROTAC molecule binds to the target protein, the E3 ubiquitin ligase is brought close, leading to the target protein being marked by ubiquitination and then degraded by the proteasome (Fig. 4). This tech has advantages like strong targeting, high efficiency, and the ability to degrade traditionally undruggable targets. It shows great potential in treating tumors, neurodegenerative diseases, and other conditions, as well as in basic biological research.

Receptor-interacting serine/threonine-protein kinase 1 (RIPK1) functions as a critical stress sentinel coordinating cell survival, inflammation, and immunogenic cell death (ICD). While RIPK1's catalytic activity is essential for triggering cell death, its non-catalytic scaffolding function mediates potent pro-survival signaling. Mannion et al. synthesized a PROTAC that specifically degrades both human and murine RIPK1 [282]. PROTAC-mediated RIPK1 depletion disrupted TNFR1 and Toll-like receptor 3/4 (TLR3/4) signaling hubs, potentiating NF-κB, MAPK, and IFN signaling outputs. This potentiated ICD, enhanced antitumor immunity, and elicited sustained therapeutic responses. Consequently, targeting RIPK1 via PROTAC technology emerges as a promising strategy to overcome resistance to radiotherapy or immunotherapy and augment anticancer therapies.

Targeted protein degradation (TPD) regulates protein levels by redirecting E3 ligases via small molecules to ubiquitinate novel substrates, marking them for proteasomal destruction. TPD has recently emerged as a pivotal strategy in drug discovery. Schröder et al. engineered a potent DCAF1-BRD9 PROTAC, providing an alternative approach to address intrinsic resistance toward VHL-based degraders [283]. This highlights the potential of DCAF1-PROTACs as a promising strategy to overcome ligase-mediated resistance in clinical settings.

Pharmacological studies have led to the discovery of vepdegestrant (ARV-471), a selective, orally available and potent estrogen receptor (ER) degrader, which is a PROTAC—based small molecule. Gough et al. used biochemical and intracellular target—binding experiments to clarify how it works [284]. They also used ER + pre—clinical breast cancer models, including those with wild—type (WT) and mutated ESR1, to confirm that ER degradation can hinder tumor growth, suggesting vepdegestrant might be a more effective foundation for estrogen therapy for ER +/HER2—breast cancer patients. RIPK has a scaffolding function, causing resistance to ICIs and emerging as a promising target to enhance cancer immunotherapy. To address the challenge of the vague binding pocket in RIPK1's intermediate domain, Yu et al. used PROTAC technology to develop the RIPK1 degrader LD4172 [285]. As reported in the study, the RIPK1 degrader LD4172 can serve both as a chemical probe to investigate the scaffolding function of RIPK1 and as a potential therapeutic agent to increase tumor response to ICIs therapy.

Triple-negative breast cancer (TNBC) is the most aggressive breast cancer subtype, marked by high heterogeneity and invasiveness, with limited treatment options. Guo et al. used PROTAC technology to develop potential protein arginine methyltransferase 5 (PRMT5) degraders in vitro and in vivo. YZ-836P, a promising compound, showed cytotoxicity against TNBC cells after 48 h, reducing PRMT5 and kruppel-like factor 5 (KLF5) protein levels [286]. These findings position YZ-836P as a strong candidate for advancing TNBC treatment.

The chemotherapeutic modulating abilities of chemotherapy drugs are promising for addressing the low immunogenicity, immunosuppressive lactic microenvironment, and adaptive immune resistance in colorectal cancer. Zhao and colleagues developed self-assembling, self-delivering nanoPROTACs (DdLD NPs) containing doxorubicin (DOX) and dBET57, stabilized and facilitated by DSPE-PEG2000 [287]. DdLD NPs enhance the stability, cellular delivery, and tumor targeting of DOX and dBET57. They effectively kill colorectal cancer cells and induce ICD. These self-delivering nanoPROTACs may pave the way for chemo-enhanced tumor immunotherapy.

Osteosarcoma (OS) is the most common malignant bone tumor, and c-MET is a recognized therapeutic target. However, traditional c-MET inhibitors are limited by acquired drug resistance and side effects. c-MET—targeted PROTACs offer better antitumor effects by overcoming drug resistance, but their safety is still a concern due to the lack of tumor—targeting ability. Fu et al. generated four AS1411—SL1 chimeras and evaluated their therapeutic effects both in vitro and in vivo [288]. These AS1411—SL1 chimeras are likely to be promising c-MET degraders in osteosarcoma targeted therapy.

In NSCLC, EGFR is a key target as 60% of cases express it. However, resistance to EGFR inhibitors and limitations in strategies for both TKI—sensitive and mutant NSCLC patients remain concerns. Vartak et al. developed a nano—lipid EGFR and BRD4 degrader PROTAC (EPRO and BPRO) for lung cancer [289]. Using refined hydrolysis, the hydrophobic molecules were encapsulated in EGFR—targeted nano—lipid carriers (T—BEPRO). In mice with tumors, T—BEPRO administered intravenously achieved a remarkable tumor growth inhibition (TGI) rate of 77.6% and had lasting tumor suppression, surpassing drug—only treatment.

Artificial intelligence-powered clinical decision support

The formidable challenge of cancer drug resistance, characterized by extreme complexity, heterogeneity, and dynamic evolution across tumor types, remains a critical bottleneck in clinical oncology. Traditional approaches to predicting resistance patterns and optimizing subsequent therapeutic strategies often struggle to harness the vast, multidimensional datasets generated in modern oncology practice, encompassing clinical records, multi-omics profiles (genomic, transcriptomic, proteomic), medical imaging (radiomics/pathomics), and real-time monitoring data. This gap has catalyzed the rapid emergence of AI-based Clinical Decision Support Systems (CDSS) as a transformative frontier in resistance management. By leveraging sophisticated machine learning (ML) and deep learning (DL) techniques, including natural language processing and multimodal data fusion, AI-CDSS platforms are increasingly being deployed to decode intricate resistance mechanisms, predict individual patient risk trajectories with enhanced accuracy and timeliness, and generate evidence-based, personalized therapeutic recommendations (Fig. 4). This rapidly evolving field is transitioning from proof-of-concept studies towards tangible clinical integration, ultimately improving patient outcomes across diverse malignancies.

ML-based clinical decision support

ML algorithms learn patterns and relationships directly from historical data without explicit programming. Within CDSS, machine learning excels at predictive modeling by identifying patients at high risk for adverse events, disease recurrence, or, crucially, treatment resistance. It further enables risk stratification of patients based on predicted outcomes or treatment responses, and facilitates pattern recognition by detecting subtle associations within structured data.

The application of ML to cancer-specific pharmacogenomic datasets shows significant promise for identifying predictive response biomarkers, thereby enabling personalized treatment. Chia et al. introduced the precision oncology platform CAN-Scan, which utilizes ML technology to analyze next-generation pharmacogenetic datasets generated from cryopreserved biobank specimens of patient-derived primary cell lines (PDCs) [290]. CAN-Scan uncovered prognostic biomarkers and alternative therapeutic strategies, particularly for patients non-responsive to first-line chemotherapy. This approach demonstrates significant potential for improving biomarker discovery and guiding personalized therapy.

ICIs have revolutionized cancer treatment for multiple tumor types. However, a substantial proportion of patients treated with CPIs fail to derive benefit or experience only transient responses. Chen et al. developed a fibroblast senescence-associated transcriptomic signature (FSS), which is highly correlated with tumor-promoting signaling pathways and immune dysregulation that foster tumor progression [291]. Utilizing the FSS, a ML framework demonstrated remarkable accuracy in predicting ICI response and survival outcomes, achieving superior AUC values across validation, testing, and internal cohorts. Most notably, the FSS consistently outperformed established signatures in predicting robustness, encompassing diverse cancer subtypes. Fomin et al. employed a machine learning-based approach utilizing the de-identified, nationwide Flatiron Health-Foundation Medicine clinico-genomic database for NSCLC to identify genomic markers predictive of clinical response to ICIs therapy [292]. Building on this, they discovered multiple genomic markers and pathways that reveal the biological mechanisms influencing ICIs therapy response, potentially enhancing response rates to CPI therapy in NSCLC patients. Ricciuti et al. performed comprehensive tumor genomic sequencing, ML-based tumor-infiltrating lymphocyte (TIL) evaluation, multi-color immunofluorescence staining, and/or HLA-I immunohistochemistry (IHC) analysis on matched pre- and post-ICI treatment tumor biopsy samples from NSCLC patients treated at the Dana-Farber Cancer Institute who developed acquired ICI resistance [293]. This multi-modal analysis revealed the genomic and immunophenotypic heterogeneity underlying ICI resistance in NSCLC. To explore mechanisms of ICIs resistance, Sahni et al. developed the Immunotherapy Resistance cell–cell Interaction Scanner (IRIS), a machine learning model designed to identify cell type-specific tumor microenvironment ligand-receptor interactions associated with ICIs resistance [294]. They proposed a robust ICIs response biomarker, highlighting the critical role of downregulating ligand-receptor interactions related to chemotaxis in suppressing lymphocyte infiltration in resistant tumors.

Given that many patients fail to derive durable benefit, Hamidi et al. leveraged machine learning to analyze RNA sequencing data, targeted DNA panel results, immunohistochemistry, and digital pathology from 2,803 UC patients across four phase III randomized trials [295]. This approach identified four distinct transcriptomic subtypes predictive of response to the PD-L1 inhibitor atezolizumab.

Neoadjuvant chemotherapy and immunotherapy aim to eliminate residual tumor cells and reduce recurrence risk. However, drug resistance during neoadjuvant treatment poses a significant obstacle. Zhou et al. integrated advanced technologies including single-cell transcriptomics, whole-genome sequencing, RNA sequencing, proteomics, machine learning, and in vitro/vivo experiments [296]. By analyzing cross-cancer cohorts, they utilized single-cell sequencing to investigate the association between the efficacy of neoadjuvant chemotherapy/immunotherapy and RNA methylation. Multi-omics analysis coupled with machine learning algorithms identified genomic variations, transcriptional dysregulation, and prognostic correlations of RNA methylation regulators (RMRs), revealing distinct molecular subtypes that guide pan-cancer stratification for neoadjuvant therapy.

Chen et al. developed a machine learning-based systematic combinatorial design strategy to identify the most promising drug combinations for patients with relapsed/refractory (R/R) acute myeloid leukemia (AML) [297]. This predictive approach leverages single-cell transcriptomic data and monotherapy response profiles from primary patient samples to identify targeted combinations capable of selectively inhibiting therapy-resistant cancer cells within individual AML patient samples. Furthermore, in initial experiments using clinical trial samples, the method predicted the clinical efficacy of venetoclax-azacitidine combination therapy in AML patients. Collectively, this combined computational and experimental approach provides a rational pathway for identifying personalized combination therapies for R/R AML patients, targeting therapy-resistant leukemia cells and thereby enhancing its potential for clinical translation. Dysregulation of alternative splicing (AS) is increasingly recognized as a pivotal factor in the pathogenesis, disease progression, and therapeutic resistance of B-cell acute lymphoblastic leukemia (B-ALL). Zhuo et al. developed a prognostic model based on 18 AS events (18-AS), refined through the sophisticated integration of bioinformatic approaches and advanced machine learning algorithms [298]. These findings illuminate the role of AS events as novel prognostic biomarkers and therapeutic targets, advancing personalized therapeutic strategies in B-ALL management. Ianevski et al. described a machine learning-based approach, scTherapy, which leverages single-cell transcriptomic profiles to prioritize screening of multi-target therapeutic regimens for individual patients with hematological malignancies or solid tumors [299]. They established a broadly applicable strategy for identifying personalized treatment regimens capable of selectively and combinatorially inhibiting malignant cells while sparing non-cancerous cells, thereby enhancing the likelihood of clinical success.

Breast cancer patients may initially benefit from cytotoxic chemotherapy but subsequently develop treatment resistance and recurrence. Chemotherapy-resistant breast cancer stem cells (BCSCs) play a pivotal role in cancer relapse and metastasis. Leveraging a machine learning strategy, Sun et al. developed an mRNA-based BCSC signature to assess cancer stemness in primary breast cancer patient samples [300]. Through this BCSC signature, they revealed the critical role of polyamine synthesis in regulating chemotherapy-induced BCSC enrichment and proposed novel therapeutic avenues for breast cancer treatment.

Over 50% of patients with refractory high-grade serous carcinoma (HGSC) retain homologous recombination proficiency, rendering them resistant to platinum-based agents and PARP inhibitors. Tamura et al. conducted a comprehensive investigation of this tumor type by integrating machine learning analysis of large public datasets with a novel mouse oviduct-based genetically engineered HGSC organoid model [301]. This approach ultimately identified combination therapy involving conventional chemotherapy and mTOR inhibitors as a potential treatment strategy for HGSC, with p62 emerging as a significant biomarker.

DNA damage repair plays a pivotal role in HCC, driving tumorigenesis, progression, and treatment response [302]. Hong et al. proposed an innovative machine learning framework for the precise assessment of DDR, leveraging both scRNA-seq and bulk RNA sequencing data [303]. This model revealed dynamic interactions between DDR and NK cells and B cells within the primary HCC microenvironment. By shaping an immunosuppressive microenvironment that promotes tumor growth through metabolic reprogramming, the framework was ultimately utilized to predict the overall survival of HCC patients and their resistance to PD-1 therapy.

The efficacy of induction chemotherapy (IC) as a first-line treatment for advanced nasopharyngeal carcinoma (NPC) remains debated, and reliable biomarkers to predict its response are lacking. Tang et al. developed an AI-based radiomics approach to identify metabolic biomarkers through a discovery cohort-based machine learning methodology, with validation in a cohort simulating clinically challenging real-world scenarios [304]. The study revealed that dysregulation of plasma lipoproteins may drive IC resistance in NPC patients. The predictive model constructed from plasma metabolite profiles demonstrated strong predictive power and real-world generalizability. These findings hold significant implications for therapeutic strategy development and may offer potential targets to enhance IC efficacy.

Leveraging recent advances in machine learning, Zhao and colleagues investigated the impact of tumor mutations on the response to common therapeutic agents (conferring resistance) [305]. The resulting predictive model integrated numerous genetic alterations distributed across multiple molecular complexes. Applied to cisplatin-treated cervical cancer patients, the model revealed the regulatory role of the RTK-JAK-STAT complex in drug resistance, enabling a quantitative and interpretable assessment of drug response.

DL-based clinical decision support

DL is a powerful subset of machine learning that utilizes multi-layered artificial neural networks. The key strength of DL lies in its ability to automatically learn hierarchical representations and complex features from raw, high-dimensional, unstructured data. For instance, in medical imaging analysis (Radiomics/Pathomics), DL extracts intricate patterns beyond human perception from computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET) scans, and histopathology slides, aiding in tumor characterization, treatment response assessment, and early detection of drug resistance. Similarly, for genomic/transcriptomic data interpretation, DL models complex interactions within high-dimensional omics data to uncover resistance mechanisms and biomarkers.

Relying solely on single-modality data often fails to comprehensively capture the complex heterogeneity among patients, including variations in resistance to anti-HER2 therapy and differences in the efficacy of combination treatment regimens, particularly in the management of HER2-positive gastric cancer [306]. Chen et al. collected multimodal data, encompassing radiological, pathological, and clinical information, from a cohort of 429 patients [307]. They further introduced a deep learning model, termed the Multimodal Mode, which integrates these diverse data streams to enable precise prediction of treatment response. This model achieved an AUC score of 0.821 for overall response prediction, with a remarkable AUC of 0.914 specifically for predicting response to combined immunotherapy. This study thus underscores the critical importance of multimodal data analysis in enhancing treatment evaluation and advancing personalized medicine for HER2-positive GC. The rapidly expanding pool of scRNA-seq data provides large-scale bulk gene expression databases for drug screening, enabling the identification of optimal clinical applications for anticancer drugs by investigating heterogeneity in drug responses across cancer cell subpopulations. Chen et al. proposed scDEAL, a deep transfer learning framework, for predicting cancer drug response at the single-cell level [308]. A key innovation of scDEAL is its integration of drug-related bulk RNA-seq data with scRNA-seq data, coupled with the transfer of models trained on bulk RNA-seq data to predict drug response in scRNA-seq data. scDEAL facilitates the investigation of cell reprogramming, drug selection, and drug repurposing to enhance therapeutic outcomes.

The identification of molecular features that mediate clinically aggressive phenotypes in prostate cancer remains a significant biological and clinical challenge [309, 310]. Elmarakeby and colleagues developed P-NET, a biologically informed deep learning model, to stratify prostate cancer patients based on treatment-resistant status and to evaluate the molecular drivers of treatment resistance through the model's full interpretability, thereby enabling therapeutic targeting [311]. They demonstrated that P-NET can utilize molecular data to predict cancer states, outperforming other modeling approaches. Their work substantiates that biologically informed, fully interpretable neural networks can enable preclinical discovery and clinical prediction in prostate cancer, with potential broad applicability across multiple cancer types. Gallagher et al. applied deep reinforcement learning (DRL) to guide adaptive drug dosing regimens [312]. They demonstrated that these regimens extended the time to disease progression by more than twofold compared to current adaptive protocols in mathematical models dynamically aligned with prostate cancer. This underscores DRL's capability to develop therapeutic strategies in novel or complex settings. Collectively, the DRL-generated personalized treatment regimens outperformed clinical standard-of-care regimens across all scenarios.

Cyclin-dependent kinase 4 and 6 inhibitors (CDK4/6i) have revolutionized the treatment landscape for breast cancer. However, objective responses are achieved in fewer than 50% of patients, and nearly all patients eventually develop resistance during therapy [313, 314]. Park et al. constructed an interpretable deep learning model, based on a reference atlas of multiprotein complexes in cancer, to predict response to the CDK4/6i drug palbociclib [315]. The predictions were applicable to both patients and PDX models, a capability unattainable with single-gene biomarkers. This study enabled the comprehensive assessment of how a tumor's genetic features modulate resistance to CDK4/6i. Furthermore, trastuzumab, a monoclonal antibody targeting HER2, is an effective therapy for metastatic breast cancer [316]. However, a subset of patients develop resistance to this treatment, making the monitoring of its therapeutic efficacy crucial [317]. Kim et al. described a deep learning-assisted method based on surface-enhanced Raman spectroscopy (SERS) immunoassay to monitor trastuzumab efficacy [318]. This method specifically targets HER2-overexpressing exosomes in murine urine. By employing SERS-deep learning analysis to monitor drug efficacy in urine exosomes from trastuzumab-treated mice, they confirmed that this monitoring system enables a proactive response to the problem of treatment resistance.

The EGFR genotype is critical for treatment decision-making in lung cancer; however, its testing results may be affected by tumor heterogeneity and biopsy procedures [319]. Importantly, not all patients harboring EGFR mutations achieve favorable outcomes following treatment with EGFR-TKIs, highlighting the necessity for stratification of EGFR mutant genotypes [320]. Wang et al. included 18,232 lung cancer patients from nine cohorts in China and the United States, all of whom underwent CT imaging and EGFR gene sequencing [272]. Based on this comprehensive dataset, they developed a fully artificial intelligence-based system (FAIS) that predicts both EGFR genotype and prognosis for EGFR-TKI therapy by analyzing whole-lung information derived from CT images.

Elucidating the molecular mechanisms underlying metabolic reprogramming to develop personalized risk prognosis assessment methods for hierarchically guided therapeutic strategies holds significant clinical importance for neuroblastoma (NB). Jin et al. employed a machine learning-based multi-step procedure to clarify the synergistic mechanisms by which metabolic reprogramming drives malignant progression in NB at both the single-cell and metabolic flux levels [321]. Subsequently, a metabolic reprogramming-related prognostic signature (MPS) based on MPS stratification, along with personalized treatment strategies, was developed and independently validated using preclinical models. This study provides profound insights into the molecular mechanisms of metabolic reprogramming-mediated malignant progression in NB. It also offers a novel perspective for developing targeted therapeutics based on innovative precision risk prediction methodologies, which are anticipated to contribute significantly to the advancement of NB treatment strategies.

Dysregulation of BCL2 family proteins plays a critical role in leukemogenesis and progression [322]. Consequently, pharmacological inhibition of these proteins is increasingly becoming a common therapeutic approach. However, its efficacy is compromised in clinical or preclinical studies due to the emergence of primary and acquired resistance [323]. Shah et al. developed a drug sensitivity prediction model based on a deep tabular learning algorithm to assess Venetoclax sensitivity in T-cell acute lymphoblastic leukemia (T-ALL) patient samples [324]. Several response biomarkers for ICIs show promise but have yet to achieve clinical-scale application. Johannet et al. developed a pipeline integrating deep learning on histological specimens with clinical data to predict ICI response in patients with advanced melanoma [325]. This multivariable classifier achieved an AUC of 0.800 on Aperio AT2 images and 0.805 on Leica SCN400 images. It accurately stratified patients into high-risk and low-risk groups, demonstrating its potential for integration into clinical practice.

Integrating multiple types of biological data is crucial for a comprehensive understanding of cancer biology; however, this task remains challenging due to data heterogeneity, complexity, and sparsity. Cai et al. proposed an unsupervised deep learning model, Multi-Omic Synthetic Augmentation model (MOSA), specifically designed to integrate and augment the Cancer Dependency Map (DepMap) [326]. MOSA successfully revealed multi-omic features critical for cell clustering and the identification of biomarkers associated with drug and genetic dependencies. The hypoxic microenvironment is often resistant to multiple therapeutic modalities, prompting the development of hypoxia-activated prodrugs (HAPs) to target these resistant regions [327]. The HAP evofosfamide (TH-302) has shown promise in preclinical and early clinical trials for sarcoma [328]. Jardim-Perassi et al. developed a DL model utilizing multiparametric MRI and registered pathology images to identify hypoxic regions and monitor the therapeutic response to TH-302 in rhabdomyosarcoma PDX models and syngeneic fibrosarcoma models. This study demonstrates that AI analysis of pre-treatment MRI images can predict hypoxia status and subsequent response to HAPs. This approach can be used to monitor treatment response and adapt therapeutic regimens to prevent the emergence of resistance. Zhang et al. leveraged a single-cell transcriptomic atlas encompassing diverse cancer and tissue types to reveal heterogeneous expression patterns within malignant cells, premalignant cells, and cancer-associated stromal and endothelial cells [329]. We propose a deep learning framework named Shennong for in vitro screening of anticancer drugs, designed to target each distinct cell population within the transcriptomic landscape. This robust and interpretable framework holds promise for accelerating the drug discovery process and enhancing the accuracy and efficiency of drug screening. The increasing availability of pharmacological data and the rapid advancement of deep learning methods have enabled the construction of models for predicting and screening drug combinations. To address the gap in virtual screening of drug combinations within large-scale databases, Ye et al. proposed the ScaffComb framework. Inspired by phenotype-directed drug design, ScaffComb integrates phenotypic information into molecular scaffolds [330]. This framework can be applied to screen drug libraries and identify high-efficacy drug combinations. Protein kinases (PKs) regulate diverse cellular functions and serve as targets for small-molecule kinase inhibitors (KIs) in cancer and other diseases. However, KI resistance has become a common clinical complication affecting multiple cancers, targeted kinases, and drugs. To address this challenge, Lin et al. leveraged multimodal features and deep hybrid learning to report an upgraded web server, Dr. Kinase, for predicting four drug resistance hotspot sites and assessing the impact of mutations on previously identified DR hotspots in PKs [331].

Translational challenges and future perspectives

Translating mechanistic insights into clinically actionable strategies for overcoming therapeutic resistance faces multifaceted challenges across biological, technological, and clinical domains. Biologically, tumor heterogeneity and evolutionary dynamics generate parallel resistance mechanisms across spatially distinct niches, while the plasticity of the tumor–immune ecosystem enables continuous immune evasion [332–334]. Technologically, conventional biopsies often fail to capture real-time clonal adaptation, and preclinical models such as immortalized cell lines lack the microenvironmental complexity necessary for accurate target validation [335]. Clinically, the compartmentalization of resistance mechanisms by cancer type complicates the development of universal therapeutic strategies, and pharmacodynamic decoupling is common when targeting non-genetic adaptive states.

Emerging solutions include functional precision medicine platforms that integrate patient-derived organotypic cultures with single-cell multi-omics, AI-driven digital twins that simulate resistance evolution, and adaptive clinical trial designs incorporating biomarker-guided treatment arms. Achieving success will require redefining therapeutic goals from maximal tumor cell eradication to ecological containment of resistant clones, leveraging evolutionary principles to design “extinction therapies” (Fig. 5).

Fig. 5.

Bridging the translational gap in drug resistance research. Schematic summary of major limitations in current preclinical models and emerging approaches to enhance translational relevance. Conventional 2D cultures and animal models lack human immune-microbiome interactions and fail to predict clinical responses. Innovative models and technologies, including patient-derived organoids, humanized mice, single-cell and spatial omics, and AI-driven prediction, provide more physiologically relevant platforms for study

Limitations of preclinical models

Conventional preclinical models have fundamental limitations in reproducing the dynamics of human therapeutic resistance, primarily due to their inability to capture the integrated tumor–immune–stromal ecosystem and the spatiotemporal heterogeneity of clinical tumors [336]. In vitro monolayer cultures fail to replicate ECM architecture, biomechanical forces, and metabolic gradients, contributing to the > 90% attrition rate of oncology drugs that demonstrate efficacy in vitro but fail in clinical trials [337, 338].

Murine models face parallel constraints. Syngeneic tumors in immunocompetent mice lack human-specific immune interactions shaped by human leukocyte antigen (HLA) diversity, whereas patient-derived xenografts (PDXs) in immunodeficient mice exclude adaptive immunity, which is essential for evaluating immunomodulatory agents [339]. Additionally, most models overlook the microbiome’s influence on drug metabolism and immune activation, as illustrated by germ-free conditions abolishing the microbiota-dependent efficacy of CTLA-4 blockade [340, 341].

Innovative approaches are addressing these deficits: patient-derived organotypic cocultures incorporating autologous immune cells and fibroblasts; computational digital twins trained on longitudinal multi-omics datasets to model resistance evolution; and humanized microbiome–immune models reconstituted with patient-derived microbiota [342]. These advances, aligned with frameworks such as the National Institutes of Health (NIH) Human Tumor Atlas Network’s spatial profiling standards, promise model systems with enhanced predictive value for resistance research (Fig. 5).

Emerging frontiers in resistance research

High-resolution technologies are reshaping our understanding of resistance by revealing spatial, temporal, and molecular dynamics previously inaccessible. Single-cell multi-omics platforms, such as paired single-cell RNA (scRNA)-seq and single-cell assay for transposase-accessible chromatin (scATAC)-seq, can track transcriptional and epigenetic reprogramming within rare resistant subclones, showing how phenotypic plasticity enables transient drug tolerance without genetic changes [343]. Spatial transcriptomics platforms map resistance niches in situ, identifying immunosuppressive stromal barriers that shelter persister cells, as observed in PDAC where fibrotic zones harbor quiescent tumor cells with high survival gene expression [344].

Epi-transcriptomic profiling, including m6A-seq and pseudouridine mapping, has uncovered RNA modification–driven adaptive pathways [345]. For example, FTO-mediated m6A demethylation in leukemia stem cells increases BCL2 mRNA stability, conferring venetoclax resistance independent of genomic mutations [346]. These insights have revealed novel vulnerability categories: metabolic dependencies in hypoxic niches, context-specific synthetic lethalities triggered by RNA-modifying enzyme dysregulation, and neoepitope-independent T cell recognition sites exposed by altered RNA modifications (Fig. 5).

Towards proactive management and preemptive strategies

Shifting from reactive to proactive resistance management represents a fundamental change in oncology. Traditional strategies often address resistance only after it becomes clinically evident, limiting treatment options. Future approaches must prioritize early detection and intervention before resistant clones become dominant.

Liquid biopsy platforms enable minimally invasive, real-time monitoring of ctDNA, circulating tumor cells (CTCs), and other biomarkers [347]. Early detection of resistance-associated mutations or clonal expansions creates an opportunity to intervene before clinical progression.

Machine learning–based risk prediction algorithms can integrate longitudinal liquid biopsy data, imaging, and clinical records to forecast resistance emergence, allowing for timely and personalized treatment adjustments.

Adaptive therapeutic cycling, alternating between agents or modalities based on molecular feedback, can suppress resistant populations by preventing any single clone from gaining a permanent advantage. This strategy aligns with evolutionary principles, aiming to control tumor growth by managing clonal dynamics rather than eradicating all cancer cells outright (Fig. 5).

Conclusion

Research into tumor drug resistance faces three core challenges: at the biological level, tumor heterogeneity and clonal evolution lead to the dynamic accumulation of resistance mechanisms, while the physical barriers of the microenvironment limit drug delivery. At the technical level, existing models struggle to simulate the human ecosystem, and static biopsies cannot capture the real-time evolution of resistant clones. At the clinical level, resistance accounts for 90% of chemotherapy failures and over 50% of targeted/immunotherapy failures, while combination strategies often fail to sustain efficacy due to cross-resistance and increased toxicity. These challenges collectively hinder improvements in treatment outcomes.

Currently, strategies to address drug resistance issues mainly focus on directly targeting drug resistance mechanisms, microenvironment remodeling, and adaptive treatment. Although the concept of precise dynamic intervention based on ctDNA early warning of drug-resistant clones and AI multi-omics model prediction of evolutionary trajectories has been proposed, the above methods still face translational challenges. Directly targeting strategies are limited by tumor adaptive escape, microenvironment regulation exhibits spatiotemporal heterogeneity, and clinical models still inadequately simulate complex mechanisms such as microbiota-immune interactions.

Therefore, future research should focus on interdisciplinary innovation to overcome current bottlenecks, including: 1) Technological innovation, developing high-resolution dynamic monitoring technologies, such as single-cell spatial multi-omics analysis of drug-resistant niches, combining liquid biopsy to track epigenetic evolution, constructing humanized immune-microbiome chimeric models, and integrating patient-derived organoids with microbiota transplantation to simulate real microenvironments. 2) Targeted expansion, delving into non-genetic adaptive mechanisms, such as the regulation of immune memory remodeling by microbial metabolites and novel PTM interaction networks. 3) Clinical paradigm innovation, promote “evolution-guided therapy” through AI-driven adaptive clinical trials, intermittent dosing to activate anti-tumor immunity, and multifunctional nanocarriers to achieve tumor ecological regulation.

These measures will help establish a proactive drug resistance early warning, targeted elimination, and ecological control prevention and control system, ultimately driving a paradigm shift in cancer treatment from passive response to drug resistance to active ecological regulation, and achieving the long-term disease management goal of controllable drug resistance.

Authors’ contributions

JL, JH, YY, HZ, YL and AJ conducted the literature search, drafted the manuscript, and prepared all figures and tables. AJ, YF and LQ were involved in the study design. AL, PL, AJ and LW reviewed the manuscript and provided critical comments. All authors have read and approved the article.

Funding

This work was supported by grants from the National Natural Science Foundation of China (Grant/Award Number: 81772740, 82173345, 81972333, 82372883, 2022YFB4700904).

Data availability

Not applicable.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

All authors state that there is no conflict of interests.