Exploiting tumor lineage features for precision cancer therapy.

Lineage-restricted dependencies in cancer
Cell lineage refers to the developmental trajectory through which stem or progenitor cells differentiate into specialized cell types, a process governed by tightly regulated molecular programs. In cancer, malignant transformation disrupts this hierarchical organization, leading to aberrant and often heterogeneous lineage architectures within tumors. These distorted differentiation pathways define distinct molecular and functional programs that sustain specialized cancer cell populations and contribute to tumor progression [1].
Understanding the structure and dynamics of these cancer-associated lineages has enhanced our ability to identify tumor liabilities that can be targeted. Historically, lineage-tracing techniques – for example, using genetic markers to track cell fate – resulted in phylogenetic insights that, when coupled with large-scale genetic sequencing across cancers and healthy tissues, provided early insights into cancer driver genes. More recently, single-cell technologies have revolutionized this field by enabling high-resolution mapping of cancer-initiating cells and their tissues-of-origin. These advances have reinforced a hierarchical model of oncogenesis, in which stepwise genetic and epigenetic alterations reshape cellular identity and drive malignant transformation [2–4]. Building on these insights, the concept of targeting lineage- and tissue-specific dependencies has gained considerable traction over recent years as a promising therapeutic strategy in oncology. These approaches take advantage of the fact that cancer cells often remain reliant on lineage-restricted survival programs of their tissue-of-origin, offering an avenue for effective treatments.
In this review, we describe how recent techniques and tools have enabled the identification of lineage-restricted features and dependencies in cancer. We then explore the critical components that define the lineage- and tissue-specific characteristics of tumors. First, we highlight that genetic alterations are preferentially selected and unevenly distributed across tissue types, leading to lineage-specific patterns of oncogenic driver representation across cancers. Second, we examine the persistence of specialized functions, established during embryonic development, which remain critical for tumor maintenance and survival. Third, we discuss how cancers remain dependent on lineage-restricted signaling pathways which are essential for the normal function of their tissue of origin, thereby creating therapeutic vulnerabilities. Together, these properties, which are shaped by both genetic and non-genetic factors, contribute to the distinct molecular profiles observed across tumor types (Figure 1). Finally, we consider how lineage-specific marker repertoires create unique vulnerabilities that can be exploited by advanced T cell-based therapies.

Technical Approaches to Uncover Lineage-Specific Features in Cancer
Early studies using morphological analyses revealed divergent cellular phenotypes in cancer cells originating from the same organ. Consequently, early cancer sub-classifications relied heavily on morphological features to identify and distinguish distinct cancer subtypes, such as non-small cell versus small cell lung cancer (NSCLC/SCLC) and non-seminomatous versus seminoma testicular germ cell tumors (NSGCT/SGCT). Researchers observed differential responses to treatment among morphologically distinct tumors, providing an early indication of lineage-specific responses to cancer therapies. The limited ability of morphological analyses to distinguish cancer subtypes based on cellular phenotype was later addressed by the implementation of bulk DNA and RNA sequencing technologies. These sequencing approaches established a foundation for understanding the relationship between tumor lineage and tumor-specific dependencies. Systematic sequencing of the exomes and whole genomes of diverse, histologically defined cancers by The Cancer Genome Atlas (TCGA) project (https://www.cancer.gov/ccg/research/genome-sequencing/tcga), the International Cancer Genome Consortium (ICGC), and other efforts revealed relationships between tumor lineages and their driver genes, mutational patterns, and structural genomic alterations. The Pan-Cancer Analysis of Whole Genomes project, expanding on TCGA and ICGC, revealed the accumulating diversification of driver genes during tumor evolution, delineating evolutionary trajectories of cancer and highlighting new opportunities to identify cancer-specific dependencies [4]. Further, despite the fact that bulk RNA sequencing provides only average gene expression data across heterogeneous cell populations, its application across histologically defined cancers revealed transcriptomic features of tumors linked to their tissue-of-origin. Finally, functional genomics resources like DepMap have married bulk DNA and RNA sequencing of tumor models with comprehensive functional analysis using genetic knockdown or knockout screens, revealing functional dependencies linked to not only mutational features, but also tumor cell lineage [5]. Similarly, databases such as the Human Cell Atlas Project depict genes expressed in normal cell types, which help to place cancers in the context of normal tissue developmental hierarchies with more accuracy and enable the potential selection of lineage- or cancer-restricted targets that minimize toxicity. Together, these approaches have been instrumental in predicting tissue- and lineage-specific cancer therapy targets [6] while also supporting the development and validation of novel lineage tracing tools [7–9].
Subsequent advances in single-cell sequencing, revealing the mutations present in single cells, enabled tracing of individual cells back to their origin. By identifying mutations that are shared (clonal) and heterogeneous (subclonal) in a tumor cell population, it is possible to map developmental trajectories and cellular hierarchies. In cancer, such maps revealed clonal evolution and key drivers of transformation. Building on this approach, CRISPR-based cellular barcoding allowed dynamic clonal tracing by integrating short DNA sequences heritable through cell divisions, tracked via high-throughput sequencing [10]. New CRISPR-based lineage-tracing (CbLT) models like DARLIN subsequently used terminal deoxynucleotidyl transferase (TdT) – a template-independent DNA polymerase – to enhance barcode diversity, improving clone identification [11]. While temporal resolution remains challenging, sequential mutations helped reconstruct developmental timelines. Sampling across time points refined our understanding of differentiation dynamics [12].
More advanced combinatorial approaches, such as integrating genomic barcoding with scRNA-seq, were then developed to enable a more accurate linking of lineage history to transcriptional states. When combined with signal-recording technologies, these methods captured lineage dynamics over time and revealed tumor evolution characteristics [13,14]. They are now used to reconstruct cell lineages in various cancers, uncover cancer-specific transcriptional programs, and identify tissue-specific drivers of transformation [15,16]. Single-cell mass spectrometry (scMS), paired with antibody-based FACS, enabled proteomic profiling across developmental stages [17]. Tools like macsGESTALT integrated transcriptomic and phylogenetic data to uncover survival-associated EMT transitions in pancreatic and lung cancers. In glioblastoma (GBM), lineage tracing with scRNA-seq has revealed both genetic and epigenetic resistance mechanisms [16,18].
To fully elucidate the cellular landscape of cancer drivers, high resolution technologies are required to sequence infrequent mutations contributing to the oncogenic network. New technologies now allow sequencing of rare mutations from both mitochondrial and nuclear genomes: Bottleneck sequencing (BotSeqS) improved sensitivity for rare somatic mutations using molecularly barcoded libraries [19], and subsequently developed nanorate sequencing (NanoSeq) reduced error rates generated with BotSeqS library preparation by 97.5%. This allowed for the reliable detection of somatic mutations in single DNA molecules and the mutation profile of non-dividing cells to be assessed, including comparison of stem cells vs terminally differentiated cells [20]. Duplex cDNA sequencing methods like CODEC identified true mutations on both DNA strands, as demonstrated in clonal hematopoiesis [21]. Perturb-Seq combined single cell RNA-seq and CRISPR-based perturbations to give insight into the transcriptional programs being controlled by cell lineage restricted components at the single cell level [22]. Further advancements in Perturb-Seq have resulted in greater power to learn genetic interactions through its integration with algorithms and machine learning techniques. The in vivo application of Perturb-Seq provided insights into the relevance of transcriptional programs in the cancerous niche [23,24]. This method of genetic perturbation at single cell level has generated rich phenotyping of gene function and regulatory networks in cancer.
Finally, more recent multiplexed single-cell lineage tracing techniques with integrated multi-omic layers have further enhanced our understanding of the complexity of lineage interactions in cancer biology. For example, in chronic lymphocytic leukemia (CLL), combining single-cell genetic, epigenetic, and transcriptomic data has refined our knowledge of how distinct lineages respond to treatment, thereby highlighting intra-tumoral lineage heterogeneity [25]. These approaches, among others, have contributed to the identification of lineage- and tissue-specific targets, and their continued development will drive the discovery of novel lineage-specific targets across cancers (Table 1).

Tissue Context Shapes the Oncogenicity of Driver Mutations
Most oncogene and tumor suppressor mutations that drive cell proliferation tend to be tumor-type biased and promote cancer only in certain tissues. Similarly, inherited mutations that predispose individuals to cancer often lead to increased risk in specific tissues rather than universally [9]. These “driver mutations” are relatively rare, occurring against a backdrop of numerous “passenger mutations” which accumulate but do not contribute to tumor growth. Notably, genome-scale overexpression screens suggest that approximately 90% of genes involved in regulating cell proliferation function in a tissue-specific manner, while only about 10% serve as core regulators with roles conserved across multiple tissues [26]. Although some core driver mutations are found across cancer types, clinical trials that target these mutations regardless of tumor origin, such as basket and umbrella trials, have had limited success, with median response rates of 14–18% [27]. This underscores the importance of incorporating tissue and lineage context into trial design. The specificity of driver mutations is shaped by interactions between genetic alterations, the cell-of-origin, and the tumor microenvironment [28], and even conventional chemotherapies show tissue-dependent efficacy, further highlighting the role of tumor origin in treatment response [29].
This important observation is further supported by genomic sequencing across tissues, which confirmed that genetic alterations are not universally oncogenic [9]. Instead, some mutations are more prevalent in cancers arising from specific cell types or tissues, indicating their role as lineage-biased oncogenic drivers that promote cancer primarily within these cell types (Table 2). For instance, the p210 BCR-ABL fusion, resulting from the t(9;22) translocation, when present in more immature hematopoietic stem cells, results in chronic myeloid leukemia (CML), and has been effectively targeted by tyrosine kinase inhibitors, transforming a once-fatal disease into a manageable chronic condition [30]. Conversely, when expressed in more committed progenitor cells the p210 BCR-ABL fusion can trigger B-cell acute lymphoblastic leukemia (B-ALL) [31]. Similarly, the fusion between two transcription factors, EWS-FLI1, which is unique to Ewing Sarcoma, reprograms neural crest-derived cells and drives tumor formation [32]. The fusion between the transcription factors PML and RARA is a unique driver feature of promyelocytic leukemia, disrupting normal cell differentiation; targeted treatment with all-trans retinoic acid and arsenic trioxide has dramatically improved patient outcomes [33]. Mutations of the tumor suppressor gene APC disrupt the delicate balance of Wnt pathway activity that is essential for normal colonic homeostasis and are observed in 80% of colon cancers yet relatively infrequently in other cancer types [34]. These examples illustrate how lineage- and tissue-specific oncogenic alterations promote cancer only in specific cellular contexts and can, in some cases, serve as precise therapeutic targets.
The genetic landscape of cancer is further complicated by the fact that, for a given oncogenic driver gene, certain alleles can promote cell growth in specific tissues while having little or no effect in others. KRAS, the most frequently altered oncogene, exemplifies this phenomenon. The A146T variant of KRAS is predominantly seen in colorectal cancer (CRC), while the G12D substitution is common in multiple cancers [35]. In CRC, different KRAS alleles were shown to drive distinct signaling profiles, with allele-specific activation of ERK2 substrates and unique proteomic and mutational patterns. In contrast, these differences were not observed in the spleen, highlighting the tissue-dependent nature of allele-specific oncogenic activity [36–38]. The KRAS G12R allele (the least common of the G12 variants) is more commonly found in pancreatic ductal adenocarcinoma (PDAC) but not in non-small cell lung cancer (NSCLC), or CRC. The context-specific oncogenic potency of the G12R variant can be attributed to its differential functional consequences in different cancer contexts. KRAS G12R is deficient in its capacity to bind the key effector PI3Kα, essential for mutated KRAS-driven cancer development, explaining the low prevalence of the KRAS G12R variant in cancer overall, including NSCLC and CRC. PDACs, however, circumvent this deficiency by overexpressing PI3Kγ, a compensatory mechanism which allows the expansion of KRAS G12R-driven PDAC [39].
Finally, the multifunctionality of cancer drivers such as the H3K27 histone methyltransferase, EZH2, can inform different therapeutic strategies depending on tissue context. High EZH2 expression is often observed in primitive malignant cells linked to cancer stem cell maintenance. In prostate cancer, EZH2 functions as a tissue-specific epigenetic driver of lineage plasticity, facilitating the transition from prostate adenocarcinoma to neuroendocrine prostate cancer (Box 1) [40]. Cellular plasticity is also reflected in the ability of cancer cells to switch between differentiation states. A notable example is neuroendocrine transdifferentiation which predominantly occurs in metastatic EGFR-mutant NSCLC treated with EGFR tyrosine kinase inhibitors. EGFR inhibition promotes dedifferentiation, thereby enabling cancer cells to acquire basal and stem cell-like characteristics before further differentiating into neuroendocrine lineage [41].
An increased understanding of mechanisms which drive cellular plasticity and subsequent therapy resistance presents the opportunity to pivot the therapeutic approach utilizing sequential or concomitant treatment strategies. Drug-tolerant persister (DTP) cells which survive targeted therapy, often exhibit a set of conserved behaviors that include mesenchymal differentiation. Several studies have demonstrated that targeting mesenchymal-selective vulnerabilities including GPX4 and PKN2, can help eliminate DTP cells [42,43]. Similarly, drugs targeting chromatin regulators, such as KDM5 enzymes, can ablate highly plastic DTP populations [44]. Alternatively, lineage plasticity mechanisms could be exploited to improve therapeutic outcomes by steering cells toward a secondary lineage or state that restores sensitivity to therapy.
Collectively, these findings highlight the notion that most cancer-driving mutations act in a manner that is templated by tissue context, influencing tumor development mainly within certain cell types. This specificity at least partially explains why targeting mutations without considering tumor origin often yields limited success. Tailoring therapies to the tissue and lineage context is therefore key to improving treatment effectiveness.

Specialized Functions of the Tissue of Origin as a Source of Lineage Vulnerability in Cancer
Cancers often co-opt the physiological functions of their cells of origin, leading to lineage-specific vulnerabilities tied to each tissue’s specialized role. Experimental models disrupting cellular homeostasis have shown that identical oncogenic events can produce distinct outcomes depending on the cell’s differentiation state or lineage context. A Sleeping Beauty transposon mutagenesis system implemented in mice introduced mutations at various stages of T-cell development revealed that the same genetic alterations can lead to different molecular and survival outcomes depending on the maturity status and function of the transformed cells. These findings provide an explanation for the presence of subtype-specific NOTCH1 mutation patterns in T-acute lymphoblastic leukemia [45]. Likewise, in pediatric AML, the lineage context in which fusion oncogenes like ETO2-GLIS2 are expressed shapes disease phenotype and aggressiveness by reprogramming lineage-defining transcription factors, thereby locking cells in an undifferentiated state that may be therapeutically reversible by targeting these transcription factors [46]. These findings demonstrate that the same mutation can drive divergent oncogenic behaviors depending on the cell functional identity and ontogeny, revealing inherently lineage-specific vulnerabilities.
Building on this, therapeutic susceptibilities often reflect the unique physiological demands of the tissue-of-origin. For instance, multiple myeloma arises from plasma cells specialized in immunoglobulin secretion. This function imposes chronic proteotoxic stress, making multiple myeloma cells highly dependent on proteasome activity. Proteasome inhibitors including the first-in-class compound bortezomib, the second-generation agent carfilzomib, and the first oral proteasome inhibitor, ixazomib, exacerbate proteostasis imbalance. As a result, these compounds constitute one of the backbone treatments for this cancer type [47,48]. This illustrates how tissue-specific functions not only shape tumor biology but also define therapeutic windows. Conversely, the absence of a clear lineage identity poses major clinical challenges. Acute leukemias of ambiguous lineage, for example, fail to commit to myeloid, B-, or T-lymphoid lineages, or display features of more than one, complicating diagnosis and treatment. This lack of lineage definition deprives clinicians of established, lineage-guided treatment strategies [49], highlighting the critical role of lineage in both disease biology and therapeutic design.
Responses to therapy can also be highly influenced by developmental stage and cell differentiation state. In AML for instance, the more advanced differentiation state of monocytic AML is associated with loss of expression of the venetoclax target, BCL-2, and increased reliance on MCL-1 for oxidative phosphorylation and survival. Therefore, resistance to venetoclax therapy is correlated to monocytic AML and phenotypically primitive AML is more sensitive to venetoclax therapy [50]. Importantly, a recent study indicates that monocytic differentiation and venetoclax resistance are two independent effects of RAS mutations. Leukemic transformation driven by mutations in RAS depends on the cellular milieu and chromatin landscape of granulo-monocytic progenitor cells, whereas venetoclax resistance is broadly conferred across hematopoietic stem and progenitor cell types by RAS mutations. This explains clinical observations in which monocytic subclones emerge during treatment: these subclones often correspond to RAS-mutant clones enriched in monocytic cells, and their expansion reflects selection at the leukemic stem cell level rather than monocytic differentiation per se driving therapy resistance [51].
Metabolism and hormone signaling represent specialized cellular functions shaped by lineage-specific gene regulation, contributing to the diverse physiological profiles of tissues and by extension, their cancers. Metabolic phenotypes vary across cancers in part due to lineage-specific expression of metabolic genes. Tumors often retain the metabolic signatures of their tissue lineage, as shown by studies integrating metabolomics, flux analysis, and functional genomics [52,53]. A notable case is the liver, whose intense metabolic activity corresponds to a high rate of somatic mutations with metabolic consequences in primary liver cancer (PLC) and in chronic liver disease contributing to PLC. Extensive metabolic rewiring in the context of liver cancer has resulted in the exploration of metformin, an inhibitor of gluconeogenesis, in this context. Metformin demonstrated substantial alleviation of the risk of hepatocellular carcinoma development and progression in vitro and in population studies [54,55]. Similarly, cancers of endocrine tissues such as breast, prostate, ovary, and endometrium sometimes retain dependence on hormone signaling, a hallmark of their lineage. Despite expressing the same nuclear hormone receptors, their therapeutic responses are lineage-specific: anti-androgen therapy, for example with enzalutamide, bicalutamide, flutamide, apalutamide, and darolutamide is effective in prostate cancer but not breast cancer. Similarly, selective estrogen receptor modulators like tamoxifen and fulvestrant benefit patients with breast but not prostate cancer [56–58].
Beyond exploiting specialized functions of their tissue of origin, cancer cells can also reawaken traits of embryonic cells, reflecting their inherent developmental plasticity. Many of these traits are not acquired anew but rather stem from the lineage-specific functions embedded in the tissue’s embryonic origin. Key embryonic signaling pathways have been shown to be reactivated during cancer development. Processes that revert cells to a more embryonic-like state that are associated with malignancy include increased ability for self-renewal in stem cells and epithelial-to-mesenchymal transition (EMT), both of which influence a disruption in cellular-lineage hierarchy in cancer [59–61]. The resulting reversion to a more stem- or mesenchymal-like state endows cells with unique, targetable vulnerabilities [62,63]. Tissues sharing a common embryonic origin often retain synapomorphies from their stem group, and their transcriptomic and proteomic profiles are more similar than those from different germ layers. Clonal asymmetry studies further suggest that somatic single nucleotide variants occurring during embryogenesis, gastrulation, and organogenesis shape long-term tissue patterning [64]. Using single-cell RNA sequencing of human pluripotent stem cells, germ-layer–specific transcriptional and drug-response programs were mapped and projected onto adult tumors. This work revealed the reactivation of embryonic lineage-specifying genes in cancers, indicating a return to early developmental states that could expose new therapeutic vulnerabilities [65]. Diffuse hemispheric gliomas, lethal brain tumors from interneuronal precursors, offer another example. These cancers segregate into subgroups defined by driver mutations tied to distinct developmental lineages. Each subgroup depends on embryonic-stage factors, and inhibiting these factors promotes differentiation and impairs tumor growth [66]. Similarly, primary central nervous system germ cell tumors likely arise from misrouted primordial germ cells and lack a dominant oncogenic driver. Instead, Genome-Wide Association Study studies revealed that multiple low-penetrance alleles disrupt pluripotency and apoptosis networks, hijacking early developmental circuits to sustain malignancy [67].
Recent advancements in sequencing technologies have improved our understanding of how developmental programming in gametes can influence cancer-inducing programs in offspring. A recent study demonstrated that positive selection of driver mutations in sperm leads to the accumulation of disease-causing mutations. The positively selected male germline mutations, whether activating or loss-of-function, were associated with developmental diseases or cancer [68]. The ability to perform extensive sequencing of gametes may reveal the extent to which specific cancers are pre-determined and the role of differential developmental programming in the emergence of cancerous lineages.
In particular molecular contexts, cancer therapies act as selective pressures resulting in release from lineage constraints and phenotypic switching. A prominent example involves the loss of differentiation regulators TP53 and RB1. In castration-resistant prostate cancer (CRPC), TP53- and RB1-deficient prostate cancers evade AR-targeted therapies by transitioning from Androgen Receptor (AR)-dependent luminal cells to AR-independent basal-like phenotypes. The loss of TP53 and RB1 results in the activation of microenvironmental signals, specifically the activation of JAK/STAT molecular program, producing AR-targeted therapy-resistant stem-like subclones. The restoration of TP53 and RB1 functions reverses this phenotypic switch and resensitizes to targeted therapy [69]. Similarly, TP53/RB1 co-mutation in lung cancer identifies a subset of patients at increased risk for small cell transformation and emergence of more stem-like clones associated with acquired resistance to targeted EGFR therapy [69,70]. These examples demonstrate how major alterations in differentiation state resulting from RB1/TP53 loss alter the expression of lineage-specific programs, leading to a more stem-like phenotype and resistance to targeted therapies in advanced cancers.
Emerging strategies aim to address the question of how to pharmacologically target embryonic dependencies such as those involved in the EMT process or regulation of the Hedgehog pathway which precision therapies have struggled to address. Early trials treating patients with MET-amplified gastric cancer using crizotinib, an inhibitor of the EMT-promoting proto-oncogene c-MET, demonstrated some clinical benefit [71]. Similarly, two inhibitors targeting the Hedgehog pathway regulator SMO, Vismodegib and Sonidegib, playing a role in embryonic development were approved by the FDA in 2012 and 2015, respectively, for the treatment of advanced basal cell carcinoma driven by PTCH gene loss-of-function alterations [72], [73]. In 2018, another SMO-targeting small-molecule inhibitor, Glasdegib, was approved for use in acute myeloid leukemia (AML) [74].

Lineage-Specific Signaling as a Source of Cancer Vulnerability
Cellular survival, proliferation, and function are regulated by a complex signaling network that maintains homeostasis within a specific lineage context. Targeting critical nodes in this lineage-specific network can thus block the progression of malignancies arising from that lineage. These mediators often exhibit lineage- or tissue-specific characteristics and distinct modes of pathway activation.
In some cases, the potential of targeting a tissue-specific signaling network has been translated into clinical outcomes. In certain lymphoid malignancies, the activation of the pro-survival NF-κB pathway is caused by a highly active BCR signaling cascade. Inhibitors that target intracellular BTK, a downstream propagator of BCR receptor signaling, have been transformative in the treatment of BCR-dependent diseases (Figure 2). In CLL, the BTK inhibitor, Ibrutinib, demonstrated a superior response rate and durability of response over chemotherapy with a significantly improved toxicity profile [75]. Genetically distinct forms of lymphoid malignancies show varying responses to BTK inhibitors. These differences reflect the underlying heterogeneity among tumor subtypes, which may appear similar phenotypically but respond differently to therapy. This variation can reveal resistance mechanisms and highlight lineage-specific trends, offering insights into the complexity of lineage networks within tissues [76].
Global signaling regulators, including G-protein coupled receptors (GPCRs) and regulators of the activity of small guanine nucleotide-binding (G) proteins such as guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs), show variable expression and functional dependency across tissues. As the largest family of transmembrane proteins, GPCRs mediate key downstream signaling, yet their receptor/ligand distribution varies significantly across cancer types, creating distinct, subtype-specific expression patterns [77]. In AML, a lineage-specific dependency has emerged on the PI3Kγ isoform for GPCR-mediated survival and proliferation. PI3Kγ is selectively expressed and essential in AML, while other PI3K isoforms are dispensable, providing an opportunity to target this isoform selectively, sidestepping the inherent limitations of pan-PI3K inhibitors which cause on-target, off-tissue toxicities that limit their use (Figure 2) [78–80]. A PROTAC-based degrader molecule of PI3Kγ derived from the PI3Kγ inhibitor, AZ2, demonstrated high specificity and efficacy in AML cells with minimal toxicity in non-AML cells. This work exemplifies how targeting lineage-specific signaling dependencies can yield potent, cancer-selective therapies (Box 2). Similarly, due to the known oncogenicity of mutations in small G protein RAS across cancers, GAPs and GEFs have been postulated to play a role in cancer development and maintenance. A recently published study utilized functional screens to identify GAP ARHGAP45 as a selective functional dependency in myeloid malignancies compared to normal HSPCs. In myeloid cells, ARHGAP45 regulates GTP-bound RhoA levels and restrains overactivation of RhoA, subsequently blocking myeloid differentiation. T cells directed towards ARHGAP45-derived antigen significantly reduced tumor burden in an in vivo AML model [81]. These examples demonstrate how alternative strategies for drug development, including PROTACs and T-cell directed therapies, can be leveraged against novel lineage-specific targets where classical pharmacological inhibitors are insufficient or absent.
As illustrated by studies on PI3Kγ, protein kinases play a central role in lineage addiction in cancer. Their impact on tumor progression arises from both kinase-dependent (catalytic phosphorylation) and kinase-independent functions, such as scaffolding, protein-protein interactions, allosteric regulation, subcellular localization, and DNA binding, functions whose relevance often varies by tissue context [82]. Further, both the signaling output of a kinase and the cancer cell’s tendency to die versus adapt to kinase inhibition are influenced by the particular signaling context present in a tissue. These factors influence the distribution of mutations impacting a given kinase across cancer types, which themselves may be differentially inhibitable with various classes of small molecule inhibitors. For example, EGFR is frequently mutated in both lung carcinoma and GBM, in each case resulting in EGFR activation. In GBM, mutations commonly occur in the extracellular domain, whereas in lung cancer, mutations more commonly impact the kinase domain. This selection for a different spectrum of EGFR mutations in GBM and lung carcinoma results in EGFR variants with context-dependent signaling functions, adaptive feedback mechanisms, and inhibitability with distinct classes of small molecules. As a result, EGFR inhibitors targeting the active kinase conformation are far more effective in lung carcinoma than in GBM (e.g. erlotinib) [83,84] (Figure 3). Similarly, BRAF-mutated CRCs and melanomas exhibit differential responses to BRAF and MEK inhibitors. BRAF mutations increase activation of the Ras-Raf-MEK-ERK (MAPK) pathway, driving cell proliferation and growth. However, BRAF mutations exist in distinct signaling contexts in these two diseases: In CRC, EGFR signaling has been well established to blunt the activity of BRAF and MEK inhibitor therapies, and constitutively active Wnt signaling may also contribute to these suboptimal responses relative to those observed in BRAF-mutant melanomas, which typically lack both. Further divergence of BRAF activity in different tissue contexts can be attributed to the presence of different BRAF variants, which vary in their representation between these tissues. The V600 variant mutation in BRAF exhibits strong kinase activity, while non-V600 variants range from RAS-independent hyperactive forms to low-activity or even ‘kinase-dead’ versions. As a result, whereas some BRAF variants yield MEK phosphorylation independently of RAS, others for example require RAS-dependent CRAF binding and activation. (Figure 4). Together, the differing BRAF signaling contexts and mutational distributions found in CRC versus melanoma lead to differential responses to BRAF and MEK inhibitor therapies, with responses in melanoma being significantly more robust [85].
The baseline essentiality of signaling pathways across tissues contributes to toxicity following their inhibition in cancer therapy. The dosing of therapies targeting essential cell signaling pathways that function as lineage dependencies requires careful optimization to effectively target malignant tissue while ensuring that non-malignant tissues can tolerate pathway inhibition. For example, androgen receptor inhibition in the treatment of prostate cancer was associated with cardiac defects, likely due to the poorly characterized mechanism of cardioprotective functions of the androgen receptor [86]. Consequently, the therapeutic window is narrowed because of these toxic effects on cardiac tissue.
Inhibiting essential signaling pathways by targeting protein isoforms with lineage-specific expression is a promising strategy to minimize toxicity in normal tissues. For example, targeting the pro-survival PI3K-AKT-mTOR pathway with PI3Kα inhibitors for the treatment of solid tumors incurs side effects in non-malignant tissues as a result of the broad importance PI3Kα plays in propagating PI3K-AKT-mTOR signaling in diverse tissues (NCT01219699). In contrast, targeting the PI3Kγ isoform offers a way to avoid such therapy-related toxicity, as its role in promoting PI3K-AKT-mTOR signaling is largely restricted to myeloid cells [80]. This example illustrates that while targeting lineage-based dependencies can pose inherent toxicity risks, selective targeting of lineage-specific protein isoforms can expand the therapeutic window and mitigate side effects, providing a clear rationale for how such strategies can be translated safely to the clinic.

The Tissue-of-Origin Antigen Repertoire of the Tumor as a Lineage-Restricted Liability for Advanced T Cell–Based Therapies
CAR-T cell therapy, which engineers patient-derived T cells to express receptors targeting tumor-associated antigens, has gained momentum. Tumor- and tissue-restricted antigens are excellent potential targets for CAR-T cell therapies, and as such a complete definition of the tumor- and tissue-restricted antigen landscape is of crucial importance (as earlier exemplified by T cells directed towards ARHGAP45-derived antigen in AML) [81,87]. Recent advances linking intratumoral T-cell phenotypes to antigen specificity, aided by multi-dimensional single-cell technologies, have pushed T-cell therapies toward more tissue- and lineage-specific precision approaches.
CD19, which is predominantly expressed across nearly all stages of B-cell development, was historically the first lineage-specific marker to be used for CAR-T cell-based therapies, with significant anti-leukemia effects in relapsed/refractory B-ALL and diffuse large B-cell lymphoma [88]. Building on this, other B-cell markers such as CD20, CD22, and CD79b, each with distinct but largely B cell-restricted expression patterns, have been explored as additional lineage-specific targets and are currently being tested in clinical trials (Table 3). CD20 is mainly expressed on mature B cells, CD22 is present on mature and some late-stage B cells, and CD79b, a component of the B-cell receptor complex, is expressed throughout most B-cell stages except plasma cells [89]. Many next-generation CAR-T systems now incorporate bispecific designs targeting combinations like CD19 and CD22, or CD19 and CD70 [90]. Following the success of CD19-based CAR-T therapies, T-lineage-associated markers such as CD5 and CD7 have also been investigated as CAR-T targets [91,92]. Of note, CD7-based CAR-T therapies often require gene-editing strategies to prevent fratricide activity [93]. In multiple myeloma, B-cell maturation antigen (BCMA) has become a major therapeutic target, with CAR-T cell therapies like idecabtagene vicleucel and ciltacabtagene autoleucel already FDA-approved [94]. GPRC5D, a newer plasma cell-associated marker with low expression in normal tissues, is now under clinical investigation, including for BCMA-refractory patients [95].
In myeloid neoplasms, CD33 and CD123 are the primary targets for CAR-T strategies. CD33 is highly expressed on normal myeloid progenitors and most AML cells, and its clinical relevance is supported by the FDA-approved ADC gemtuzumab ozogamicin [96]. However, its expression on normal precursors limits this agent’s therapeutic index. CD123, which is more selectively expressed on leukemic stem cells and blasts than on normal pluripotent hematopoietic progenitors, represents a potentially safer and more targeted therapeutic option. This is exemplified by ongoing trials using CD123-directed CAR-T cells and the FDA-approved recombinant fusion protein tagraxofusp, which combines the CD123 ligand IL-3 with a truncated diphtheria toxin for the treatment of blastic plasmacytoid dendritic cell neoplasm [97]. Dual antigen targeted CAR-T therapies can compensate for the heterogeneity of single antigen expression, resulting in improved specificity and efficacy. In AML, dual-targeted CD33 and CD-123 CAR-Ts were able to eradicate AML from patient derived xenograft models [98]. Multi-specific engagers show promise to improve the selectivity of CD-123 based T cell therapy for cancer cells relative to normal cells, reducing the risk of cytokine release [99].
Antigens like CD19 are homogeneously expressed in hematologic malignancies but solid tumors often display variable and patchy antigen expression, a factor that has made them more difficult to target with antigen-specific T cell based therapies. Approaches such as NGS have been applied to identify better tumor-specific targets. In CRPC, this led to development of CAR-T cells against PSMA and PSCA [100,101]. To minimize risk of antigen escape in solid tumors due to heterogeneous antigen marker expression, newer CAR T-cell strategies such as the Multi‑4SCAR‑T trial (NCT04430595) in breast cancer are testing simultaneous targeting of HER2, GD2, and CD44v6 (Table 3). By focusing on multiple antigens linked to both lineage and cancer stem cells, this approach reduces the risk of immune evasion and toxicity to healthy tissues and enhances tumor clearance. Lineage-specific markers have proven essential for the precision and efficacy of CAR T-cell therapies, especially in hematologic cancers. Exploiting these markers remains key to enhancing therapeutic impact and limiting off-target toxicity, even in the more complex context of solid tumors.
An elegant approach to improve antigen-specific immunotherapy in the absence of cancer restricted surface markers involves the engineering of healthy hematopoietic stem and progenitor cells to delete or modify the target antigen, thereby reducing the risk of toxicity caused by immune destruction of healthy cells. For example, in AML, CD33-targeted CAR-T therapy is restricted by the destruction of healthy myeloid cells due to the widespread expression of CD33 in healthy and malignant myeloid compartments. The engraftment of HSPCs with CD33-deleted but otherwise normal functions was shown to yield effective cancer targeting while sparing the normal, CD33-deficient donor cells that comprised the reconstituted hematopoietic system [102]. This method has since been applied to markers beyond CD33, including the combinatorial targeting of FLT3, KIT and CD123 through the alteration of key epitopes in donated normal HSPCs [103]. In principle, epitope engineering of T cell target antigens in healthy compartments has the potential to be applied beyond hematological malignancy in solid cancers whereby CAR-T therapies remain limited, largely due to the unrestricted expression of lineage-related antigens in essential solid tissues.

CONCLUDING REMARKS AND FUTURE DIRECTIONS
Lineage-tracing techniques have helped uncover a new generation of lineage- and tissue-specific dependencies across cancers. Recent advances in lineage-tracing and single-cell technologies have enabled high-resolution mapping of these programs, leading to the development of targeted therapies that exploit the molecular context of the cell-of-origin. Continued efforts to define these lineage-specific targets and dependencies are likely to expand the repertoire of precision cancer therapies. Achieving this goal requires a deeper understanding of how cellular hierarchies evolve and diverge in healthy versus malignant tissues (see outstanding questions). Improved lineage-tracing methods, particularly when coupled with single-cell and multi-omics approaches, are helping clarify how genetic alterations interact with non-genetic features across tissue contexts. In immunotherapy, these tools have enabled the identification of tissue-specific antigen expression patterns for selective T cell-based targeting. Finally, recent advancements in artificial intelligence and deep learning methods have identified novel causal drivers of lineage-specificity and plasticity in the cancer context [104].
This emerging precision strategy marks a shift from broad cytotoxic treatments to more selective interventions that improve efficacy and reduce toxicity. However, current lineage-tracing methods have limitations. For instance, they often rely on assumptions about the rarity of identical mutations arising independently, potentially introducing bias in clonal relationships. Moreover, the interaction of intrinsic and extrinsic factors—including genetic and epigenetic plasticity—can enable cancer cells to escape lineage-targeted therapies.
Careful mining of existing data from resources like DepMap has identified several historically ‘undruggable’ lineage-specific targets, most notably including transcription factors [5]. Excitingly, emerging drug design strategies employing chemoproteomics, molecular glues, PROTACs, and other technologies are rapidly changing the landscape of druggability, suggesting that these lineage-selective targets may soon become pharmacologically addressable. Cellular plasticity also remains a major obstacle in this field. One therapeutic approach involves using combination regimens to steer cells into a fixed lineage state, thereby increasing sensitivity to lineage-specific cytotoxic agents. Future efforts should then focus on improving lineage-tracing resolution, understanding the molecular basis of plasticity, and developing novel delivery platforms for lineage-restricted therapies (Box 3; see outstanding questions). This interdisciplinary frontier will benefit from collaboration between chemists, biologists, and clinicians to translate biological specificity into therapeutic precision.