Heterogeneity in cancer: molecular mechanisms and therapeutic strategies

Abstract

Cancer heterogeneity drives therapeutic failure, resistance, and relapse and encompasses variation across genetic, cellular, tissue, organ, and systemic scales. Existing frameworks define malignant capabilities and tumor–microenvironment interactions but less explicitly address how heterogeneity propagates across scales and informs therapeutic design. Here, we propose the Holistic Integrative Tumor Ecosystem Theory (HITET), a five-dimensional framework connecting these scales through functional interfaces. Its organizing principle is the niche–flow–feedback triad: niches define selective habitats; flows transmit cells, resources, and signals across scales; and feedback loops amplify or constrain heterogeneity by reshaping niches and flows. HITET generates several provisional predictions: interventions targeting three or more active dimensions may produce more durable responses than narrower strategies; tumors with strong vascular–immune coupling may be particularly sensitive to combined vascular normalization and immune activation; and clonal or phenotypic diversification may correlate with the number and strength of positive feedback loops. We use HITET to synthesize how genetic alterations, cellular plasticity, metabolic programs, immune and stromal interactions, spatial gradients, organ-specific environments, microbiota, and host physiology jointly shape cancer heterogeneity. Therapeutically, HITET organizes strategies as vertical integration within a dimension, horizontal integration across dimensions, and modulation of host–tumor interfaces. HITET does not replace existing cancer frameworks or constitute a validated predictive model. Its current contribution is organizational and hypothesis-generating, providing a structured basis for linking mechanisms and measurements and developing predictions that require prospective experimental and clinical validation.

Introduction

Cancer heterogeneity is a major cause of therapeutic failure, acquired resistance, and relapse. It occurs in patients with the same cancer type, between different lesions in one patient, and across spatial regions of a single tumor.^1,2 Although driver mutations and clonal evolution have long provided a central framework for understanding this diversity,^3,4 genetic alterations alone cannot fully explain why tumors with the same actionable driver often respond differently depending on cancer type, anatomical sites, or treatment context.^5–7 Tumor behavior is shaped by interactions between molecular alterations, cellular states, tissue microenvironments, organ-specific constraints, and systemic host physiology.^8,9 Environmental and host-related factors, including diet, microbiota, inflammation, metabolism, endocrine signals, and neural regulation further influence tumor fitness and therapeutic response.^10–17 Thus, cancer heterogeneity should be considered an emergent cross-scale property of the entire tumor–host system rather than just a tumor cell-intrinsic feature.

This view builds on, rather than replaces, existing cancer frameworks, which differ in their primary analytical focus. The Hallmarks of Cancer comprise the set of functional capabilities acquired by malignant cells^18,19; models of tumor microenvironment (TME) emphasize local interactions among tumor, immune, stromal, vascular, metabolic, and extracellular matrix (ECM) components²⁰; evolutionary frameworks focus on clonal selection, adaptation, and temporal change²¹; cancer-ecosystem models conceptualize tumors as ecological communities shaped by competition, cooperation, and environmental selection^6,9; and systems-oncology approaches integrate molecular, cellular, clinical, and computational data into network-level models.²² Because these frameworks differ in their principal units and scales of analysis, they cannot provide a shared structure for tracing the transmission and reshaping of heterogeneity across genetic, cellular, tissue, organ, and systemic levels.

To systematically organize these relationships and generate testable cross-scale hypotheses, we propose the Holistic Integrative Tumor Ecosystem Theory (HITET) as a five-dimensional integrative framework (Fig. 1). HITET categorizes cancer heterogeneity into five interacting dimensions: genetic, cellular, tissue, organ, and systemic; it overlaps with existing frameworks and does not present any novel individual biological components. The proposed contribution of the categories is narrower and focuses on the interfaces through which heterogeneous states are transmitted, filtered, stabilized, amplified, or constrained across biological scales. These five dimensions are intended as heuristic analytical categories, rather than mutually exclusive compartments or a formal taxonomy. The mechanisms are discussed according to the dominant scale at which they are generated, measured, or exert their principal effects in a given context, while cross-dimensional effects and context-dependent reassignments are expected. Therefore, the boundaries between dimensions represent functional interfaces rather than fixed biological divisions.

Fig. 1.

The HITET framework for organizing multiscale cancer heterogeneity. This schematic summarizes the Holistic Integrative Tumor Ecosystem Theory (HITET) as an organizational framework for cancer heterogeneity. Cancer heterogeneity is divided into five interacting dimensions: genetic, cellular, tissue, organ, and systemic. The genetic dimension represents heritable alterations and epigenetic states that provide substrates for tumor evolution. The cellular dimension reflects heterogeneous cell states, signaling programs, and interactions among tumor and non-tumor cells. The tissue dimension captures local microenvironmental organization, including vascular, stromal, immune, biochemical, and physical features. The organ dimension highlights site-specific anatomical and physiological contexts that influence metastatic colonization and treatment response. The systemic dimension includes host-level regulatory processes, such as immune, metabolic, endocrine, neural, and microbiota-related factors. The arrows between dimensions indicate functional interactions rather than a strict linear hierarchy. The lower part of the figure links this multidimensional view to current therapeutic modalities and future directions, including mechanism integration, precision treatment, and systemic intervention. This overview provides the conceptual basis for the more detailed discussion of mechanisms and therapeutic strategies in the review. This figure was originally designed and created by the authors using Adobe Illustrator

At the center of HITET is the niche–flow–feedback triad, which links local environments with cross-scale exchange and reciprocal regulation. Its operationalization is context-specific. A “niche” is a spatial microhabitat defined by measurable cellular, biochemical, and physical features, such as oxygen tension, nutrient availability, immune composition, and ECM structure and stiffness, which can be assessed using spatial profiling, imaging, or biochemical and biophysical measurements. A “flow” is the directional transfer of cells, resources, or signals, quantified using fluxes, gradients, trafficking rates, perfusion or drainage parameters, and longitudinal changes. “Feedback” occurs when altered niches modify flows, and altered flows subsequently reshape niches; its direction depends on whether the response amplifies or counteracts the initiating change, whereas its strength is determined by estimating the response magnitude and persistence using time-resolved or perturbation data and causal or dynamic models. Across dimensions, individual measurements should be linked through biologically specified interfaces rather than collapsing into a single composite score, for example, via tracing genetic alterations through cellular states and niche changes. These principles provide a provisional measurement logic, rather than a standardized workflow or validated quantitative model.

By formalizing these relationships, HITET generates several provisional testable predictions. First, interventions that simultaneously constrain three or more dimensions may produce more durable responses than those targeting only one or two dimensions. Second, tumors with strong vascular–immune coupling may be more sensitive to therapies that combine vascular normalization with immune activation. Third, the rate of clonal or phenotypic diversification may correlate with the number and strength of active positive feedback loops within the tumor regions. Fourth, in some clinical contexts, only a subset of dimensions may dominate the treatment response, which enables dimension-specific monitoring strategies when multi-scale profiling is not practical. These propositions are intended as testable hypotheses rather than empirically established conclusions. At present, HITET should be regarded primarily as an organizational framework that generates cross-scale questions and predictions; its explanatory and clinical utility require prospective validation in experimental models, longitudinal patient cohorts, and intervention studies.

This approach also provides a roadmap for therapeutic design. HITET distinguishes three complementary intervention modes: vertical integration, horizontal integration, and ecosystem modulation. Vertical integration refers to a multitarget intervention within one dimension, such as co-targeting a driver’s pathway and its compensatory nodes. Horizontal integration combines therapies across dimensions, such as targeted therapy with immunotherapy, or radiotherapy with immune checkpoint blockade. Ecosystem modulation aims to reshape host-level conditions that influence tumor evolution, including microbiome modulation, metabolic rebalancing, neuroimmune regulation, endocrine intervention, exercise-based conditioning, and chronotherapeutic dosing. Together, these strategies seek to restrict adaptive escape while reshaping the tumor–host environment.

In this review, we used HITET to organize cancer heterogeneity across five interconnected dimensions: genetic, cellular, tissue, organ, and systemic (Fig. 2). We first examine how tumor-intrinsic alterations generate diversity and then discuss how cellular programs, tissue habitats, organ-specific contexts, and systemic host regulation shape its evolution. Finally, we analyze how this multidimensional framework can guide therapeutic design, disease monitoring, and ecosystem-informed clinical trials.

Fig. 2.

Tumor heterogeneity in a multiscale tumor ecosystem framework. This schematic depicts tumor heterogeneity from a multiscale ecosystem perspective. Tumor heterogeneity is shown as an emergent property arising from the dynamic coupling of intrinsic genetic alterations and extrinsic ecological constraints across biological scales. Heterogeneity occupies the center, surrounded by interconnected dimensions arranged radially. The genetic dimension includes driver mutations, epigenetic variation, and chromosomal abnormalities underlying clonal diversification. The cellular dimension reflects signaling networks and tumor–host interactions that shape cell-state plasticity. The tissue dimension captures spatial and physical heterogeneity within the tumor microenvironment, including extracellular matrix properties and peritumoral niches. The organ dimension represents organ-specific anatomical and physiological constraints. The systemic dimension integrates whole-body regulatory processes, such as neuro–tumor interactions, metabolic disruption, cachexia, and the gut microbiota–host axis. At their intersection, therapy is conceptualized as a cross-scale ecological perturbation, reshaping selective pressures and heterogeneity across levels. Together, this framework summarizes how tumor heterogeneity is generated and maintained across scales and how it influences tumor progression and treatment response. This figure was created using BioRender (https://biorender.com/)

Genetic dimension: the heritable substrate of heterogeneity

Within the HITET framework, the genetic dimension represents a heritable substrate that constrains but does not fully determine tumor evolution (Fig. 3). It includes somatic mutations that confer selective advantages, and broader genomic and epigenomic alterations that expand the range of possible adaptive states.^23,24 Driver mutations,²⁵ epigenetic reprogramming,^26,27 and chromosomal structural variations²⁸ create substrates that can be selected under cellular and tissue-level pressures. Under selective pressures imposed by immune surveillance, metabolic stress, hypoxia, and therapy, genetic heterogeneity is continuously filtered and reshaped, favoring the expansion of clones that are compatible with particular cellular states or tissue niches.^29,30 Thus, the genetic layer should be understood not as a deterministic blueprint but as an evolving set of heritable constraints and possibilities whose phenotypic consequences depend on higher-dimensional contexts.

Fig. 3.

Genetic and epigenetic substrates of cancer heterogeneity. This figure summarizes the genetic dimension of HITET as a heritable substrate that constrains, but does not fully determine, tumor evolution. Three major categories are shown. First, driver gene mutations and clonal architecture include activating alterations in proto-oncogenes and inactivation of tumor suppressor genes, which can alter proliferation, survival, differentiation, metabolism, and immune evasion. Second, epigenetic plasticity includes DNA methylation, histone modifications, and chromatin remodeling, which can expand the range of accessible transcriptional and phenotypic states without changing DNA sequence. Third, chromosomal instability and structural variation include deletions, amplifications, inversions, aneuploidy, and chromosomal translocations, which can reshape gene dosage, genome organization, and clonal diversity. Together, these alterations provide heritable constraints and adaptive possibilities on which cellular states, tissue niches, organ-specific environments, and systemic host pressures can act. Thus, the genetic dimension is presented not as a deterministic blueprint, but as a context-dependent substrate for cross-dimensional selection within the HITET framework. This figure was created using BioRender (https://biorender.com/)

Driver mutations and clonal architecture

Driver mutations are important genetic events that initiate malignant transformation and shape clonal architecture. Mutations, amplifications, and deletions in oncogenes or tumor suppressor genes can alter core cellular programs,^31–33 including proliferation, differentiation, survival, metabolism, and immune evasion.⁷

Representative oncogenes and tumor suppressor genes illustrate this cross-dimensional role. Although epidermal growth factor receptor (EGFR) and Kirsten rat sarcoma viral oncogene homolog (KRAS) alterations promote clonal expansion by activating growth and survival pathways, their therapeutic relevance depends on the cellular state, lineage context, tissue microenvironment, and compensatory signaling.^34–37 Similarly, TP53 alterations contribute to genomic instability and stress response remodeling, but their effects on prognosis, immune interactions, and treatment response vary across cancer types and disease stages.^38–41 Therefore, rather than providing detailed descriptions of individual driver genes and genomic alterations in the main text, we have summarized representative genetic and epigenetic events, their major functions, and their cross-dimensional interfaces in Table 1.^{3,34–39,42–54}

Table 1.

Representative driver genes and genomic alterations: functions and cross-dimensional interfaces

Genetic or epigenetic event	Category	Major function	Cross-dimensional interface	Role in cancer heterogeneity
EGFR alteration34–36	Oncogene activation/amplification	Activates growth and survival signaling	Cellular signaling; tissue growth-factor niches; therapeutic pressure	Promotes clonal expansion, pathway dependency, and context-dependent response to EGFR-targeted therapy
KRAS mutation37	Oncogene activation	Activates MAPK and PI3K-related signaling; supports metabolic adaptation	Cellular metabolism; stromal remodeling; organ-specific metastatic niches	Drives clonal diversification, therapy resistance, and site-dependent tumor behavior
TP53 alteration38,39	Tumor suppressor loss or mutation	Impairs genome surveillance, stress response, and apoptosis	Cellular stress adaptation; immune interaction; systemic metabolic stress	Increases genomic instability, phenotypic diversity, and heterogeneous treatment response
BRAF V600E42,43	Oncogene activation	Activates MAPK signaling	Cellular pathway dependency; lineage context; feedback signaling	Produces context-dependent therapeutic vulnerability, with different responses across melanoma and CRC
BRCA1/2 mutation25,44	DNA repair defect	Impairs homologous recombination repair	Cellular DNA-damage response; immune recognition; therapy-induced selection	Creates DNA repair vulnerability and heterogeneous sensitivity to platinum agents or PARP inhibition
IDH1/2 mutation46,47	Metabolic/epigenetic driver	Produces 2-hydroxyglutarate and reshapes DNA/histone methylation	Cellular metabolism; lineage state; tissue context	Links genotype, metabolic state, epigenetic remodeling, and tumor lineage identity
MGMT promoter methylation48,49	Epigenetic alteration	Reduces DNA repair capacity through promoter silencing	Cellular DNA-damage response; therapy response	Stratifies treatment sensitivity and survival in glioma-related contexts
HLA-I loss/antigen-presentation loss50,52	Immune-evasion genomic alteration	Reduces antigen presentation to cytotoxic T cells	Tissue immune niche; systemic immunotherapy response	Enables immune escape and contributes to heterogeneous response to immune checkpoint blockade
ecDNA3,51,311	Structural genomic alteration	Enables focal oncogene amplification and flexible gene dosage	Cellular oncogene dosage; therapy-induced selection	Promotes rapid adaptive plasticity, intratumoral heterogeneity, and reversible drug resistance
Whole-genome doubling53,54	Large-scale genomic event	Increases genomic content and tolerance for subsequent rearrangements	Cellular fitness; metastatic evolution	Provides a permissive state for chromosomal diversification and clonal evolution

Pan-cancer analyses further clarified the relationship between driver events and intratumoral heterogeneity. Whole-genome analyses of 2658 tumors across 38 cancer types showed that most evaluable tumors contained a subclonal architecture, indicating that coexisting clones and subclones with distinct genomic features are a common occurrence. These patterns vary according to the cancer type and are often accompanied by large-scale copy-number alterations that contribute to genetic and transcriptional heterogeneity.⁵⁵ Thus, driver events and their genomic context help establish a clonal framework upon which further selection can act.

Concurrently, clinically relevant driver events are often shared across major tumor populations. A systematic analysis of seven untreated epithelial cancers showed that approximately 97% of functional driver mutations in multiregional primary tumors were clonal, and matched metastatic lesions largely retained shared trunk drivers from the primary tumor. Most patients showed concordant responses to targeted therapies across multiple metastatic sites.⁵⁶ These findings suggest that intratumoral heterogeneity often coexists with relatively stable functional drivers.

Taken together, these observations support a balanced view of driver mutations in HITET. Driver events provide an inherited clonal framework and define important therapeutic vulnerabilities; however, they do not fully determine the tumor phenotype or treatment response. Their consequences are influenced by cellular plasticity, tissue niches, organ-specific environments, and systemic pressures. This duality explains why trunk driver-directed therapies can be effective in some contexts, while resistance and relapse frequently emerge through higher-dimensional adaptations.

Epigenetic plasticity

Epigenetic regulation provides an important interface between the genetic substrate and environmentally shaped tumor cell states.^57,58 Unlike fixed DNA sequence alterations, epigenetic states are reversible, context-dependent, and responsive to tissue-level pressures such as hypoxia, inflammation, metabolic stress, and therapy. Through changes in DNA methylation, chromatin accessibility, and histone modifications, tumor cells with similar genetic backgrounds can acquire distinct transcriptional programs, lineage states, and adaptive phenotypes.^59,60 Thus, epigenetic plasticity expands the phenotypic space available to a clone and helps explain why genetically related tumor cells may differ in invasiveness, immune interaction, and treatment tolerance.

DNA methylation is a clinically relevant aspect of this plasticity.⁶¹ Rather than acting only as a static molecular marker, methylation patterns can reflect both tumor lineage and adaptive state.^62,63 In gliomas, O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation is associated with survival and treatment stratification, including in 1p/19q-codeleted tumors, supporting the clinical relevance of methylation-based heterogeneity.⁶⁴ In neuroendocrine prostate cancer, elevated DNA methyltransferase activity and global methylation dysregulation are linked to neuroendocrine lineage programs, and the loss of DNA methyltransferases (DNMT) genes can reduce neuroendocrine marker expression and impair tumor progression in vivo.⁶⁵ These examples indicate that methylation remodeling can connect the genomic background with lineage identity, therapeutic vulnerability, and interpatient variation.

Histone modifications provide another mechanism by which tumor cells integrate intrinsic and extrinsic signals. Altered acetylation, methylation, phosphorylation, and ubiquitination can reshape chromatin states, thereby influencing epithelial-mesenchymal transition (EMT), stemness, metabolism, invasion, and therapy response.⁶⁶ For example, histone deacetylase (HDAC)-dependent regulation of H3K9 acetylation at the NEDD9 promoter is linked to focal adhesion kinase (FAK) activation and breast cancer metastasis.⁶⁷ AMP-activated protein kinase (AMPK)-mediated phosphorylation of PHD finger protein 2 (PHF2) can reduce H3K9me2 and suppress EMT-related programs in lung cancer.^68,69

In addition to histone-directed regulation, ubiquitination of non-histone proteins can affect tumor-relevant cellular programs. For instance, the ubiquitination-associated regulation of beclin-1 (BECN1) affects autophagic flux and breast cancer progression.^70,71 This does not indicate a histone modification mechanism, but illustrates that post-translational regulation can influence metabolic adaptation, survival programs, and tumor progression. Together, these mechanisms show how chromatin levels and non-chromatin post-translational regulation contribute to phenotypic plasticity and treatment-response heterogeneity.

From the HITET perspective, epigenetic plasticity should therefore be viewed not as an independent layer separated from the tumor ecosystem, but as a regulatory interface that enables genetic clones to respond to changing niches and flows. Microenvironmental conditions can reshape epigenetic states, while epigenetically defined cell states can alter immune evasion, stromal remodeling, metabolic exchange, and therapeutic responses.

Chromosomal instability and structural variation

Chromosomal instability and structural variation provide another source of heritable diversity in cancer. Alterations such as copy-number changes, translocations,⁷² deletions,⁷³ amplifications,⁷² inversions,⁷⁴ aneuploidy,^75,76 and whole-genome doubling reshape gene dosage and genome organization at both focal and chromosome-wide scales.⁷⁷ These changes can alter oncogenic activity, tumor suppressor function, transcriptional programs, and cellular fitness, and their importance in the context of HITET lies not only in generating genetic diversity but also in creating substrates that can be selected by tissue-level pressures such as hypoxia, immune surveillance, nutrient limitation, stromal interaction, and therapy.

Extrachromosomal DNA (ecDNA) illustrates how structural variations can result in flexible oncogene dosage. Lee et al. analyzed 780 breast cancer genomes and proposed a translocation–bridge amplification mechanism in which interchromosomal translocations generate dicentric bridges that break during mitosis and produce fragments that can circularize into ecDNA, contributing to the amplification of oncogenes, such as erb-b2 receptor tyrosine kinase 2 (ERBB2) and cyclin D1 (CCND1).⁷² Similarly, seismic amplification driven by chromothripsis can generate complex ecDNA architectures that either persist as circular elements or reintegrate into chromosomes.⁷⁸ These processes produce diverse amplification patterns across and within tumors, enabling heterogeneous oncogene expression and treatment responses.

The clinical relevance of chromosome-level alterations is also evident in copy number and deletion patterns. For example, the loss of heterozygosity at 1p36 or 19q13, particularly the combined loss of both loci, is associated with prognosis in diffuse infiltrative WHO grade 2 glioma.⁷⁹ This example supports the value of structural alterations in molecular stratification while also illustrating that their clinical effects must be interpreted within the tumor type, lineage, and treatment context.

Recent functional studies have linked chromosomal alterations with phenotypic plasticity and niche-dependent selectivity. ecDNA can mediate focal high-level amplification of oncogenes, generate graded MYC dosage, and promote intratumoral phenotypic heterogeneity in pancreatic ductal adenocarcinoma (PDAC); its copy number can be remodeled in response to selective pressure, although ecDNA may impose fitness costs in the absence of such pressure.⁵¹ Aneuploidy and chromosomal instability (CIN) can generate subclonal diversity by altering chromosomal segregation and gene dosage, whereas whole-genome doubling can facilitate subsequent chromosomal rearrangements and clonal diversification.⁸⁰

Taken together, ecDNA-mediated dosage plasticity, CIN-associated aneuploidy, and whole-genome doubling expand the genetic space available for tumor evolution. However, their consequences are not determined by genome structure alone. Whether a structurally altered clone expands, remains dormant, or is eliminated, depends on its compatibility with cellular programs and tissue habitats. Therefore, chromosomal instability and structural variations serve as dynamic substrates for selection across higher HITET dimensions, particularly at the tissue level.

Beyond the driver–passenger dichotomy

The traditional distinction between driver and passenger mutations is useful for identifying genetic events that confer selective advantages during tumor evolution.^81–83 However, this binary classification does not fully capture the context-dependent effects of mutations within the tumor ecosystem. Some passenger mutations may not directly initiate malignant transformation, but they can persist as evolutionary “hitchhikers” when carried by expanding clones. In some contexts, they may also indirectly influence tumor behavior by shaping the neoantigen load, modifying local immune recognition, or cooperating with other genomic and epigenomic alterations.^83,84

Conversely, canonical oncogenic mutations do not necessarily cause malignant diseases. For example, BRAF V600E is frequently detected in benign melanocytic nevi; however, most nevi remain growth-arrested rather than progressing to melanoma.⁸⁵ Similarly, isocitrate dehydrogenase 1 (IDH1) R132H can appear in early or premalignant glioma-associated contexts, but its oncogenic consequences depend on additional genetic events, epigenetic remodeling, lineage state, and the surrounding brain microenvironment.⁸⁶ These examples indicate that the presence of an oncogenic mutation alone is not sufficient to determine tumor initiation, progression, or therapeutic vulnerability.

Within the HITET framework, these observations can be explained by dimensional interactions. A mutation may generate a genetic possibility; however, its phenotypic outcome can be filtered through cellular programs, tissue-level constraints, organ-specific niches, and systemic host conditions. In benign nevi, oncogene-induced senescence, immune surveillance, and local tissue restriction may constrain the expansion of BRAF-mutant cells. In glioma evolution, IDH1 mutations may reshape metabolism and the epigenetic state; however, malignant progression requires additional cellular and microenvironmental permissiveness. Thus, HITET reframes mutations not as isolated deterministic events, but as context-dependent perturbations whose consequences emerge through cross-dimensional selection and feedback.

This perspective has potential therapeutic implications. Targeting a trunk driver may be effective when the driver remains functionally coupled to tumor fitness; however, resistance can emerge when higher-dimensional contexts allow alternative cell states, compensatory pathways, or protective niches to sustain survival. Therefore, driver-focused therapy should be interpreted together with cellular state, tissue habitat, organ context, and systemic host regulation.

HITET synthesis: context-dependent substrate

In summary, the genetic dimension within HITET should be understood not as a static catalogue of mutations but as a context-dependent substrate for tumor evolution. Germline and somatic alterations establish inherited constraints and vulnerabilities; epigenetic plasticity expands the range of accessible cell states; chromosomal instability and structural variations remodel gene dosage and genome organization; and the distinction between driver and passenger events is shaped by evolutionary and ecological contexts. Together, these processes generate heritable variations through which selection can act.

However, genetic variation alone does not determine the tumor phenotype or therapeutic response. The consequences are filtered through cellular programs, tissue habitats, organ-specific environments, and systemic host pressures. Immune surveillance, stromal interactions, metabolic stress, hypoxia, and therapy can help identify the clones that expand, remain dormant, disseminate, or are eliminated. Thus, the genetic layer provides a necessary, but inadequate, basis for understanding cancer heterogeneity. Within HITET, the full meaning emerges only when interpreted through higher-dimensional niches, flows, and feedback loops that shape tumor evolution.

Cellular dimension: functional diversity and social behaviors

Within the HITET framework, the cellular dimension describes how genetic and epigenetic variations are translated into functional tumor behaviors (Fig. 4). At this level, differences in signaling activity, metabolic state, cell-state plasticity, and intercellular communication generate heterogeneity in proliferation, survival, invasion, immune evasion, and therapeutic responses. These cellular programs are not determined by genetic alterations alone; they are continuously shaped by nutrient availability, hypoxia, therapeutic pressure, and signals from neighboring tumors, immune cells, and stromal cells. Thus, the cellular dimension serves as an interface through which inherited alterations become dynamic phenotypes, which in turn, remodel the local TME.

Fig. 4.

Cellular dimension: functional diversity and intercellular behaviors. This figure summarizes how the cellular dimension of HITET connects genetic and epigenetic variation with functional tumor behaviors. a Cellular social behaviors include clonal competition, clonal expansion, collective invasion with leader–follower organization, extracellular matrix remodeling, and metastatic dissemination. These behaviors are shaped by nutrient availability, oxygen tension, spatial constraints, and local cell–matrix interactions. b Tumor–immune/stromal interactions include immune checkpoint signaling, T-cell inhibition, hypoxia- and VEGF-associated immune escape, mesenchymal stem cell-derived inflammatory factors, fibroblast-associated matrix remodeling, and endothelial-cell-mediated vascular regulation. These interactions help generate local differences in immune activity, stromal support, vascular access, invasion, and therapeutic sensitivity. c Core signaling networks, represented here by the PI3K/AKT/mTOR and Wnt/β-catenin pathways, translate upstream growth-factor or niche-derived signals into metabolic activity, protein synthesis, cell growth, proliferation, and transcriptional programs. Together, these cellular programs illustrate how tumor and non-tumor cells interact, adapt, and organize local microenvironments, providing a mechanistic bridge from the genetic substrate to tissue-level heterogeneity. This figure was created using BioRender (https://biorender.com/)

Core signaling networks

Intracellular signaling networks translate genetic and epigenetic alterations into functional cellular states. In cancer, heterogeneity often arises not from the presence or absence of a pathway alone but from differences in pathway activity, pathway dependency, compensatory crosstalk, and downstream transcriptional output. In HITET, signaling networks serve as key cellular interfaces that connect the genetic substrate to metabolic adaptation, cell-state plasticity, immune interactions, and therapeutic vulnerability. Detailed molecular components of the representative pathways are summarized in Table 1.

The phosphoinositide 3-kinase (PI3K)/protein kinase B (AKT)/mammalian target of rapamycin (mTOR) axis illustrates how graded pathway activity generates clinically relevant functional heterogeneity. Aberrant activation of this pathway occurs across multiple cancer types; however, its intensity, biological consequences, and therapeutic dependency vary according to lineage, molecular subtype, and cellular context.^87,88 PI3K/AKT/mTOR signaling activity is prevalent in IDH-mutant diffuse gliomas and is associated with grade progression in specific molecular subtypes and poorer prognosis in WHO grade 2 diffuse glioma.⁸⁹ These findings suggest that even within a defined molecular class, tumors can be stratified into biologically and clinically distinct groups based on differences in pathway activity. Similarly, patient-derived xenograft models of metaplastic breast cancer show activation of both the PI3K-AKT-mTOR and receptor tyrosine kinases (RTK)-mitogen-activated protein kinase (MAPK) pathways, and combined PI3K and MEK inhibition can induce tumor regression in mutation-driven models.⁹⁰ This example highlights that pathway-level heterogeneity is not only descriptive, but can also define context-specific therapeutic dependencies.

Wnt/β-catenin signaling provides another example of pathway activity heterogeneity, particularly in colorectal cancer (CRC). Rather than functioning as a uniform oncogenic program, Wnt signaling varies across tumor regions and cellular states, contributing to differences in stemness, proliferation, differentiation, invasion, and treatment response.⁹¹ Aberrant Wnt/β-catenin activity is closely linked to cancer stem cell maintenance, recurrence, and drug resistance in CRC. Moreover, Wnt output is shaped not only by tumor-intrinsic mutations but also by microenvironmental cues that modulate β-catenin activity and stemness-associated programs.⁹² Thus, Wnt signaling connects genetic alterations with cellular plasticity and tissue-level niche interactions.

Together, PI3K/AKT/mTOR and Wnt/β-catenin signaling demonstrate that cellular heterogeneity is shaped by dynamic differences in signaling intensity and network coupling rather than by pathway activation alone. These signaling states influence metabolic programs, stemness, immune evasion, and treatment sensitivity, thereby providing a mechanistic bridge from the genetic dimension to metabolic and social behaviors discussed below.

Metabolic heterogeneity across interfaces

Although metabolic outputs can reshape tissue gradients and systemic host physiology, metabolic heterogeneity is discussed here within the cellular dimension, because metabolic reprogramming is first executed as a cell-state program. Differences in nutrient uptake, biosynthesis, redox balance, energy production, and stress adaptation between tumor cells and neighboring non-tumor cells represent a major component of cellular functional heterogeneity. These cellular metabolic states also provide natural interfaces for higher HITET dimensions. Their local outputs shape tissue habitats, whereas diffusible metabolites and inflammatory mediators can influence systemic regulation.

Therefore, metabolic heterogeneity links cellular programmes, tissue habitats, and systemic host regulation. At the cellular level, oncogenic signaling and epigenetic states can rewire glucose, glutamine, and lipid metabolism, allowing tumor cells to adjust their biosynthetic activity, redox balance, energy production, and survival under stress.⁹³ However, this is not a consequence of fixed genetic programs alone. Single-cell, spatial multi-omics, and isotope-tracing studies indicate that intratumoral metabolic states are shaped by nonclonal regulatory differences and microenvironmental gradients.⁹⁴

At the cellular interface, signaling–metabolism coupling creates distinct metabolic dependencies among tumor subpopulations. For example, PI3K/AKT signaling can promote glycolytic programs by regulating glucose transporter localization and the activity of metabolic enzymes and transcription factors.^95–97 Differences in glucose transporter 1 (GLUT1) expression or membrane localization can generate variations in glucose uptake and proliferative capacity among tumor cells.^98,99 Similar principles apply to glutamine and lipid metabolism, where the pathway activity and cellular state determine whether tumor cells rely on anaplerosis, fatty acid synthesis, fatty acid oxidation, or antioxidant programs. These differences create metabolic vulnerabilities that vary across clones, cell states, and treatment contexts.

In addition to intracellular signaling–metabolism coupling, intercellular mitochondrial transfer provides another mechanism through which metabolic states can be redistributed across cell populations. Mitochondria can be transferred between tumor cells and neighboring stromal or immune cells through tunneling nanotubes, extracellular vesicles, or other contact-dependent routes.¹⁰⁰ A single-cell deconvolution study using mitochondrial-enabled reconstruction of cellular interactions (MERCI) identified near-unidirectional mitochondrial transfer from T cells to cancer cells in human tumors, which is associated with energy production programs, increased cancer-cell cycling, T cell depletion, and poor clinical outcome.¹⁰¹ Conversely, mitochondrial transfer from bone marrow stromal cells to CD8-positive T cells through nanotubular connections enhanced T cell respiration, reduced exhaustion, improved tumor infiltration, and strengthened antitumor activity.¹⁰² Thus, mitochondrial transfer can redistribute metabolic fitness between tumor and immune cells in a donor-recipient-dependent manner, linking cellular communication to metabolic adaptation and therapeutic responses.

At the tissue interface, cellular metabolism helps generate local gradients that reshape the TME. An uneven blood supply, oxygen tension, nutrient availability, lactate accumulation, extracellular acidity, and waste clearance result in distinct spatial metabolic niches. These niches can influence immune cell function, stromal remodeling, invasion, and drug sensitivity. Tumor-associated macrophages (TAM) provide an example of this coupling: single-cell analyses have identified metabolically distinct TAM states, including purine metabolism-enhanced subsets associated with protumor phenotypes and poor immunotherapy response.^103,104 Thus, metabolic heterogeneity is not restricted to malignant cells but also involves non-tumor cells within the local ecosystem.

At the systemic interface, tumor- and host-derived metabolic products can propagate beyond the local niche. Lactate, lipids, amino acid metabolites, inflammatory mediators, and microbiota-related metabolites may influence immune tone, endocrine regulation, cachexia, and distant organ environments. Thus, metabolic heterogeneity links the cellular pathway activity to tissue-level gradients and systemic host physiology. Mapping these metabolic interfaces within HITET may help identify context-specific vulnerabilities and guide combination strategies that integrate signaling inhibition, metabolic targeting, immune modulation, and ecosystem-level intervention.^105–107

Cellular social behaviors

Within the HITET framework, cellular social behaviors refer to how tumor cells compete, cooperate, migrate, enter dormancy, acquire stem-like states, and interact with neighboring cells. These behaviors are shaped by intrinsic genetic and epigenetic states, but also strongly influenced by niche-derived signals, resource availability, therapeutic pressure, and intercellular communication. Therefore, cellular heterogeneity is not limited to differences in molecular profiles, but also includes differences in how tumor cells act within a shared ecological space. Clonal competition,¹⁰⁸ collective migration,¹⁰⁹ and cancer stem cells (CSCs) plasticity¹¹⁰ are the three major behavioral modes through which tumor cells generate invasion, recurrence, and therapeutic resistance.

Clonal competition reflects the selective expansion of tumor subpopulations with growth, survival, or niche-adaptive advantages. Different clones may compete for limited nutrients, oxygen, space, and access to supportive signals, leading to clonal expansion, elimination, or dormancy. In CRC, for example, Apc-mutant cells can achieve fixation by overexpressing the secreted WNT antagonist NOTUM, which suppresses WNT signaling in neighboring wild-type crypt cells, restricts their proliferation, and promotes their differentiation. Inhibition of NOTUM impairs the expansion of Apc-mutant cells and prevents adenoma formation.^111,112 This illustrates that clonal competition is driven not just by cell-autonomous growth advantages, but also depends on secreted factor-mediated reshaping of the local niche.

Collective migration is another form of cellular social behavior. During invasion and metastasis, tumor cells may migrate as coordinated groups rather than as isolated single cells. In such collectives, leader cells at the invasive front often display stronger polarity, matrix remodeling capacity, and migratory ability, whereas follower cells advance along the tracks created by leader cells.^113,114 Leader cells can remodel the ECM, communicate with follower and stromal cells, and activate pathways such as RHO, MAPK, PI3K, transforming growth factor beta (TGFβ), vascular endothelial growth factor (VEGF)A–VEGFR, Notch, and gap-junction-mediated signaling to coordinate collective invasion.^113,115 Rather than listing these pathways as independent mechanisms, HITET views collective migration as a division of labor among plastic cellular states, in which invasive leader-like cells and supportive follower-like cells cooperate to promote spatial expansion.¹¹⁶

CSC plasticity further illustrates the dynamic regulation of cellular behavior. Glioblastoma stem cells (GSCs) self-renew while giving rise to multiple tumor cell lineages. This dual capacity contributes to intratumoral heterogeneity, tumor progression, treatment resistance, and posttreatment recurrence.¹¹⁷ However, increasing evidence suggests that CSCs should not be viewed only as a fixed-cell population. Instead, stem-like properties can be reversibly acquired or lost under the influence of genetic and epigenetic regulation, therapeutic stress, hypoxia, and TME signals.¹¹⁸ CSC states with different degrees of EMT or mesenchymal features may occupy distinct spatial niches, such as invasive fronts or tumor cores, and signaling axes, such as Notch–Jagged–IL-6, can stabilize intermediate plastic states and expand the CSC fraction.^110,119 Non-coding RNA-mediated regulation, including miRNAs and lncRNAs, further tunes these plastic cell state programs by modulating pathway output, EMT, stemness, and therapy-associated adaptation.^120,121

CSC-like states can also remodel the tumor immune microenvironment, thereby promoting immune evasion and invasion. In glioblastoma multiforme (GBM) and other cancers, reciprocal interactions among stemness programs, EMT-like states, immune suppression, and niche signals can reinforce functional heterogeneity and increase therapeutic resilience.¹²² Thus, cellular social behavior provides a mechanistic bridge between intracellular programs and tissue-level ecosystem organization. They explain how tumor cells in distinct states compete, cooperate, migrate, and adapt to local niches, ultimately shaping invasion, relapse, and resistance.

Tumor–immune/stromal interactions

Tumor cells continuously interact with immune and stromal cells, and these interactions play a key role in spatial and functional heterogeneity within the tumor microenvironment. Through immune checkpoint engagement, cytokine and chemokine signaling, ECM remodeling, angiogenic regulation, and exosome-mediated communication, tumor and non-tumor cells jointly create local niches with distinct immune activity, stromal density, vascular access, and therapeutic sensitivity. Within the HITET framework, these interactions provide an important bridge from cellular behavior to tissue-level ecosystem organization.

Immune checkpoint signaling illustrates how tumor–immune interactions can generate heterogeneous therapeutic responses. The programmed cell death protein 1 (PD-1)/PD-L1 axis suppresses T cell activity and can be clinically targeted in several cancers. In non-small cell lung cancer (NSCLC), PD-1/PD-L1 blockade is an important treatment strategy that can be combined with chemotherapy, targeted therapy, or additional immunomodulatory agents to improve efficacy and overcome resistance.^123–126 This indicates that immune checkpoint dependence is not uniform but varies according to tumor genotype, immune infiltration, treatment context, and compensatory immune pathways. In contrast, PDAC has a limited response to PD-1/PD-L1 blockade.¹²⁷ Single-cell analyses of PDAC circulating tumor cells and tumor biopsies suggest that PD-1/PD-L1 interactions may be weak or nearly absent in some contexts, whereas alternative mechanisms, such as platelet-mediated protection of circulating tumor cells and HLA-E–CD94/NKG2A-dependent natural killer (NK) cell suppression, may support immune escape.¹²⁸ These findings suggest that NSCLC and PDAC rely on distinct immune-evasion architectures, resulting in tumor-type-specific immune niches and divergent responses to checkpoint blockade.

Beyond checkpoint signaling, immunosuppressive cytokine and chemokine networks further shape local immune niches. Tumor cells and non-tumor cells can recruit suppressive myeloid populations, inhibit effector T cells and NK cells, and promote macrophage polarization. For example, beta-2-microglobulin (B2M)-mediated activation of PI3K/AKT/mTOR signaling and TGFβ1 secretion can support glioma stem cell survival and self-renewal while promoting M2-like macrophage polarization, thereby remodeling the tumor immune microenvironment toward an anti-inflammatory and tumor-promoting state.^129,130 In breast cancer, C-C motif chemokine ligand 20 (CCL20)-driven recruitment and expansion of polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs) can enhance cancer cell stemness through C-X-C motif chemokine receptor 2 (CXCR2)-dependent mechanisms.¹³¹ These examples show how immune interactions can form spatially uneven “hot” and “cold” immune regions, generating local heterogeneity in immune activity and therapeutic response.

Stromal cells also contribute to heterogeneous niche formation.^132,133 Cancer-associated fibroblasts (CAFs) remodel the ECM, regulate angiogenesis, and modulate immune exclusion through secreted factors and matrix-associated signals. Therefore, different CAF states may create tumor regions with distinct invasive capacities, vascular access, drug penetration, and immune infiltration.^134–136 Mesenchymal stem cells (MSCs) provide another stromal interface. Through cytokines, such as IL-6 and IL-17A, or through exosome-mediated transfer of regulatory RNAs, MSCs can support tumor growth, drug resistance, stemness maintenance, and clonogenic potential in specific cancer contexts.^137,138 These CAF- and MSC-mediated circuits illustrate how stromal cells help convert cellular communication into structured tissue niches.

Together, tumor–immune and tumor–stromal interactions transform cellular heterogeneity into spatially organized microenvironments. Immune checkpoint activity, myeloid recruitment, CAF-mediated matrix remodeling, and MSC-derived cytokines or exosomal signaling do not operate uniformly across tumors. Instead, they create regional niches with varying levels of immunosuppression, stromal support, vascular access, invasion potential, and sensitivity to treatment. This spatial organization provides a direct conceptual transition to the tissue dimension, where these local interactions are further integrated into the vascular, stromal, hypoxic, and immune landscapes.

HITET synthesis: from cellular programs to tissue niches

Within the HITET framework, the cellular dimension explains how heritable alterations are translated into functional tumor behaviors. Signaling networks convert genetic and epigenetic variation into heterogeneous pathway activity and therapeutic dependencies; metabolic programs determine how tumor and non-tumor cells use nutrients, maintain redox balance, and adapt to stress; cellular social behaviors such as clonal competition, collective migration, and CSC plasticity shape invasion, recurrence, and resistance; and tumor–immune/stromal interactions further organize immune suppression, matrix remodeling, and niche support.

Together, these processes show that cellular heterogeneity is not only a matter of molecular identity but also a matter of functional behavior. Tumor cells in different states compete, cooperate, migrate, enter dormancy, evade immunity, and adapt to treatment in a context-dependent manner. Importantly, these behaviors are not confined to individual cells. Through cytokines, metabolites, ECM remodeling, immune cell recruitment, exosomes, and cell–cell contact, cellular programs generate spatially organized microenvironments. Thus, the cellular dimension provides a mechanistic bridge between the genetic substrate and tissue habitats, as discussed in the next section.

Tissue dimension: the heterogeneous TME and beyond

Within the HITET framework, the tissue dimension represents the local habitat in which tumor cells interact with the structural, biochemical, vascular, stromal, and immune components of the surrounding tissue (Fig. 5). This dimension extends beyond the aggregate behavior of individual cells and emerges from the spatial organization of tumor cells and non-tumor compartments within the TME.^139,140 The TME is not a uniform entity but a spatial mosaic of micro-niches characterized by heterogeneous distributions of oxygen, nutrients, immune cells, stromal elements, ECM architecture, vascular access, and physical forces.^141,142 This local habitat diversity, including the peritumoral microenvironment (PME), acts as a selective filter that influences the expansion, persistence, invasion, and therapy response of cellular phenotypes. Therefore, understanding the tissue dimension is essential for linking cellular behaviors to the spatial patterns of immune escape, invasion, and therapeutic resistance.

Fig. 5.

Tissue dimension: spatial heterogeneity of the tumor microenvironment and peritumoral niche. This schematic illustrates how spatial immune organization, biochemical gradients, and physical properties collectively shape tissue-level heterogeneity within the Holistic Integrative Tumor Ecosystem Theory (HITET) framework. a Spatial tumor microenvironment (TME) heterogeneity: Tumor regions may exhibit inflamed, immune-excluded, or immune-cold phenotypes along a continuum from immune infiltration to immunosuppression. These spatial immune states are influenced by the peritumoral microenvironment, including regional differences in perfusion and oxygenation, metabolites and pH, chemokine distribution, and stromal architecture. b Biochemical gradients: Hypoxia and hypoxia-inducible factor (HIF) signaling, metabolic reprogramming, and aberrant angiogenesis generate spatial variations in oxygen availability and other biochemical conditions within and around the tumor. c Physical heterogeneity: Cancer cells and fibroblasts participate in extracellular matrix remodeling, while matrix-derived mechanical stimulation reciprocally influences cellular behavior. Together, these interconnected processes establish spatially distinct tissue habitats that contribute to regional differences in immune infiltration, tumor progression, and therapeutic response. This figure was created using BioRender (https://biorender.com/)

Spatial immune heterogeneity

Spatial immune heterogeneity reflects how immune cells are distributed, retained, excluded, or functionally suppressed across tumor regions. At the tissue level, immune cell infiltration is shaped by biochemical gradients, stromal architecture, vascular permeability, chemokine fields, and physical barriers. These features generate three commonly described immune phenotypes: inflamed, excluded, and cold tumors.^143,144 Inflamed or “hot” tumors contain abundant immune-cell infiltration and inflammatory signals and are generally more responsive to immune checkpoint blockade. Cold tumors show limited immune infiltration and dominant immunosuppressive programs, which reduce the likelihood of effective antitumor immunity. Excluded tumors occupy an intermediate state, in which immune cells accumulate at the tumor margin or stromal interface but fail to penetrate the tumor core, often because of CAF-associated TGFβ signaling, abnormal vasculature, and dense ECM barriers.¹⁴⁵

These immune phenotypes should not be viewed as static categories, but as spatial readouts of the underlying tissue organization. For example, oxygen tension, lactate accumulation, extracellular acidity, chemokine distribution, stromal density, and vascular access jointly determine the entry and function of cytotoxic T, myeloid, and other immune cell populations. Mechanistic evidence further supports the plasticity of these immune responses. Castiglioni et al. showed that dual blockade of TGFβ and PD-L1 promotes the expansion and intratumoral entry of interferon gamma (IFNγ)-high CD8 T cells, reprograms myeloid, stromal, and tumor niches, and converts immune-excluded tumors into immune-inflamed tumors.¹⁴⁶ These findings illustrate that spatial immune heterogeneity can be therapeutically remodeled by targeting both immune checkpoints and stromal exclusion mechanisms.

Biochemical gradients

Biochemical gradients are a major organizing principle for tissue-level heterogeneity. These gradients arise from mismatched supply and demand, rapid tumor growth, abnormal vascular remodeling, uneven perfusion, stromal compression, and heterogeneous metabolic activity. Spatial variations in oxygen tension, nutrient availability, metabolites, pH, ions, cytokines, and chemokines create microhabitats that differentially constrain and instruct tumor and non-tumor cells.^142,147 In HITET, biochemical gradients represent a tissue-level interface through which cellular metabolism, vascular flow, and immune functions are coupled into spatially organized niches.

Hypoxia is one of the most important biochemical gradients in solid tumors. Because vascular distribution and perfusion are uneven, hypoxia-inducible factor (HIF) signaling is activated in a region-specific manner, generating heterogeneous hypoxic niches with distinct invasive, angiogenic, metabolic, and immune-suppressive features.^148–150 In hepatocellular carcinoma (HCC), loss of glutathione S-transferase zeta 1 (GSTZ1) promotes succinylacetone accumulation, inhibits PHD2 activity, stabilizes HIF-1α, and increases VEGF expression, thereby promoting abnormal angiogenesis.¹⁵¹ In cholangiocarcinoma, WDR5 enhances HIF-1α accumulation through c-Myc-dependent transcriptional regulation and HDAC2-mediated suppression of PHD2, promoting epithelial-mesenchymal transition (EMT), invasion, and metastasis.¹⁵² These examples show that although hypoxia/HIF signaling is a shared tissue-level stress axis, its upstream regulation and downstream consequences are tumor type-specific.

Vascular remodeling further reinforces the biochemical gradients. Tumor-driven VEGF signaling promotes angiogenesis; however, the resulting vessels are often structurally abnormal, leaky, poorly perfused, and spatially uneven.^153,154 These vascular abnormalities cause regional differences in oxygenation, nutrient delivery, drug penetration, immune cell infiltration, and waste clearance. Hypoxia can reactivate VEGF signaling, creating a feedback loop that maintains abnormal vasculature and regional immune suppression.^153,155 Studies in prostate cancer models and metastatic renal cell carcinoma further illustrate the context-dependence of this axis. VEGF–neuropilin-2 signaling can influence immune regulation in prostate cancer models, whereas resistance to VEGF-targeted agents remains a major therapeutic challenge in metastatic renal cell carcinoma.^156,157 These studies provide mechanistic or translational evidence regarding vascular regulation and therapeutic resistance rather than evidence of a uniform clinical benefit of VEGF-targeted therapy across cancer types.

Metabolic gradients also contribute to local niche formation, but here the focus is on tissue-level consequences rather than on the metabolic pathways themselves. Lactate accumulation, extracellular acidification, nutrient depletion, and altered amino acid or lipid availability can suppress immune cell function, promote invasion, and generate therapy-resistant regions.^158–164 Together, hypoxia, abnormal perfusion, lactate/pH gradients, and cytokine or chemokine fields create interlocking biochemical landscapes that shape immune activity, invasive behavior, and treatment response across tumor regions.

Physical heterogeneity

Physical heterogeneity refers to regional differences in ECM composition, tissue stiffness, solid stress, interstitial fluid pressure, and mechanical force transmission. These physical features determine how tumor cells move, how immune cells infiltrate, how blood vessels collapse or remain functional, and how drugs are distributed through the tissue.^165,166 Therefore, the mechanical microenvironment is not simply a passive scaffold but an active organizer of tissue-level heterogeneity.

ECM remodeling is a central source of physical heterogeneity. Changes in collagen deposition, fibronectin organization, matrix cross-linking, and remodeling enzyme activity alter tissue stiffness and cell–matrix signaling.^167–172 Matrix metalloproteinases (MMPs) contribute to invasion by degrading ECM barriers and releasing bioactive matrix fragments, and MMP activity is closely associated with malignant progression in multiple cancers.^173,174 CAFs further amplify matrix heterogeneity by producing collagen, fibronectin, angiogenic factors, and immune-modulatory signals.¹⁷⁵ In pancreatic cancer, targeting and reprogramming CAFs can alleviate hypoxia and improve chemotherapy sensitivity.¹⁷⁶ In bladder cancer, an SLC14A1-positive CAF subset enhances tumor stemness through WNT5A and is associated with treatment resistance and poor prognosis, whereas inhibition of this CAF subset improves chemotherapy efficacy.¹⁷⁷ These findings indicate that matrix remodeling is driven by functionally distinct stromal programs, rather than by a uniform fibrotic process.

Solid stress provides another layer of mechanical heterogeneity. It arises from tumor cell proliferation, stromal expansion, abnormal ECM deposition, and confinement by surrounding tissues. The reported levels of solid stress vary markedly across cancer types, ranging from relatively low levels in GBM to much higher levels in PDAC, illustrating strong organ- and tumor-type specificity.¹⁷⁸ In GBM, low solid stress may coexist with diffuse infiltration through soft brain parenchyma, whereas in PDAC, high stiffness and solid stress contribute to vascular compression, hypoxia, poor drug penetration, and immune exclusion.^178,179 These examples show how physical properties can shape distinct tissue habitats even when similar cellular programs are present.

Interstitial fluid flow and pressure further stratify the tissue habitat. Leaky tumor vessels, impaired lymphatic drainage, and compression of blood and lymphatic vessels increase interstitial fluid pressure and generate directional flow at the tumor margins.^180,181 This flow can impose shear stress, alter cell–matrix interactions, create chemical gradients, and influence immune-cell recruitment and drug distribution. Consequently, tumor cores and invasive fronts may experience different mechanical, biochemical, and therapeutic conditions. ECM stiffness integrates many of these physical features and can promote malignant behavior when tumor cells migrate through stiff or confined matrices.^182,183 Emerging approaches combining spatial transcriptomics, mechanical mapping, and atomic force microscopy may help define mechanical niches and identify strategies to overcome physical barriers to therapy.^184,185

Peritumoral microenvironment (PME)

Beyond the tumor core, the PME, which is a dynamic interface between the tumor and normal parenchyma, has emerged as a critical habitat governing recurrence and therapeutic resistance. Characterized by unique immune landscapes, metabolic gradients, and stromal activation, the PME represents a therapeutically targetable ecological niche that is distinct from the tumor core.¹⁸⁶

The PME possesses distinct physical and immunological properties that promote tumor progression through interrelated mechanical cues and immune activities. Considering GBM as an example, tumor recurrence predominantly occurs in the peritumoral brain zone (PBZ). Within the PBZ, a high blood perfusion interface (HBI) is characterized by increased vascularization, dense immune cell infiltration, and an active tumor transcriptional profile, suggesting that HBI may serve as a critical source of recurrence and a promising target for postoperative therapeutic intervention.¹⁸⁷ Similarly, in HCC, peritumoral hepatocytes reshape the local immune microenvironment through the m⁶A reader protein YTHDF2, thereby promoting CD8⁺ T cell recruitment and activation and enhancing tumor responsiveness to immunotherapy. Mechanistically, oxaliplatin induces YTHDF2 expression in hepatocytes by activating the cyclic GMP–AMP synthase–stimulator of interferon genes (cGAS–STING) pathway and stabilizes the chemokine Cx3cl1 transcript in an m⁶A-dependent manner, amplifying antitumor immune responses.¹⁸⁸ These findings further highlight the critical role of PME in shaping immunotherapeutic efficacy and support the rationale for targeting PME as a precision therapeutic strategy.

Beyond serving as a driving force for peripheral invasion and local dissemination, the PME is a major organizer of spatial and functional tumor heterogeneity at the tumor margin.¹⁸⁶ By creating gradients and mosaics in vascular density, immune cell infiltration, stromal composition, metabolic stress, and mechanical stiffness, the PME segregates the tumor edge into distinct niches with divergent proliferative, invasive, and immune-evasive phenotypes.^188,189 These niche-specific properties not only determine which tumor cell subsets are preferentially selected during local evolution but also critically shape the accessibility, distribution, and efficacy of immunotherapy and other therapeutic interventions.^187,190 Therefore, understanding the cellular and molecular architecture, regulatory dynamics, and niche-level heterogeneity of PME is important to overcome locally confined resistance and margin-dominant relapses.

HITET synthesis: the habitat layer

In summary, the tissue dimension represents the habitat layer of HITET. This explains how cellular behaviors are spatially organized into local immune, biochemical, vascular, stromal, mechanical, and peritumoral niches. Immune phenotypes, such as hot, excluded, and cold tumors, reflect the combined effects of biochemical gradients and stromal barriers. Hypoxia, lactate, acidity, vascular abnormalities, ECM stiffness, solid stress, and interstitial fluid pressure further define regional habitats with distinct invasion patterns and therapeutic sensitivities. PME extends this habitat logic to the tumor–host boundary, where recurrence and treatment resistance often emerge.

Therapeutically, the tissue dimension suggests that local habitats can be remodeled rather than just targeted at the individual cell level. Strategies, such as vascular normalization, reversal of immune exclusion, stromal reprogramming, matrix remodeling, reduction of physical barriers, and PME-directed intervention, may alter the selective pressures that sustain resistant tumor states. Thus, the tissue dimension provides a spatial basis for understanding why therapies succeed in some tumor regions but fail in others. However, once tumor cells disseminate beyond the primary lesion, they encounter anatomical, physiological, immune, metabolic, and barrier constraints. This transition extends the framework from local habitats to organ-specific and systemic contexts, as discussed below.

Organ and systemic dimensions: macro-scale contexts of tumor heterogeneity

Within the HITET framework, the organ and systemic dimensions describe how local tumor evolution is shaped by organ-specific habitats and whole-host physiology (Figs. 6, 7). Tumors with similar molecular drivers can behave differently across anatomical sites because each organ has a distinct vascular architecture, stromal composition, immune tone, metabolic substrate availability, barrier properties, and neural or endocrine inputs. At the systemic level, microbiota-derived metabolites, endocrine and metabolic states, inflammatory programs, and neural regulation can synchronize or de-synchronize selective pressure across organs. Thus, these dimensions explain why tumor behavior and therapeutic response cannot be fully inferred from genotype alone, but must be interpreted in relation to metastatic site, host state, and inter-organ communication.

Fig. 6.

Organ dimension: organotropism, pre-metastatic niche formation, and organ-specific metastatic tropism. This schematic illustrates how organ-level conditions shape metastatic heterogeneity within the Holistic Integrative Tumor Ecosystem Theory (HITET) framework. a Molecular basis of organotropism: Tumor lineage, molecular subtype, and genotype are associated with distinct patterns of metastatic dissemination, as illustrated by the preferential involvement of bone by prostate cancer and estrogen receptor-positive (ER-positive) breast cancer, lung by ER-negative breast cancer, and lung by KRAS-mutant colorectal cancer. These organ-specific metastatic patterns are associated with differences in metastatic burden and clinical outcome. b Pre-metastatic niche formation: Primary tumors release exosomes and cytokines that act on distant organs, recruit myeloid-derived suppressor cells (MDSCs) and regulatory T cells (Tregs), and promote tissue fibrosis, thereby contributing to the establishment of a pre-metastatic niche (PMN). c Organ-specific metastatic tropism: Hemodynamic routes, immune tolerance, cytokine signaling, organ-specific microenvironments, and immunosuppressive niches collectively influence the survival and expansion of disseminated tumor cells in particular organs. Together, these processes determine organ-selective metastatic colonization and contribute to heterogeneity in metastatic progression and treatment response. This figure was created using BioRender (https://biorender.com/)

Fig. 7.

Systemic dimension: interconnected host-level regulatory interfaces in cancer heterogeneity. This schematic illustrates the systemic dimension of the Holistic Integrative Tumor Ecosystem Theory (HITET), in which the tumor and its local microenvironment are embedded within a network of interconnected host-level processes. The surrounding sectors depict six major systemic interfaces. Cancer-associated cachexia links chronic inflammation with progressive loss of body mass and systemic functional decline. Endocrine, metabolic, and energy dysregulation involves disruption of the hypothalamic–pituitary–adrenal (HPA), hypothalamic–pituitary–thyroid (HPT), and hypothalamic–pituitary–gonadal (HPG) axes, thereby connecting neuroendocrine regulation with altered energy homeostasis. Systemic metabolic disorders encompass disturbances in glucose, amino-acid, and lipid metabolism that can influence nutrient availability, inflammatory status, and tumor-supporting host conditions. The brain–body interface represents bidirectional communication through which neural and systemic signals influence tumor behavior, while tumor-associated signals may alter central nervous system function. Neuro–tumor interactions connect neural signaling and inflammation with tumor progression and neurological dysfunction. The gut microbiota–immune–metabolic axis links intestinal microbial communities with host immunity, systemic metabolism, and tumor-associated immune regulation. Together, these interacting systemic processes contribute to interpatient heterogeneity in tumor progression, treatment tolerance, toxicity, and therapeutic response and provide potential opportunities for host-directed intervention. This figure was created using BioRender (https://biorender.com/)

Organ-specific metastasis

Metastasis is not a random dissemination process, but is shaped by compatibility between disseminated tumor cells and organ-specific microenvironments.^191–193 Organotropism reflects both the intrinsic tumor properties and extrinsic features of target organs, including vascular routes, immune surveillance, stromal support, metabolic resources, and pre-existing or tumor-induced niche states.

Molecular basis of organotropism

The clinical patterns of metastasis illustrate the relationship between the tumor subtype and organ-specific colonization. In breast cancer, estrogen receptor (ER)-positive tumors preferentially metastasize to the bone, whereas human epidermal growth factor receptor 2 (HER2)-positive tumors more frequently involve the brain, liver, and lungs; lung metastases are especially common in ER-negative disease.¹⁹⁴ In CRC, KRAS-mutant tumors are more likely to develop lung metastases than KRAS wild-type tumors, whereas BRAF-mutant tumors have a higher tendency toward peritoneal dissemination.¹⁹⁵ Prostate cancer most commonly metastasizes to bone, which is generally associated with more favorable survival than liver or lung metastases.¹⁹⁶ These examples suggest that metastatic site selection is influenced by tumor lineage, genotype, and compatibility with organ-specific niches.

Genomic alterations also contribute to organotropic behavior and metastatic progression. Analysis of primary and metastatic tumor samples shows that copy number alteration burden is increased in metastases, particularly in breast, pancreatic, and prostate cancers. Whole-genome doubling is also more frequent in metastatic lesions and may promote chromosomal instability, clonal diversification, and adaptation to distant sites. In addition, metastatic tumors often acquire or select immune-evasion features, including CDKN2A deletion, human leukocyte antigen class I (HLA-I) loss of heterozygosity, and alterations in antigen presentation pathways.¹⁹⁷ Thus, organotropism is shaped by both organ-specific ecological filtering and genotype-dependent evolutionary capacity.

The specific organ context also influences treatment response. The same actionable pathway may produce different therapeutic outcomes, depending on the tumor type and metastatic site. For example, EGFR inhibitors are effective in selected EGFR-mutant NSCLC, whereas EGFR-targeted strategies have shown more limited benefit in pancreatic cancer despite EGFR pathway involvement.^35,36,198 Similarly, immune checkpoint inhibitors show strong activity in melanoma and subsets of NSCLC but have limited efficacy in pancreatic cancer and many CRCs.^199,200 In metastatic NSCLC treated with nivolumab, clinical outcomes differ according to metastatic site, with lymph node metastases showing more favorable responses than liver or lung metastases.²⁰¹ These observations suggest that genotype-based treatment selection should be interpreted in the context of metastatic sites and local tissue environments.

Pre-metastatic niche formation

The seed-and-soil concept is now understood not only as passive compatibility between tumor cells and target organs but also as active preconditioning of distant tissues. Primary tumors can induce pre-metastatic niches through secreted factors, extracellular vesicles, inflammatory mediators, and stromal or immune-cell remodeling, thereby increasing the likelihood that disseminated tumor cells will survive and expand.²⁰²

Lung metastasis of breast cancer is a representative example. Cyclooxygenase-2 (COX-2)-positive pulmonary fibroblasts promote pre-metastatic niche formation by producing prostaglandin E2 (PGE2), which induces dendritic cell dysfunction and myeloid immune suppression. IL-1β-driven inflammation further reinforces this immunosuppressive lung environment. Deletion of Ptgs2 or inhibition of the PGE2 receptors EP2/EP4 reverses immune suppression, enhances the antimetastatic efficacy of dendritic cell therapy and PD-1 blockade, and reduces lung metastasis.²⁰³ This example illustrates how tumor-derived inflammatory programs and organ-specific stromal responses cooperate to create metastatic-permissive niches.

Organ-specific metastatic tropism

Different target organs impose distinct ecological constraints upon the arrival of tumor cells. Bone metastases, which are frequently observed in breast, prostate, and lung cancers, are shaped by osteoclast activation, bone marrow immune regulation, and growth factor release from the bone matrix. Tumor-derived PTHrP promotes osteolytic remodeling, whereas TGFβ and CSF1 can support myeloid-derived suppressor-cell differentiation and immune suppression.²⁰⁴ Lung metastases, common in breast cancer and melanoma, involve inflammatory and stromal remodeling, and tumor-derived CCL2 and IL-6 promote fibroblast activation and immunosuppressive niche formation.^205,206 These examples show that metastatic heterogeneity is not only determined by disseminated tumor cells but also by the resident cells of the target organ and the immune architecture.

Hemodynamics further shape metastatic tropism. In PDAC, liver metastasis is facilitated by the portal venous route, and portal vein circulating tumor cells (CTCs) may represent an intermediate state during dissemination.²⁰⁷ Eliminating CTCs from the portal vein effectively suppresses PDAC metastasis.¹²⁸ Multi-omics analyses further indicate that liver-tropic PDAC is characterized by persistent replication stress, high primary organotropism (pORG) gene-set expression, and poor immune infiltration, whereas lung-tropic or liver-averse tumors show lower replication stress and richer tumor-infiltrating lymphocytes.²⁰⁸

Metastatic trajectories can diverge in the same patient. Multi-region sequencing of CRC metastases has shown that liver, lung, and other organ lesions can arise from distinct clonal branches and carry heterogeneous driver gene constellations with sequential, branching, or discrete dissemination patterns.^209,210 These findings support lesion-specific sampling and molecular profiling of different metastatic sites, particularly when treatment resistance emerges in a site-restricted manner.

Systemic ecological interfaces

Systemic heterogeneity is organized by host-level interfaces that connect whole-body physiology with local tumor habitats. These include the gut microbiota–immune–metabolic axis, endocrine and metabolic systems, inflammatory programs, and neural regulation. Rather than directly changing the tumor genotype, these interfaces influence the immune tone, nutrient availability, hormonal signaling, tissue barriers, stress responses, and treatment tolerance. Therefore, they help explain the inter-patient variability in progression, metastatic patterns, toxicity, and therapeutic response.

Gut microbiota–immune–metabolic axis

The gut microbiota can be regarded as a functional organ-like ecosystem because it produces metabolites, microbial components, and inflammatory signals that regulate host immunity, metabolism, and distant tissue environments.^211–215 Pan-cancer multi-omics studies indicate that intratumoral microbiota in metastatic lesions are shaped by both primary tumor type and metastatic organ, and are associated with hypoxia, inflammation, ECM remodeling, immune-cell infiltration, and response to immune checkpoint blockade.²¹⁶ Thus, microbial heterogeneity contributes to both systemic and local tumor heterogeneity.

In CRC, dysbiosis can reshape the TME through microbial metabolites that modulate epithelial and immune-cell functions.^217,218 In multifocal hepatocellular carcinoma, intrahepatic lesions can harbor highly heterogeneous microbial communities. Intrahepatic metastases (IM)-HCC and multicentric occurrence (MO)-HCC show systematic differences in microbiota composition, and a nine-species bacterial signature can distinguish between these patterns. Bacteria enriched in IM nodules, such as Enterococcus faecium and Streptococcus anginosus, are associated with EMT activation and an immunosuppressive microenvironment, and can promote HCC migration and invasion in orthotopic models.²¹⁹ These findings extend the concept of the microbiome from the gut to organ-level tumor heterogeneity.

Endocrine and metabolic systems

Systemic metabolic and endocrine regulation provides another host-level interface that shapes cancer heterogeneity. Tumor metabolism begins at the cellular level, contributes to tissue gradients, and eventually causes systemic metabolic disturbances. In parallel, endocrine axes regulate stress, sex hormones, thyroid function, appetite, muscle metabolism, adipose tissue, and immune–metabolic coupling. These processes vary across tumor types, stages, treatments, and host backgrounds, creating patient-specific systemic states that influence tumor progression and therapy tolerance.

Multiple metabolic disorders

Cancer-associated metabolic dysregulation includes glucose, amino acid, and lipid metabolism. Increased tumor glycolysis and lactate production can reduce glucose availability in peripheral tissues and contribute to insulin resistance, especially when combined with tumor-derived inflammatory cytokines that impair insulin signaling.^160,220 Tumor-induced lipolysis increases circulating free fatty acids, which can further worsen insulin resistance and systemic metabolic stress.²²¹ These changes connect cellular metabolic reprogramming with whole-body glucose dysregulation.

Amino-acid metabolism is the second layer of systemic metabolic heterogeneity. Tumor cells may increase glutamine uptake to support energy production, nucleotide synthesis, redox control, and adaptation to hypoxia, and glutamine-derived α-ketoglutarate enters the tricarboxylic acid cycle.²²² Leucine uptake enabled ER-positive breast cancer cells to alleviate nutrient stress, while branched-chain amino acids contribute to tumor metabolism in several contexts.^223,224 Differences in dependence on glutamine, leucine, or branched-chain amino acids create heterogeneous metabolic vulnerabilities across cancer types and cell states.

Lipid metabolism is also frequently altered. Tumors may increase fatty-acid synthesis to support membrane production and signaling, whereas fatty-acid oxidation can provide an adaptive energy source under metabolic stress.^225,226 Cholesterol metabolism is relevant not only for membrane structure and steroid synthesis but also for immune regulation. Increased cholesterol synthesis and HMG CoA reductase (HMGCR) expression support tumor growth, and elevated cholesterol levels are linked to prostate cancer progression.^227,228 Cholesterol accumulation in the TME can promote M2-like macrophage polarization and suppress CD8-positive T cell activity.²²⁹ Host factors such as diet, obesity, age, and physical activity further modulate immune-cell metabolism and contribute to inter-patient heterogeneity in disease progression and treatment response.²³⁰

Endocrine system disruption

Both cancer and cancer therapy can disrupt hypothalamic–pituitary–target gland axes, producing tumor type-, stage-, and treatment-specific endocrine phenotypes. The hypothalamic–pituitary–adrenal axis is often affected by stress and paraneoplastic hormone production. In small-cell lung cancer, ectopic adrenocorticotropic hormone (ACTH) secretion can elevate cortisol levels and produce Cushing-like manifestations, including hyperglycemia, muscle wasting, immunosuppression, and mood disturbance.²³¹

The hypothalamic–pituitary–gonadal axis can be altered by tumors and cancer therapies through the suppression of gonadotropin secretion or direct gonadal injury, leading to reduced sex hormone levels and impaired reproductive function.^232,233 Sex hormone signaling also contributes to cancer risk and progression; genetic variations in sex hormone signaling and metabolic pathways are associated with CRC risk.^234,235 Prolactin can further interfere with gonadal function by suppressing gonadotropin-releasing hormone (GnRH) in specific contexts.²³⁶

The hypothalamic–pituitary–thyroid axis is particularly relevant in thyroid cancer. Thyroidectomy or tissue destruction can cause hormone deficiency and require lifelong replacement therapy, whereas suppression of thyroid-stimulating hormone (TSH) with high-dose levothyroxine is often used to reduce recurrence risk in differentiated thyroid cancer; however, long-term treatment may cause osteoporosis and metabolic disturbances.^237,238 Radioiodine therapy also requires precise manipulation of TSH levels through levothyroxine withdrawal or administration of recombinant human TSH.²³⁹ Primary intracranial tumors, including craniopharyngiomas and pituitary adenomas, can directly disrupt hypothalamic–pituitary regulation and cause single- or multi-axis endocrine dysfunction.^240,241 Together, these examples demonstrate how endocrine disruption adds a system-level layer of heterogeneity that affects host physiology and treatment tolerance.

Cancer-associated cachexia

Cancer-associated cachexia is a systemic manifestation of tumor-induced metabolic, endocrine, and inflammatory dysregulation. It is characterized by an ongoing loss of skeletal muscle mass, with or without loss of fat mass, that cannot be fully reversed by conventional nutritional support.^242,243 Its timing and severity vary across cancer types, tumor burden, organ involvement, treatment exposure, inflammatory states, and baseline host metabolism. Cachexia reduces quality of life and lowers tolerance to chemotherapy, radiotherapy, and immunotherapy.²⁴⁴

Chronic inflammation is the central driver of cachexia. Tumor- and microenvironment-derived cytokines promote systemic inflammatory responses, accelerate muscle and adipose tissue depletion, increase protein breakdown, suppress muscle synthesis, and enhance adipose catabolism.^244,245 These effects are partly mediated by the ubiquitin–proteasome system and autophagy–lysosome pathways.²⁴⁶ In the context of HITET, cachexia can be viewed as a systemic phenotype characterized by the sustained coupling of metabolic stress, endocrine resetting, and inflammatory amplification. Its management requires dynamic monitoring and intervention across the metabolism–endocrine–inflammation continuum, rather than nutrition alone.

The brain–body axis

The brain–body axis represents a systemic regulatory interface linking tumor biology to neural, immune, endocrine, and behavioral physiology. Through afferent sensing of inflammatory and metabolic signals and efferent regulation via the autonomic, neuroendocrine, and vagal pathways, the nervous system can influence immune tone, metabolic balance, vascular function, pain, fatigue, appetite, and stress responses.^247,248 Structural neuro–tumor interactions, including perineural invasion, neurite ingrowth, and neuron–tumor synapses, further contribute to organ- and tumor-specific heterogeneity.^249,250

Brain–body interface

Cancer is increasingly being recognized as a systemic disease rather than a purely localized lesion. The nervous system participates in this systemic regulation by coordinating immune responses, metabolism, behavioral outputs, and stress adaptation. The concept of immunoception emphasizes that the nervous system can detect inflammatory states and regulate immune responses rather than acting separately from immunity.²⁵¹ In cancer, tumor-derived cytokines such as IL-6 and TNF-α can activate afferent vagal pathways and central circuits, including the nucleus of the solitary tract and hypothalamus, contributing to appetite suppression, fatigue, and depressive-like symptoms.²⁵²

Direct neurotumor interactions also shape tumor behavior. Glioma cells can form functional synaptic connections with neurons, and neuron–tumor communication can promote coexisting glioma states, including synaptic coupling, network integration, and neuron-like motility.^253,254 In prostate cancer, distant metastasis depends on sympathetic nervous regulation.²⁵⁵ Neural signals can also influence angiogenesis, immune evasion, and remodeling of the TME.²⁵⁶ Chronic stress further suppresses antitumor immunity: β-adrenergic signaling can promote T cell exhaustion and impair metabolic and antitumor function, reducing immunotherapy efficacy.²⁵⁷ In addition, pro- and anti-inflammatory cytokines can activate distinct vagal neuron subsets, enabling central feedback control of peripheral immune responses.²⁴⁸ These findings suggest that the brain–body axis is a dynamic regulator of tumor–immune and tumor–metabolic heterogeneity.

Neuroinvasion and cancer-induced neuropathy

Perineural invasion refers to tumor spread along the nerve fibers and is associated with pain, neurological dysfunction, motor impairment, and poor clinical outcomes.^258,259 Tumor cells and neural tissues can form reciprocal signaling loops through neurotrophic factors, neurotransmitters, and receptor upregulation, promoting tumor dissemination along nerve bundles.²⁶⁰ Cancer treatments can further damage the peripheral nervous system. Chemotherapeutic agents, such as paclitaxel, cisplatin, and vinca alkaloids, can induce axonal degeneration, myelin injury, and mitochondrial dysfunction, leading to numbness, neuropathic pain, and sensory loss.²⁶¹ Radiotherapy may also damage the adjacent neural tissue and cause radiation-induced neuritis or neuropathy.²⁶² Differences in pre-existing neural involvement, treatment exposure, local microenvironment, and host susceptibility generate heterogeneous trajectories of cancer-associated neuropathy.

Neural regulation of tumor progression

Autonomic regulation can influence tumor progression in a branch-specific manner. Sympathetic nervous activity promotes the release of norepinephrine and epinephrine, which act through β-adrenergic receptors on tumor and stromal cells to support proliferation, invasion, metastasis, angiogenesis, and immune modulation.^263,264 In cachexia-inducing tumors, type 2 immune responses in the adipose tissue can enhance peripheral sympathetic activity, increase catecholamine release, induce adipose tissue browning, and accelerate energy expenditure.²⁶⁵ In contrast, parasympathetic signaling, represented mainly by vagal acetylcholine release, can exert context-dependent tumor-suppressive effects through the cholinergic pathways.²⁶³ In breast cancer, tumor-derived neuroactive molecules stimulate local nerve growth and branching.²⁶⁶ Therefore, differences in autonomic activity, neural remodeling, and receptor expression contribute to heterogeneity in tumor growth, invasion, cachexia, and sensitivity to neuroendocrine-immune interventions.

Host neurological protection and tumor evasion

The blood–brain barrier (BBB) is the central protective interface of the nervous system. It maintains central nervous system homeostasis and limits the entry of harmful agents; however, in cancer, it can restrict drug delivery and immune-cell access.²⁶⁷ The degree of BBB disruption and remodeling varies by tumor type, disease stage, and lesion region, making the neuroprotection–tumor escape axis a source of organ- and stage-specific heterogeneity.^268,269

Tumors can disrupt or remodel the BBB through VEGF, HIF-1α, and TGFβ signaling. VEGFA can disassemble endothelial tight junctions and increase vascular permeability, whereas hypoxia-induced HIF-1α upregulates VEGF expression and promotes endothelial remodeling.^270–272 Targeted downregulation of HIF-1α in GBM can enhance the efficacy of sonodynamic therapy, and TGFβ signaling can disrupt endothelial homeostasis and promote immunosuppression.^273,274 In GBM, BBB dysfunction is spatially heterogeneous; some regions show increased permeability whereas others remain protected, limiting therapeutic delivery. This heterogeneity contributes to immune exclusion, restricted drug penetration, and treatment failure.^275,276 A preclinical study of an immune exosome–nanomicelle delivery system suggested that enhanced BBB penetration may improve antitumor activity in GBM models; however, its clinical efficacy and effect on recurrence in patients remain to be established.²⁷⁷

HITET synthesis: macro-scale regulatory interfaces

In summary, the organ and systemic dimensions represent macroscale regulatory interfaces within HITET. Organotropism, pre-metastatic niche formation, and lesion-specific molecular evolution explain why tumors with similar drivers behave differently across anatomical sites. Systemic interfaces, including the microbiota–immune–metabolic axis, endocrine and metabolic systems, inflammatory cachexia programs, and brain–body axis, further shape tumor behavior by regulating immune tone, resource availability, barrier properties, stress responses, and treatment tolerance.

Clinically, this perspective suggests that treatment decisions should account for lesion location, local tissue conditions, and host-level factors rather than relying only on tumor genotype or a single biomarker. Lesion-specific biopsy and spatial or multi-omics profiling may be required when different metastases show divergent behaviors. Multicompartment biomarkers, including circulating, stool-based, endocrine, metabolic, inflammatory, and neurophysiological markers, may help capture host-level heterogeneity. Therapeutically, microbiome modulation, metabolic and endocrine rebalancing, anti-cachexia strategies, neuromodulation, vascular or barrier normalization, and organ-specific drug-delivery approaches may complement local tumor-directed therapy. Thus, at this dimension, the goal is not only to eliminate tumor cells but also to modify the organ and systemic conditions that determine where tumor cells survive, how they evolve, and whether the therapy remains durable.

Therapeutic implications: targeting multidimensional heterogeneity

Within the HITET framework, cancer therapies can not only be understood as tumor-directed interventions but also as selective pressures that act across genetic, cellular, tissue, organ, and systemic dimensions. Surgery, radiotherapy, chemotherapy, targeted therapy, immunotherapy, and endocrine therapy are the essential components of cancer treatment. However, their effects extend beyond direct tumor killing; they can reshape clonal competition, cell-state composition, immune niches, vascular access, stromal organization, organ function, and host physiology. Therefore, therapeutic design should consider both antitumor efficacy and treatment-induced adaptive changes.

Here, we organize the therapeutic implications into three complementary strategies. Vertical integration combines interventions within one biological dimension to suppress the compensatory escape. Horizontal integration combines interventions across dimensions, such as pairing tumor cell-directed therapy with immune- or tissue-level modulation. Ecosystem modulation aims to reshape local and systemic host–tumor interfaces, including the vascular, stromal, microbiota, metabolic, endocrine, neural, behavioral, and temporal regulatory axes. These strategies are not mutually exclusive; in practice, persistent disease control may require different combinations, depending on the dominant sources of heterogeneity in each patient and lesion (Fig. 8).

Fig. 8.

Therapeutic integration for multidimensional cancer heterogeneity. This schematic illustrates therapeutic strategies derived from the Holistic Integrative Tumor Ecosystem Theory (HITET), with the central goal of targeting multidimensional cancer heterogeneity. The upper sector emphasizes that conventional cancer treatments—including surgery, radiotherapy, chemotherapy, targeted therapy, and immune-based therapy—act as ecological perturbations. Although these treatments reduce tumor burden, they may also reshape clonal composition, selective pressures, immune states, stromal organization, and subsequent evolutionary trajectories. The right sector illustrates vertical integration within a single biological dimension, in which multipoint blockade of related signaling or resistance nodes, exemplified by BRAF- and EGFR-associated pathways, is used to restrict compensatory escape and create evolutionary bottlenecks. The lower sector depicts horizontal integration across biological dimensions, combining tumor-directed treatments with modulation of the tumor microenvironment, gut-associated processes, or other host-level factors to generate cross-dimensional therapeutic synergy. The left sector represents ecosystem modulation at the host–tumor interface, incorporating niche engineering, temporal treatment ecology, epigenetic–microenvironmental interactions, the microbiota–immune axis, and the behavior–host interface. Together, these complementary strategies shift treatment from isolated tumor–cell targeting toward coordinated intervention across genetic, cellular, tissue, organ, and systemic dimensions, thereby supporting adaptive and context-specific therapeutic planning. This figure was created using BioRender (https://biorender.com/)

Rethinking conventional cancer therapies: the double-edged sword of ecological perturbation

Conventional therapies remain indispensable, but they should also be recognized as strong perturbations of the tumor–host system. Surgery can remove tumor masses and provide tissue for molecular and spatial profiling; however, residual cells at invasive margins or peritumoral niches may seed relapse. Radiotherapy and chemotherapy can reduce the tumor burden, but also create spatial and systemic selection pressures that favor resistant clones, alter stromal and immune compartments, and affect normal tissue functions.^278–284 Thus, their long-term impact depends not only on cytotoxic efficacy, but also on the response of surviving tumor and non-tumor compartments.

Chemotherapy illustrates this dual effect. In addition to killing proliferating tumor cells, chemotherapy can induce myelosuppression, impair immune competence, damage mucosal barriers, and disrupt the gut or oral microbiota.^285–290 These effects may influence infection risk, inflammation, quality of life, and tolerance to treatment. Similarly, radiotherapy promotes local tumor control and tissue remodeling. Collateral injury to the surrounding parenchyma, vasculature, and neural structures can lead to fibrosis, organ dysfunction, and neurocognitive and sensory deficits.^291–293 Spatial dose gradients and fractionation schedules may also allow hypoxic or intrinsically radioresistant subclones to survive in partially protected regions, contributing to a spatially structured recurrence.²⁹⁴

Immunotherapy adds another layer of complexity. Immune checkpoint inhibitors can restore antitumor T cell activity, but they may also produce tissue-specific immune-related adverse events, including endocrine, gastrointestinal, pulmonary, dermatologic, rheumatologic, and neurological toxicities.²⁹⁵ These toxicities reflect the systemic consequences of immune reactivation and must be considered when designing drug combinations. Overall, conventional therapies should not be dismissed; rather, they should be integrated with molecular, spatial, and host-level monitoring to reduce the tumor burden while limiting adaptive escape and host damage.

Vertical integration: constructing evolutionary bottlenecks within dimensions

Vertical integration refers to a combination of interventions within the same biological dimension that restricts compensatory adaptation. Instead of targeting a single dominant lesion or pathway, vertically integrated strategies aim to block driver pathways along with feedback circuits, resistance nodes, or state-dependent vulnerabilities.

The BRAF-mutant CRC model is a clear example. BRAF V600E is therapeutically actionable in melanoma, where BRAF inhibition can produce substantial responses. However, in CRC, BRAF inhibitor monotherapy has limited efficacy because EGFR feedback activation rapidly restores downstream signaling.²⁹⁶ This context-dependent resistance supports the co-targeting of BRAF and EGFR with or without MEK inhibition. Clinical data in BRAF-mutant metastatic CRC further support the combined pathway blockade, including in patients with deficient mismatch repair (dMMR)/microsatellite instability-high (MSI-H) disease progression after immune checkpoint inhibitors.²⁹⁷ This example illustrates how vertical integration can convert a weak single-agent vulnerability into a more constrained signaling bottleneck.

Resistance bottlenecks can also be identified by functional screening. Genome-wide CRISPR analyses have shown that chemotherapy resistance routes vary by lineage and drug mechanism; however, shared vulnerabilities may still exist within heterogeneous resistance landscapes. In CRC models resistant to oxaliplatin or irinotecan, polo-like kinase 4 (PLK4) has emerged as a druggable node whose inhibition restores oxaliplatin sensitivity across models.²⁹⁸ These findings suggest that vertical integration can be guided by resistance-state mapping rather than by the baseline genotype alone.

Cell-state heterogeneity is another vertical target. In PDAC, the response to KRAS inhibition depends not only on the KRAS mutation status, but also on the tumor cell state. Basal-like cells are sensitive to KRAS blockade, whereas the classical epithelial state may persist as a less-responsive reservoir that supports relapse. Depleting this resistant state, genetically or through chemotherapy-mediated approaches, deepens the response to KRAS inhibition and delays recurrence in mouse and patient-derived xenograft (PDX) models.²⁹⁹ Thus, vertical integration can operate within the cellular dimension by co-targeting oncogene-dependent signaling and resistant cell states that sustain adaptive escape.

Horizontal integration: building synergies across dimensions

Horizontal integration refers to therapeutic combinations that connect different HITET dimensions. The goal is not only to intensify tumor killing, but also to align tumor cell targeting with immune, stromal, vascular, organ-level, or systemic modulation. This approach is particularly relevant when resistance is driven by protective niches rather than tumor-intrinsic mutations alone.

Recent studies on tertiary lymphoid structures (TLSs) provide a concrete example of horizontal integration between systemic immune checkpoint blockade and tissue-level immune niche engineering. In advanced renal cell carcinoma, tumors with high TLS abundance and low tissue-resident exhausted CD8⁺ T cells show superior responses to first-line nivolumab, whereas the opposite pattern is associated with resistance, underscoring TLS architecture as both a determinant and a potential lever for PD-1 efficacy.³⁰⁰ Complementing this, a previous study on PDAC demonstrated that IL-33-activated, lymphoneogenic ILC2s can be pharmacologically harnessed to induce TLS neogenesis, convert immunologically “cold” tumors into TLS-rich immune niches, and enhance antitumor immunity, revealing a druggable IL-33–ILC2–TLS axis that reshapes local immune ecology.³⁰¹

Taken together, these findings support a cross-dimensional strategy that couples systemic PD-1/PD-L1 inhibition with agents that promote TLS formation or remodel the immune microenvironment, thereby aligning tissue-level microecology with whole-body immune interventions as a representative model of horizontal integration within the HITET cancer ecosystem.

Ecosystem modulation: reprogramming the host–tumor interface

Ecosystem modulation focuses on therapeutic strategies that alter the biological context in which tumors evolve. These approaches do not replace tumor-directed therapy but may expand their therapeutic window by reshaping vascular, stromal, immune, metabolic, microbial, endocrine, behavioral, or temporal conditions. This section highlights four actionable interfaces: behavior–host regulation, microbiota–immune modulation, epigenetic–microenvironment remodeling, and temporal scheduling.

At the behavior–host interface, preclinical evidence suggests that exercise may interact with tumor immunity and treatment response. In mouse models of established breast cancer, aerobic exercise training promoted tumor vascular normalization, alleviated hypoxia, increased CD8⁺ T cell infiltration and effector function, improved tumor control, and sensitized tumors to immune checkpoint blockade,³⁰² indicating that exercise is a potential modifier of immune–vascular–metabolic coupling in this experimental context. However, the study did not establish exercise as a clinically validated or generalizable adjuvant for immunotherapy across patients and cancer types, and prospective clinical evaluations are required.

Along the microbiota–immune axis, baseline gut microbiota composition is associated with response to PD-1 blockade in melanoma. Responders show enrichment of commensal species, such as Bifidobacterium longum, Collinsella aerofaciens, and Enterococcus faecium, and fecal transfer from responders improved antitumor immunity and anti-PD-L1 efficacy in germ-free mice.³⁰³ Clinical studies on fecal microbiota transplantation combined with immune checkpoint blockade in melanoma and CRC further support microbiome modulation as a potential adjunct to immunotherapy in selected patients.^304–306 These findings provide a practical example of systemic ecosystem modulation.

Epigenetic therapies can reshape the immune microenvironment. Combined HDAC and DNMT inhibition can increase IFNγ-positive CD8-positive T cells, NK cells, and NKT cells, while reducing MDSCs and PD1-high CD4-positive T cells, thereby reversing immunosuppressive features of the TME.³⁰⁷ In this setting, chromatin-level interventions function as a bridge between molecular reprogramming and tissue-level immune activation, supporting combinations with immune checkpoint blockade.

Temporal scheduling is a modifiable therapeutic variable. Dose-dense chemotherapy improves disease-free or progression-free outcomes in several cancers, including high-risk early breast cancer, recurrent ependymoma, and muscle-invasive bladder cancer.^308–310 From a HITET perspective, timing and dose density shape selection pressure, recovery windows, host tolerance, and resistance evolution. Therefore, treatment schedules should consider the dynamic components of multidimensional therapy, rather than neutral delivery formats.

Adaptive treatment designs can be made more quantitative by tracking dynamic target states and patient-specific signaling dependencies. In glioblastoma, resistance to EGFR tyrosine kinase inhibitors can arise through reversible loss of mutant EGFR from extrachromosomal DNA, with clonal EGFR alterations reappearing after drug withdrawal. This indicates that therapeutic targets may shift under treatment pressure and support longitudinal monitoring rather than one-time target assignment.³¹¹ In triple-negative breast cancer (TNBC), phosphoproteomic profiling, combined with patient-specific signaling signature analysis, can be used to identify unbalanced signaling processes in patient-derived tumors and guide individualized combination therapies. In this setting, anti-EGFR monotherapy is often expected to be insufficient, whereas patient-specific combinations show stronger experimental performance.³¹² These examples support a shift from biomarker-only treatment selection to an adaptive combination design based on measurable signaling states, resistance dynamics, and treatment-induced feedback.

Methodologically, several emerging platforms may help quantify tumor ecosystem states and treatment-induced changes. Label-free digital holographic microscopy combined with machine learning has been used in mouse- and patient-derived PDAC models to track single-cell phenotypic heterogeneity, subtype differentiation, and EMT-associated changes under treatment pressure.³¹³ Spatially resolved approaches that integrate spatial transcriptomics, single-nucleus RNA sequencing, and multiplexed protein imaging can further map tumor micro-niches, spatial subclones, metabolic programs, antigen presentation states, and immune hot–cold regions within the tissue architecture.³¹³ These tools are not substitutes for clinical judgment or conventional sampling; however, they can provide quantitative readouts for identifying dominant resistance states and selecting adaptive combination strategies.

Translationally, conventional clinical trials often rely on static and linear outcome metrics, lacking composite indicators that capture the dynamic feedback of the tumor–host-environment interface. Looking forward, cutting-edge technologies, such as spatial transcriptomics, single-cell multi-omics, and digital twin simulations, may enable the construction of predictive, integrative models of tumor–host–environment symbiosis.^314–317 Such measurements could support a practical detect–intervene–reassess cycle, in which therapy is adjusted according to evolving tumor states, host tolerance, and treatment-induced feedback.

From static protocols to dynamic ecological management

The implementation of HITET-guided therapy requires careful monitoring; however, this should not be interpreted as rejecting conventional treatment or requiring unlimited invasive sampling. Instead, when clinically indicated, surgical and biopsy specimens can provide spatial and molecular information that can inform the therapy. Liquid biopsies, circulating cytokines, microbiota profiles, imaging, endocrine and metabolic markers, and selected neurophysiological readouts may complement tissue-based assays by capturing longitudinal and systemic changes. The practical goal is to combine high-resolution tissue information with lower-burden longitudinal monitoring.

Therefore, an adaptive therapeutic design should follow a detect–intervene–reassess logic. First, the dominant sources of heterogeneity, such as driver pathway escape, resistant cell states, immune exclusion, vascular barriers, microbiota-associated immune suppression, metabolic dysfunction, or systemic inflammation, should be identified. Second, the treatment should combine direct tumor targeting with the most relevant ecosystem-modulating interventions. Third, response and toxicity should be reassessed over time to adjust the therapy before resistant states become dominant.

In summary, HITET does not replace existing therapeutic principles, but provides a framework for organizing combination strategies according to the dimensions that sustain resistance. Vertical integration restricts adaptation within a dimension; horizontal integration links tumor cell-directed therapy with immune, tissue, or organ-level modulation; and ecosystem modulation adjusts host–tumor interfaces that shape treatment durability. This framework may provide a practical basis for integrating tumor genotype, cellular state, lesion location, local tissue barriers, and host-level factors into treatment planning while balancing durable disease control against treatment-related toxicity.

Conclusions and perspectives

Cancer heterogeneity is not determined by genetic alterations alone, but emerges from interactions among heritable variation, cellular programs, tissue habitats, organ-specific environments, and systemic host physiology. In this review, we used the HITET framework to organize these interactions across five dimensions: genetic, cellular, tissue, organ, and systemic. HITET builds on, rather than replaces, the Hallmarks of Cancer, the TME, evolutionary, cancer-ecosystem, and systems-oncology frameworks. Its present contribution is primarily organizational and hypothesis-generating: it places established mechanisms within a shared multiscale structure and uses the niche–flow–feedback principle to examine how heterogeneous states are transmitted, filtered, stabilized, amplified, or constrained across scales. This perspective helps explain why the same molecular driver can produce different phenotypes across tumor types or metastatic sites, why treatment may control one lesion but fail in another, and why microbiota, metabolism, endocrine state, neural regulation, and host inflammation can modify local tumor behavior. However, these cross-scale relationships remain provisional and should not be interpreted as an empirically validated theory, predictive model, or clinical rule.

A major challenge is measuring these dynamic ecosystems with sufficient spatial and temporal resolution while maintaining an acceptable clinical burden. Static biopsies remain essential; however, they capture only limited regions and time points, whereas repeated deep tissue profiling is invasive and often impractical. A feasible strategy is adaptive sampling, which combines clinically indicated tissue acquisition with lower-burden longitudinal approaches, including circulating tumor DNA, extracellular vesicles, circulating immune and cytokine profiles, microbiota signatures, metabolic markers, and functional imaging. Positron emission tomography (PET) tracers, hyperpolarized magnetic resonance imaging (MRI), and other related methods may help monitor metabolic, hypoxic, immune, and vascular states without requiring repeated tissue acquisition. When tissue is available, spatial transcriptomics, single-cell or single-nucleus sequencing, multiplexed protein imaging, spatial metabolomics, and epigenomic profiling can define how clones, cell states, immune niches, stromal programs, metabolic gradients, and invasive fronts are organized within the tissue architecture.³¹⁸ Organoids, assembloids, organ-on-a-chip systems, ECM-mimetic scaffolds, and 3D bioprinting can be used to reconstruct selected ecosystem components under controlled conditions.³¹⁹ Although these platforms cannot reproduce the entire patient ecosystem, they can test whether a resistant state depends on a specific niche, whether microbial or stromal signals alter the immune response, and whether combined interventions limit adaptive escape more effectively than restricted targeting.

The next challenge is the conversion of multiscale measurements into useful predictions and treatment decisions. Cross-modal artificial intelligence can infer cellular, immune, or spatial states from routine histopathological or lower-dimensional data, allowing ecosystem-level information to be extended to cohorts in which full multi-omics is unavailable.³²⁰ Patient-specific digital twins may further integrate tumor genotypes, signaling states, spatial niche structures, immune statuses, organ sites, treatment histories, and host variables to simulate schedules or combinations.³¹⁵ These approaches should remain transparent and updatable decision-support tools, rather than deterministic substitutes for clinical judgment. Prospective translation also requires trial designs that capture multidimensional heterogeneity without becoming impractical. Adaptive platform trials could incorporate biomarkers, such as resistant cell states, immune phenotype, spatial niche classification, circulating tumor DNA, microbiota composition, metabolic or endocrine markers, inflammatory status, and organ-specific imaging. In such studies, HITET suggests a provisional detect–intervene–reassess logic: identify the dominant and measurable sources of resistance, select tumor-directed and ecosystem-directed interventions that are biologically actionable and tolerable, and reassess the response, adaptation, toxicity, and host state over time. Therapeutic strategies can then be organized as vertical integration within a dimension, horizontal integration across dimensions, or modulation of host–tumor interfaces. These categories do not replace standard therapy or constitute a validated ranking algorithm but provide a structure for testing how existing and emerging interventions may be combined.

However, this study has important limitations. HITET is neither a mathematical model capable of predicting tumor evolution de novo nor a prescriptive clinical algorithm. The five dimensions are heuristic analytical categories rather than mutually exclusive compartments and their boundaries may shift with the biological question, measurement strategy, and clinical context. Some mechanisms span several dimensions, and the relative contribution of each dimension to the resistance or response may be difficult to estimate. Similarly, niche, flow, and feedback are measurable working concepts rather than fully standardized variables. Their operationalization requires cancer-specific metrics, thresholds, longitudinal sampling, perturbation experiments, causal inference, and dynamic modeling. Full five-dimensional profiling is not feasible or necessary for all patients. Therefore, future studies should determine when dimension-rich profiling adds value, when dimension-sparse monitoring is sufficient, and whether HITET-informed models outperform established frameworks or simpler molecular and clinical models according to prespecified measures of prediction, treatment selection utility, feasibility, toxicity, patient burden, and outcome. Several hypotheses can be proposed to guide this evaluation. Tumors with strong positive feedback between hypoxia, abnormal vasculature, and immune exclusion may be particularly sensitive to combined vascular normalization and immune activation. Lesions with divergent organ-specific niches may require site-specific sampling or treatment adaptation despite sharing trunk drivers. Tumors with reversible target states or ecDNA-mediated oncogene plasticity may benefit from longitudinal monitoring and adaptive scheduling, rather than fixed target assignment. More broadly, interventions that constrain three or more active dimensions may produce more durable responses than narrower strategies, and the rate of clonal or phenotypic diversification may correlate with the number and strength of positive feedback loops. These propositions remain unvalidated and should be prospectively evaluated against simpler alternative explanations or models. Systemic inflammatory, metabolic, microbiota, endocrine, or neural states that reinforce immune suppression may identify contexts in which carefully selected host-directed interventions add value to tumor-directed therapy.

In conclusion, HITET offers a structured method to connect established mechanisms of cancer heterogeneity with measurable biological states, cross-scale hypotheses, and future therapeutic studies. Its long-term value will depend not on conceptual breadth alone but on whether it can define informative interfaces, guide parsimonious measurements, generate falsifiable predictions, and improve decisions beyond existing approaches. Progress will require longitudinal and spatial datasets, experimentally tractable ecosystem models, transparent computational methods, biomarker-defined prospective trials, and thorough consideration of feasibility and toxicity. If progressively refined and prospectively validated, HITET may help support a more context-sensitive and adaptive form of precision oncology in which treatment is adjusted as the tumor and host states evolve.

Author contributions

W.-L.J. and Y.-L.H. were involved in the conception and design of the review. Y.-L.H., Y.-M.R., G.-Q.Y., and L.N. wrote the paper and prepared the figures. W.-L.J. and Y.-W.P. critically reviewed and edited the manuscript. All authors have read and approved the article.

Competing interests

The authors declare no competing interests.