Skip to main content
Breast Cancer : Targets and Therapy logoLink to Breast Cancer : Targets and Therapy
. 2026 Nov 2;18:657423. doi: 10.2147/BCTT.S657423

Two Niches, One Disease: Brain and Leptomeningeal Metastases in Breast Cancer: A Narrative Review

Hongxu Ren 1,*, Wencheng Chi 2,*, Xiao Han 2, Linfeng Guo 1, Junlin Li 1,✉, Jiakang Jiang 2,✉
PMCID: PMC13742005  PMID: 42997441

Abstract

Central nervous system (CNS) metastasis is an increasingly important challenge in advanced breast cancer, yet brain parenchymal metastases (BM) and leptomeningeal metastases (LM) are often considered within a single framework despite occupying biologically distinct CNS compartments. In this narrative review, we synthesize breast cancer-specific mechanistic and clinical evidence to examine how these two forms of CNS disease differ in metastatic entry, local adaptation, molecular evolution, biomarker representation, and therapeutic vulnerability. Parenchymal BM primarily arise through haematogenous dissemination and subsequent adaptation to a neurovascular, neural, and immune-rich environment, where successful colonization requires more than blood–brain barrier traversal and is supported by reciprocal tumour–host interactions and metabolic plasticity. By contrast, LM develops within a fluid-dominated, CSF-facing compartment characterized by distinct meningeal and blood–CSF interfaces, constrained resource availability, and dynamic immune conditions. C3–C3aR-mediated barrier remodeling and LCN2–SLC22A17-dependent iron acquisition are supported mainly by experimental LM models and selected patient-derived observations, including mixed-cancer cohorts, and require broader breast cancer-specific clinical validation. Genomic and phenotypic studies further indicate that BM and LM can share metastatic ancestry while undergoing branched and compartment-associated evolution, although observed divergence may also reflect pre-existing heterogeneity, treatment selection, elapsed time, and sampling effects. Therapeutic efficacy therefore depends not only on CNS drug access but also on target expression, tumour state, niche support, disease geometry, and prior treatment. By integrating metastatic dissemination, niche adaptation, molecular evolution, biomarker representation, and therapeutic vulnerability within a single comparative framework, this review links compartment-associated biology to molecular sampling, experimental modeling, and clinical trial design. Clinically, this framework may inform compartment-specific molecular sampling and trial stratification by BM treatment status and LM presentation, although its utility for individual treatment selection requires prospective validation. Future studies combining paired multi-compartment sampling, lineage-resolved models, and serial CSF and plasma profiling will be needed to distinguish shared ancestry from niche-associated adaptation and treatment-driven evolution.

Keywords: breast cancer, brain metastasis, leptomeningeal metastasis, CNS compartmentalization, tumour microenvironment, clonal evolution, cerebrospinal fluid

Introduction

Central nervous system (CNS) involvement has become an increasingly important challenge in metastatic breast cancer. The risk is strongly subtype-dependent: a systematic review and meta-analysis reported pooled cumulative incidences of brain metastases (BM) of 31% in HER2-positive metastatic breast cancer, 32% in triple-negative breast cancer (TNBC), and 15% in HR-positive/HER2-negative disease.1 More recent real-world data from 18,075 patients further showed that BM prevalence increased across successive lines of therapy in all major breast cancer subtypes.2 BM remains a major source of neurological morbidity and requires increasingly complex integration of local and systemic treatments.3,4 Leptomeningeal metastasis (LM) is less common but represents a particularly challenging manifestation of advanced breast cancer, with substantially more limited high-quality prospective evidence.5,6

The distinction between BM and LM extends beyond anatomical location. BM develop within a specialized neurovascular and neural microenvironment comprising endothelial cells, pericytes, astrocytes, microglia, neurons and recruited immune populations.7,8 Tumour progression disrupts the normal blood–brain barrier (BBB) and produces a heterogeneous blood–tumour barrier (BTB), whose permeability varies substantially between metastatic lesions.7,9,10 By contrast, LM involves the leptomeninges and subarachnoid space and therefore exposes tumour cells to cerebrospinal fluid (CSF), meningeal surfaces, the choroid plexus and blood–CSF interfaces.11,12 The leptomeningeal compartment is further distinguished by CSF circulation, relative nutrient scarcity and distinct immune constraints.12,13 These compartmental distinctions also have practical clinical implications. Differences in tissue accessibility, imaging and CSF-based assessment, therapeutic drug exposure, and response evaluation mean that grouping BM and LM under a single category of CNS metastasis may obscure clinically relevant heterogeneity and complicate the interpretation of therapeutic evidence.

Importantly, neither compartment is biologically homogeneous. Breast cancer BM exhibits substantial lesion-to-lesion heterogeneity, including marked variation in vascular permeability,9,10 whereas LM encompasses linear, nodular and mixed patterns with variable cranial and spinal involvement.11 We therefore use the term “two niches” as a conceptual shorthand for two major CNS metastatic compartments, not as an exhaustive classification of CNS metastatic presentations or a claim that either compartment represents a uniform biological environment. This overarching compartmental framework is summarized in Figure 1. Although previous reviews have provided important insights into leptomeningeal dissemination, microenvironmental interactions, and emerging therapeutic approaches,12,14 a central challenge remains in interpreting brain parenchymal metastases and leptomeningeal metastases within a unified biological and clinical framework. Compartment-associated differences in tumour phenotypes, molecular alterations, and treatment responses cannot be fully understood without considering metastatic ancestry, pre-existing heterogeneity, treatment history, and the distinct sampling constraints of brain tissue and cerebrospinal fluid. The present review therefore extends beyond a description of the two metastatic environments by integrating four interconnected dimensions: routes of dissemination, compartment-specific adaptation, molecular evolution and biomarker representation, and therapeutic vulnerability. Particular emphasis is placed on distinguishing breast cancer-specific observations from evidence derived from other tumour types or experimental models, and on separating observed molecular divergence from demonstrated niche-driven selection. By connecting these biological distinctions with longitudinal sampling strategies, experimental model selection, and compartment-specific clinical trial design, this comparative framework aims to identify testable questions and improve the interpretation of CNS metastatic disease rather than propose a single universal mechanism.

Figure 1.

A conceptual diagram of breast cancer spread to brain parenchymal and leptomeningeal compartments. Primary breast cancer at the top center is labeled, Primary breast cancer, with a callout labeled, Primary-tumour heterogeneity, pointing to multiple tumour foci in a breast cross section. Below, a banner reads, Disseminating metastatic clones, above three separate tumour cell clusters. Left section titled, Brain parenchymal metastasis (B M), shows a brain outline at the top with a highlighted lesion. A schematic labeled, blood-tumor barrier, shows a vessel with labels, endothelial cells and, pericytes. Surrounding labeled cell types include, astrocytes, neurons, microglia and, infiltrating macrophages. A central mass is labeled, metastatic colony. Three boxed phrases below read, Neurovascular adaptation, Neural-immune crosstalk and, Metabolic plasticity. A footer line reads, cell-dense, neurovascular, neural. Right section titled, Leptomeningeal metastasis (L M), includes a brain cross section at the top right and a circular inset labeled, blood-CSF barrier and, choroid plexus, connected by dashed lines to the brain cross section. Below, a meningeal cross section is labeled, arachnoid and, pia, with a dashed rectangle labeled, subarachnoid space. Inside the rectangle are labels, floating tumour cells, leptomeningeal tumour nodule, flowing CSF and, leptomeningeal tumour cells. Three boxed phrases below read, Fluid-dominated ecology, Interface remodeling and, Resource competition and immune adaptation. A footer line reads, CSF-facing, fluid-dominated, resource-constrained. Between the two sections, text reads, possible secondary dissemination in selected cases, above a dashed arrow pointing from the left section toward the right section, followed by the line, left parenthesis Distinct but interconnected CNS compartments right parenthesis. Along the bottom, a sentence reads, One metastatic disease, two major CNS compartments, multiple adaptive states.

Two Niches, One Disease: a compartmental framework for breast cancer CNS metastasis. Primary-tumour heterogeneity gives rise to disseminating metastatic clones that can seed two major central nervous system (CNS) compartments. Brain parenchymal metastasis (BM) develops within a cell-dense neurovascular and neural environment, where neurovascular adaptation, neural–immune crosstalk and metabolic plasticity shape metastatic growth. Leptomeningeal metastasis (LM), by contrast, occupies a cerebrospinal fluid (CSF)-facing, fluid-dominated and resource-constrained compartment involving the meninges, choroid plexus and blood–CSF interface, with floating, surface-associated and nodular tumour states. BM and LM are therefore depicted as distinct but interconnected manifestations of the same metastatic disease. The dotted BM-to-LM arrow denotes possible secondary dissemination in selected cases and does not imply an obligatory linear progression.

This narrative review synthesized evidence on breast cancer BM and LM identified through PubMed/MEDLINE, Web of Science Core Collection, and Scopus, with searches conducted on 8 September 2026, covering the period from the inception of each database to that date. No language restrictions were applied to the database searches. BM and LM were searched separately, using terms covering CNS dissemination, barrier biology, tumour–host interactions, molecular evolution, biomarkers, treatment and preclinical models. Representative search concepts included “breast cancer” or “breast neoplasm” combined with “brain metastasis”, “brain parenchymal metastasis”, “leptomeningeal metastasis”, “leptomeningeal disease”, or “meningeal carcinomatosis”, together with topic-specific terms addressing metastatic dissemination, the blood–brain and blood–CSF barriers, tumour microenvironment, metabolic adaptation, clonal evolution, cerebrospinal fluid biomarkers, and therapeutic response. The database-specific search strategies and supplementary topic-specific search terms are detailed in Supplementary Table S1. Priority was given to original mechanistic and clinical studies, particularly breast cancer-specific, paired and longitudinal patient data; mixed solid-tumour or other cancer evidence was used only when breast cancer-specific evidence was limited and was identified as such in the text. Recent high-quality reviews, guidelines and consensus statements were used mainly for established concepts and clinical context. Studies were evaluated for relevance to CNS metastatic biology, the specificity of their experimental or clinical setting, and their contribution to understanding differences or potential relationships between BM and LM. Evidence was organized by disease compartment, study population or experimental model, and the strength and limitations of breast cancer-specific validation. Conflicting, null and context-dependent findings were actively sought to reduce confirmation bias, and evidence from human samples, animal models and in vitro systems was distinguished when interpreting mechanistic claims. Because the aim was a critical, evidence-weighted synthesis rather than exhaustive study identification or quantitative pooling, this review was not conducted as a systematic review and did not follow a PRISMA-based selection process. To facilitate comparison across heterogeneous evidence sources, Table 1 summarizes the principal biological and clinical differences between BM and LM, the supporting study populations or experimental models, the extent of breast cancer-specific validation, and the main limitations of the available evidence.

Table 1.

Comparative Evidence on the Biology, Molecular Evolution, Biomarkers, and Therapeutic Vulnerabilities of Breast Cancer Brain and Leptomeningeal Metastases

Comparative Domain Brain Metastases (BM) Leptomeningeal Metastases (LM) Evidence Basis and Breast Cancer Specificity Principal Limitation
Metastatic entry Predominantly haematogenous entry through cerebral microvessels and the BBB Multiple potential routes, including choroid plexus-associated entry and secondary dissemination Breast cancer experimental models support parenchymal extravasation and ICAM2–ICAM1-mediated LM entry in TNBC Relative contributions of LM entry routes remain uncertain
Local microenvironment Cell-dense neurovascular and neural environment CSF-facing, fluid-dominated environment involving meningeal and blood–CSF interfaces Human anatomical observations and experimental studies; substantial breast cancer-specific evidence for BM Considerable heterogeneity within both compartments
Tumour–host interactions Astrocyte, microglial, macrophage, and neuronal signalling influences colonization Barrier remodeling, inflammatory signalling, and immune–metabolic competition may support persistence Breast cancer-specific preclinical evidence for several BM pathways; selected LM mechanisms supported by broader experimental models Mechanistic findings may be model- and disease-context dependent
Metabolic adaptation Flexible glucose, mitochondrial, amino acid, and lipid metabolism Resource limitation and iron acquisition through LCN2–SLC22A17 in experimental LM models Multiple breast cancer-specific BM studies; LM iron-acquisition evidence includes broader cancer-model findings No universal metabolic program has been established
Molecular evolution Genomic and receptor discordance between primary tumours and BM CSF-derived tumour cells and ctDNA may reveal distinct molecular populations Human breast cancer genomic studies and selected longitudinal CSF cohorts Shared ancestry does not establish direction of spread or niche-driven selection
Biomarker assessment Resected tissue supports spatial, genomic, and single-cell analyses CSF cytology, CTCs, and ctDNA support compartment-focused molecular assessment Breast cancer-specific tissue and CSF investigations Tissue sampling is spatially restricted; CSF does not reliably represent every parenchymal lesion
Therapeutic vulnerability Local therapy combined with subtype-directed systemic treatment; heterogeneous BTB permeability Treatment depends on disease distribution, CSF exposure, molecular subtype, and prior therapy Relatively mature clinical evidence for HER2-positive BM; more limited prospective LM evidence BM treatment efficacy cannot automatically be extrapolated to LM

Breast Cancer Reaches the CNS Through Partly Shared and Distinct Routes

CNS-Tropic Clones are Selected Before Colonization

Breast cancer dissemination to the CNS is unlikely to be a purely stochastic consequence of systemic spread. Primary breast cancers are genetically and phenotypically heterogeneous, and only a fraction of tumour cells appear capable of completing the steps required for CNS metastasis. Genomic comparison of a basal-like primary breast cancer with its matched brain metastasis showed enrichment of mutations that were present only at low frequency in the primary tumour, consistent with metastatic outgrowth from a minor subpopulation rather than proportional dissemination of the primary tumour as a whole.15 Experimental selection studies provide complementary evidence: serial isolation of brain-seeking derivatives from heterogeneous breast cancer cell populations generated sublines with markedly enhanced brain-metastatic activity.16 Together, these findings support the existence of selective bottlenecks that favour tumour cells with sufficient metastatic fitness to survive dissemination and initiate CNS colonization. Importantly, however, such fitness is unlikely to reflect a single universal “brain-tropism” program; different models identify partly distinct molecular determinants, suggesting that CNS competence can arise through multiple phenotypic routes.

The marked subtype dependence of brain metastasis is consistent with this biologically non-random pattern. In metastatic breast cancer, the cumulative incidence of brain metastasis is substantially higher in HER2-positive and TNBC than in HR-positive/HER2-negative disease.1 This association should not be interpreted as evidence that HER2 expression or the triple-negative phenotype alone determines CNS tropism, because subtype is also linked to differences in genomic background, systemic therapy, extracranial disease control and survival. Nevertheless, it indicates that tumour-intrinsic biology strongly influences the probability of CNS dissemination. Heterogeneity is also evident among circulating tumour cells (CTCs). A patient-derived CTC population expressing HER2, EGFR, HPSE and NOTCH1 showed increased invasive and brain-metastatic capacity in experimental models.17 Given the small patient series and the ability of these cells to form extracranial metastases, these findings are better interpreted as evidence that CTCs differ in metastatic competence than as a validated brain-specific CTC signature.

Arrival in the cerebral circulation introduces a further layer of selection. Breast cancer cells that reach the brain commonly arrest within capillaries and post-capillary venules, but vascular lodging does not guarantee metastatic outgrowth. In experimental models, many arrested cells remain intravascular for days, disappear, or fail to progress, whereas only a subset successfully extravasates and persists within the brain parenchyma.18 Thus, vascular arrest represents more than passive mechanical trapping: survival during endothelial contact, adhesion, transendothelial migration and early perivascular persistence create additional bottlenecks that discriminate between cells that merely reach the brain and those capable of establishing a metastatic foothold. This distinction is important because it separates delivery to the CNS from true metastatic competence.

A central unresolved issue is whether CNS-competent clones are primarily pre-existing in the primary tumour or are progressively generated and selected during disease evolution. Evidence supports both scenarios. The enrichment of low-frequency variants from the primary tumour in matched BM indicates that pre-existing subclones can contribute to CNS seeding.15 Conversely, genomic analyses of matched primary tumours and BM show branched evolution after metastatic divergence, with clinically relevant alterations emerging in brain lesions that were not detected in the corresponding primary tumour.19 Receptor discordance between primary breast tumours and BM provides further clinical evidence of this divergence.20 However, these differences cannot be attributed solely to the CNS niche: systemic therapy, temporal evolution, pre-existing heterogeneity and sampling limitations may all contribute. The relative importance of each selective pressure therefore remains unresolved.

Overall, the available evidence suggests that CNS dissemination is shaped by successive selective bottlenecks rather than representing a purely random process. Primary-tumour heterogeneity supplies potential metastatic founders, whereas survival in the circulation, vascular arrest and subsequent colonization progressively restrict which clones persist. Some CNS-relevant traits may therefore pre-exist before dissemination, while others may emerge through continued evolution and treatment selection. Notably, most lineage and functional evidence supporting this model comes from parenchymal brain metastasis; comparable breast cancer-specific evidence for leptomeningeal dissemination remains limited. This distinction becomes important when considering whether metastatic cells encounter the brain parenchymal and leptomeningeal compartments through the same, or only partly overlapping, routes.

Multiple CNS Interfaces Shape Metastatic Entry

Metastatic access to the CNS is governed by several anatomically distinct interfaces rather than by the BBB alone. For haematogenous seeding of the brain parenchyma, CTCs typically encounter the cerebral microvascular endothelium. Unlike peripheral vessels, brain endothelial cells are joined by specialized junctional complexes, exhibit low rates of vesicular transcytosis and function together with pericytes, basement membranes and astrocytic endfeet as part of the neurovascular unit.21 These properties tightly regulate the movement of cells and solutes from blood into the neural parenchyma. Breast cancer cells destined for parenchymal metastasis must therefore progress from vascular arrest to endothelial interaction and transendothelial migration before establishing themselves beyond the vascular compartment. Although numerous tumour-derived factors have been implicated in this process, their contribution varies among experimental models, arguing against a single conserved molecular program for BBB traversal.22

The BBB should, however, be distinguished from the BTB. The two terms describe related but temporally different vascular states. The BBB is most relevant during the initial interaction between CTCs and the cerebral vasculature, whereas the BTB emerges as established metastatic lesions remodel pre-existing brain vessels and develop abnormal tumour-associated vasculature. Importantly, this remodelling does not simply abolish barrier function. Contemporary analyses emphasize that the BTB retains variable degrees of endothelial restriction and displays substantial spatial heterogeneity within and between BM.22,23 The BTB should therefore not be conceptualized as an additional anatomical gateway that tumour cells must cross after the BBB; rather, it is an evolving property of established parenchymal metastases. Its major implications for drug exposure and therapeutic response are considered later in this review.

Leptomeningeal dissemination presents a different anatomical problem. The CSF-filled leptomeningeal space is separated from the systemic circulation by barrier systems that are distinct from the BBB. One major interface is the choroid plexus, where fenestrated capillaries permit relatively free exchange between blood and the choroidal stroma, whereas tight junctions between choroid plexus epithelial cells regulate passage from this stroma into ventricular CSF.24,25 A second barrier is formed by the arachnoid barrier cell layer, which restricts exchange between meningeal vasculature and the subarachnoid CSF. These anatomical arrangements are important because they provide potential routes into the CSF without requiring tumour cells first to cross the parenchymal BBB.

The possible routes into the leptomeningeal compartment are correspondingly diverse. Anatomical analyses support access through the choroid plexus and meningeal vascular interfaces, but also describe routes involving perivascular spaces, direct extension from established intracranial disease and other meningeal corridors.24 The relative contribution of these routes in breast cancer remains poorly defined, and their importance may differ between patients and over the course of disease. Accordingly, the presence of breast cancer cells in the leptomeninges should not be interpreted as evidence that they have necessarily traversed the BBB or followed the same sequence of events as cells forming parenchymal brain metastases.

Thus, CNS dissemination is better understood as compartment-dependent engagement with multiple vascular and meningeal interfaces, rather than as a single BBB-crossing event. The BBB remains a major gateway for haematogenous entry into the brain parenchyma, but it is only one component of a broader CNS barrier system that also includes the BTB of established metastases and the blood–CSF and meningeal interfaces relevant to leptomeningeal disease. This anatomical distinction provides an essential basis for considering why parenchymal and leptomeningeal metastases may arise through partly shared but non-identical dissemination routes.

Parenchymal and Leptomeningeal Dissemination Follow Partly Distinct Routes

Breast cancer can reach the brain parenchyma and leptomeningeal compartment through overlapping but non-identical routes. Parenchymal BM arise predominantly through haematogenous dissemination, in which CTCs arrest in the cerebral microvasculature and subsequently extravasate into the neural parenchyma. LM can also originate from the systemic circulation, but access to the CSF does not necessarily require passage through the parenchymal blood–brain barrier. The choroid plexus provides one alternative portal, with its fenestrated capillaries allowing circulating cells to access the stromal compartment before encountering the epithelial blood–CSF barrier. Importantly, this route has experimental support in breast cancer rather than being solely an anatomical inference. In a TNBC model, ICAM2 expression promoted adhesion to ICAM1 on choroid plexus epithelial cells and enhanced trans-blood–CSF barrier migration and leptomeningeal colonization.26 Nevertheless, this mechanism has been demonstrated in a specific experimental context and should not be generalized as the dominant route of LM across breast cancer subtypes.14,26,27

Other potential routes broaden the distinction between parenchymal and leptomeningeal dissemination. Perivascular spaces extend between the subarachnoid compartment and penetrating cerebral vessels and can contain metastatic deposits, raising the possibility that tumour cells associated with established parenchymal disease may access CSF-filled perivascular corridors. Perineural migration along cranial or spinal nerves provides another anatomically plausible route to meningeal surfaces. However, the evidence supporting these pathways in solid tumours is substantially stronger at the anatomical and pathological level than at the level of direct breast cancer-specific lineage tracing. In particular, perineural access is considered more relevant to tumours arising near cranial nerves or spinal roots than to breast cancer in general. Accordingly, perivascular and perineural spread should be regarded as possible routes of breast cancer LM rather than established dominant mechanisms.12 Recent reviews likewise emphasize haematogenous choroid plexus entry, retrograde venous dissemination and direct seeding from intracranial lesions as better-supported routes of solid-tumour LM.27

LM can also arise secondarily from established CNS metastases. Parenchymal lesions abutting the pial surface or ventricular system can potentially breach local tissue boundaries and gain access to the CSF, while dural, skull or vertebral metastases may extend into adjacent meningeal structures. Clinical observations after brain-metastasis surgery provide indirect but informative support for this route. In a recent breast cancer-specific cohort, postoperative leptomeningeal seeding was associated with TNBC and piecemeal resection, suggesting that tumour biology and mechanical disruption of an established intracranial lesion can both increase the likelihood of meningeal dissemination.28 A large multicentre analysis of 2,673 patients undergoing brain-metastasis resection similarly found that breast cancer was associated with an increased risk of postoperative LM, with incomplete resection emerging as an independent risk factor.29 These data demonstrate that established BM can contribute to subsequent LM under certain circumstances, but postoperative dissemination represents a treatment-altered setting and cannot be assumed to reproduce the natural history of spontaneous LM.

Genomic evidence further cautions against interpreting the temporal coexistence of brain and leptomeningeal disease as proof of a uniform sequential pathway. In a breast cancer-specific study using CSF cell-free DNA, disseminated tumour cells and matched tumour material, Fitzpatrick and colleagues reconstructed the evolutionary histories of breast cancer LM.30 Among assessable cases, metastatic seeding was predominantly monophyletic, indicating that LM and other metastatic sites often descended from the same primary-tumour subclone, although polyphyletic dissemination was also observed. Particularly informative was a patient with a brain metastasis resected three years before LM development: the brain and leptomeningeal lesions shared two ancestral clones but subsequently acquired distinct subclones. This observation establishes clonal relatedness between the two CNS compartments in at least some patients, but it does not demonstrate that the leptomeningeal lesion necessarily arose directly from the preceding brain metastasis. Phylogenetic ancestry and anatomical direction of spread are related but distinct questions and require longitudinal sampling to resolve.

The available evidence therefore favours a pluralistic model of CNS dissemination. Parenchymal and leptomeningeal metastases can both arise through haematogenous spread, but LM has additional potential routes involving the choroid plexus, meningeal and venous interfaces, perivascular or perineural corridors, and contiguous or secondary spread from established CNS lesions. The contribution of each route is likely to vary according to tumour biology, anatomical context and treatment history, and several remain incompletely validated specifically in breast cancer. Brain and leptomeningeal metastases can consequently share metastatic ancestry without necessarily representing successive stages of one anatomical pathway. Current evidence does not support a single obligatory model in which parenchymal brain metastasis uniformly precedes and gives rise to leptomeningeal disease. These partly shared but non-identical routes of CNS dissemination are summarized in Figure 2.

Figure 2.

Flow diagram of circulating tumour cell routes to parenchymal brain and leptomeningeal metastasis. Circulation at left shows a vertical blood vessel labeled, Circulating tumour cells (CTCs), with multiple tumour cells inside and arrows indicating movement within systemic circulation. Cns interfaces at top center contains two main pathways. Parenchymal brain metastasis (BM), blood brain barrier slash neurovascular unit: a left to right sequence starting from a labeled CTC, then cerebral vascular arrest, then brain endothelial interaction, then transendothelial migration slash extravasation, then brain parenchymal seeding. A rightward arrow continues into the metastatic compartment. Metastatic compartment at top right, brain parenchyma: a rightward arrow from brain parenchymal seeding leads to Established lesion, then a rightward arrow leads to heterogeneous BTB. Leptomeningeal metastasis (LM) occupies the lower central area and connects to a boxed metastatic compartment at right labeled, Leptomeningeal compartment, CSF and meninges, containing a brain cross section with surrounding meninges. Four labeled route bars on the left feed into LM with dotted arrows: Systemic circulation, Perivascular corridors, Perineural spread and Contiguous extension. From Systemic circulation, dotted arrows lead to choroid plexus-associated entry and to Blood CSF barrier (choroid plexus), with dotted arrows continuing toward the leptomeningeal compartment box. Additional dotted arrows converge on a structure labeled meningeal vascular interfaces, which also connects by dotted arrows to the leptomeningeal compartment box. A dotted vertical arrow drops from the Established lesion in the brain parenchyma region down toward the leptomeningeal compartment, indicating a connection from established parenchymal disease to the leptomeningeal compartment.

Partly shared and distinct routes of breast cancer CNS dissemination. Circulating tumour cells (CTCs) provide a common haematogenous source for central nervous system (CNS) metastasis. Parenchymal brain metastasis (BM) follows a relatively established sequence of cerebral vascular arrest, endothelial interaction, transendothelial migration/extravasation and brain parenchymal seeding across the blood–brain barrier (BBB)/neurovascular unit; the heterogeneous blood–tumour barrier (BTB) is shown as a feature of established BM rather than a second entry barrier. Leptomeningeal metastasis (LM), in contrast, may arise through multiple routes involving the choroid plexus/blood–CSF barrier, meningeal vascular interfaces, perivascular or perineural spread, contiguous extension and, in selected cases, secondary dissemination from established BM. Thus, shared haematogenous origin does not imply identical CNS entry routes, and the relative contribution of individual LM pathways remains incompletely defined. The evidence supporting these routes differs substantially. The principal steps of haematogenous parenchymal seeding are supported by breast cancer-specific experimental studies, while choroid plexus-associated entry has been demonstrated in selected breast cancer models. Secondary dissemination from established CNS lesions is supported by clinical observations, whereas perivascular and perineural routes remain anatomically plausible but lack comparable direct breast cancer-specific validation. The non-solid arrows illustrate alternative or potentially context-dependent routes rather than a formal grading of evidence strength or their relative frequency.

Brain Metastases Adapt to the Parenchymal Niche

Neurovascular Adaptation Enables Parenchymal Colonization

Crossing the blood–brain barrier is necessary for haematogenous brain metastasis, but it does not by itself establish a metastatic colony. After arrest in cerebral microvessels, breast cancer cells must interact productively with brain endothelial cells, complete transendothelial migration and survive the abrupt transition from the vascular lumen to the neural parenchyma. Recent breast cancer-specific work indicates that the endothelium can actively participate in this process rather than functioning only as a passive barrier. In brain-tropic TNBC models, brain endothelial cells induced EGFR signalling and a DOCK4–RAC1-dependent morphological program that facilitated tumour-cell intercalation and extravasation. Notably, DOCK4 depletion impaired entry into the brain without affecting tumour-cell survival or proliferative capacity, experimentally separating the machinery required for extravasation from that required for subsequent growth.31

The selective pressure persists after extravasation. Newly arrived tumour cells frequently remain closely associated with pre-existing cerebral vessels, using the vascular basement membrane as a substrate for adhesion, survival and early expansion. Experimental and human pathological studies have shown that early micrometastatic colonies are strongly enriched around existing vessels, with β1-integrin-dependent adhesion contributing to vascular co-option.32 In breast and lung cancer models, Valiente and colleagues further showed that many cancer cells entering the brain fail to survive, whereas successful metastatic cells remain attached to capillaries and exploit L1CAM-dependent vascular co-option. Brain-metastatic cells expressing plasminogen-activator inhibitory serpins were protected from stromal plasmin activity, thereby preserving L1CAM and reducing exposure to astrocyte-derived FasL-mediated death signals.33 These findings place perivascular survival, rather than extravasation alone, among the critical early bottlenecks of brain colonization.

Astrocytes add another layer of selection to this early neurovascular niche. Their role is not uniformly supportive: reactive astrocytes can contribute to antimetastatic defence, yet tumour cells that survive this initial pressure can subsequently co-opt astrocytic programs that favour metastatic persistence. A STAT3-activated subset of reactive astrocytes has been shown experimentally to support brain metastasis through local immunomodulatory effects, whereas genetic or pharmacological disruption of astrocytic STAT3 reduced metastatic burden.34 Importantly, a recent breast cancer-specific analysis of 100 resected BM provided human evidence for the relevance of this state: high pSTAT3 expression in surrounding reactive astrocytes was observed in 57% of lesions, was particularly frequent in TNBC, and was associated with increased intracranial recurrence and shorter intracranial progression-free survival.35 These clinical observations do not establish that STAT3-positive astrocytes initiate colonization, but they support the broader concept that successful BM maintain biologically consequential interactions with the astrocytic compartment.

Early colonization is also not uniform across breast cancer subtypes. Using breast cancer models together with human tissue, Gan and colleagues recently identified distinct modes of incipient brain growth in TNBC and HER2-positive disease. TNBC-derived colonies preferentially adopted a perivascular, infiltrative architecture with extensive contact with astrocytes and microglia, whereas HER2-positive colonies more often formed compact spheroids supported by tumour-derived tenascin C and spatially segregated surrounding stromal cells.36 This work is important because it moves beyond the assumption that all breast cancer cells use the same post-extravasation strategy. Instead, successful colonization can be achieved through different combinations of vascular dependence, stromal engagement and tumour-autonomous support, even within breast cancer.

Thus, BBB traversal should be regarded as an entry event rather than the endpoint of metastatic competence. Once within the brain, disseminated breast cancer cells must survive perivascular stress, exploit or remodel the existing neurovasculature, withstand early stromal defence and establish a local growth-supportive relationship with surrounding brain cells. The ability to colonize the parenchyma therefore constitutes a major selective bottleneck beyond extravasation itself. Current evidence further suggests that this bottleneck is not solved by a single universal program: TNBC and HER2-positive disease can adopt distinct spatial and stromal strategies during early brain colonization.31,33,36

Neural and Immune Crosstalk Supports Brain Colonization

Successful brain colonization depends not only on tumour-intrinsic fitness but also on reciprocal communication with neural and immune cells in the metastatic niche. Astrocytes, brain-resident microglia, recruited macrophages and neurons do not act as independent stromal components; instead, signals exchanged among these populations can alter tumour-cell survival, inflammatory tone and the capacity of disseminated cells to expand after extravasation. Importantly, these interactions are not uniformly tumour-promoting. Host responses can initially restrict metastatic growth, whereas tumour cells that persist can subsequently exploit or reprogramme the same cellular networks. This context dependence is particularly evident in the divergent functions reported for astrocytes and myeloid populations during breast cancer brain metastasis.

Direct tumour–astrocyte communication provides one of the clearest examples of such reciprocal adaptation. Chen and colleagues showed that brain-metastatic breast and lung cancer cells use protocadherin-7 (PCDH7) to establish connexin-43 (Cx43)-containing gap junctions with astrocytes. These junctions were dispensable for initial extravasation but became important during post-extravasation outgrowth. Tumour-derived cGAMP transferred through the gap junctions activated STING signalling in astrocytes, inducing interferon-α and tumour necrosis factor production; these cytokines subsequently activated STAT1 and NF-κB in metastatic cells and enhanced growth and chemoresistance.37 This mechanism is notable because it illustrates a bidirectional circuit rather than a conventional one-way stromal signal: metastatic cells first engage a neural support network and then convert its innate immune response into a survival advantage.

Astrocyte–tumour communication can in turn reshape the local myeloid compartment. In breast cancer models, astrocyte-derived exosomal microRNAs reversibly suppressed PTEN expression in metastatic tumour cells, increasing tumour-cell CCL2 secretion and recruitment of IBA1-positive myeloid cells; these recruited cells reciprocally enhanced tumour proliferation and reduced apoptosis. Human paired primary and brain-metastatic samples in the same study showed an association between PTEN loss, increased CCL2 and greater IBA1-positive myeloid infiltration, providing clinical support for this multicellular circuit.38 The biological significance lies less in PTEN loss alone than in the sequence of reciprocal reprogramming: an astrocytic signal alters the metastatic cell, which then changes the composition of its surrounding immune niche.

The myeloid compartment itself is heterogeneous, however, and should not be reduced to a uniformly protumoural “macrophage/microglia” population. CSF1R-expressing tumour-associated macrophages and microglia are abundant in human and experimental BM, and pharmacological CSF1R inhibition in breast-to-brain models reduced early metastatic establishment and tumour growth. Yet established lesions adapted to CSF1R blockade through a compensatory CSF2Rβ–STAT5 programme in tumour-associated myeloid cells, which promoted neuroinflammatory and wound-repair states associated with tumour recurrence.39 Conversely, lineage-resolved single-cell and genetic studies have shown that brain-resident microglia can exert a distinct antitumour function. Evans and colleagues found that microglial depletion increased breast cancer brain-metastatic burden and impaired natural killer- and T-cell responses, whereas inflammatory microglial signatures were conserved in human samples and associated with more favourable outcomes.40 These observations caution against assigning fixed tumour-promoting functions to microglia or extrapolating macrophage biology directly to resident microglia. Rather, the contribution of the myeloid compartment appears to depend on cellular origin, activation state and disease stage.

Neurons provide another source of niche-derived signals that metastatic breast cancer can appropriate. Zeng and colleagues demonstrated that breast-to-brain metastatic cells express functional N-methyl-D-aspartate receptors (NMDARs) but produce insufficient glutamate for autonomous receptor activation. Instead, tumour cells localize near glutamatergic synapses and form pseudo-tripartite structures that enable access to neuron-derived glutamate. NMDAR signalling promoted metastatic colonization rather than initial brain seeding, directly linking neuronal activity to post-entry tumour expansion.41 More recent breast cancer-specific work broadens this concept. Transcriptomic analyses of matched primary and brain-metastatic breast cancers identified enrichment of multiple glutamatergic and GABAergic receptor components in brain lesions, while pharmacological interference with glutamatergic or GABAergic transmission suppressed metastatic outgrowth in experimental models.42 Together, these findings suggest that metastatic cells can acquire features of neural signal responsiveness without necessarily forming bona fide neuron-to-tumour synapses of the type described in primary gliomas.

Neural communication is also intertwined with glial biology rather than operating as an isolated tumour–neuron interaction. A breast cancer study showed that extracellular vesicle-associated miR-199b-5p from brain-metastatic cells targeted transporters in both astrocytes and neurons, disrupting normal neuron–astrocyte coupling and creating an extracellular environment that favoured tumour growth.43 Although the metabolic consequences of this process are discussed in the next section, the study reinforces a broader point: metastatic breast cancer can exploit pre-existing communication networks between host brain cells, rather than interacting with each cellular population independently.

Current evidence therefore supports brain colonization as a product of dynamic tumour–host crosstalk rather than passive stromal support. PCDH7–Cx43-mediated communication couples tumour cells to astrocytic innate signalling; astrocyte-induced tumour reprogramming can alter myeloid recruitment; CSF1R-dependent and other myeloid states exert stage- and lineage-specific effects; and metastatic cells can appropriate neuronal neurotransmission to sustain post-seeding growth. At the same time, the presence of antitumour microglial responses argues against a simple model in which all brain-resident cells are converted into metastatic accomplices. A more accurate view is that successful breast cancer cells selectively exploit, evade or remodel different components of the neural–immune ecosystem during colonization.

Metabolic Plasticity Supports Brain Persistence

Metabolic adaptation in brain metastasis is better understood as plasticity across multiple nutrient and bioenergetic pathways than as a fixed shift toward either glycolysis or oxidative metabolism. Early breast cancer-specific work already suggested this complexity. Proteomic analysis of brain-metastatic breast cancer cells showed coordinated increases in glycolytic enzymes, tricarboxylic acid cycle activity and oxidative phosphorylation, together with activation of the pentose phosphate pathway and glutathione-related antioxidant systems.44 These changes were associated with enhanced survival in the brain microenvironment and resistance to perturbations of cellular redox balance. Thus, glucose metabolism in breast cancer brain metastasis cannot be reduced to a conventional Warburg phenotype: glycolytic flux can coexist with substantial mitochondrial oxidation, while glucose-derived intermediates can also be diverted toward biosynthesis and NADPH production.

The capacity to switch carbon sources becomes particularly important when glucose availability or utilization is constrained. In breast cancer models, brain-metastatic cells acquired the ability to survive under glucose-limited conditions by increasing gluconeogenic activity and oxidizing glutamine and branched-chain amino acids, thereby supporting the non-oxidative pentose phosphate pathway and purine synthesis. Fructose-1,6-bisphosphatase suppression impaired metastatic-cell viability, while human breast cancer BM showed higher FBP expression and glycogen accumulation than corresponding primary tumours, providing partial clinical support for this adaptive state.45 More recent work in HER2-positive breast cancer further demonstrated that different brain-tropic states can use distinct metabolic solutions. Aggressive synchronous metastatic cells produced and metabolized lactate, which also suppressed innate immune surveillance, whereas latent and metachronous cells relied more heavily on glutamine oxidation and the cystine–glutamate transporter xCT to maintain redox homeostasis.46 xCT expression was increased in matched metachronous BM relative to primary breast tumours, and xCT inhibition reduced residual disease and late recurrence in preclinical models. These findings suggest that glucose, lactate and amino-acid metabolism are not interchangeable components of one uniform programme; their relative importance changes with metastatic state and proliferative fitness.

Lipid metabolism provides an equally clear example of context-dependent adaptation. In HER2-positive breast cancer models, tumours growing in the brain displayed increased de novo fatty-acid synthesis compared with the same breast cancer cells growing at extracranial sites. Importantly, Ferraro and colleagues linked this phenotype to reduced availability of extracellular lipids accessible to tumour cells in the brain environment, rather than to an absence of lipid from brain tissue itself; genetic or pharmacological inhibition of fatty-acid synthase selectively impaired intracranial tumour growth.47 In contrast, latent brain-metastatic breast cancer cells can exploit fatty-acid oxidation. Parida and colleagues showed that DRP1-dependent mitochondrial fragmentation facilitated fatty-acid oxidation, sustained bioenergetics and preserved redox balance in latent cells, whereas limiting mitochondrial plasticity increased lipid-droplet accumulation and reduced metastatic persistence.48 These observations are not necessarily contradictory. Rather, they indicate that proliferating and latent tumour cells may solve lipid stress differently, with some states increasing lipid synthesis and others increasing lipid oxidation.

Recent breast cancer-specific evidence strengthens the link between oncogenomic state and lipid adaptation. Analysis of breast cancer BM revealed frequent p53 pathway inactivation, and functional experiments showed that loss of p53 increased SCD1 expression and fatty-acid synthesis, both of which contributed to brain-metastatic growth.49 SCD1 loss abolished the growth advantage conferred by p53 inactivation, while astrocyte-derived factors further enhanced fatty-acid metabolism in a p53-dependent manner. These findings are important because they connect metabolic flexibility to both tumour-intrinsic genotype and local stromal input, rather than treating metabolism as a passive response to nutrient availability. They also reinforce the broader principle that the metabolic phenotype selected in the brain may differ among molecular subsets of breast cancer.

Evidence from broader brain-metastasis models supports several additional metabolic constraints, although these mechanisms should not automatically be generalized to breast cancer. A multi-cancer study that included aggressive TNBC brain-metastatic models showed that low interstitial serine and glycine availability in the brain increased dependence on PHGDH-mediated de novo serine synthesis; PHGDH inhibition preferentially suppressed brain metastatic growth while having less effect on extracranial tumours.50 Separately, isotope-tracing studies in patients with glioblastoma and BM demonstrated oxidation of acetate in the tricarboxylic acid cycle, including in one patient with breast cancer brain metastasis.51 The latter finding establishes that acetate can contribute to oxidative metabolism in human brain tumours, but the very small number of metastatic specimens precludes treating acetate dependence as an established breast cancer-specific phenotype. Accordingly, PHGDH dependence observed in a multi-cancer experimental setting that included TNBC cannot be extrapolated to all breast cancer subtypes, while acetate oxidation detected in a limited human cohort does not establish a clinically validated metabolic dependency in breast cancer BM.

Importantly, recent evidence also argues against an overly simple model in which the abundance of any single nutrient dictates breast cancer organotropism. In a study that quantified 124 metabolites across tissues and experimentally imposed defined nutrient auxotrophies on metastatic breast cancer cells, Abbott and colleagues found that no single nutrient was sufficient to explain the organ-specific distribution of breast cancer metastases.52 Brain tumours showed increased labelling into several amino acids, including serine and glycine, but increased biosynthetic flux did not consistently translate into a stronger site-specific dependency. This distinction is critical: metabolic activity, metabolic requirement and metastatic fitness are not equivalent measures. The brain therefore imposes a combination of substrate availability, cellular competition, biosynthetic demand and oxidative stress rather than one uniform state of “nutrient deprivation”.

Current breast cancer-specific evidence consequently supports a model in which persistence in the brain depends on the capacity to reconfigure fuel use, biosynthesis, mitochondrial function and redox control according to metastatic state and local nutrient context. Glucose oxidation and glycolysis can coexist; glutamine and branched-chain amino acids can compensate when glucose use is constrained; lactate can function as both a metabolic substrate and an immunomodulatory metabolite; and lipid synthesis or fatty-acid oxidation can predominate in different tumour states. Broader brain-metastasis studies add plausible dependencies on serine synthesis and alternative oxidative substrates, but these mechanisms require breast cancer-specific validation before they can be considered general features of breast cancer BM. Metabolic plasticity, rather than reliance on any single pathway, therefore appears to be the more defensible explanation for how disseminated breast cancer cells endure metabolic stress and persist within the brain.

Several components of the parenchymal-niche framework are supported by analyses of human breast cancer BM, but most mechanistic relationships remain grounded primarily in preclinical models. Evidence from other tumour types further broadens the range of plausible adaptive mechanisms, although these findings cannot be assumed to apply uniformly across breast cancer subtypes. Importantly, the literature also reveals context-dependent and sometimes opposing host responses: astrocytes and myeloid cells can exert either restrictive or tumour-supportive effects depending on cellular state, disease stage and experimental setting, while metabolic dependencies likewise vary across metastatic states. These differences argue against a single conserved programme of brain colonization and instead support a model of adaptive heterogeneity whose relevance in human breast cancer still requires broader validation. The major neurovascular, neural–immune and metabolic adaptations supporting post-entry brain colonization are summarized in Figure 3.

Figure 3.

Three-part pathway diagram of neurovascular, neural immune and metabolic adaptations in brain colonization. A) Flow diagram titled Neurovascular adaptation and early colonization. A row of boxes reads Vascular arrest to Endothelial interaction to Extravasation, then a downward arrow to Perivascular survival, then a leftward arrow to Early micrometastatic growth. Below, a schematic labeled TNBC-like breast cancer cell shows cells along a Brain microvessel and into Brain parenchyma, with labels Astrocyte endfoot, Endothelial cell, Pericyte and Early micrometastasis. A dashed box labeled blood brain barrier or neurovascular unit contains Brain endothelial signals pointing to EGFR DOCK4 RAC1, which points to Tumour cell intercalation or extravasation. A section titled Perivascular survival and niche interactions shows vessel to beta subscript 1 integrin, adhesion, to Perivascular growth. L1CAM points to Vascular co option or survival, which points to Tumour survival. FasL slash plasmin points toward Tumour survival and Serpins points toward plasmin. The words restrictive signals are placed near an astrocyte drawing. A bottom sentence reads, blood brain barrier traversal is an entry event; durable colonization requires post entry survival. B) Pathway titled Neural immune crosstalk. A banner reads Tumour astrocyte gap junction and innate signalling. An astrocyte and a metastatic breast cancer cell connect through PCDH7 slash Cx43 gap junction, leading to cGAMP, then to STING. A dashed arrow labeled interferon alpha slash tumour necrosis factor points toward Tumour STAT1 slash nuclear factor kappa B, which points to Survival slash chemoresistance. Exosomal micro ribonucleic acids point to PTEN, then to CCL2, then to myeloid recruitment. A cluster labeled Infiltrating slash TAM like myeloid cell connects to CSF1 slash CSF1R, which leads by dashed arrows to protumoural myeloid state and CSF1R inhibition. A separate dashed pathway shows CSF2Rbeta STAT5 compensation leading to recurrence associated myeloid state. A box reads Reduced early metastatic establishment. Along the bottom, Resident microglia points to immune activation, which points to NK slash T cell response, which points to tumour. A strip titled Neuronal signaling includes the words pseudo tripartite slash synaptic proximity and shows Glutamatergic neuron to Glutamate to tumour NMDAR, leading to post entry colonization. C) Module diagram titled Metabolic plasticity. A box titled Glucose metabolism shows Glycolysis feeding into tricarboxylic acid cycle and oxidative phosphorylation, with a branch labeled pentose phosphate pathway slash nicotinamide adenine dinucleotide phosphate. A box titled Alternative carbon use lists Glucose limitation, gluconeogenic adaptation slash FBP and glutamine plus branched chain amino acid oxidation, followed by the note model supported; partial human validation. A box titled State dependent lactate slash glutamine is split into two tracks by a dotted vertical divider: Synchronous slash aggressive state to lactate production slash use to innate immune suppression and Latent slash metachronous state to glutamine oxidation to xCT to redox homeostasis. A box titled Lipid adaptation shows Proliferative lesions to de novo fatty acid synthesis slash FASN and Latent cells to DRP1 dependent mitochondrial plasticity to fatty acid oxidation. A footer line reads state dependent alternatives.

Multilevel adaptation supports parenchymal brain colonization. Successful breast cancer brain colonization requires more than blood–brain barrier (BBB) traversal. Neurovascular adaptation encompasses vascular arrest, endothelial interaction and extravasation, followed by perivascular survival and early micrometastatic growth; representative mechanisms include endothelial EGFR–DOCK4–RAC1 signalling, β1-integrin-mediated adhesion, L1CAM-dependent vascular co-option, and tumour serpins that counteract restrictive plasmin-associated signals. Neural–immune crosstalk includes tumour–astrocyte PCDH7/Cx43–cGAMP–STING signalling, astrocyte-associated PTEN loss and CCL2-mediated myeloid recruitment, context-dependent myeloid and microglial responses, and glutamate–NMDAR signalling that supports post-entry colonization. CSF1R inhibition is depicted as reducing early metastatic establishment in the relevant experimental setting, whereas compensatory CSF2Rβ–STAT5 signalling is associated with a distinct recurrence-related myeloid state. These findings should not be interpreted as indicating a uniform effect of CSF1R inhibition across different stages of brain metastasis. Metabolic adaptation is similarly heterogeneous, with parallel use of glucose, alternative carbon sources, lactate/glutamine pathways and lipid metabolism; proliferative and latent states may favour distinct metabolic programmes. Overall, durable brain colonization depends on coordinated post-entry survival, host-cell interactions and state-dependent metabolic plasticity.

Leptomeningeal Metastases Adapt to a Distinct CNS Environment

The Leptomeningeal Space Imposes Distinct Ecological Constraints

LM occupies a markedly different ecological setting from parenchymal brain metastasis. The leptomeninges consist of the pia and arachnoid membranes and enclose the CSF-filled subarachnoid space, which extends across the entire craniospinal axis. Once tumour cells enter this compartment, they are exposed to continuous CSF circulation rather than the relatively fixed extracellular matrix and dense cellular architecture of the brain parenchyma. Pulsatile CSF flow can redistribute free tumour cells throughout the neuraxis, allowing dissemination over considerable anatomical distances but also subjecting them to a spatially dynamic environment in which local attachment, nutrient access and drug exposure can vary.13 Thus, LM should not be viewed simply as a superficial form of brain metastasis; it occupies a fluid-dominated compartment with its own physical and biological constraints.

The degree of cellular and matrix support within this compartment is also heterogeneous. LM cells can exist as freely floating cells or clusters within CSF, but they can also adhere to the pia and grow as surface-associated plaques or nodules. Experimental models derived from both triple-negative and HER2-positive breast cancer reproduce these floating and adherent states, and prolonged fractionation studies indicate substantial plasticity between them. Floating LM cells showed lower ATP content and reduced proliferation under standard in vitro conditions yet rapidly colonized the leptomeningeal space after intracisternal implantation, illustrating that phenotypes poorly suited to conventional tissue culture can nevertheless be highly fit within the CSF environment.53 These observations are important because they qualify the notion of “sparse cellular support”: some LM cells survive with little direct stromal anchorage, whereas others acquire access to meningeal surfaces and local cellular interactions. LM is therefore not a uniformly free-floating disease.

Resource availability imposes an additional selective pressure. Compared with plasma and most tissue interstitial fluids, CSF contains relatively low concentrations of many nutrients and micronutrients, while leptomeningeal tumour cells lack the dense tumour-associated capillary network available to established parenchymal lesions. Direct evidence for the biological importance of this constraint comes from single-cell analysis of CSF from patients with LM. Chi and colleagues identified a high-affinity LCN2–SLC22A17 iron-acquisition programme in tumour cells, but not CSF macrophages, and showed that this system enabled cancer cells to compete for limited extracellular iron; disruption of iron availability suppressed LM growth in experimental models.54 Importantly, the patient cohort included LM from breast cancer, although the study was not restricted to breast cancer alone. The significance of this finding extends beyond iron itself: it demonstrates that survival in LM can depend on mechanisms that compensate for resource scarcity rather than on the metabolic conditions encountered in vascularized solid metastases.

At the same time, the CSF should not be treated as a chemically uniform or universally nutrient-depleted medium. Tumour growth, inflammation, blood–CSF barrier disruption and treatment can all alter its composition. A recent targeted metabolomic analysis of 117 CSF samples, including 47 from patients with LM, identified substantial changes in lipid, acylcarnitine and amino-acid profiles compared with healthy controls and patients with parenchymal brain tumours. Notably, several metabolites were increased rather than depleted in LM, and conventional nutrient classes did not show a simple uniform pattern.55 These data caution against reducing the leptomeningeal niche to a single state of “starvation”. A more accurate interpretation is that LM cells occupy a resource-constrained but dynamically remodelled fluid environment, in which specific nutrients may become limiting while others accumulate as a consequence of tumour and host metabolism.

The anatomical form of LM further reflects this ecological heterogeneity. Radiographically, disease may appear as linear or plaque-like enhancement along meningeal surfaces, discrete nodules, or a mixture of both. In a breast cancer-specific cohort of 161 patients with leptomeningeal disease, 53% had nodular involvement, 36% classical disease, and 6% mixed disease.56 These patterns should not be interpreted as merely descriptive imaging categories: they likely represent different degrees of tumour attachment, local tissue interaction and access to supportive structures, although their underlying biological determinants remain incompletely resolved.

Overall, the leptomeningeal compartment imposes a combination of fluid dynamics, limited stromal support, constrained nutrient access and spatially heterogeneous opportunities for attachment. These pressures differ fundamentally from those encountered within the cellular and vascularized brain parenchyma. Successful LM therefore requires more than simple survival in CSF; metastatic cells must remain viable while circulating or surface-adherent, compete for limiting resources and adapt to local meningeal interfaces. This ecological setting provides the context for the active niche-remodelling and resource-acquisition mechanisms discussed in the following sections.

Tumor Cells Remodel the Blood–CSF Interface and Local Milieu

The choroid plexus is not merely an anatomical gateway to the CSF; it is also a dynamic regulator of CSF composition and therefore a potential target of metastatic niche remodeling. Its fenestrated capillaries expose the stromal compartment to circulating macromolecules, whereas tight junctions between choroid plexus epithelial cells restrict their subsequent passage into the ventricular CSF. In LM, this regulatory interface can become functionally altered, changing the availability of circulating proteins, nutrients and signaling molecules within an otherwise resource-constrained compartment. Contemporary models therefore view the blood–CSF interface as an active component of LM biology rather than a static barrier that is either intact or disrupted.14,57

A representative example is the complement C3–C3aR mechanism described by Boire and colleagues. Using breast- and lung-cancer LM models together with patient-derived CSF, the investigators found that leptomeningeal metastatic cells upregulated and secreted complement component 3 (C3). Tumour-derived C3 activated C3aR expressed by choroid plexus epithelial cells, altered epithelial tight-junction organization and increased blood–CSF barrier permeability. Genetic or pharmacological interference with this pathway reduced leptomeningeal tumour growth in experimental models.58 Importantly, these experiments primarily demonstrate how tumour cells already adapted to the leptomeningeal compartment can modify the choroid plexus to support subsequent growth; they do not establish C3–C3aR signaling as a universal mechanism of initial LM entry. The pathway is therefore more appropriately interpreted as a proof of principle for tumour-driven niche remodeling than as a central organizing mechanism of LM.

The consequence of barrier remodeling is equally important as the disruption itself. C3-mediated changes in the choroid plexus permitted selected plasma-derived components to enter the CSF, effectively allowing metastatic cells to draw on resources that are normally less accessible within this compartment. Amphiregulin provided the clearest experimental example: it was markedly enriched in the CSF of many patients with LM, promoted growth of LM-derived cells under mitogen-poor conditions and enhanced leptomeningeal growth in vivo. However, patient CSF profiles were heterogeneous. In longitudinal samples, multiple growth-promoting factors increased during LM progression, including amphiregulin, platelet-derived growth factor, glial cell line-derived neurotrophic factor, persephin, artemin and fibroblast growth factor 4, with different combinations observed among individual patients.58 This heterogeneity argues against a single plasma-derived mitogen being universally required. Instead, barrier remodeling may provide a broader ecological advantage by increasing access to a variable pool of circulating trophic factors.

Evidence beyond complement signaling further supports the idea that LM is accompanied by active remodeling of extracellular and barrier-associated processes. Analysis of CSF from patients with neoplastic meningitis showed substantially increased activities of MMP-9, ADAM8 and ADAM17 compared with patients who had parenchymal BM without leptomeningeal involvement and with neurological controls. These proteolytic signatures were associated with barrier impairment and are consistent with increased extracellular-matrix turnover in the leptomeningeal compartment, although the study included heterogeneous tumour types and did not establish the choroid plexus as the specific source of these enzymes.59 More broadly, recent reviews of LM biology emphasize that remodelling can involve not only the blood–choroid plexus barrier but also the pia and surrounding meningeal interfaces.14 Thus, complement-dependent permeability changes should be regarded as one experimentally well-defined example within a wider and still incompletely characterized spectrum of tumour–host interactions.

The available evidence therefore supports a broader niche-remodeling principle: leptomeningeal tumour cells are not confined to passively tolerating the native CSF environment but can modify surrounding interfaces in ways that increase access to systemic resources and reshape local signaling conditions. C3–C3aR provides the most mechanistically resolved example of this process, but its relevance across breast cancer subtypes and across different forms of LM remains incompletely established. This distinction is important because barrier permeability, plasma-factor influx and extracellular remodeling are likely to vary between patients and over time. Such remodeling may partially compensate for the ecological constraints of the leptomeningeal space, setting the stage for the resource-acquisition and immune-adaptive mechanisms considered in the next section.

Resource Competition and Immune Adaptation Contribute to Leptomeningeal Persistence

Persistence within the leptomeningeal compartment requires tumour cells to solve two closely related problems: obtaining sufficient resources from a relatively constrained extracellular environment and remaining viable in the presence of local immune surveillance. These pressures are mechanistically intertwined because tumour and immune cells can compete for the same metabolites and micronutrients, while inflammatory signals generated by host cells can themselves alter tumour-cell nutrient acquisition. The strongest experimental example is iron competition, but this should be viewed as a general principle of adaptation to ecological scarcity rather than a universal metabolic programme of LM.

The LCN2–SLC22A17 system illustrates this principle particularly well. Single-cell RNA sequencing of CSF from five patients with LM, including three with breast cancer, showed that tumour cells expressed the high-affinity iron-binding protein lipocalin-2 (LCN2) and its receptor SLC22A17, whereas these molecules were not comparably expressed by CSF macrophages.54 Human tissue and CSF analyses confirmed tumour-cell expression, and breast and lung LM models demonstrated that LCN2 promoted iron acquisition and metastatic growth under the metabolically restrictive conditions of the leptomeningeal space. LCN2 depletion reduced LM growth, whereas limiting extracellular iron pharmacologically also suppressed disease in vivo.54 Importantly, these findings do not establish LCN2–SLC22A17 as an obligatory metabolic dependency of all breast cancer LM. Rather, they demonstrate that metastatic cells can acquire specialized high-affinity resource-capture mechanisms when an essential micronutrient becomes limiting.

Iron competition also reveals how metabolic adaptation can intersect with immune function. CSF macrophages produced inflammatory cytokines, including IL-6, IL-8 and IL-1β, that increased LCN2 expression in LM cells, creating an apparently paradoxical interaction in which a host inflammatory response enhanced tumour-cell access to a limiting resource.54 In experimental models, this competitive advantage reduced iron availability to CSF macrophages and was accompanied by impaired phagocytosis and reactive oxygen species generation; tumour-cell LCN2 depletion partially restored these macrophage functions. Thus, resource acquisition can provide more than a direct proliferative advantage: competition for a shared micronutrient may simultaneously weaken innate immune activity. The mechanism provides a concrete example of metabolic competition and immune escape becoming coupled within the same ecological interaction.

Patient-derived CSF studies support the presence of an altered immune ecosystem in breast cancer LM, although they also indicate that macrophage states are more heterogeneous than a simple immunosuppressive phenotype would imply. Single-cell analysis of CSF from breast and lung LM identified enrichment of monocyte/macrophage populations with anti-inflammatory or M2-associated transcriptional features, together with regulatory T cells and extensive predicted communication between macrophages and tumour cells.60 In a separate breast cancer-specific study, high-dimensional flow cytometry of patients with metastatic breast cancer showed that confirmed LM was associated with fewer CSF CD45-positive cells and CD8-positive T cells, more regulatory T cells, and an increased fraction of partially exhausted CD8-positive T cells compared with patients without LM.61 These human observations are consistent with a relatively permissive immune environment, but they are based on small cohorts and remain largely correlative. Moreover, the inflammatory macrophage activity that induces LCN2 in experimental systems and the anti-inflammatory macrophage signatures observed in patient CSF should not be forced into a single polarization model; they more likely reflect functional heterogeneity, temporal plasticity and differences in sampling context.

More recent experimental work further suggests that meningeal immune adaptation extends beyond macrophages already present in the CSF. Zhao and colleagues identified a population of dura-derived LM-associated macrophages that entered the CSF in an MMP14-dependent manner and were recruited through tumour-derived SPP1. In LM mouse models, these cells contributed to an immunosuppressive environment characterized by impaired and exhausted T-cell states, whereas interference with the SPP1–MMP14 programme reduced macrophage recruitment, restrained tumour growth and prolonged survival.62 This study provides strong mechanistic support for the concept that tumour cells can actively reshape immune-cell trafficking at CNS borders. However, this pathway has not yet been established as a breast cancer-specific or universally conserved mechanism of human LM and should therefore be interpreted as an additional example of tumour–immune niche adaptation rather than a defining LM axis.

The available evidence therefore supports a broader model in which resource acquisition, metabolic competition and immune adaptation are interconnected strategies for persistence in the leptomeningeal space. LCN2-mediated iron capture provides the most experimentally resolved example, but direct evidence for analogous competitive dependencies involving glucose, amino acids or lipids in human breast cancer LM remains limited. Likewise, CSF macrophages and lymphocytes should not be regarded as uniformly suppressed populations; their states appear heterogeneous and potentially dynamic. A longitudinal multimodal analysis of CSF from a patient with HER2-positive breast cancer LM demonstrated substantial treatment-associated shifts in tumour and immune-cell composition and inflammatory cytokine profiles, although the single-patient design precludes broader inference.63 Overall, the more defensible conclusion is not that LM is governed by one metabolic or immune pathway, but that successful metastatic cells can exploit ecological scarcity while simultaneously modifying the immune context in which that competition occurs.

Mechanistic evidence for breast cancer LM remains much less mature than that for parenchymal brain metastasis. C3–C3aR-mediated barrier remodeling has been investigated in breast- and lung-cancer LM models with supporting patient-derived observations, whereas LCN2–SLC22A17-mediated iron acquisition is supported by mixed-cancer CSF samples, including breast cancer cases, and experimental LM models. The SPP1–MMP14-associated macrophage recruitment pathway is supported primarily by experimental LM models and has not been established as a breast cancer-specific mechanism in patients. Although these findings provide important mechanistic insights, their prevalence, subtype specificity, and clinical applicability in breast cancer LM remain insufficiently established. Overall, the current framework should be interpreted as an evidence-weighted synthesis of emerging principles, not a fully validated mechanistic model. The ecological, interface-remodeling and resource–immune adaptations that support leptomeningeal persistence are summarized in Figure 4.

Figure 4.

Three-part schematic of leptomeningeal metastasis: fluid ecology, blood to CSF remodeling and immune competition. A) Schematic titled, Fluid-dominated leptomeningeal ecology. Labels show, Arachnoid, Subarachnoid space slash CSF, CSF flow, Pia slash brain surface and Brain parenchyma. Three tumour states are labeled, Floating cells slash clusters, Linear surface-associated disease and Nodular disease. A flow diagram below shows two inputs, CSF flow and Limited stromal anchorage, feeding into LM survival pressure. Two outputs from LM survival pressure are labeled, Resource constraint and Variable access to supportive surfaces. B) Schematic titled, Blood to CSF interface remodeling. A layered interface is labeled from left to right, Blood slash fenestrated capillary, Choroid plexus stroma, Choroid plexus epithelium with tight junctions and Ventricular CSF slash LM cells. Within the interface are labels, plasma, stromal cells and epithelial cells. On the CSF side, an LM tumour cell is shown with labels, C3 secretion, C3aR and C3aR activation. Downstream boxes read, tight-junction integrity and Barrier permeability up arrow. A lower section titled, Plasma-derived trophic factors, lists bullet points, Amphiregulin and other trophic factors, with a label, into CSF, leading to a box labeled, LM growth support. A footer line reads, Representative niche-remodeling mechanism not a universal LM axis. C) Schematic titled, Resource competition and immune adaptation. At upper left are labels, CSF macrophage, CD8 T cells and Treg. A dashed callout box reads, Immune-permissive CSF states observed in patient cohorts, followed by bullets, fewer CD8 T cells, more Tregs, partially exhausted CD8 state. Near the LM tumour cell are labels, IL-6 slash IL-8 slash IL-1 beta and LCN2 expression. A note reads, limited extracellular iron, with a note, high-affinity iron acquisition. A pathway under the LM tumour cell shows a box labeled, LCN2 to SLC22A17, then Iron acquisition, then Reduced macrophage iron, leading to two boxes, phagocytosis down arrow and ROS generation down arrow. On the right side, a separate pathway labeled, Tumour SPP1, leads to text, dura-derived macrophage recruitment slash MMP14-dependent entry, then to T-cell dysfunction. A bottom label reads, preclinical slash pan-LM evidence.

Leptomeningeal persistence emerges from fluid ecology, interface remodeling, and resource–immune competition. Leptomeningeal metastasis (LM) occupies a cerebrospinal fluid (CSF)-facing, fluid-dominated compartment in which floating, linear surface-associated and nodular tumour states experience CSF flow, limited stromal Anchorage, resource constraints and variable access to supportive surfaces. Tumour-derived C3–C3aR signalling illustrates how LM cells can remodel the choroid plexus blood–CSF interface, increasing barrier permeability and access to plasma-derived trophic factors. Under iron-limited conditions, LCN2–SLC22A17-mediated iron acquisition may confer a competitive advantage on LM tumour cells. The revised pathway depicts this acquisition as a positive process rather than an inhibitory interaction. Competition for iron may reduce macrophage iron availability and contribute to impaired phagocytosis and reactive oxygen species (ROS) generation in the relevant experimental setting. Patient CSF studies further indicate immune-permissive states characterized by reduced CD8 T-cell representation, increased regulatory T cells (Tregs), and partially exhausted CD8 T-cell phenotypes. These observations derive from different experimental and clinical settings, and the depicted mechanisms should be regarded as representative, context-dependent adaptations rather than universally established pathways in breast cancer LM.

CNS Metastases Undergo Compartment-Associated Molecular Evolution

CNS Disease Reflects Clonal Selection and Branched Evolution

CNS metastases should not be regarded as static molecular replicas of the primary breast tumour. Primary breast cancers are themselves genetically heterogeneous, providing multiple subclonal populations on which metastatic selection can act. Phylogenetic analyses of multiregion primary and metastatic breast cancer have identified more than one route to systemic dissemination: in some patients, distant metastases descend from a common metastatic precursor, whereas in others, distinct metastatic lesions are seeded by different subclones already present within the primary tumour.64 These observations support the possibility that traits contributing to CNS competence may exist in pre-existing subclones before clinically detectable dissemination. They do not, however, imply that a fully specified “CNS clone” is already present in the untreated primary tumour; subsequent selection and evolution remain important after dissemination.

Direct studies of BM provide strong evidence for such post-seeding divergence. In a landmark whole-exome analysis of matched primary tumours and BM across several cancer types, including breast cancer, Brastianos and colleagues found that clonally related primary and brain lesions shared a common ancestor but subsequently followed branched evolutionary trajectories.19 BM frequently contained potentially actionable alterations that were not detected in the sampled primary tumour, while extracranial metastases could themselves diverge substantially from the intracranial lesion. Breast cancer-specific datasets reinforce this principle while also showing that divergence is incomplete rather than universal. In longitudinal paired samples, Cosgrove and colleagues identified clinical and intrinsic subtype switching together with subtype-specific transcriptional and genomic changes in breast cancer BM.65 More recent genomic profiling likewise showed that many alterations remain conserved between brain and extracranial tumours, even though lesion-specific somatic mutations, copy-number changes and pathway differences are also detectable.66 Thus, branched evolution does not mean complete molecular replacement: CNS lesions generally retain substantial ancestral information while acquiring or selecting additional features during progression.

Phenotypic evolution provides a parallel indication that the metastatic state is dynamic. A breast cancer-specific analysis of 75 brain metastases, including matched primary tumours, identified recurrent changes in gene expression and intrinsic subtype between primary and intracranial disease, with several paired cases showing a shift toward HER2-enriched phenotypes.67 Such changes may reflect a combination of pre-existing subclone selection, transcriptional plasticity and subsequent genomic evolution. Importantly, they should not automatically be attributed to the brain microenvironment. Differences between an archival primary tumour and a later brain metastasis also integrate the effects of elapsed time, systemic disease evolution, intervening therapies and sampling of different tumour subclones.

Leptomeningeal disease provides complementary evidence for compartment-associated evolution. In breast cancer LM, Fitzpatrick and colleagues found substantial genomic divergence between CSF-derived tumour DNA and both primary and extracranial disease, with phylogenetic reconstruction suggesting that leptomeningeal lineages often diverged relatively early from the primary tumour and subsequently continued to evolve independently.30 Among matched CSF and plasma samples, most evaluable cases showed divergent rather than strictly linear evolution, although linear co-evolution was also observed in a minority, emphasizing that no single phylogenetic pattern applies to all patients. More recent breast cancer-specific data provide a useful counterbalance. Wessels and colleagues detected CSF-restricted alterations in only 20% of evaluable patients with LM, while alterations identified in the primary tumour or systemic metastases were generally retained in CSF.68 These apparently different degrees of divergence are not necessarily contradictory; they may reflect differences in sequencing depth, sample composition, disease subtype, treatment history and evolutionary timing. Together, they indicate that CNS compartmentalization can coexist with substantial preservation of ancestral clones.

Therapy adds another selective layer to this evolutionary process. In a large analysis of paired primary and metastatic samples from breast, lung and colorectal cancers, treated metastases were more likely than untreated metastases to contain private driver alterations, consistent with therapy-associated clonal remodeling.69 Breast cancer-specific data from the AURORA programme likewise showed increased clonality of selected drivers, enrichment of alterations such as ESR1, PTEN and RB1 in metastatic disease, and intrinsic subtype switching between paired primary and metastatic samples.70 These studies were not CNS-specific, however, and therefore establish treatment as a general source of metastatic selection rather than proving that therapy generates particular CNS clones. Indeed, breast cancer LM provides evidence that treatment pressure itself may be spatially uneven: chemotherapy-associated mutational signatures were more prominent in plasma than in CSF-derived tumour DNA, whereas the extent of CSF-specific genomic divergence was not significantly associated with the number of previous treatment lines.30

The resulting model is therefore one of shared ancestry followed by variable clonal selection and branched evolution. Some CNS-fated populations may originate from subclones already present within the heterogeneous primary tumour, but their subsequent composition can be reshaped by metastatic bottlenecks, local adaptation, therapy and time. Brain and leptomeningeal lesions may consequently retain common ancestral alterations while acquiring different genomic or phenotypic states, and the magnitude of this divergence varies substantially among patients. CNS disease is therefore better understood as an evolving branch of metastatic breast cancer than as a static copy of the primary tumour.

Molecular Phenotypes Differ Across Metastatic Compartments

Molecular discordance between primary breast cancer and CNS metastases is well documented, but its biological meaning requires careful interpretation. Changes in estrogen receptor (ER), progesterone receptor (PR) and HER2 status are among the most clinically visible manifestations. In a series of 105 paired primary breast cancers and resected BM, immunohistochemical discordance was observed in 44% of patients, most commonly as loss of ER or PR expression, whereas increased HER2 expression accounted for a smaller proportion of changes.71 Such observations indicate that the biomarker profile used to classify the primary tumour may not always represent the subsequent intracranial lesion. They do not, however, demonstrate that the brain microenvironment directly induces receptor conversion. The interval between primary diagnosis and brain surgery often encompasses systemic progression, multiple therapies and additional metastatic bottlenecks, each of which can alter the abundance of pre-existing subclones.

HER2 provides a particularly instructive example because both genuine biological change and evolving diagnostic categories contribute to apparent discordance. In 197 matched primary–brain metastasis pairs, Pereslete and colleagues found that HER2-positive primaries were highly stable, with 97% remaining HER2-positive in the brain, whereas HER2-0 primaries could acquire HER2-low or, less frequently, HER2-positive expression; HER2-low tumours were considerably more variable, with approximately half changing category in the brain.72 These findings are consistent with enrichment of HER2 expression in selected CNS lesions but argue against a uniform tendency toward HER2 gain. They also highlight a methodological issue: once HER2-negative disease is subdivided into HER2-0 and HER2-low categories, relatively small differences in immunohistochemical intensity can generate apparent phenotypic switching. Consequently, HER2 gain or loss may reflect a combination of clonal evolution, treatment selection, intratumour heterogeneity and assay variability rather than a CNS-specific adaptive programme.

Genomic alterations show a similarly mixed pattern. The PI3K–AKT–PTEN pathway is frequently implicated in breast cancer BM, but matched-pair studies do not support a single directional change. In a targeted sequencing study of 37 evaluable matched primary–brain metastasis pairs, Thulin and colleagues found PIK3CA mutations in 22% of primary tumours and 25% of BM, while the overall mutational pattern was discordant in approximately one quarter of pairs.73 Another paired analysis examining TP53, PIK3CA, ESR1, ERBB2 and AKT1 detected differences in at least one of these genes in 62.5% of 32 primary–brain metastasis pairs, yet the overall frequencies of several individual alterations were similar between the two compartments.74 Conversely, large unpaired genomic datasets have reported enrichment of ERBB2, ESR1 and other alterations in BM compared with primary-breast-cancer cohorts.75 These apparently different results are informative rather than mutually exclusive: population-level enrichment does not necessarily imply that every individual tumour acquires the same alteration during CNS progression.

Leptomeningeal disease provides further evidence that compartment-associated molecular differences can emerge, while also illustrating their limited penetrance. In 27 patients with breast cancer LM, Wessels and colleagues identified CSF-restricted genomic alterations in 5 of 25 evaluable patients; several involved the PTEN–PI3K–AKT pathway, particularly in triple-negative disease.68 At the same time, all alterations detected in available primary or systemic metastatic tissues were also detected in CSF, demonstrating substantial preservation of ancestral genomic features. Thus, the data support selective divergence in a subset of LM, not wholesale molecular replacement of the systemic tumour. This distinction is particularly important because the presence of a CSF-specific PTEN, PIK3CA or AKT alteration cannot by itself establish that the leptomeningeal environment generated or selected that alteration; unobserved extracranial subclones and differences in assay sensitivity remain plausible alternatives.

ESR1 further illustrates the difficulty of separating compartmental selection from treatment-associated evolution. A large genomic series of 733 breast cancer BM found ESR1 alterations more frequently than in an unpaired primary-breast-cancer comparison cohort.75 Yet ESR1 mutations are also a well-recognized consequence of endocrine selection in advanced hormone receptor-positive disease, making treatment history an essential competing explanation. Consistent with this, Wessels and colleagues identified an ESR1 alteration in the CSF of one patient with LM that was also present in a lymph-node metastasis but absent from the primary tumour; importantly, no ESR1 alteration restricted to the CSF was identified among the evaluable HR-positive/HER2-negative cases.68 This pattern is more consistent with systemic metastatic evolution under endocrine pressure than with a uniquely leptomeningeal mechanism. Accordingly, detection of ESR1 in a CNS specimen should not be interpreted as evidence of CNS-specific adaptation without temporally and spatially matched samples.

Several methodological factors further complicate interpretation. A primary breast tumour is spatially heterogeneous, and the archived block selected for comparison captures only a fraction of that diversity. Brain-metastasis studies are likewise biased toward surgically accessible lesions, while unresected lesions and LM may harbour different clones. ER and PR measurements depend on staining quality and positivity thresholds, and low-level HER2 classification is particularly vulnerable to pre-analytical variation, interobserver differences and sampling of heterogeneous expression.76 Differences in sequencing platform, panel breadth, tumour purity and variant-detection thresholds can similarly determine whether an alteration appears “private” to one compartment. Temporal separation compounds these effects because the compared specimens often represent different stages of disease rather than contemporaneous samples.

Molecular differences across primary, extracranial, brain and leptomeningeal compartments should therefore be interpreted as compartment-associated phenotypes, not automatically as compartment-driven evolution. ER/PR loss, HER2 gain or loss, alterations in PI3K–AKT–PTEN signalling and ESR1 mutations all demonstrate that clinically relevant biomarkers can change during CNS progression, but none is uniquely attributable to the CNS niche on the basis of current evidence. Their distribution reflects an unresolved combination of pre-existing spatial heterogeneity, clonal selection, elapsed time, systemic treatment and local environmental pressures. Establishing genuine niche-specific evolution will require longitudinal, multiregion sampling of primary, extracranial and CNS disease with harmonized molecular assays.

CNS Sampling Reveals Evolving Disease States

Understanding CNS evolution requires more than comparing an archival primary tumour with a single late metastatic specimen. Surgically resected BM remain the most direct source for interrogating tumour genotype, transcriptional state, spatial architecture and the surrounding neuroimmune microenvironment. Recent single-cell profiling of human breast cancer BM has resolved substantial inter-patient heterogeneity among metastatic tumour and myeloid populations, while paired genomic analyses can identify alterations that are shared with or divergent from the primary tumour.77 However, surgical tissue represents a spatially restricted snapshot, is available only from selected resectable lesions and is rarely amenable to repeated sampling. Tissue therefore provides the highest-resolution view of an individual parenchymal lesion, but by itself cannot readily reconstruct how intracranial disease changes during treatment or across multiple CNS sites.

CSF provides a complementary window into CNS disease because it can be sampled repeatedly and contains both intact tumour cells and tumour-derived molecular material. Conventional cytology remains important for demonstrating malignant cells in LM, but its binary or semiquantitative readout captures relatively little of the molecular state of those cells. CSF CTCs extend this information by allowing enumeration and phenotypic characterization. In a prospective breast cancer study, Darlix and colleagues detected CSF CTCs in all patients with positive cytology and also in a subset with negative cytology; notably, HER2-positive CTCs were identified in the CSF of some patients whose breast cancers had been classified as HER2-negative.78 Such findings illustrate that CSF-derived cells can reveal phenotypic states not apparent from archival extracranial tissue. More importantly for evolutionary monitoring, serial CTC measurements can in principle track changes in tumour burden and cellular phenotype rather than merely establish the presence of LM.

CSF circulating tumour DNA (ctDNA) adds a genomic dimension to this longitudinal view. In breast cancer LM, Fitzpatrick and colleagues showed that CSF ctDNA could be quantified serially during treatment and that rising ctDNA fractions could precede clinically recognized progression, whereas sustained suppression was associated with more favourable outcomes.79 Because ctDNA can encode somatic mutations, copy-number alterations and changing clone fractions, serial CSF analysis offers a means of observing genomic evolution without repeated neurosurgical biopsy. The potential compartment specificity of this approach is illustrated by a HER2-positive breast cancer case in which CSF and plasma ctDNA moved in opposite directions during therapy: extracranial molecular markers declined in plasma while TP53-, PIK3CA- and ERBB2-associated signals increased in CSF, paralleling intracranial progression.80 Although a single-patient observation cannot establish general performance, it demonstrates how parallel plasma and CSF sampling can separate systemic from CNS evolutionary trajectories.

The informational value of CSF increases further when cellular and acellular analytes are integrated. A breast cancer-specific study combined single-cell RNA sequencing of brain metastatic tissue, adjacent tissue, CSF, and circulating cells with genomic profiling of matched primary and brain-metastatic specimens, revealing marked differences in the cellular composition and transcriptional states captured across sampling compartments.77 In another breast cancer pilot study, single-cell genomic analysis and multiplexed proteomics of CSF CTCs identified clonally related but phenotypically heterogeneous tumour cells, while copy-number profiles derived from CTCs and CSF cell-free DNA were not completely concordant.81 This lack of equivalence is biologically informative: CTCs, ctDNA, immune cells and soluble proteins capture different layers of CNS disease rather than serving as interchangeable surrogates.

Longitudinal multi-omic sampling provides an early indication of what a more dynamic CNS monitoring strategy could achieve. In a patient with refractory HER2-positive breast cancer LM, serial CSF samples obtained through an Ommaya reservoir were analysed using cytopathology, CTC enumeration, single-cell RNA sequencing, and cytokine profiling before and after treatment modification.63 Changes in tumour-cell abundance, HER2-expressing populations, immune-cell composition and inflammatory mediators were observed over time, demonstrating that treatment response can involve coordinated shifts across several biological layers. The single-patient design precludes generalization, but the study illustrates an important principle: CNS evolution is unlikely to be captured fully by any single genomic or cellular readout.

This principle is especially relevant when distinguishing LM from parenchymal brain metastasis. Because LM directly involves the CSF-containing compartment, CSF CTCs and ctDNA are particularly well positioned to sample leptomeningeal tumour populations. The same assumption cannot be made for all parenchymal lesions. Breast cancer-focused reviews and CNS consensus analyses emphasize that CSF ctDNA detection from parenchymal metastases depends on tumour burden, anatomical proximity to CSF spaces and sampling location; conversely, parenchymal lesions themselves may shed DNA into CSF, meaning that CSF can represent a composite signal when brain and leptomeningeal disease coexist.82 Thus, CSF should not be treated as a molecular census of every intracranial metastasis. Direct tissue remains indispensable for spatially resolved analysis of accessible brain lesions, whereas CSF is most informative for LM and for selected parenchymal disease with sufficient molecular shedding.

CNS sampling should therefore be viewed as a longitudinal, multi-compartment strategy for observing evolving disease states rather than as a competition between diagnostic technologies. Resected brain tissue provides spatial and molecular depth; cytology and CTCs reveal cellular presence and phenotype; ctDNA tracks genomic composition; and single-cell, proteomic and multi-omic approaches resolve cellular states and tumour–host interactions. Integrating these sources across time offers the clearest route to distinguishing persistent ancestral features from newly emerging CNS states, while preserving the critical distinction that CSF most directly represents leptomeningeal disease and cannot be assumed to capture the full heterogeneity of parenchymal BM.

Niche Context Influences Therapeutic Vulnerability

Drug Access Alone Does Not Explain CNS Efficacy

Systemic therapy must reach an intracranial lesion before it can exert a direct antitumour effect, but drug access is only one component of CNS efficacy. For parenchymal BM, the relevant vascular interface is no longer a uniformly intact BBB, but a spatially heterogeneous BTB whose permeability varies between lesions, within individual lesions and over time. Across five experimental breast cancer brain-metastasis models, passive permeability differed markedly among metastases and showed no consistent relationship with lesion size.83 Human pharmacokinetic data reinforce this heterogeneity. In patients receiving capecitabine or lapatinib immediately before surgical resection, both agents reached breast cancer BM, but intratumoural concentrations varied substantially between patients and, for lapatinib, relative tumour-to-serum exposure was highly variable.84 Thus, radiographic evidence of barrier disruption—or even measurable drug within a resected lesion—does not imply uniform pharmacologically effective exposure throughout intracranial disease.

Conversely, limited passive permeability does not fully predict therapeutic failure. Drug properties determine not only whether a compound reaches a lesion, but also how it is transported, retained and activated once there. This distinction has become particularly relevant with antibody–drug conjugates (ADCs), whose molecular size would traditionally be considered unfavourable for CNS delivery. In HER2-positive breast cancer brain-metastasis models, Khan and colleagues recently showed that trastuzumab deruxtecan (T-DXd) could cross experimental BTB systems partly through FcRn-associated endothelial transcytosis, although intratumoural ADC distribution remained heterogeneous. Importantly, antitumour effects extended beyond regions with the highest detectable ADC accumulation, consistent with contributions from payload release and bystander activity, whereas residual lesions in one model showed reduced HER2 expression.85 These findings illustrate why “penetration” should not be treated as a single pharmacokinetic variable: mode of transport, intralesional distribution, target-dependent internalization, payload properties and bystander effects can all modify the relationship between tissue concentration and tumour response.

Target availability is therefore as important as physical drug access. The strong intracranial activity of T-DXd in HER2-positive disease provides a clinically relevant example. In DESTINY-Breast12, which prospectively enrolled 263 patients with stable or active BM, T-DXd produced substantial and durable intracranial activity despite being a large antibody-based therapeutic.86 This does not demonstrate that the BBB or BTB is unimportant; rather, it shows that molecular size or presumed brain permeability alone cannot predict CNS efficacy when a highly active drug is matched to an expressed tumour target and possesses favourable pharmacodynamic properties. At the same time, HER2 expression itself can vary across metastatic compartments and within intracranial disease, as discussed above. A drug may therefore reach two lesions to a similar extent yet produce different responses if target abundance, target accessibility or downstream signalling differs between them.

The biological state of the metastatic cell adds another layer of variability. Drugs that reach a lesion at adequate concentrations still encounter tumour cells with different proliferative, survival and stress-response programmes. Experimental breast cancer models provide direct evidence that the brain microenvironment can modify this response. Kim and colleagues showed that direct interaction with astrocytes induced AKT/MAPK signalling and upregulated survival-associated genes in metastatic breast cancer cells, thereby reducing susceptibility to several cytotoxic agents; importantly, the protective phenotype depended on continued tumour–astrocyte contact and was reversible.87 Tumour state itself can also influence treatment sensitivity. In a three-dimensional breast cancer brain-metastasis model, cells maintained in a matrix environment associated with a non-proliferative, dormant-like state were more resistant to paclitaxel and lapatinib than proliferating cells, with p38–SGK1 signalling contributing to this phenotype.88 These observations remain predominantly preclinical, but they demonstrate an important principle: adequate exposure is not equivalent to cellular susceptibility.

Treatment history further complicates this relationship. A CNS lesion sampled after endocrine therapy, HER2-directed therapy, chemotherapy or local radiation is not biologically equivalent to the same lesion before treatment. Previous systemic therapies can select resistant clones or alter target expression, while local treatment can modify vascular integrity, inflammatory states and tumour composition. Temporal evolution therefore changes both sides of the therapeutic equation: the lesion that a drug must reach and the tumour population that the drug encounters. This is particularly relevant when comparing untreated, previously treated but stable, and actively progressing BM. Differences in clinical response among these states should not automatically be attributed to drug penetration, because treatment history is entangled with clonal selection, target evolution, lesion architecture and local microenvironmental adaptation.

The same conceptual distinction applies to niche support. Astrocytes, microglia, macrophages, extracellular matrix and vascular cells can provide survival signals that alter the concentration of drug required for effective killing or activate bypass pathways despite successful target inhibition. Consequently, two metastases with comparable drug exposure and nominally identical receptor status may still respond differently because their cellular states and host interactions differ. This also explains why pharmacokinetic optimisation alone is unlikely to eliminate CNS resistance: improved delivery can overcome insufficient exposure, but it cannot by itself reverse tumour-intrinsic resistance or microenvironment-mediated protection.

CNS efficacy should therefore be understood as the product of drug exposure, lesion-specific permeability, target expression, tumour state, niche-dependent survival and prior therapeutic selection, rather than as a direct readout of BBB penetration. BBB or BTB access remains indispensable when inadequate exposure is the limiting factor, but it is neither uniformly limiting across metastases nor sufficient to guarantee response once a drug reaches the tumour. The translational question is consequently not simply whether an agent “crosses the BBB”, but whether pharmacologically active drug reaches the relevant CNS compartment, engages the appropriate target, and remains effective against the biological state of the tumour within its local niche.

Brain Metastasis Care Integrates Local and Subtype-Specific Systemic Therapy

Management of breast cancer BM increasingly requires coordinated use of local and systemic therapy rather than treating the intracranial compartment in isolation. Local treatment remains particularly important for symptomatic disease, lesions causing mass effect, and situations in which rapid intracranial control is required. Surgical resection is most relevant for large or symptomatic accessible lesions and can simultaneously provide tissue for contemporary molecular characterization. For patients with limited intracranial disease, stereotactic radiosurgery (SRS) has become the preferred focal radiation strategy in many settings, including treatment of suitable postoperative cavities, because it achieves local control while limiting exposure of uninvolved brain tissue. Whole-brain radiotherapy (WBRT) remains appropriate when intracranial disease is too diffuse or extensive for focal approaches; when WBRT is indicated in patients with a favourable prognosis, hippocampal avoidance and memantine should be considered to reduce neurocognitive toxicity.89,90

The emergence of CNS-active systemic therapy has modified, but not eliminated, the role of local treatment. In selected patients with small, asymptomatic BM and a systemic therapy with established intracranial activity, initial systemic treatment with close MRI surveillance may permit local therapy to be deferred. In contrast, symptomatic lesions or those threatening neurological function generally still require prompt surgery or radiotherapy irrespective of the systemic regimen.89 The relevant clinical question has therefore shifted from whether local or systemic therapy is preferable to how they should be sequenced according to symptoms, lesion size and number, extracranial disease status, prior treatment and the expected intracranial activity of the available systemic therapy.

The strongest evidence for this integrated strategy currently comes from HER2-positive breast cancer. HER2CLIMB provided a practice-changing demonstration that systemic therapy can improve clinically meaningful CNS outcomes in patients with established BM. Among 291 patients with baseline BM, including both active and stable disease, addition of tucatinib to trastuzumab and capecitabine reduced the risk of intracranial progression or death by 68%, prolonged median CNS progression-free survival from 4.2 to 9.9 months, and improved overall survival. Among patients with measurable intracranial disease, the confirmed intracranial response rate was 47.3% with tucatinib versus 20.0% with the control regimen.91 Importantly, HER2CLIMB included patients with untreated or progressing BM who had often been excluded from earlier registration trials, establishing randomized evidence that a subtype-directed systemic regimen can directly alter the course of active intracranial disease.

T-DXd has further expanded the systemic treatment landscape for HER2-positive BM. Early prospective evidence from TUXEDO-1 demonstrated a high intracranial response rate in a small cohort with untreated or progressing BM, and the much larger DESTINY-Breast12 study subsequently evaluated 263 patients with stable or active BM.86,92 In DESTINY-Breast12, the 12-month CNS progression-free survival rate was 58.9%, and the CNS objective response rate was 71.7% overall, including 62.3% in patients with active BM.86 Because DESTINY-Breast12 was a single-arm study, these results should not be interpreted as a direct comparison with tucatinib-based therapy. Rather, tucatinib and T-DXd now provide two well-supported HER2-directed strategies whose sequencing depends on previous treatment, intracranial disease status, extracranial disease burden, toxicity considerations and whether immediate local control is required.

This therapeutic success should not, however, be generalized across breast cancer subtypes. In HR-positive/HER2-negative BM, endocrine therapy and molecularly targeted agents remain important for systemic disease, but prospective CNS-specific evidence is considerably less mature. Abemaciclib provides an instructive example: despite achieving pharmacologically relevant concentrations in resected brain-metastasis tissue, a dedicated Phase II study reported an intracranial objective response rate of only 5.2% in the HR-positive/HER2-negative cohort, although a larger proportion achieved intracranial clinical benefit.93 In TNBC, sacituzumab govitecan is an established systemic therapy, but the ASCENT trial enrolled only patients with stable BM, and this small subgroup showed substantially less convincing benefit than the extracranial trial population.94 Taken together, the evidence indicates subtype dependence at distinct biological and therapeutic levels. TNBC and HER2-positive BM can exhibit different patterns of early tumour–brain interaction in selected experimental models, whereas HER2-positive BM has the most mature prospective evidence supporting systemic intracranial treatment. Comparable mechanistic and treatment-response evidence for HR-positive/HER2-negative and TNBC CNS disease, particularly LM, remains limited. Thus, molecular subtype should inform interpretation of CNS-specific evidence without being treated as proof of a uniform compartment-specific mechanism or therapeutic response.

Current care should therefore be understood as subtype-specific systemic therapy superimposed on individualized local CNS control. Surgery and SRS remain central for lesions in which anatomical burden or neurological risk requires focal intervention, while WBRT retains a selective role in more extensive intracranial disease. At the same time, tucatinib-based therapy and T-DXd have made systemic treatment an increasingly important component of intracranial control in HER2-positive disease.86,91 International advanced-breast-cancer consensus recommendations similarly emphasize integrating local management with tumour subtype and available CNS-active systemic therapy.95 Crucially, however, the most mature systemic CNS evidence in breast cancer remains disproportionately concentrated in HER2-positive disease. The marked progress achieved in this subgroup should not be interpreted as evidence that breast cancer BM as a whole have become uniformly controllable; substantial therapeutic gaps remain for HR-positive/HER2-negative and triple-negative CNS disease.

Leptomeningeal Disease Requires Compartment-Specific Treatment

LM cannot simply be treated according to the therapeutic logic developed for parenchymal BM. BM usually form discrete lesions for which surgery or stereotactic radiation can provide durable local control, whereas LM extends along CSF-containing surfaces throughout the craniospinal axis and may coexist as free-floating tumour cells, linear deposits and nodular disease. Treatment must therefore address not only tumour subtype and systemic disease, but also CSF exposure, neuraxial distribution, disease geometry and CSF flow. Systemic therapy remains important because LM commonly coexists with extracranial and parenchymal CNS disease and because CTCs can continue to reseed the leptomeningeal compartment. However, a systemically active drug is not automatically an effective LM therapy: its activity also depends on access to the blood–CSF and meningeal compartments, target expression and prior treatment history.96

The importance of compartmental pharmacokinetics is illustrated most clearly in HER2-positive disease. The TBCRC049 phase II study prospectively evaluated tucatinib, trastuzumab, and capecitabine in 17 women with newly diagnosed HER2-positive LM. Tucatinib and its active metabolite were measurable in CSF at pharmacologically relevant concentrations, while the regimen produced a composite LM response in 5 of 13 evaluable patients and neurological improvement in 7 of 12; median overall survival was 10 months.97 These findings provide direct evidence that a systemic regimen can achieve clinically meaningful activity within the leptomeningeal compartment. At the same time, the small, single-arm study should not be interpreted as showing that CSF penetration alone explains efficacy. CSF concentration measures drug exposure in the fluid compartment, but may not fully represent delivery into adherent plaques or nodular deposits, and pharmacological exposure must still coincide with tumour sensitivity and adequate target expression.

Direct intra-CSF therapy addresses the same problem from the opposite direction by bypassing systemic barrier limitations and generating high local drug concentrations. Its utility, however, depends strongly on the physical form of LM. Intra-CSF drugs are best positioned to encounter floating tumour cells and diffuse linear disease, whereas penetration from CSF into larger nodules is limited and abnormal CSF flow can prevent homogeneous drug distribution; intra-CSF treatment is therefore inappropriate in the presence of significant CSF-flow obstruction.11 Breast cancer-specific clinical evidence supports this strategy without establishing it as universally beneficial. In the randomized DEPOSEIN study, adding intra-CSF liposomal cytarabine to systemic therapy prolonged LM-related progression-free survival from 2.2 to 3.8 months, but did not significantly improve overall survival.98 More specifically, a phase II study of weekly intrathecal trastuzumab in HER2-positive breast cancer LM found that 74% of 19 patients remained free of neurological progression at 8 weeks, with a median LM-related progression-free survival of 5.9 months and overall survival of 7.9 months.99 These studies support intra-CSF treatment as a compartment-directed component of multimodal therapy, rather than a substitute for systemic disease control.

Radiotherapy likewise serves a different function in LM than in focal brain metastasis. Focal cranial or spinal irradiation is primarily used for symptomatic, bulky or nodular sites and can also be considered when local disease contributes to CSF-flow obstruction; it does not treat tumour distributed elsewhere along the neuraxis. Breast cancer-specific radiotherapy guidance therefore recommends integrating focal radiation with systemic treatment rather than assuming that regional irradiation alone can control disseminated LM.100 Craniospinal irradiation addresses this anatomical limitation by treating the entire neuraxis. In the final analysis of a randomized phase II trial enrolling patients with breast cancer or non-small-cell lung cancer LM, proton craniospinal irradiation improved median CNS progression-free survival to 8.2 months compared with 2.3 months with photon involved-field radiotherapy and improved median overall survival from 4.9 to 11.3 months.101 Because the randomized population combined two primary tumour types, these effect estimates should not be interpreted as breast cancer-specific, and patient fitness, extracranial disease, marrow reserve and access to proton therapy remain important constraints.

The development of HER2-directed therapies is also beginning to blur the historical distinction between systemic and compartment-directed treatment. T-DXd provides the clearest emerging antibody–drug conjugate example. Cohort 5 of the phase II DEBBRAH trial enrolled seven patients with pathologically confirmed LM from HER2-positive or HER2-low advanced breast cancer. Median progression-free and overall survival were 8.9 and 13.3 months, respectively, and five of seven patients achieved disease stabilization lasting at least 24 weeks.102 These results are encouraging because they demonstrate that an intravenously administered ADC can have activity in clinically established LM, but the cohort is far too small to define a new general treatment standard or to determine whether HER2-positive and HER2-low LM derive equivalent benefit. Moreover, the study did not establish that a particular level of free ADC in CSF is responsible for response. Target expression, lesion permeability, payload release and disease morphology may all influence activity.

LM treatment should therefore be conceptualized as matching therapeutic route and spatial coverage to the biology and geometry of the involved CNS compartment. Systemic therapy can simultaneously control extracranial disease, parenchymal metastases and leptomeningeal populations when sufficient CNS activity is present; intra-CSF therapy directly targets tumour cells exposed to CSF but is less suited to bulky nodular disease or disrupted CSF circulation; and radiotherapy is used either to control selected symptomatic sites or, in carefully selected patients, to encompass the broader neuraxis. As with BM, the strongest breast cancer-specific therapeutic evidence remains concentrated in HER2-driven disease, but the LM evidence base is smaller and consists largely of modest prospective cohorts. The success of HER2-directed treatment should therefore not be generalized into a single treatment algorithm for breast cancer LM as a whole.

Treatment Imposes New Selective Pressures

Therapy does not simply reduce CNS tumour burden; it also changes the biological conditions under which residual disease must persist. Radiotherapy can eliminate radiosensitive cells within treated lesions and produce a local population bottleneck, while systemic therapies preferentially suppress tumour populations that remain dependent on their therapeutic targets. However, direct lineage-based evidence that radiotherapy selects defined resistant clones in human breast cancer BM remains limited. Clinically, local control of visible disease clearly does not eliminate the capacity for subsequent intracranial relapse. In a multicentre series of 538 patients treated for breast cancer BM, 37.4% subsequently developed new BM, with recurrence patterns differing by molecular subtype and being particularly frequent in triple-negative disease.103 Such recurrence may reflect survival of pre-existing microscopic disease, continued reseeding, differential systemic drug exposure or treatment-resistant tumour states; it should not automatically be interpreted as de novo radiation-induced clonal evolution.

Systemic treatment creates a parallel selective pressure. This is particularly relevant to ADCs, whose efficacy depends on target abundance, internalization, linker–payload properties and cellular sensitivity to the released cytotoxic agent. Recent paired-sample analyses of metastatic breast cancer before and after T-DXd provide direct evidence that treatment can reshape target availability: decreased HER2 expression was observed in almost half of post-treatment samples at progression, with complete HER2 loss in a substantial subset, while ERBB2 mutations affecting trastuzumab binding could also confer resistance.104 These findings establish target loss and altered target binding as mechanisms of T-DXd resistance, but they are not CNS-specific and therefore cannot be assumed to explain intracranial progression in every patient. In the CNS, relapse after an ADC may similarly reflect target evolution, payload resistance, heterogeneous drug exposure or survival of pre-existing low-target populations; distinguishing among these possibilities will require paired pre- and post-treatment CNS sampling.

Treatment sequencing adds another layer of selection because local and systemic therapies expose different tumour populations to different pressures at different times. SRS may eradicate one or several macroscopic lesions while leaving untreated microscopic deposits elsewhere in the brain, whereas an effective systemic therapy may control extracranial disease more completely than intracranial disease—or, with newer CNS-active agents, suppress both compartments to different degrees. Consequently, a later CNS relapse is not necessarily the product of a newly evolved “brain-resistant clone”. It may represent outgrowth of a previously occult population that was never exposed to equivalent drug concentrations, a pre-existing resistant clone that survived therapy, or a tumour state that changed during treatment. The sequence and timing of surgery, radiotherapy and systemic therapy therefore influence not only immediate disease control but also which tumour populations remain available for subsequent expansion.

The interaction between radiotherapy and ADCs illustrates how treatment pressure can also reshape the host compartment and generate new toxicity constraints. In a retrospective cohort of patients receiving stereotactic radiotherapy with T-DM1, T-DXd or sacituzumab govitecan, concurrent ADC exposure was associated with a higher risk of symptomatic radiation necrosis, particularly in previously irradiated or larger lesions.105 Breast cancer-specific data with T-DM1 support this concern: a multicentre study found that T-DM1 exposure within one year of SRS was independently associated with increased risks of both overall and symptomatic radionecrosis.106 These observations raise an important translational issue: combining two individually effective CNS treatments may alter normal-tissue tolerance, making the timing of therapy itself biologically and clinically relevant.

Importantly, the ADC–radiotherapy literature is not uniform. In a multicentre series of 34 patients receiving T-DXd and SRS, 12-month local control was 97%, while the cumulative incidence of symptomatic radiation necrosis was 2.1% per lesion at 24 months, with no clear excess risk attributed to concurrent treatment.107 A larger retrospective analysis likewise reported no radionecrosis events among 29 patients treated with T-DXd and found favourable intracranial control compared with patients receiving other anti-HER2 therapies.108 In contrast, a systematic review and meta-analysis of seven retrospective studies involving 444 patients estimated a pooled symptomatic radionecrosis rate of 20% among ADC recipients and found higher odds of radionecrosis with ADC exposure, particularly when treatment was concurrent with radiotherapy; notably, more recent studies reported lower toxicity rates.109 Several methodological differences may explain these apparently conflicting findings. The studies evaluated different ADC agents, including T-DM1 and T-DXd, which should not automatically be assumed to have identical interactions with radiation. Definitions of concurrent exposure, prior irradiation, lesion volume, radiation dose and fractionation, and follow-up duration may also differ substantially. Furthermore, symptomatic radionecrosis rates reported per lesion are not directly interchangeable with patient-level estimates, while differences in imaging criteria and clinical adjudication may affect event ascertainment. Importantly, the absence of radionecrosis events in a small retrospective cohort does not exclude a clinically relevant late risk, just as elevated pooled estimates from observational studies cannot establish a causal effect of concurrent ADC administration. These limitations highlight the need for agent-specific, adequately powered prospective investigations with standardized exposure windows, lesion-level dosimetry, consistent radionecrosis definitions, and sufficient follow-up before optimal treatment sequencing can be established.

Radiation necrosis itself further complicates subsequent treatment because it represents a treatment-altered CNS state rather than tumour progression alone. Vascular injury, tissue inflammation and delayed necrosis can mimic recurrent metastasis radiographically, require corticosteroids, bevacizumab or surgery, and influence whether additional radiotherapy or systemic therapy can be delivered. Although it is biologically plausible that treatment-induced changes in the neurovascular and immune microenvironment could modify the fitness of residual tumour cells, direct evidence that radionecrosis or post-radiation inflammation drives specific resistant clones in breast cancer remains insufficient. This distinction is important: treatment changes both tumour and host, but not every post-treatment change should be interpreted as tumour evolution.

Therapy therefore represents the next selective environment encountered by an already adapted CNS metastasis. Local treatment changes lesion architecture and surrounding tissue, systemic therapy selects among molecularly heterogeneous tumour populations, and the order in which these interventions are delivered determines which populations and niches remain under pressure. Subsequent CNS relapse can then arise from residual, previously occult or newly resistant disease. In this sense, treatment does not terminate the biological process described throughout this review; it feeds back into it. CNS metastatic progression is best viewed as a continuing cycle in which niche adaptation and tumour evolution shape therapeutic response, while therapy itself creates new selective conditions that influence the next state of disease.

Key Knowledge Gaps Define a Testable Research Agenda

The Determinants of Parenchymal and Leptomeningeal Tropism Remain Unresolved

A central unresolved question is why a disseminated breast cancer clone establishes within the brain parenchyma in some circumstances but the leptomeningeal compartment in others. Clinical coexistence of BM and LM does not establish a common route or obligatory sequence. In the German BM in Breast Cancer Registry, 859 patients developed LM, yet 42.5% had LM without documented parenchymal BM, indicating that clinically apparent BM is not a prerequisite for leptomeningeal disease.110 Experimental evidence likewise suggests that the two destinations can select distinct metastatic phenotypes. Breast cancer populations serially selected from the same parental lines for leptomeningeal or parenchymal growth acquired contrasting CNS tropism and transcriptional states after haematogenous dissemination.58 However, such selection models cannot determine whether compartment preference was encoded in rare pre-existing clones, acquired during dissemination or reinforced only after arrival, because iterative selection itself can enrich phenotypes that may not reflect natural disease evolution.

Patient phylogenetics has begun to narrow these possibilities but has not resolved them. Genomic reconstruction of breast cancer LM indicates that leptomeningeal and extracranial metastases can share a common ancestral subclone while subsequently following divergent evolutionary trajectories.30 In the limited setting in which temporally separated BM and LM material was available from the same patient, the two CNS lesions shared ancestral clones but subsequently diverged, demonstrating relatedness without establishing the anatomical direction of spread. A decisive clinical strategy would therefore require matched, longitudinal sampling across compartments: primary breast tumour, extracranial metastases, resected BM and CSF- or tissue-derived LM collected from the same patients, preferably at closely matched time points and before and after major treatment transitions. Somatic mutations could reconstruct clonal ancestry, while single-cell transcriptomic, epigenomic and spatial analyses could determine whether compartment-associated states pre-exist dissemination or emerge after colonization. Treatment history and sampling interval must be incorporated explicitly, because otherwise therapy-induced selection and temporal evolution can be mistaken for intrinsic CNS tropism.

Functional resolution will require lineage tracing in models that preserve the natural metastatic sequence. Genetic barcoding of primary breast cancer cells followed by longitudinal recovery of the same clones from CTCs, BM, meninges and CSF would allow direct comparison of parenchymal and leptomeningeal fates within the same host. The feasibility of this strategy has recently been demonstrated outside the CNS: single-cell barcoding combined with transcriptomics traced breast cancer clones from the primary tumour through dissemination and showed that cells sharing the same lineage can adopt different transcriptional states in different metastatic organs.111 Applying an analogous design to CNS metastasis, ideally in orthotopic models that develop spontaneous dissemination, would distinguish true compartmental tropism from simple differences in post-entry survival. Existing brain-metastasis model collections provide a useful starting point, but many models were generated by repeated in vivo selection and differ substantially in inoculation route and metastatic behaviour.112 Intracardiac models can isolate later haematogenous steps and intra-CSF models can interrogate leptomeningeal survival, but neither alone can establish how an untreated primary clone naturally chooses between the two CNS compartments.

These approaches also generate testable predictions. If CNS tropism is largely pre-existing, BM- or LM-biased lineages and associated molecular states should be detectable in the primary tumour before dissemination and should reproducibly favour the same compartment across independent hosts. If tropism is predominantly niche-induced, the same barcode should be capable of entering both compartments but subsequently acquire different molecular states according to local conditions. If LM arises secondarily from an established BM in a given patient, longitudinal phylogenies should place the LM lineage within the evolutionary branch of the preceding BM; independent CNS seeding should instead produce separate sister branches descending from a shared ancestor. Candidate mechanisms should therefore be validated not merely by reducing total CNS metastasis, but by demonstrating a reproducible shift in the relative probability of parenchymal versus leptomeningeal colonization. Resolving this distinction would transform CNS tropism from a descriptive concept into a directly testable biological property.

The Routes Linking Brain and Leptomeningeal Disease Remain Incompletely Defined

The evolutionary relationship between parenchymal BM and LM remains unresolved. At least four scenarios are biologically plausible: the two compartments may be seeded independently by different metastatic clones; they may arise as sister branches from a common metastatic ancestor; an established BM may subsequently seed the leptomeninges; or therapy may preferentially permit expansion of an LM-capable resistant population. These possibilities are not mutually exclusive and may occur in different patients. Importantly, temporal order alone cannot distinguish them. An LM diagnosed after a BM does not establish that the BM was its source, just as the absence of a clinically detectable BM does not prove direct independent seeding of the leptomeninges.

The limited phylogenetic evidence available in breast cancer illustrates this uncertainty. Fitzpatrick and colleagues reconstructed breast cancer LM evolution using matched primary tumour, extracranial tissue, plasma and CSF DNA and found predominantly monophyletic metastatic seeding, although polyphyletic dissemination also occurred. In the particularly informative patient for whom an archived BM resected three years before LM was available, BM and LM shared two ancestral clones but each contained distinct subsequent subclones.30 This establishes common ancestry and later divergence, but not the direction of dissemination: the LM could have arisen from an unsampled common precursor rather than directly from the earlier BM. More recent CSF sequencing similarly identified LM-restricted alterations in only a subset of patients while showing substantial retention of mutations present in primary or systemic disease.68 Thus, molecular similarity alone cannot distinguish direct BM-to-LM spread from parallel evolution of clonally related lesions.

There is nevertheless clinical evidence that established intracranial disease can contribute to secondary leptomeningeal dissemination under some conditions. Breast cancer-specific series have reported LM after BM resection, with piecemeal surgery and selected tumour characteristics associated with increased risk.28 A multicentre study of 2,673 patients undergoing BM resection likewise found that incomplete resection was independently associated with subsequent LM and that breast cancer was among the primary tumour types associated with increased risk.29 These observations support a route from established intracranial disease to the leptomeningeal compartment, particularly in a treatment-altered setting, but they cannot establish that spontaneous LM generally develops from BM. Surgical manipulation, microscopic meningeal disease already present at the time of surgery and continued systemic dissemination remain competing explanations.

Treatment-selected LM populations are even more difficult to establish. Serial CSF tumour-cell profiling provides evidence that leptomeningeal phenotypes can change during therapy: in a retrospective breast cancer cohort with repeated CSF testing, HER2 amplification status changed over time in 4 of 14 longitudinally evaluable patients.113 These observations are compatible with changing clonal composition under treatment pressure, but repeated FISH measurements cannot by themselves distinguish true clonal selection from spatial sampling, fluctuations in tumour-cell abundance or technical variability. Moreover, existing genomic data provide an important counterpoint to a simple treatment-selection model. In the Fitzpatrick cohort, the proportion of CSF-specific variants was not significantly associated with the number of previous treatment lines, and chemotherapy-associated mutational signatures were more prominent in plasma than in CSF.30 Thus, therapy almost certainly contributes to metastatic evolution, but current evidence does not demonstrate that LM usually represents a treatment-selected descendant of an established BM.

Resolving these alternatives will require phylogenetic reconstruction coupled to longitudinal, multi-compartment sampling. The most informative design would collect the primary tumour, temporally matched extracranial metastases, individual resected BM lesions and serial CSF tumour cells or ctDNA before LM diagnosis, at LM emergence and during subsequent therapy. Deep genomic profiling could then determine whether an LM lineage is nested within a pre-existing BM branch, consistent with secondary dissemination, or instead forms a separate sister branch arising from an earlier common ancestor. Serial CSF sampling can additionally determine whether particular subclones expand only after therapeutic exposure; existing breast cancer LM studies have already shown that CSF ctDNA can be followed longitudinally during treatment, providing a practical foundation for such analyses.79 Whenever possible, single-cell DNA/RNA profiling and spatial sampling of multiple CNS lesions should accompany bulk sequencing to reduce the risk of mistaking unsampled intratumour heterogeneity for true evolutionary divergence.

The key objective is therefore not simply to establish whether BM and LM are molecularly related, but to determine the direction, timing and treatment context of that relationship. A BM-derived LM should retain defining private alterations from the antecedent BM and appear phylogenetically downstream of it; independent seeding should generate distinct CNS branches from an earlier ancestor; and a treatment-selected LM population should show longitudinal expansion of a pre-existing or newly detectable resistant lineage after exposure. Until datasets containing this level of temporal and anatomical resolution become available, BM-to-LM progression should remain one possible route among several rather than the default model of breast cancer CNS dissemination.

Longitudinal Biomarkers Must Capture CNS Evolution

The next challenge for CNS biomarkers is not simply to detect metastatic disease, but to determine whether they can track its evolution with sufficient temporal and compartmental resolution to change clinical management. This is particularly relevant to LM, where a single cytological or molecular result provides only a snapshot of a disease that can change rapidly during treatment. Breast cancer-specific proof-of-concept data are encouraging: serial CSF ctDNA measurements in patients with breast cancer LM showed that molecular suppression during therapy was associated with longer survival, whereas rising ctDNA could be detected up to 12 weeks before clinically recognized progression.79 However, longitudinal evidence remains sparse. A systematic review of solid-tumour LM identified 14 studies involving 963 patients with baseline CSF ctDNA data, but only two studies comprising 26 patients evaluated longitudinal ctDNA kinetics.114 The priority is therefore prospective validation of serial molecular dynamics rather than further demonstration that tumour DNA can be detected in CSF.

Future studies should collect CSF and plasma concurrently at predefined longitudinal time points. Plasma predominantly reflects systemic disease, whereas CSF may provide a more direct readout of CNS disease, particularly LM. This distinction is illustrated by a HER2-positive breast cancer case in which systemic tumour-associated signals decreased in plasma during treatment while TP53, PIK3CA and ERBB2-associated signals increased in CSF in parallel with intracranial progression.80 Although a single case cannot establish general performance, it provides a testable framework: paired plasma–CSF profiling should determine whether apparently discordant systemic and CNS responses reflect genuinely different evolutionary trajectories rather than assay variability. Future cohorts should therefore distinguish LM-only, parenchymal BM-only and mixed CNS disease and record extracranial response concurrently.

Molecular monitoring also needs to be anchored to serial imaging and neurological assessment rather than developed as an independent laboratory endpoint. CSF should ideally be obtained close to standardized contrast-enhanced MRI examinations and clinical evaluations, allowing prospective assessment of molecular response, molecular progression and lead time relative to conventional endpoints. A prospective NSCLC brain-metastasis study provides a useful methodological precedent: paired CSF and plasma were collected at baseline, after eight weeks of therapy and at progression, and reductions in CSF ctDNA correlated more closely with intracranial radiographic response and intracranial progression-free survival than corresponding plasma changes.115 Because this evidence comes from lung cancer, the quantitative thresholds cannot be transferred directly to breast cancer, but the study design is highly relevant. Similarly, a recent pan-cancer CSF study found that falling variant allele frequencies generally accompanied clinical stabilization, whereas persistent or rising levels were associated with progression or emerging resistance.116 These observations require prospective breast cancer-specific replication using harmonized assays and predefined response criteria.

Longitudinal sampling should also be designed to identify what changes during treatment, rather than merely whether total tumour burden rises or falls. Serial sequencing can determine whether established driver alterations contract, resistant subclones expand or new compartment-associated alterations become detectable CSF CTCs provide an additional cellular layer, permitting changes in receptor expression and transcriptional state to be studied alongside ctDNA. A longitudinal analysis of HER2-positive breast cancer LM combined serial CTC enumeration, single-cell RNA sequencing, and cytokine profiling, documenting coordinated changes in tumour and immune-cell populations during treatment alongside MRI evidence of response.63 Because this study involved a single patient, it demonstrates feasibility rather than validation; future studies should integrate ctDNA, CTCs and selected proteomic or single-cell measurements in larger longitudinal cohorts to determine which signals consistently anticipate resistance.

This framework yields a clear prospective prediction. If serial CSF molecular profiling truly reflects LM evolution, reproducible increases in tumour fraction, variant allele frequencies or resistant subclones should precede or coincide with radiographic and neurological progression, whereas sustained molecular suppression should accompany durable response. The most informative biomarker would additionally provide information not already available from MRI, cytology or plasma—for example, identifying CNS-specific resistance before overt progression and predicting which patients subsequently fail a given therapy. Demonstrating such added value will require prespecified sampling intervals, uniform CSF processing, quantitative molecular thresholds and blinded comparison with longitudinal clinical and imaging outcomes.

The interpretation must nevertheless remain compartment-specific. CSF is anatomically well positioned to sample LM, but it cannot be assumed to represent every parenchymal brain metastasis, particularly lesions remote from CSF spaces or lesions with limited shedding. The future goal is therefore not a universal “CNS liquid biopsy”, but a longitudinal multi-compartment monitoring strategy in which CSF, plasma, imaging and, when available, metastatic tissue are integrated according to the disease being sampled. Such an approach could convert CNS biomarkers from static diagnostic adjuncts into tools for observing treatment response, clonal evolution and impending resistance in real time.

Models and Trials Must Represent CNS Compartmentalization

Progress in CNS metastasis research depends on whether experimental models reproduce the biological step that a study claims to interrogate. This distinction is particularly important in breast cancer because commonly used inoculation routes capture different portions of the metastatic cascade. Intracardiac injection introduces tumour cells into the arterial circulation and therefore bypasses primary-tumour invasion and intravasation while retaining systemic dissemination, cerebral vascular arrest, blood–brain barrier interaction, extravasation and subsequent colonization. Intracarotid injection further enriches delivery to the cerebral circulation and reduces competing extracranial metastatic burden; it is therefore well suited to studying endothelial interaction, extravasation and early parenchymal colonization, but it likewise cannot address the natural escape of tumour cells from the breast or the determinants of initial systemic dissemination.117,118 These models are valuable experimental metastasis assays, but their results should not be interpreted as evidence for the entire natural metastatic sequence.

Direct intracranial inoculation answers a different question. By depositing tumour cells within the brain parenchyma, it bypasses vascular arrest, blood–brain barrier traversal and extravasation and is therefore primarily a model of post-entry survival, niche interaction, tumour outgrowth and treatment response.117,118 Conversely, intra-CSF approaches, including intrathecal or intracisternal inoculation, place tumour cells directly into the CSF-containing leptomeningeal compartment. These models are useful for investigating CSF survival, meningeal attachment, neuraxial dissemination, resource competition and therapeutic response after leptomeningeal entry, but they cannot establish how tumour cells naturally reach the leptomeninges or which blood–CSF or meningeal interfaces they traverse.13 Thus, a mechanism discovered after direct intracranial or intra-CSF implantation should not automatically be described as a mechanism of metastatic entry.

No single model can therefore represent the entire CNS metastatic process. Spontaneous dissemination from an orthotopic mammary tumour offers greater biological continuity from primary growth to distant colonization, but brain or leptomeningeal involvement is often infrequent, delayed and accompanied by extensive extracranial disease.117 Experimental injection models provide greater reproducibility but progressively remove earlier steps of dissemination. Future studies should use complementary models matched to the biological question: vascular-delivery models when testing CNS entry and extravasation, direct CNS implantation when testing post-entry fitness or therapeutic vulnerability, and spontaneous or lineage-traced models when making claims about organotropism or the natural relationship between BM and LM. Ideally, key findings should be reproduced across more than one model and, where possible, validated in patient-derived material.

Clinical trials require the same degree of compartmental resolution. Historically, patients with CNS metastases were frequently excluded or grouped together despite markedly different disease states. A foundational ASCO–Friends of Cancer Research framework distinguished treated/stable BM from active disease, with active BM defined as newly diagnosed or progressive intracranial disease.119 Future breast cancer trials should go further by separating treated and stable BM, untreated BM, and previously treated but progressing BM. These groups are not interchangeable: untreated lesions have not yet been exposed to local CNS selection, whereas progressive lesions after surgery or radiotherapy represent disease that has persisted or recurred under previous therapeutic pressure. Pooling them can obscure whether an apparent intracranial effect reflects activity against treatment-naïve lesions, salvage activity in resistant disease or simply continued stability after effective local therapy.

Recent trial landscapes show that this problem has not been fully resolved. The BMScope analysis of 4,921 BM/LM studies found that, despite increasing systemic-therapy research, symptomatic patients remained underrepresented and many studies lacked dedicated intracranial endpoints; even trials enrolling untreated or progressive CNS disease often did not use a CNS-specific primary endpoint.120 A SNO–ASCO consensus similarly emphasized that advances in intracranial metastasis research will require trial designs that explicitly address CNS populations rather than treating intracranial disease as a secondary eligibility characteristic.23 Accordingly, future studies should prospectively report CNS disease status, prior surgery and radiotherapy, corticosteroid requirement, measurable intracranial disease, extracranial disease status and CNS-specific response and progression endpoints.

LM requires still finer stratification. It should not be embedded within a generic “brain metastasis present” category or added as a few incidental patients to a BM cohort. EANO–ESMO criteria distinguish linear LM, nodular LM, mixed linear–nodular disease and disease without typical MRI abnormalities.11 These phenotypes differ in geometry, measurability, tumour–CSF exposure and suitability for compartment-directed treatment; for example, diffuse linear disease with CSF tumour cells presents a different therapeutic problem from a bulky surface-associated nodule. The SNO–ASCO LM trial-design consensus specifically identifies the persistent exclusion of patients with LM and the need for dedicated, standardized clinical research as major barriers to progress.57 LM trials should therefore prospectively document linear, nodular or mixed morphology, CSF cytological or molecular status, cranial versus spinal distribution, coexisting parenchymal BM and previous compartment-specific therapy rather than analysing all LM as a homogeneous entity.

The same principle should connect preclinical and clinical development. A therapy that succeeds after direct intracranial implantation has demonstrated activity against established parenchymal tumour growth, not necessarily an ability to prevent BBB traversal or initial brain colonization. A therapy effective after intracisternal inoculation has demonstrated activity against post-entry leptomeningeal disease, not necessarily against natural LM dissemination. Similarly, a clinical response observed only in treated/stable BM should not automatically be extrapolated to active untreated BM, progressing BM or LM. Future translational studies should therefore define CNS disease by compartment, route, stage and treatment state, replacing the binary designation of “CNS metastasis: yes/no”. Only when experimental models and clinical cohorts describe the same biological state will mechanistic discoveries and therapeutic signals be meaningfully comparable. This integrated framework for CNS evolution, treatment-driven selection, longitudinal monitoring and future validation is summarized in Figure 5.

Figure 5.

A multi-part infographic on CNS metastasis evolution, sampling, therapy selection, models and trials. A) A linear arrow sequence reads, heterogeneity, compartmental dissemination, adaptation, evolution, treatment, new selective conditions, longitudinal reassessment. B) Clonal selection and branched CNS evolution. Text reads, Heterogeneous primary breast tumour. A cluster is labeled, Clone A, Clone B, Clone C. Four numbered scenarios are shown. 1 Common ancestry with branched evolution, possible evolutionary scenario, with Primary tumour leading to Metastatic ancestor, then branching to BM subclones and LM subclones. 2 Independent or parallel seeding, with separate dashed arrows from the primary tumour to BM and to LM. 3 possible secondary dissemination, with BM connected by a dashed arrow to LM. 4 hypothesized treatment selected lineage, with Therapy exposed metastatic clone connected by a dashed arrow to LM. A note reads, Potential evolutionary relationships not mutually exclusive. Shared ancestry does not establish direction of spread. C) Molecular divergence and longitudinal sampling. A box labeled Primary or extracranial tissue lists, ER or PR, HER2, PI3K AKT PTEN, ESR1. A label reads, Observed molecular discordance, with contributing labels, pre existing heterogeneity, clonal selection, elapsed time, systemic therapy, spatial sampling, technical sensitivity. A line of text reads, Primary diagnosis to systemic metastasis to BM or LM to treatment to progression. A section labeled CNS specimens, BM tissue or CSF lists, ER or PR discordance, HER2 gain or loss, PI3K AKT PTEN alterations, ESR1 alterations. A section labeled BM resected tissue shows boxes, spatial analysis, genomic profiling, scRNA seq and a note, high spatial resolution, limited longitudinal accessibility. A section labeled BM lesion includes text, variable shedding into CSF and paired compartmental monitoring, with Plasma and CSF labels. A section labeled LM CSF based assessment contains boxes, cytology, CTCs, ctDNA, proteomics or multi omics, single cell and a note, repeatable, especially informative for LM. A statement reads, CSF does not represent all parenchymal BM. D) Therapy creates new selective conditions. Local therapy lists, surgery, SRS, WBRT or radiotherapy. Systemic or compartment directed therapy lists, tucatinib based therapy, T DXd or ADCs, subtype directed therapy, intra CSF therapy for selected LM. A central label reads, CNS efficacy, surrounded by labels, lesion permeability, drug access, target expression, tumour state, niche support, treatment history. A note reads, Most mature systemic CNS evidence, HER2 positive disease, tucatinib based therapy, T DXd. Another note reads, HR HER2 minus and TNBC, less mature CNS specific evidence. E) Treatment selection and evolution. A sequence reads, response or population bottleneck, residual disease, resistant or altered states, CNS relapse. F) Testable research agenda. A section titled Tropism or lineage question shows boxes, paired primary extracranial BM LM samples, matched BM LM phylogenetics, lineage tracing, genetic barcoding, longitudinal models, serial treatment stage sampling. A central box reads, What determines BM versus LM fate. Lower boxes read, Pre existing tropism, Niche induced adaptation, Combined model. G) Model question matching. A list of routes reads, Intracardiac, Intracarotid, Intracranial, Intrathecal, Intracisternal. H) Clinical trial stratification. A top box reads, CNS metastasis yes no. Two columns are labeled BM and LM. Under BM, items read, Stable treated BM, Active progressive BM, Untreated BM. Under LM, items read, linear LM, nodular LM, mixed LM, cranial spinal involvement. A box reads, CNS specific trial design, with items, intracranial response, CNS PFS, neurological outcome, compartment specific pharmacokinetics, serial molecular monitoring. A final statement reads, If CSF truly tracks LM evolution, rising ctDNA, VAF, resistant subclones should precede or coincide with progression, with an arrow to radiographic or neurological progression.

CNS evolution, treatment, and a testable translational framework. Breast cancer central nervous system (CNS) metastasis is depicted as a dynamic process linking tumour heterogeneity, compartmental dissemination, adaptation, molecular evolution, treatment-associated selection and longitudinal reassessment. In the upper-left panel, Clone A, Clone B, and Clone C represent illustrative subclonal populations within a heterogeneous primary breast tumour rather than defined molecular subtypes or individually validated patient-derived clones. The numbered labels indicate four potential evolutionary scenarios: (1) common ancestry followed by branched evolution into brain parenchymal metastasis (BM) and leptomeningeal metastasis (LM) subclones; (2) independent or parallel seeding of BM and LM; (3) possible secondary dissemination from established BM to LM; and (4) a hypothesized treatment-selected metastatic lineage. The numbering distinguishes alternative scenarios and does not imply a chronological sequence or an established hierarchy of evolutionary mechanisms. Human genomic and phylogenetic studies support clonal relatedness and branched divergence in selected metastatic breast cancer settings, but do not establish a predominant evolutionary relationship or the direction of dissemination between BM and LM. Secondary dissemination is supported by selected clinical observations, whereas treatment-selected lineages remain hypothetical and require direct longitudinal validation. These evolutionary scenarios are potentially overlapping, and the connecting arrows illustrate possible relationships rather than a formal hierarchy of evidence or an obligatory sequence of metastatic progression. Molecular discordance across primary, extracranial and CNS compartments may reflect pre-existing heterogeneity, clonal selection, treatment exposure, elapsed time and sampling effects rather than niche-driven evolution alone. BM tissue, CSF and plasma provide complementary but incompletely overlapping molecular information. Within CSF-based assessment, cytology, circulating tumour cell (CTC) analysis, circulating tumour DNA (ctDNA) profiling, proteomics/multi-omics and single-cell analysis are depicted as complementary approaches rather than sequential diagnostic or analytical procedures. Therapeutic efficacy depends on CNS drug access together with lesion permeability, target expression, tumour state, niche support and treatment history, while local and systemic treatments may introduce additional selective pressures. The proposed translational framework integrates matched multi-compartment sampling, lineage-resolved experimental models and CNS-specific clinical trial stratification, while recognizing differences between parenchymal and leptomeningeal disease. Longitudinal CSF biomarkers require prospective validation against radiographic and neurological progression before they can be considered reliable indicators of LM evolution or treatment response.

Conclusions

Current evidence supports the view that brain parenchymal metastases and leptomeningeal metastases represent two major CNS metastatic compartments of breast cancer that are exposed to distinct anatomical, cellular and metabolic constraints. Although they can share metastatic ancestry and may coexist within the same patient, their routes of dissemination, modes of local adaptation, patterns of molecular evolution and therapeutic vulnerabilities are not fully interchangeable. This distinction provides a more biologically informative framework than treating all CNS involvement as a single metastatic state. By bringing metastatic dissemination, niche adaptation, molecular evolution, biomarker representation and therapeutic vulnerability into a unified comparison, this review provides an evidence-weighted framework for interpreting BM–LM differences across biological and clinical studies.

At the same time, current data do not establish the CNS niche as the sole causal source of the molecular and therapeutic differences observed between compartments. The key unresolved issue is the relative contribution of pre-existing clonal heterogeneity, treatment-associated selection and local adaptation to the observed divergence; current cross-sectional sampling and predominantly preclinical mechanistic evidence cannot reliably disentangle these influences. Accordingly, compartment-associated differences should not automatically be interpreted as compartment-driven evolution, and several proposed mechanisms remain supported mainly by preclinical or pan-cancer evidence, particularly in leptomeningeal disease.

The main value of the two-compartment framework is therefore not to impose a complete mechanistic model, but to generate testable questions. Clinically, it supports compartment-aware interpretation of imaging, tissue and CSF findings, and cautions against directly extrapolating BM therapeutic evidence to LM or across breast cancer subtypes and treatment states. Future studies should combine paired primary, extracranial, brain and leptomeningeal sampling with longitudinal CSF and plasma monitoring, lineage-resolved models that distinguish metastatic entry from post-entry growth, and clinical trials that explicitly stratify previously treated/stable BM, untreated BM, and previously treated/progressing BM, while separately accounting for linear, nodular and mixed LM presentations. Such studies are needed to determine whether compartment-specific biomarkers and candidate therapeutic vulnerabilities provide reproducible information that can improve clinical trial design and ultimately inform treatment decisions.

Funding Statement

This study was supported by the Natural Science Foundation of Heilongjiang Province (Project No. PL2025H277).

Author Contributions

All authors made a significant contribution to the work reported, whether in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising, or critically reviewing the article; gave final approval of the version to be published; agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.

References

  • 1.Kuksis M, Gao Y, Tran W, et al. The incidence of brain metastases among patients with metastatic breast cancer: a systematic review and meta-analysis. Neuro-Oncol. 2021;23(6):894–36. doi: 10.1093/neuonc/noaa285 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Sammons SL, Sanglier T, Leone JP, et al. Prevalence by therapy line and incidence of breast cancer brain metastases in 18 075 patients. J Natl Cancer Inst. 2025;117(8):1565–1572. doi: 10.1093/jnci/djaf048 [DOI] [PubMed] [Google Scholar]
  • 3.Le Rhun E, Guckenberger M, Smits M, et al. EANO-ESMO clinical practice guidelines for diagnosis, treatment and follow-up of patients with brain metastasis from solid tumours. Ann Oncol off J Eur Soc Med Oncol. 2021;32(11):1332–1347. doi: 10.1016/j.annonc.2021.07.016 [DOI] [PubMed] [Google Scholar]
  • 4.Aizer AA, Lamba N, Ahluwalia MS, et al. Brain metastases: a society for neuro-oncology (SNO) consensus review on current management and future directions. Neuro-Oncol. 2022;24(10):1613–1646. doi: 10.1093/neuonc/noac118 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Franzoi MA, Hortobagyi GN. Leptomeningeal carcinomatosis in patients with breast cancer. Crit Rev Oncol Hematol. 2019;135:85–94. doi: 10.1016/j.critrevonc.2019.01.020 [DOI] [PubMed] [Google Scholar]
  • 6.Mollica L, Leli C, Puglisi S, Sardi S, Sottotetti F. Leptomeningeal carcinomatosis and breast cancer: a systematic review of current evidence on diagnosis, treatment and prognosis. Drugs Context. 2021;10. doi: 10.7573/dic.2021-6-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Arvanitis CD, Ferraro GB, Jain RK. The blood-brain barrier and blood-tumour barrier in brain tumours and metastases. Nat Rev Cancer. 2020;20(1):26–41. doi: 10.1038/s41568-019-0205-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Schulz M, Salamero-Boix A, Niesel K, Alekseeva T, Sevenich L. Microenvironmental regulation of tumor progression and therapeutic response in brain metastasis. Front Immunol. 2019;10:1713. doi: 10.3389/fimmu.2019.01713 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Lockman PR, Mittapalli RK, Taskar KS, et al. Heterogeneous blood-tumor barrier permeability determines drug efficacy in experimental brain metastases of breast cancer. Clin Cancer Res off J Am Assoc Cancer Res. 2010;16(23):5664–5678. doi: 10.1158/1078-0432.CCR-10-1564 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Babak MV, Zalutsky MR, Balyasnikova IV. Heterogeneity and vascular permeability of breast cancer brain metastases. Cancer Lett. 2020;489:174–181. doi: 10.1016/j.canlet.2020.06.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Le Rhun E, Weller M, van den Bent M, et al. Leptomeningeal metastasis from solid tumours: EANO-ESMO clinical practice guideline for diagnosis, treatment and follow-up. ESMO Open. 2023;8(5):101624. doi: 10.1016/j.esmoop.2023.101624 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Remsik J, Boire A. The path to leptomeningeal metastasis. Nat Rev Cancer. 2024;24(7):448–460. doi: 10.1038/s41568-024-00700-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Wilcox JA, Chukwueke UN, Ahn MJ, et al. Leptomeningeal metastases from solid tumors: a society for neuro-oncology and American society of clinical oncology consensus review on clinical management and future directions. Neuro-Oncol. 2024;26(10):1781–1804. doi: 10.1093/neuonc/noae103 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Ozair A, Wilding H, Bhanja D, et al. Leptomeningeal metastatic disease: new frontiers and future directions. Nat Rev Clin Oncol. 2025;22(2):134–154. doi: 10.1038/s41571-024-00970-3 [DOI] [PubMed] [Google Scholar]
  • 15.Ding L, Ellis MJ, Li S, et al. Genome remodelling in a basal-like breast cancer metastasis and xenograft. Nature. 2010;464(7291):999–1005. doi: 10.1038/nature08989 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Bos PD, Zhang XHF, Nadal C, et al. Genes that mediate breast cancer metastasis to the brain. Nature. 2009;459(7249):1005–1009. doi: 10.1038/nature08021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Zhang L, Ridgway LD, Wetzel MD, et al. The identification and characterization of breast cancer CTCs competent for brain metastasis. Sci Transl Med. 2013;5(180):180ra48. doi: 10.1126/scitranslmed.3005109 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Lorger M, Felding-Habermann B. Capturing changes in the brain microenvironment during initial steps of breast cancer brain metastasis. Am J Pathol. 2010;176(6):2958–2971. doi: 10.2353/ajpath.2010.090838 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Brastianos PK, Carter SL, Santagata S, et al. Genomic characterization of brain metastases reveals branched evolution and potential therapeutic targets. Cancer Discov. 2015;5(11):1164–1177. doi: 10.1158/2159-8290.CD-15-0369 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Kotecha R, Tonse R, Rubens M, et al. Systematic review and meta-analysis of breast cancer brain metastasis and primary tumor receptor expression discordance. Neuro-Oncol Adv. 2021;3(1):vdab010. doi: 10.1093/noajnl/vdab010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Profaci CP, Munji RN, Pulido RS, Daneman R. The blood-brain barrier in health and disease: important unanswered questions. J Exp Med. 2020;217(4):e20190062. doi: 10.1084/jem.20190062 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Zhao H, Wang L, Ji X, Zhang L, Li C. Biology of breast cancer brain metastases and novel therapies targeting the blood brain barrier: an updated review. Med Oncol. 2023;40(6):181. doi: 10.1007/s12032-023-02047-0 [DOI] [PubMed] [Google Scholar]
  • 23.Sharma A, Kennedy LB, Van Swearingen AED, et al. Intracranial metastases from solid tumors: call to action and consensus from the society for neuro-oncology and American society of clinical oncology collaborative. Neuro-Oncol. 2026;28(2):319–333. doi: 10.1093/neuonc/noaf226 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Freret ME, Boire A. The anatomic basis of leptomeningeal metastasis. J Exp Med. 2024;221(4):e20212121. doi: 10.1084/jem.20212121 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Gelb S, Lehtinen MK. Snapshot: choroid plexus brain barrier. Cell. 2023;186(16):3522–3522.e1. doi: 10.1016/j.cell.2023.07.015 [DOI] [PubMed] [Google Scholar]
  • 26.Pan JK, Lin WD, Kuo YL, et al. ICAM2 initiates trans-blood-CSF barrier migration and stemness properties in leptomeningeal metastasis of triple-negative breast cancer. Oncogene. 2023;42(39):2919–2931. doi: 10.1038/s41388-023-02769-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Sener U, Wilcox JA, Boire AA. Leptomeningeal disease: current approaches and future directions. Curr Neurol Neurosci Rep. 2025;25(1):25. doi: 10.1007/s11910-025-01412-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Hong JH, Lee J, Shin K, et al. Risk factors of developing leptomeningeal seeding after resection of brain metastasis in patients with breast cancer: defining the indication for preoperative SRS. Vivo. 2025;39(2):1094–1103. doi: 10.21873/invivo.13914 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Mühlhausen L, Kocher M, Karadachi HH, et al. Leptomeningeal disease (LMD) after resection of brain metastases: results of the multicenter SUBAROMA study. J Neurooncol. 2026;178(3):99. doi: 10.1007/s11060-026-05700-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Fitzpatrick A, Iravani M, Mills A, et al. Genomic profiling and pre-clinical modelling of breast cancer leptomeningeal metastasis reveals acquisition of a lobular-like phenotype. Nat Commun. 2023;14(1):7408. doi: 10.1038/s41467-023-43242-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Galloni C, Egnuni T, Zahed Mohajerani S, et al. Brain endothelial cells promote breast cancer cell extravasation to the brain via EGFR-DOCK4-RAC1 signalling. Commun Biol. 2024;7(1):602. doi: 10.1038/s42003-024-06200-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Carbonell WS, Ansorge O, Sibson N, Muschel R. The vascular basement membrane as “soil” in brain metastasis. PLoS One. 2009;4(6):e5857. doi: 10.1371/journal.pone.0005857 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Valiente M, Obenauf AC, Jin X, et al. Serpins promote cancer cell survival and vascular co-option in brain metastasis. Cell. 2014;156(5):1002–1016. doi: 10.1016/j.cell.2014.01.040 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Priego N, Zhu L, Monteiro C, et al. STAT3 labels a subpopulation of reactive astrocytes required for brain metastasis. Nat Med. 2018;24(7):1024–1035. doi: 10.1038/s41591-018-0044-4 [DOI] [PubMed] [Google Scholar]
  • 35.Pellerino A, Priego N, Bertero L, et al. STAT3 expression in brain metastases from breast cancer is correlated with molecular subtype and impacts clinical outcome. Neuro-Oncol. 2025;27(12):3132–3144. doi: 10.1093/neuonc/noaf187 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Gan S, Macalinao DG, Shahoei SH, et al. Distinct tumor architectures and microenvironments for the initiation of breast cancer metastasis in the brain. Cancer Cell. 2024;42(10):1693–1712.e24. doi: 10.1016/j.ccell.2024.08.015 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Chen Q, Boire A, Jin X, et al. Carcinoma-astrocyte gap junctions promote brain metastasis by cGAMP transfer. Nature. 2016;533:7604):493–498. doi: 10.1038/nature18268 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Zhang L, Zhang S, Yao J, et al. Microenvironment-induced PTEN loss by exosomal microRNA primes brain metastasis outgrowth. Nature. 2015;527:7576):101–104. doi: 10.1038/nature15376 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Klemm F, Möckl A, Salamero-Boix A, et al. Compensatory CSF2-driven macrophage activation promotes adaptive resistance to CSF1R inhibition in breast-to-brain metastasis. Nat Cancer. 2021;2(10):1086–1101. doi: 10.1038/s43018-021-00254-0 [DOI] [PubMed] [Google Scholar]
  • 40.Evans KT, Blake K, Longworth A, et al. Microglia promote anti-tumour immunity and suppress breast cancer brain metastasis. Nat Cell Biol. 2023;25(12):1848–1859. doi: 10.1038/s41556-023-01273-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Zeng Q, Michael IP, Zhang P, et al. Synaptic proximity enables NMDAR signalling to promote brain metastasis. Nature. 2019;573:7775):526–531. doi: 10.1038/s41586-019-1576-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Mondal J, Nylund P, Malgulwar PB, Johnson WE, Huse JT. Synaptic transmission promotes brain metastatic outgrowth in breast cancer. JCI Insight. 2025;10(19):e193119. doi: 10.1172/jci.insight.193119 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Ruan X, Yan W, Cao M, et al. Breast cancer cell-secreted miR-199b-5p hijacks neurometabolic coupling to promote brain metastasis. Nat Commun. 2024;15(1):4549. doi: 10.1038/s41467-024-48740-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Chen EI, Hewel J, Krueger JS, et al. Adaptation of energy metabolism in breast cancer brain metastases. Cancer Res. 2007;67(4):1472–1486. doi: 10.1158/0008-5472.CAN-06-3137 [DOI] [PubMed] [Google Scholar]
  • 45.Chen J, Lee HJ, Wu X, et al. Gain of glucose-independent growth upon metastasis of breast cancer cells to the brain. Cancer Res. 2015;75(3):554–565. doi: 10.1158/0008-5472.CAN-14-2268 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Parida PK, Marquez-Palencia M, Nair V, et al. Metabolic diversity within breast cancer brain-tropic cells determines metastatic fitness. Cell Metab. 2022;34(1):91–105.e7. doi: 10.1016/j.cmet.2021.12.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Ferraro GB, Ali A, Luengo A, et al. Fatty acid synthesis is required for breast cancer brain metastasis. Nat Cancer. 2021;2(4):414–428. doi: 10.1038/s43018-021-00183-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Parida PK, Marquez-Palencia M, Ghosh S, et al. Limiting mitochondrial plasticity by targeting DRP1 induces metabolic reprogramming and reduces breast cancer brain metastases. Nat Cancer. 2023;4(6):893–907. doi: 10.1038/s43018-023-00563-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Laue K, Pozzi S, Zerbib J, et al. p53 inactivation drives breast cancer metastasis to the brain through SCD1 upregulation and increased fatty acid metabolism. Nat Genet. 2026;58(1):116–131. doi: 10.1038/s41588-025-02446-1 [DOI] [PubMed] [Google Scholar]
  • 50.Ngo B, Kim E, Osorio-Vasquez V, et al. Limited environmental serine and glycine confer brain metastasis sensitivity to PHGDH inhibition. Cancer Discov. 2020;10(9):1352–1373. doi: 10.1158/2159-8290.CD-19-1228 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Mashimo T, Pichumani K, Vemireddy V, et al. Acetate is a bioenergetic substrate for human glioblastoma and brain metastases. Cell. 2014;159(7):1603–1614. doi: 10.1016/j.cell.2014.11.025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Abbott KL, Subudhi S, Ferreira R, et al. Nutrient requirements of organ-specific metastasis in breast cancer. Nature. 2026;649(8099):1292–1301. doi: 10.1038/s41586-025-09898-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Remsik J, Chi Y, Tong X, et al. Leptomeningeal metastatic cells adopt two phenotypic states. Cancer Rep. 2022;5(4):e1236. doi: 10.1002/cnr2.1236 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Chi Y, Remsik J, Kiseliovas V, et al. Cancer cells deploy lipocalin-2 to collect limiting iron in leptomeningeal metastasis. Science. 2020;369:6501):276–282. doi: 10.1126/science.aaz2193 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Jang S, Gwak HS, Lee KY, et al. Exploratory profiling of metabolites in cerebrospinal fluid using a commercially available targeted LC-MS based metabolomics kit to discriminate leptomeningeal metastasis. Cancer Metab. 2025;13(1):2. doi: 10.1186/s40170-024-00367-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Id Said B, Jerzak KJ, Chen H, et al. Survival outcomes among patients with breast cancer and leptomeningeal disease. Sci Rep. 2025;15(1):24170. doi: 10.1038/s41598-025-07191-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Piña Y, Umeh Garcia M, Boire A, et al. Breaking barriers in leptomeningeal disease: a SNO-ASCO consensus review on clinical trial design and basic science integration. Neuro-Oncol. 2026:noag126. doi: 10.1093/neuonc/noag126 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58.Boire A, Zou Y, Shieh J, Macalinao DG, Pentsova E, Massagué J. Complement component 3 adapts the cerebrospinal fluid for leptomeningeal metastasis. Cell. 2017;168(6):1101–1113.e13. doi: 10.1016/j.cell.2017.02.025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59.Conrad C, Dorzweiler K, Miller MA, Lauffenburger DA, Strik H, Bartsch JW. Profiling of metalloprotease activities in cerebrospinal fluids of patients with neoplastic meningitis. Fluids Barriers CNS. 2017;14(1):22. doi: 10.1186/s12987-017-0070-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Ruan H, Wang Z, Sun Z, et al. Single-cell RNA sequencing reveals the characteristics of cerebrospinal fluid tumour environment in breast cancer and lung cancer leptomeningeal metastases. Clin Transl Med. 2022;12(6):e885. doi: 10.1002/ctm2.885 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61.Im KW, Huppert LA, Malevanchik L, et al. High-dimensional immune cell profiling of cerebrospinal fluid from patients with metastatic breast cancer and leptomeningeal disease. Npj Breast Cancer. 2023;9(1):22. doi: 10.1038/s41523-023-00526-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62.Zhao J, Zeng R, Li X, et al. Dura immunity configures leptomeningeal metastasis immunosuppression for cerebrospinal fluid barrier invasion. Nat Cancer. 2024;5(12):1940–1961. doi: 10.1038/s43018-024-00858-2 [DOI] [PubMed] [Google Scholar]
  • 63.Rahbarlayegh E, Wojcinski A, Nomura N, Juarez TM, Blouw B, Kesari S. Multimodal evaluation of cerebrospinal fluid from breast cancer leptomeningeal metastases using circulating tumor cell detection, single-cell sequencing, and proteomics. J Transl Med. 2025;23(1):773. doi: 10.1186/s12967-025-06671-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64.Brown D, Smeets D, Székely B, et al. Phylogenetic analysis of metastatic progression in breast cancer using somatic mutations and copy number aberrations. Nat Commun. 2017;8:14944. doi: 10.1038/ncomms14944 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65.Cosgrove N, Varešlija D, Keelan S, et al. Mapping molecular subtype specific alterations in breast cancer brain metastases identifies clinically relevant vulnerabilities. Nat Commun. 2022;13(1):514. doi: 10.1038/s41467-022-27987-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66.Van Swearingen AED, Lee MR, Rogers LW, et al. Genomic and immune profiling of breast cancer brain metastases. Acta Neuropathol Commun. 2025;13(1):99. doi: 10.1186/s40478-025-02001-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67.Griguolo G, Dieci MV, Fineberg S, et al. Gene expression profiling in a retrospective real-world cohort of breast cancer brain metastases and paired primary tumors identifies biological changes with potential therapeutic implications. ESMO Open. 2025;10(8):105507. doi: 10.1016/j.esmoop.2025.105507 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Wessels PH, Cornelissen S, Monkhorst K, et al. Divergence of DNA tumor mutations in cerebrospinal fluid in metastatic breast cancer patients with leptomeningeal metastases. Breast. 2026;85:104674. doi: 10.1016/j.breast.2025.104674 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 69.Hu Z, Li Z, Ma Z, Curtis C. Multi-cancer analysis of clonality and the timing of systemic spread in paired primary tumors and metastases. Nat Genet. 2020;52(7):701–708. doi: 10.1038/s41588-020-0628-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 70.Aftimos P, Oliveira M, Irrthum A, et al. Genomic and transcriptomic analyses of breast cancer primaries and matched metastases in Aurora, the breast international group (BIG) molecular screening initiative. Cancer Discov. 2021;11(11):2796–2811. doi: 10.1158/2159-8290.CD-20-1647 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 71.de S CG, Rodrigues FR, Fernandes PV, Iá S, da Silva JL, Araujo LHL. Real-world evidence of immunohistochemical discordance in breast cancer brain metastases. Clin Breast Cancer. 2026;26(5):112–120.e3. doi: 10.1016/j.clbc.2026.03.002 [DOI] [PubMed] [Google Scholar]
  • 72.Pereslete AM, Hughes ME, Martin AR, et al. Analysis of HER2 expression changes from breast primary to brain metastases and the impact of HER2-low expression on overall survival. Neuro-Oncol. 2025;27(1):184–194. doi: 10.1093/neuonc/noae163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 73.Thulin A, Andersson C, Werner Rönnerman E, et al. Discordance of PIK3CA and TP53 mutations between breast cancer brain metastases and matched primary tumors. Sci Rep. 2021;11(1):23548. doi: 10.1038/s41598-021-02903-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 74.Giannoudis A, Sartori A, Eastoe L, et al. Genomic profiling using the UltraSEEK panel identifies discordancy between paired primary and breast cancer brain metastases and an association with brain metastasis-free survival. Breast Cancer Res Treat. 2021;190(2):241–253. doi: 10.1007/s10549-021-06364-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 75.Huang RSP, Haberberger J, McGregor K, et al. Clinicopathologic and genomic landscape of breast carcinoma brain metastases. Oncologist. 2021;26(10):835–844. doi: 10.1002/onco.13855 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 76.Dowling GP, Keelan S, Cosgrove NS, et al. Receptor discordance in metastatic breast cancer; a review of clinical and genetic subtype alterations from primary to metastatic disease. Breast Cancer Res Treat. 2024;207(3):471–476. doi: 10.1007/s10549-024-07431-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 77.Shi H, Chen J, Wu Z, et al. Single-cell profiling reveals distinct populations of tumor-associated macrophages and metastatic tumor cells in breast cancer brain metastasis. Cell Death Dis. 2026;17(1):553. doi: 10.1038/s41419-026-08807-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 78.Darlix A, Cayrefourcq L, Pouderoux S, et al. Detection of circulating tumor cells in cerebrospinal fluid of patients with suspected breast cancer leptomeningeal metastases: a prospective study. Clin Chem. 2022;68(10):1311–1322. doi: 10.1093/clinchem/hvac127 [DOI] [PubMed] [Google Scholar]
  • 79.Fitzpatrick A, Iravani M, Mills A, et al. Assessing CSF ctDNA to improve diagnostic accuracy and therapeutic monitoring in breast cancer leptomeningeal metastasis. Clin Cancer Res off J Am Assoc Cancer Res. 2022;28(6):1181–1191. doi: 10.1158/1078-0432.CCR-21-3017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 80.Siravegna G, Geuna E, Mussolin B, et al. Genotyping tumour DNA in cerebrospinal fluid and plasma of a HER2-positive breast cancer patient with brain metastases. ESMO Open. 2017;2(4):e000253. doi: 10.1136/esmoopen-2017-000253 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 81.Shishido SN, Marvit A, Pham D, et al. Multi-omic characterization of single cells and cell-free components detected in the cerebrospinal fluid of patients with leptomeningeal disease. Cancers. 2024;16(22):3746. doi: 10.3390/cancers16223746 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 82.Morganti S, Parsons HA, Lin NU, Grinshpun A. Liquid biopsy for brain metastases and leptomeningeal disease in patients with breast cancer. Npj Breast Cancer. 2023;9(1):43. doi: 10.1038/s41523-023-00550-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 83.Adkins CE, Mohammad AS, Terrell-Hall TB, et al. Characterization of passive permeability at the blood-tumor barrier in five preclinical models of brain metastases of breast cancer. Clin Exp Metastasis. 2016;33(4):373–383. doi: 10.1007/s10585-016-9784-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 84.Morikawa A, Peereboom DM, Thorsheim HR, et al. Capecitabine and lapatinib uptake in surgically resected brain metastases from metastatic breast cancer patients: a prospective study. Neuro-Oncol. 2015;17(2):289–295. doi: 10.1093/neuonc/nou141 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 85.Khan I, Kumar D, Zhang W, et al. Trastuzumab deruxtecan efficacy, FcRn-facilitated penetration of the blood-tumor barrier, and distribution to tumor cells in HER2+ brain metastasis model systems. Neuro-Oncol. 2026;28(4):952–967. doi: 10.1093/neuonc/noag002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 86.Harbeck N, Ciruelos E, Jerusalem G, et al. Trastuzumab deruxtecan in HER2-positive advanced breast cancer with or without brain metastases: a Phase 3b/4 trial. Nat Med. 2024;30(12):3717–3727. doi: 10.1038/s41591-024-03261-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 87.Kim SJ, Kim JS, Park ES, et al. Astrocytes upregulate survival genes in tumor cells and induce protection from chemotherapy. Neoplasia. 2011;13(3):286–298. doi: 10.1593/neo.11112 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 88.Yakati V, Shevde LA, Rao SS. Matrix stiffness influences response to chemo and targeted therapy in brain metastatic breast cancer cells. Biomater Sci. 2024;12(15):3882–3895. doi: 10.1039/d4bm00342j [DOI] [PubMed] [Google Scholar]
  • 89.Vogelbaum MA, Brown PD, Messersmith H, et al. Treatment for brain metastases: ASCO-SNO-ASTRO guideline. J Clin Oncol off J Am Soc Clin Oncol. 2022;40(5):492–516. doi: 10.1200/JCO.21.02314 [DOI] [PubMed] [Google Scholar]
  • 90.Gondi V, Bauman G, Bradfield L, et al. Radiation therapy for brain metastases: an ASTRO clinical practice guideline. Pract Radiat Oncol. 2022;12(4):265–282. doi: 10.1016/j.prro.2022.02.003 [DOI] [PubMed] [Google Scholar]
  • 91.Lin NU, Borges V, Anders C, et al. Intracranial efficacy and survival with tucatinib plus trastuzumab and capecitabine for previously treated HER2-positive breast cancer with brain metastases in the HER2CLIMB trial. J Clin Oncol off J Am Soc Clin Oncol. 2020;38(23):2611–2619. doi: 10.1200/JCO.20.00775 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 92.Bartsch R, Berghoff AS, Furtner J, et al. Trastuzumab deruxtecan in HER2-positive breast cancer with brain metastases: a single-arm, Phase 2 trial. Nat Med. 2022;28(9):1841–1847. doi: 10.1038/s41591-022-01935-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 93.Tolaney SM, Sahebjam S, Le Rhun E, et al. A phase II study of abemaciclib in patients with brain metastases secondary to hormone receptor-positive breast cancer. Clin Cancer Res off J Am Assoc Cancer Res. 2020;26(20):5311–5319. doi: 10.1158/1078-0432.CCR-20-1764 [DOI] [PubMed] [Google Scholar]
  • 94.Hurvitz SA, Bardia A, Punie K, et al. Subgroup analyses from the phase 3 ASCENT study of sacituzumab govitecan in metastatic triple-negative breast cancer. Npj Breast Cancer. 2024;10(1):33. doi: 10.1038/s41523-024-00635-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 95.Cardoso F, Paluch-Shimon S, Schumacher-Wulf E, et al. 6th and 7th international consensus guidelines for the management of advanced breast cancer (ABC guidelines 6 and 7). Breast. 2024;76:103756. doi: 10.1016/j.breast.2024.103756 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 96.Bartsch R, Jerzak KJ, Larrouquere L, Müller V, Le Rhun E. Pharmacotherapy for leptomeningeal disease in breast cancer. Cancer Treat Rev. 2024;122:102653. doi: 10.1016/j.ctrv.2023.102653 [DOI] [PubMed] [Google Scholar]
  • 97.Murthy RK, O’Brien BJ, Berry DA, et al. Tucatinib-trastuzumab-capecitabine for treatment of leptomeningeal metastasis in women with HER2+ breast cancer: TBCRC049 phase 2 study results. Nat Cancer. 2026;7(3):424–434. doi: 10.1038/s43018-026-01120-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 98.Le Rhun E, Wallet J, Mailliez A, et al. Intrathecal liposomal cytarabine plus systemic therapy versus systemic chemotherapy alone for newly diagnosed leptomeningeal metastasis from breast cancer. Neuro-Oncol. 2020;22(4):524–538. doi: 10.1093/neuonc/noz201 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 99.Oberkampf F, Gutierrez M, Trabelsi Grati O, et al. Phase II study of intrathecal administration of trastuzumab in patients with HER2-positive breast cancer with leptomeningeal metastasis. Neuro-Oncol. 2023;25(2):365–374. doi: 10.1093/neuonc/noac180 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 100.Borm KJ, Behzadi ST, Hörner-Rieber J, et al. DEGRO guideline for personalized radiotherapy of brain metastases and leptomeningeal carcinomatosis in patients with breast cancer. Strahlenther Onkol Organ Dtsch Rontgengesellschaft Al. 2024;200(4):259–275. doi: 10.1007/s00066-024-02202-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 101.Yang JT, Yerramilli D, Pentsova E, et al. Proton craniospinal irradiation for patients with leptomeningeal metastasis: a randomized clinical trial. JAMA Oncol. 2025;11(11):1293–1301. doi: 10.1001/jamaoncol.2025.3007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 102.Vaz Batista M, Pérez-García JM, Garrigós L, et al. The DEBBRAH trial: trastuzumab deruxtecan in HER2-positive and HER2-low breast cancer patients with leptomeningeal carcinomatosis. Med. 2025;6(1):100502. doi: 10.1016/j.medj.2024.08.001 [DOI] [PubMed] [Google Scholar]
  • 103.Kim JS, Kim K, Jung W, et al. New brain metastases after whole-brain radiotherapy of initial brain metastases in breast cancer patients: the significance of molecular subtypes (KROG 16-12). Breast Cancer Res Treat. 2021;186(2):453–462. doi: 10.1007/s10549-020-06043-0 [DOI] [PubMed] [Google Scholar]
  • 104.Chen W, Gupta A, Mai N, et al. Trastuzumab deruxtecan resistance via loss of HER2 expression and binding. Cancer Discov. 2026;16(2):235–249. doi: 10.1158/2159-8290.CD-25-0647 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 105.Lebow ES, Pike LRG, Seidman AD, Moss N, Beal K, Yu Y. Symptomatic necrosis with antibody-drug conjugates and concurrent stereotactic radiotherapy for brain metastases. JAMA Oncol. 2023;9(12):1729–1733. doi: 10.1001/jamaoncol.2023.4492 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 106.Chun SJ, Kim K, Kim YB, et al. Risk of radionecrosis in HER2-positive breast cancer with brain metastasis receiving trastuzumab emtansine (T-DM1) and brain stereotactic radiosurgery. Radiother Oncol J Eur Soc Ther Radiol Oncol. 2024;199:110461. doi: 10.1016/j.radonc.2024.110461 [DOI] [PubMed] [Google Scholar]
  • 107.Khatri VM, Mestres-Villanueva MA, Yarlagadda S, et al. Multi-institutional report of trastuzumab deruxtecan and stereotactic radiosurgery for HER2 positive and HER2-low breast cancer brain metastases. Npj Breast Cancer. 2024;10(1):100. doi: 10.1038/s41523-024-00711-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 108.Chun SJ, Kim K, Chang WI, et al. Impact of trastuzumab deruxtecan (T-DXd) and brain stereotactic radiosurgery on intracranial control and radionecrosis risk in HER2-positive or -low breast cancer brain metastases. Breast. 2026;87:104751. doi: 10.1016/j.breast.2026.104751 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 109.Zare A, Zare A, Kiani I, et al. Timing matters: radiation necrosis risk of antibody-drug conjugates (ADCs) combined with radiation therapy in breast cancer brain metastases: a systematic review and comparative meta-analysis. Neurosurg Rev. 2026;49(1):252. doi: 10.1007/s10143-025-04056-1 [DOI] [PubMed] [Google Scholar]
  • 110.Laakmann E, Schmidt M, Lübbe K, et al. Clinical characteristics and prognostic factors in patients with breast cancer and leptomeningeal metastases from a large registry of BMBC. Breast. 2025;81:104433. doi: 10.1016/j.breast.2025.104433 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 111.Miao K, Zhang A, Yang X, et al. Lymphatic system is the mainstream for breast cancer dissemination and metastasis revealed by single-cell lineage tracing. Mol Cancer. 2025;24(1):75. doi: 10.1186/s12943-025-02279-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 112.Valiente M, Van Swearingen AED, Anders CK, et al. Brain metastasis cell lines panel: a public resource of organotropic cell lines. Cancer Res. 2020;80(20):4314–4323. doi: 10.1158/0008-5472.CAN-20-0291 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 113.Kumthekar PU, Blouw B, Corkos P, et al. The HER2 flip-HER2 amplification of tumor cells in the cerebrospinal fluid of breast cancer patients with leptomeningeal disease: implications for treating the LM tumor with anti-HER2 therapy. Front Oncol. 2024;14:1402651. doi: 10.3389/fonc.2024.1402651 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 114.Sankar BS, Johnson D, Boasiako PA, et al. Cerebrospinal fluid ctDNA as a prognostic and disease-activity biomarker in leptomeningeal metastases: systematic review, meta-analysis, and implications for CSF-guided care. J Neurooncol. 2026;177(2):75. doi: 10.1007/s11060-026-05520-8 [DOI] [PubMed] [Google Scholar]
  • 115.Li M, Chen J, Zhang B, et al. Dynamic monitoring of cerebrospinal fluid circulating tumor DNA to identify unique genetic profiles of brain metastatic tumors and better predict intracranial tumor responses in non-small cell lung cancer patients with brain metastases: a prospective cohort study (GASTO 1028). BMC Med. 2022;20(1):398. doi: 10.1186/s12916-022-02595-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 116.Udhane V, Larson A, Adams JN, et al. Variant allelic frequency to track therapy response and evaluate leptomeningeal disease in metastatic central nervous system cancers. Diagnostics. 2026;16(6):851. doi: 10.3390/diagnostics16060851 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 117.Knier NN, Pellizzari S, Zhou J, Foster PJ, Parsyan A. Preclinical models of brain metastases in breast cancer. Biomedicines. 2022;10(3):667. doi: 10.3390/biomedicines10030667 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 118.Liu Z, Dong S, Liu M, et al. Experimental models for cancer brain metastasis. Cancer Pathog Ther. 2024;2(1):15–23. doi: 10.1016/j.cpt.2023.10.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 119.Lin NU, Prowell T, Tan AR, et al. Modernizing clinical trial eligibility criteria: recommendations of the American society of clinical oncology-friends of cancer research brain metastases working group. J Clin Oncol off J Am Soc Clin Oncol. 2017;35(33):3761–3773. doi: 10.1200/JCO.2017.74.0761 [DOI] [PubMed] [Google Scholar]
  • 120.Cheng VWT, Heywood R, Zakaria R, et al. BMScope: a scoping review to chart the evolving clinical study landscape in brain and leptomeningeal metastasis. Neuro-Oncol. 2024;26(12):2193–2207. doi: 10.1093/neuonc/noae140 [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Breast Cancer : Targets and Therapy are provided here courtesy of Dove Press

RESOURCES