2.1 The application of new technologies in the era of AI

With the development of science and technology, the integration of medicine and engineering has become a prominent trend, significantly advancing the diagnosis and treatment of tumors.^{73, 74} The successful application of AI in medical imaging has enabled AI-based cancer imaging analysis technologies to address complex clinical needs, such as predicting the cancer prognosis, predicting treatment responses, distinguishing between benign and malignant lesions, identifying abnormal tumor responses, and predicting mutations and molecular features.^{75, 76} In the field of LCBM, AI and imaging omics have proven effective in distinguishing BM from different primary tumors.⁷⁷ Gao et al.⁷⁸ developed a deep learning model for the automatic identification and classification of 18 types of brain tumors. This model, using T1-weighted gradient-echo MRI scans, can detect nearly all BM that are 6 mm or larger with a low false positive rate.⁷⁹ A systematic review and meta-analysis conducted by Wang et al.,⁸⁰ which included 42 studies demonstrated that deep learning models, particularly U-Net and its variants, excel in segmentation accuracy.⁸¹ Yun applied these deep learning models in prospective studies, noting improvements in detection performance, though challenges remain for small BM. Enhancing the detection sensitivity for small metastatic lesions, may require larger training datasets and refined network designs.^{79, 82, 83}

Deep learning also plays a crucial role in identifying the primary sources of BM. Jiao's retrospective study, which included BM patients from various tumors (100 SCLC patients, 125 NSCLC patients, 116 BC patients, and 108 gastrointestinal cancer patients), utilized a three-dimensional residual network (3D-ResNet) to identify the tumor origin. The model could distinguish between LC and non-LC, SCLC, and NSCLC, BC and gastrointestinal cancer using MRI sequences like T1WI, DWI, and CE-T1WI. However, combining MRI sequences such as CE-T1WI + T2WI + DWI improved differentiation between BC and gastrointestinal cancer. but did not accurately distinguish LC from non-LC.⁸⁴

To predict the incidence of LCBM, enable early detection and accurate classification, clinical, pathological, imaging, and other omics data are integrated. A deep learning algorithm has achieved an 87% accuracy rate in predicting BM development, significantly outperforming the average accuracy of four pathologists (57.3%).⁸⁵ Jeong et al.⁸⁶ found that sensitivity for detecting high-risk patients was 95%.⁸⁷ Targeted therapy for LCBM often relies on pathological information from the LC. However, the discrepancies between BM and primary LC histology have led to the use of multitask deep learning networks to predict molecular classifications such as epidermal growth factor receptor (EGFR) wild-type and mutant types.^{88, 89} Additionally, deep learning models for EGFR 19Del/21L858R mutations and wild type have achieved AUC values above 0.97, though precautions are needed to address potential issues such as deceptive strategies and overfitting in AI.⁸⁶

Combining clinical information with MRI-based deep learning facilitates the segmentation of BM gross tumor volume and predicts RT efficacy.^{88, 90-94} Deep learning radiomics and EGFR status are used to predict survival after SRS for LCBM. Combining deep learning with imaging data before and after whole-brain radiation therapy (WBRT), Rammohan et al.⁹⁵ demonstrated that the aging rate of the brain and changes in brain substructure accelerated after WBRT, which are associated with neurocognitive function and could guide clinical treatment. Beyond RT, AI is also employed to predict the efficacy of targeted therapy, immunotherapy, and other treatments.^96-98 Recent advancements include the use of deep learning for intraoperative brain tumor identification and near-real-time diagnosis using simulated Raman histology and deep neural networks.^{99, 100} Radiogenomics, which examines the relationship between genomics and imaging phenotypes, has been instrumental in addressing tumor heterogeneity and predicting immune responses and progression.¹⁰¹ However, unveiling the “black box” of radiomics imaging will contribute to the development of precision medicine research.¹⁰²

In BCBM, AI primarily focuses on predicting the occurrence of BM, identifying the primary lesions, and predicting molecular subtypes. A multivariate logistic regression model using clinical variables at diagnosis achieved AUC values of 0.95, 0.94, 0.77, and 0.61 for BC, melanoma, and NSCLC/SCLC, respectively.^{2, 103} Deep learning models have been used to identify the primary lesions of BM by analyzing anatomical distribution differences.⁴⁴ HER2 status has been predicted with high accuracy based on relative cerebral blood volume, achieving a model accuracy of 0.98.¹⁰⁴ Preoperative brain MRI combined with deep learning has predicted ER, PR, and HER2 status with accuracies of 0.89, 0.88, and 0.87, respectively.^{105, 106} However, Strotzer et al.¹⁰⁷ noted limitations in predicting BM histology based on MRI. Additionally, AI also aids in forecasting prognosis and guiding treatment for BM patients, with the XGBoost model predicting the 6-month to 3-year prognosis for BC patients, achieving AUC values exceeding 0.8.¹⁰⁸ Pandey et al.¹⁰⁹ developed a deep learning framework to optimize SRS dose planning for BCBM patients using multiparametric MRI images.

Despite promising prospects, machine learning and radiomics may not always be reliable. Deep learning exhibits limitations such as deceptive predictions and overfitting.¹¹⁰ The auxiliary role of AI is powerful, but whether it can be independently applied to the diagnosis and treatment of BM remains an open question.⁸⁶ Future development will focus on creating more practical and accurate algorithms, adopting refined imaging techniques, and continuing the integration of medicine and engineering.⁸⁰

In general, diagnosing BM remains relatively straightforward, whether during the initial treatment or throughout the course of the primary tumor. In the era of AI, machine learning is primarily used to predict the occurrence and prognosis of BM, locate primary lesions, and assess treatment responses. However, prediction models’ accuracy can be limited by small sample sizes and noninnovative algorithms. Additionally, the lack of explainability in AI restricts their further application.

3.1 LCBM

The treatment of LCBM mainly include surgery, RT, chemotherapy, targeted therapy, and immunotherapy.^{29, 111} Here, we briefly describe the treatment strategies.

3.2 BCBM

3.3 Differences in efficacy and mechanisms of primary tumor and BM

As a physical and biological barrier, the BBB creates a unique microenvironment for the brain. The efficacy of chemotherapy for BM is relatively poor compared with that for primary tumors, which can be partly attributed to the low intracranial drug concentration caused by the BBB. While therapeutic antibodies were traditionally believed to be unable to penetrate the BBB, but real-world evidence shows that anti-PD-L1 and PD-1 antibodies have therapeutic effects in LCBM.²²⁵ Although there is a moderate consistency in HLA class 1 expression between LC and BM, nearly a quarter of patients exhibit inconsistent HLA expression. Antigen presentation loss may represent one of the many potential mechanisms for inconsistent responses to ICIs therapy.²²⁶ Furthermore, the heterogeneous microenvironment of tumors greatly reduces the therapeutic efficacy of ICIs.²²⁷ Multiple immunofluorescence and spatial transcriptomics studies revealed that ICB reduces a unique population of CD206+ macrophages in the perivascular space, which may regulate T cell entry into BM. Biomimetic codelivery strategies can potentially reverse osimertinib resistance by inhibiting macrophage-mediated innate immunity.²²⁸

LC cells travel long distances and grow within the brain. Tumor cells in BM exhibit distinct characteristics, including specific molecular expressions that influence therapeutic outcomes. For example, cells with high expression of S100A9 may evade osimertinib-induced killing and promote tumor recurrence. Mechanistically, S100A9 upregulates the expression of ALDH1A1 and activates the retinoic acid (RA) signaling pathway.²²⁹ Similarly, the S100A9/RAGE interaction also mediates radiation resistance in LCBM.²³⁰

Metabolic abnormalities are a crucial mechanism underlying chemotherapy resistance in BM. Tumor metabolism, a hallmark of cancer, plays a pivotal role in mediating various therapeutic resistances.²³¹ For instance, GPX4 activates the WNT/NR2F2 signaling pathway by regulating GSTM1, leading to high glutathione consumption and consequent resistance of BM to platinum-based chemotherapy.²³² Warburg originally observed that cancer tissue sections in vitro utilize large amounts of glucose to produce lactate even in the presence of oxygen, a phenomenon known as aerobic glycolysis or the Warburg effect.²³¹ Aldo-keto reductase family 1 B10 (AKR1B10) in BM promotes Warburg metabolism by regulating lactate dehydrogenase, ultimately contributing to pemetrexed resistance in BM.

The brain's unique metabolic feature is the coupling of neurons and astrocytes through glutamate, glutamine, and lactate.²³³ Metabolic pathways, including glycolysis, alanine, aspartate, and glutamate metabolism, as well as arginine biosynthesis, are significantly altered in BM patients.²³⁴ Additionally, there are notable differences in BMs depending on the primary lesion or the timing of metastasis, such as synchronous, latent, and metachronous metastases.²³⁵ More importantly, metabolic disorders related to fat synthesis and decomposition are prevalent in BCBM. Fat synthesis enables tumor cells to adapt to the brain's low-lipid microenvironment, facilitating their colonization and growth.²³⁶ Targeting fat metabolism in BC cell has emerged as an effective treatment strategy.²³⁶

The treatment of BM differs significantly from that of primary tumors. One of the most significant differences is the BBB. The BBB is dynamic during tumorigenesis, and its conversion to the blood–tumor barrier (BTB) is common. As a biophysical barrier, the BBB maintains the relative stability of the intracranial microenvironment but also limits drug penetration. Utilizing new drug carriers, nanotechnology, and physical methods to overcome the BBB may be effective treatment strategies. Additionally, the differences between tumor cells in primary tumors and BM are substantial. Variations in gene expression and metabolism might account for the differences in treatment efficacy between primary lesions and BM. Employing multiomics approaches and preclinical studies to explore the mechanisms of BM will aid in developing potential treatments.

4.1 Cell and animal models

Cell and animal models form the foundational framework in BM research.²⁴² Currently, in vitro studies often utilize commercially available tumor cells, while animal models enhance the credibility of findings in BM research.^{243, 244} Recently, the development of BM cell lines has gained prominence. Valiente et al.²⁴⁵ coordinated 19 laboratories to establish a panel of cells with brain tropism, providing comprehensive insights into experimental models of BM. This collaborative effort has significantly advanced BM as a distinct research field.²⁴⁵ Animal models predominantly involve mice, with additional utilization of rat and zebrafish models.^245-247 Common models involve injecting human cells into immunodeficient mice, utilizing species-specific mouse-derived cells like LLC and 4T1, which facilitates robust platforms for studying BM immunotherapy.²⁴⁵ Orthotopic tumor models in animals typically encompass BC, LC, and melanoma.²⁴⁵ Various inoculation methods include systemic and local approaches.²⁴⁸ Local inoculation, which directly injects tumor cells into the brain parenchyma via a syringe, is straightforward for BM modeling but lacks adequate interaction with the brain microenvironment and fails to replicate the metastatic cascade, diminishing its relevance in BM research.²⁴⁹ Systemic inoculation, involving the introduction of tumor cells via routes such as the tail vein, carotid artery, or intracardiac injection, accurately mimics tumor cells circulation, BBB breach, and colonization in brain parenchyma. Nonetheless, systemic inoculation presents challenges like complexity, incomplete representation of tumor cells escapes from primary sites, potential extracranial tumor formation, and increased animal disease burden.^{248, 250}

Spontaneous BM models require tumor cells to independently complete all metastatic cascade steps from spontaneously arising or orthotopically implanted tumors, faithfully reflecting BM progression.²⁵¹ However, high costs and lengthy timelines restrict their widespread use in BM research. Patient-derived tumor xenograft (PDX) models involve implanting tumor tissue or primary cells from patients into immunodeficient mice to retain parental tumor histopathology, molecular traits, and drug responses.²⁵² Despite these advantages, PDX models face challenges such as low success rates, extended experimental cycles, and species differences, limiting their applicability in BM research.²⁵³

Most in vitro studies in BM employ tumor cell lines derived from primary tumors, which somewhat undermines the robustness of conclusions in BM research. Establishing a cell model of BM often relies on refined animal models. The current research paradigm involves continuously adapting tumor cells using animal models to enhance their propensity for BM. Enhancing animal model construction techniques and employing humanized mice could potentially advance preclinical BM models.²⁵⁴ Systemic therapies like chemotherapy, targeted therapy, and immunotherapy primarily rely on models of primary tumor, with scant exploration into their efficacy against BM in preclinical settings. Local treatment parameters, such as specific RT parameters, play crucial roles in influencing the trajectory of BM research.^{248, 255} Shi et al.²⁴⁸ summarized findings in BM cells and models concerning RT, detailing advancements from model establishment to therapeutic strategies integrating RT. Their work furnishes comprehensive and meticulous data for refining RT-based local treatments for BM.²⁴⁸ Nonetheless, variability in specific RT parameters such as dosage and dose rate in preclinical BM models warrants greater attention.²⁴⁸

4.2 Organoids

Organoids are tissue analogs with defined spatial structures formed by three-dimensional culture of adult or pluripotent stem cells in vitro.⁷⁶ They effectively preserve molecular, cellular, and histological phenotypes of original tumors, thereby maintaining patient-specific tumor heterogeneity, which confers unique advantages in disease modeling and precision tumor therapy.²⁵⁶ However, the complexity of brain tumor biology and the unique brain microenvironment have somewhat hindered the full development of organoid models. Bridging the gap to realistically reflect nerve–tumor interactions observed in patients remains a significant challenge for current research using conventional cell and animal models.²⁵⁷ In 2018, Bian et al.²⁵⁸ pioneered the establishment of an organoid model capable of simulating brain tumors, marking a pivotal advancement. The utilization of fetal tissue brain organoids combined with Crispr-Cas9 technology has further enabled sophisticated brain tumors modeling.²⁵⁹ Despite several years of progress, organoids remain predominantly employed in studying brain tumors such as gliomas and meningiomas.^{257, 260} Recent studies by Qu et al.²⁴⁴ have explored the interaction between SCLC and astrocytes using assembloids composed of SCLC aggregates and human cortical organoids. Fitzpatrick et al.²⁶¹ successfully developed organoids derived from BC patients with leptomeningeal metastasis. Choe et al.²⁶² established a three-dimensional in vitro model that more accurately replicates BM using tumor cells and brain organoids derived from human embryonic stem cells (metastatic brain cancer brain organoids). Quaranta and Linkous²⁶³ expanded on their primary brain tumor models to create an authentic in vitro model of brain development for studying BM. Currently, organoid models are increasingly employed as surrogate models for BM in vitro experiments, drawing from insights gained in brain tumor research, particularly gliomas.²⁶⁴ Understanding the biological underpinnings and the TME specific to LCBM holds the key to further advancing organoid models in this context.^{263, 265}

To simulate the interaction between BC and the brain microenvironment, Wang et al.²⁶⁶ cocultured various BC cells with brain organoids derived from human embryonic stem cells to construct an organoid model. They discovered that MDA-MB-231 and SUM159PT cells could form tumor colonies within human brain tissue.²⁶⁶ Organoid model have also facilitated the development of drug screening platforms, advancing the new treatment strategies for BM.^{261, 267}

4.3 Microphysiological systems

Microfluidic chips represent a powerful technical tool offering advantages such as replicating the in vivo microenvironment, low sample consumption, high automation, and seamless integration.²⁶⁸ Their capability to construct metastasis cascade models is pivotal in cancer research.^{269, 270} For instance, Kim's team developed a three-dimensional microfluidic platform integrating astrocytes, brain endothelial cells (BECs), and patient-derived NSCLC cells to simulate BM.²⁷¹ Cancer metastasis accounts for 90% of cancer-related deaths, underscoring the importance of constructing metastasis cascade models in microfluidic chips to study vascularization, tumor cells invasion, and simulate processes like intravasation and extravasation.^{261, 272} CTCs play a critical role in mediating tumor metastasis. CD44+CD74+ CTCs are prevalent in BM patients and serve as effective indicators for diagnosing BM.²⁷³ Microfluidic technology facilitates the enrichment of CTCs, enhancing our understanding of their role in metastasis.²⁷⁴ Moreover, microfluidic chips are frequently employed to simulate the BBB.^{271, 275} For instance, Lim et al.²⁷⁶ developed a choroid plexus-on-a-chip utilizing oscillatory flow to mimic the human brain choroid plexus, offering a novel model for studying BM.

Using microfluidic technology, Lim et al.²⁷⁶ integrated features such as capillaries, epithelial layers, and secretory components to accurately simulate the characteristics and dynamics of the human brain choroid plexus.

4.4 Hydrogel model

The hydrogel model serves not only to study the interaction between BC cells and the extracellular matrix (ECM) but also to reversibly simulate the dormancy of BC cells in the brain.^277-281 Yakati et al.²⁸² cultured BC cells on soft hyaluronic acid (HA) hydrogels (0.4 kPa), which mimicked a dormant phenotype, and on stiff HA hydrogels (4.5 kPa), which mimicked a proliferative phenotype. They found that cells on soft HA hydrogels exhibited chemotherapy resistance through the p38–SGK1 signaling pathway.²⁸²

Model construction of BM is complex and challenging, which hinders the progress of preclinical research to some extent. Currently, the most widely used models in preclinical studies are cell and animal models. Cell models are primarily derived from brain tropism cells. The methods for establishing animal models are diverse, with researchers often weighing the pros and cons based on specific research goals. Additionally, the organoid model, despite its difficulty and cost, is noteworthy because it closely resembles clinical reality and can significantly enhance precise, individualized treatment. Moreover, tumor dormancy is a distinctive feature of BM. The use of hydrogel models to simulate BC dormancy has also advanced preclinical research. Additionally, Li et al.²⁸³ developed and validated a physiologically based pharmacokinetic model to predict plasma and central nervous system pharmacokinetics using a four-compartment permeability finite brain model.

5.1 Bulk RNA transcriptome

Bulk RNA transcriptome analysis has been a powerful tool in elucidating molecular mechanisms, past and present, owing to its cost-effectiveness. Also, it assists in exploring the TME of BM (Figure 3A). After animal models simulate LC metastasis and obtain tissues or metastatic cells of distant metastasis (such as lymph node metastasis, bone metastasis, and BM, etc.), bulk RNA transcriptomics revealed that miR-660-5p may be a key driver molecule of NSCLC and distant metastasis.²⁸⁴ Similarly, the miR-17-5p/HOXA7 axis may induce LCBM through ferroptosis.²⁸⁵ Finding key molecules that drive BM, such as miRNAs, lncRNAs, and circRNA, is also a hot topic.^{199, 286-288} TCR sequencing by Zhou and Chen's²⁸⁹ team showed a unique pattern of stronger oligoclonal T cell expansion, weakened CD8+TIL infiltration in BM, and CD8+TIL was an independent positive indicator of OS.

Driver gene mutations represent a major characteristic of NSCLC. Studies have consistently demonstrated that patients harboring these mutations exhibit poorer responses to immunotherapy. Consequently, gaining a deeper understanding of the immune microenvironment in NSCLC across different driver gene subtypes has become a forefront research area. Zhou's team conducted a comprehensive analysis of the TME in EGFR/ALK-positive/-negative LCBM.³¹ They observed a decrease in CD8+ T cells and cytotoxic lymphocytes alongside an increase in M2 macrophages, which collectively contribute to an immunosuppressive TME.³¹ This disparity underscores why patients with positive driver genes experience limited efficacy with immunotherapy. Moreover, compared with EGFR wild-type malignant adenomas, EGFR-mutated LCBM exhibit upregulation of multiple immune-related pathways.²⁸⁷

Although obtaining paired paraffin-embedded sections of LC lesions and LCBM presents challenges, such specimens are crucial for convincingly illustrating the differences between primary tumors and metastatic lesions.^{31, 290, 291} Chen et al.²⁹² delved into this by analyzing paired specimens from 43 patients, revealing significant immune heterogeneity between synchronous and metachronous LCBM. The TME is more complex in the metachronous group.²⁹² Additionally, conflicting findings regarding expressions and correlations of PD-L1 between BM and LC exist.^{289, 292, 293} While PD-L1 expression is generally low in LCBM, its relationship with the time interval of BM in NSCLC has been reported.²⁹⁰ Tsakonas et al.²⁹⁴ conducted a study using 25 paired pathological specimens to identify differentially expressed miRNAs in LCBM, aiming to uncover potential biomarkers and therapeutic targets.

Similar to LC, one of the current research trends is to discover metastasis-related genes and predict the TME and immune checkpoints through bulk RNA transcriptomics. Xiao et al.²⁹⁵ confirmed that oxidative phosphorylation (OXPHOS) utilization is increased in BCBM patients and that the immune microenvironment is inhibitory. Based on the bulk RNA-seq, B7-H3 was identified as a checkpoint for T cell immunosuppression. B7-H3 expressed in 90% of BCBM but relatively low in BMs from colorectal and renal cancers, indicating it may be a potential target for immunotherapy.²⁹⁶ Zhang et al.²⁹⁷ used transcriptomics to explore the mechanisms of organ-specific metastasis and demonstrated that the neuroactive ligand–receptor interaction pathway plays a significant role in BCBM.

5.2 Single-cell transcriptome

Single-cell transcriptome, in simple terms, is a technology that sequences and analyzes the genome, transcriptome, and proteome of individual cells. It addresses challenges such as low sample quantity and cell variability, which are particularly relevant in research areas like stem cells, cancer, and immunity.^{298, 299} In BM research, single-cell transcriptome identifies specific LC cells driving metastasis to the brain, offering potential therapeutic targets for LCBM (Figure 3B). Wang et al.³⁰ utilized 11 LUAD and 10 BM samples to identify S100A9+ BM-associated epithelial cells through pseudotime trajectory analysis, predicting BM risk with machine learning algorithms. Chen et al.³⁰⁰ employed single-cell RNA sequencing (scRNA-seq) on seven patients, highlighting changes in adhesion, ECM, and VEGF pathways in RAC1-high LC cells which maybe a promoter of BM and potential therapeutic targets. Liang et al.³⁰¹ characterized the TME via single-cell transcriptomics, revealing a higher presence of malignant epithelial cells in LCBM, particularly those expressing high levels of DDIT4, which are associated with increased invasiveness. Wu et al.³⁰² demonstrated that BM-associated epithelial cells expressing SPP1, SAA1, and CDKN2A originate from aneuploid LC cells, mediating BM occurrence. Ongoing research continues to uncover additional genes driving BM, such as CKAP4, SERPINA1, SDC2, and GNG11, offering promising targets for treatment.²⁹⁷ Single-cell transcriptomics uniquely enables pseudo-time-series and cell-to-cell communication analyses, providing critical insights into how BM interacts with other cells in the TME.²⁶⁶

ScRNA-seq can explore the developmental trajectory of tumor metastasis, uncover gene functions, and provide a comprehensive understanding of metastatic development in BCBM. Xie et al.³⁰³ identified ILF2 as specifically related to BCBM and a possible therapeutic target through scRNA-seq. Zou et al.³⁰⁴ conducted scRNA-seq on mouse lesions of BCBM, identifying microglia markers and demonstrating the conservation of their proinflammatory response. Their combination of scRNA-seq and multiple immunofluorescences confirmed the presence of immunosuppressive cells, such as FOXP3+ Tregs and LGALS1+ microglia, in the BCBM microenvironment.³⁰⁴ Additionally, integrating single-cell and bulk RNA transcriptomics, previously unrecognized immune regulatory subtypes enriched in BMs were identified, representing potential targets for new therapeutic strategies.³⁰⁴ They also found that the PD-1 receptor and PD-L1/2 ligand may not be the primary immune checkpoint signaling pathways in liver and brain metastases of BC.³⁰⁴ Interestingly, Bejarano et al.³⁰⁵ focused on endothelial cells and parietal cells in the vascular system.

5.3 Spatial transcriptome

Spatial transcriptome analyzes gene expression data with spatial resolution, providing simultaneous insights into the spatial organization and transcriptional profiles of cells. Named the “technology of the year” by Nature Methods in 2020, this advanced technique has become indispensable in fields such as tumor biology, immune infiltration, pathology, and disease mechanisms.³⁰⁶ In LCBM, spatial transcriptomics has enabled a more detailed characterization of the TME.³⁰⁷ Zhang et al.³² utilized spatial transcriptomics to identify regions of immunosuppression and fibrosis (Figure 3C). Specifically, they observed decreased presence of antigen-presenting cells, B cells, and T cells, along with increased numbers of neutrophils, M2 macrophages, immature microglia, and reactive astrocytes.³² These findings align with similar conclusions drawn by Liang et al.³⁰¹ in their single-cell transcriptome study, which highlighted increased myofibroblast cancer-associated fibroblasts (CAFs) and enhanced angiogenic capabilities of endothelial cells.^{32, 289, 301}

5.4 Genomics

Cancer genomics contribute to explore the molecular mechanisms underlying carcinogenesis, tumor heterogeneity, classification, and personalized treatment strategies.³⁰⁸ Despite its critical role in understanding metastatic progression, which is the primary cause of cancer-related deaths, the genomic mechanisms driving metastasis remain poorly understood (Figure 3D).³⁰⁹ Nguyen et al.³⁰⁹ conducted a prospective genomic analysis of 25,000 patients, revealing distinct genomic characteristics in BM originating from different primary tumors such as LC, prostate cancer, and BC. LUAD BM were found to exhibit a higher frequency of TP53 mutations, TERT amplification, and EGFR mutations, but a lower frequency of RBM10 mutations.³⁰⁹ In LCBM, Huang and Smyth confirmed frequent genomic alterations in EGFR, KRAS, TP53, RB1, MYC, CDKN2A, CDKN2B, STK11, NKX2-1, KEAP1, MAPK, PI3K, mTOR, and cell cycle-related pathways.^{310, 311} A systematic review and meta-analysis involving 9058 NSCLC patients identified ALK positivity and RET translocation as the most prevalent genomic alterations, accounting for 34.9 and 32.2%, respectively.⁴ The genomic profile may indicate a predisposition for tumor metastasis. For instance, KRAS/NRAS mutations are particularly associated with brain and bone metastases, whereas PD-L1 expression and TP53 mutations may influence metastasis to other organs in NSCLC.^{287, 312} Analysis of 3600 cases of SCLC BM revealed frequent alterations in PTEN.³¹³

Chromosomal instability is closely associated with the metastatic burden of LUAD and BC.³⁰⁹ Genome sequencing has revealed that most mutations are acquired before metastasis.³⁰⁹ For instance, new copy number events in the MCL1 gene were observed in 75% of cases, may promote BM.³¹⁴ Deletions of CDKN2A/B and alterations in the cell cycle pathway were also enriched in BM.³¹⁵ Recent evidence suggests that LCBM exhibit increased genomic instability and complexity, potentially due to deficiencies in homologous recombination or other mechanisms involving somatic copy number changes during metastasis.³¹⁶ However, whole exome sequencing has shown that the overall mutational signature remains relatively consistent between primary tumors and BM. Key driver genes such as EGFR and TP53 appear to be largely conserved across these tumor stages.³¹⁷ Furthermore, genome sequencing has delineated distinct mutational signatures in synchronous versus metachronous LCBM, underscoring the potential benefits of CSF biopsy where alterations in EGFR and TP53 are frequently observed.³¹⁸ Deng et al.³¹⁹ also highlighted the significance of CSF genomics.

During the progression of BC, primary tumors and BMs often exhibit discordant genetic mutations.^{105, 320-322} Comparisons between BMs and primary tumors have revealed genes with increased mutation frequency in BMs, including TP53, ATR, and APC (in LUAD); ARID1A and FGF10 (in SCLC); PIK3CG, NOTCH3, and TET2 (in lung squamous cell carcinomas); ERBB2, BRCA2, and AXL1 (in BC).³²³

Bhogal et al.³²⁴ analyzed 822 BCBM tissues and identified nine common structural rearrangement gene mutations, with CDK12 being the most prevalent. Lu et al.’s³²⁵ genomic mutation analysis of the Chinese population revealed that somatic mutations in TP53 (82%), PIK3CA (35%), and MLL2 (22%) were the most common. The most frequent copy number alterations were HER2 (64%), RAD21 (36%), and CCND1 (32%).³²⁵ Genome detection by ctDNA showed that ESR1 and BRCA2 were more commonly mutated in BCBM patients.⁶⁹ Huang et al.³²⁶ conducted a meta-analysis of genomic information from 37,218 BCBM patients and confirmed that the mutation incidence of ESR1, ERBB2, EGFR, PTEN, BRCA2, and NOTCH1 in BM patients was significantly higher compared with those with extra-brain metastases.³²⁷

Mutated genes in BMs are primarily involved in regulating gene transcription, cell cycle, and DNA repair processes.³²¹ Genome sequencing of BMs from various primary tumors revealed mutations in genes associated with the cyclin-dependent kinase (CDK) pathway, including CDKN2A, RB1, CCND2, and CCND3, as well as genes related to the PI3K–AKT–mTOR and MAPK pathways, such as BRAF, KRAS, and MAP2K1.^{328, 329} Additionally, potential targets for BCBM, such as PARP and ATM, also offer new therapeutic prospects.^{321, 329, 330}

5.5 Metabolomics

Metabolically flexible disseminated tumor cells utilize nutrients from distant organs to sustain their survival and growth (Figure 3E).³³¹ This adaptability underscores the complex metabolic alterations that occur during metastasis. In a study examining metabolome changes in 88 BCBM patients, significant alterations were observed in metabolic pathways, including alanine, aspartate, and glutamate metabolism, as well as arginine biosynthesis. A predictive model for BM occurrence constructed using 15 metabolites achieved a remarkable accuracy of 96.6%.²³⁴ Bulk RNA-seq further revealed that genes related to metabolic stress pathways, particularly those involved in glycolysis and OXPHOS, were upregulated in BCBM and LCBM compared with their primary cancers.^{295, 332} This suggests that these metabolic pathways play a crucial role in the progression and sustenance of metastatic cells.

Metabolomics offers insights into the timing of metastatic formation. HER2+ BC patients can present with synchronous, latent, or metachronous BMs. Parida et al.²³⁵ explored the metabolic differences among these patterns. In synchronous BMs, lactate secreted by invasive metastatic cells was found to hinder innate immune surveillance and promote metastasis.²³⁵ Conversely, latent or metachronous BMs exhibited metabolic adaptations such as oxidizing glutamine, and maintaining cellular redox homeostasis via the anionic amino acid transporter xCT.²³⁵ Further experiments indicated that fragmented mitochondrial puncta in latent BM cells oxidize fatty acids to support cellular bioenergetics and redox homeostasis, facilitating metabolic reprogramming for survival.³³¹ Additionally, RA receptor responder 2, a multifunctional adipokine and chemokine, is downregulated in TNBC cells and regulates rapamycin levels and triglyceride levels through the PTEN–mTOR–SREBP1 signaling pathway.³³³

Future research trends emphasize multiomics approaches to unravel metastasis cascades and develop therapeutic targets.³³⁴ Integrating whole exome/genome sequencing with RNA sequencing to investigate the immune microenvironment and immune checkpoint composition of tumors is essential for advancing treatment strategies.³³⁵

5.6 Proteomics

Proteomics has proven to be a valuable tool in identifying biomarkers for predicting the development, prognosis and therapeutic efficacy of BM (Figure 3F). Li et al.³³⁶ utilized plasma exosome proteome sequencing to analyze the protein composition of exosomes in LCBM and revealed significant heterogeneity between SCLC BM and NSCLC BM. They discovered that exosomal proteins in LCBM are predominantly calcium-dependent/S100 proteins.³³⁶ Deng et al.³³⁷ developed a proteomic-based predictive model for BM occurrence in patients with EGFR mutations, achieving an impressive AUC of 0.9401. Integrating proteomics with other datasets enhances our understanding of the BM microenvironment. For instance, [64Cu] [Cu (ATSM)] PET imaging combined with proteomic analysis revealed changes in protein expression related to hypoxia and oxidative stress in BM.³³⁸ High levels of proteins such as NES, ALdh6a1, Cathepsin F, and Fibulin-1 were identified as emerging diagnostic markers.^{339, 340}

Quantitative proteomic analysis based on high-resolution mass spectrometry has revealed that tenascin C levels are significantly elevated in young mice with BM. This elevation promotes tumor cells proliferation and migration, which may explain increased propensity for BM observed in younger BC patients.^{341, 342}

5.7 DNA methylation sequence

Epigenetics is believed to be closely linked to the onset, progression, and treatment response of BM.³⁴³ DNA methylation, a form of chemical modification that alters gene expression without changing the DNA sequence, plays a significant role in these processes (Figure 3G). Distinct methylation patterns are observed in primary LC tissues with and without BM. A comprehensive analysis of methylation profiles across normal lung tissue, primary LC, and LCBM has revealed that methylation patterns, such as H3k9me3 and bivalent marks like H3k27me3 and H3K4me1 may drive disease progression and worsen prognosis in BM.³⁴⁴ Furthermore, a model utilizing DNA methylation data achieved an impressive AUC value of 0.94 for predicting the risk of BM.³⁴⁵

Methylation array analysis revealed that hypermethylation of RP11-713P17.4, MIR124-2, and NUS1P3, as well as hypomethylation of MIR3193, CTD-2023M8.1, and MTND6P4, may be linked to the BM metastasis of BC.³⁴⁶ Additionally, dysregulation of DNA methylation has been observed in both primary tumors and BMs.^{346, 347} Moreover, experiments have shown that the positive rate of the m6A reader IGF2BP3 increases with BC progression and is significantly associated with BCBM.³⁴⁸

5.8 Microbiomes

Microbiomes have emerged as novel tumor markers and a focal point in cancer research over the past decade. The microbiome, comprising intestinal microorganisms, other mucosal organ microorganisms, and intratumoral microorganisms, influences tumor progression through modulation of tumor growth, inflammatory responses, immune evasion, genomic stability, and resistance to treatment.³⁴⁹ The concept of the gut–brain axis has spurred investigation into the microbiome's role in BM (Figure 3H).^{350, 351} Dong et al.³⁵² investigated alterations in intestinal and sputum microbiota in NSCLC with distant metastasis, revealing distinct microbial compositions across different metastatic sites, notably elevated Pseudomonas aeruginosa levels in BM patients. Jiang et al.³⁵³ analyzed the intestinal microbiome and fecal short-chain fatty acid levels in healthy individuals, early LC patients, and those with LCBM. They identified reduced Firmicutes and Actinobacteria associated with altered short-chain fatty acid content, suggesting an impact on lipid metabolism and possibly influencing LCBM occurrence.³⁵³ A model based on the microbiome for distinguishing BM patients achieved an impressive AUC of 0.88.³⁵³ Nonetheless, the microbiome remains a cutting-edge and underexplored area in BM research. Most findings are presented in microbiome maps, and the underlying mechanisms yet to be fully elucidated. Robust evidence is still needed to clarify the microbiome's exact role in BM.

5.9 Glycomics

Glycomics is a discipline focused on studying the structure and function of sugar chains. As a new field emerging after genomics and proteomics, glycomics offers fresh insights into research areas such as diseases, cancer, and immunity.³⁵⁴ Alterations in glycosylation play a crucial role in BC development (Figure 3I). Onigbinde et al.³⁵⁵ reported that the isomers of the O-glycan structure HexNAc1Hex1NeuAc1 exhibited significant changes across BC cell lines CRL-1620 and the BCBM cell line MDA-MB-231BR. These findings highlight the potential of glycomics in understanding BC progression and BM.

The integration of various omics, leveraging their complementary advantages, represents a significant avenue to unravel the invasion patterns and immune landscapes for BM.^356-358 By elucidating the key molecules and cell subpopulations driving BM, monitoring the dynamics of metastatic tumor cells, and exploring interactions between tumor cells and components within metastatic lesions, we can gain insights into the mechanisms of BM and advance clinical applications.²⁶⁶ Comparative analyses between primary tumors and BM, different metastatic sites, intracranial primary tumors, and synchronous/metachronous BM provide multifaceted perspectives essential for unraveling the complexities of BM. Currently, joint analyses of bulk RNA transcriptomes and publicly available single-cell transcriptomes are mainstream, facilitating comprehensive investigations. TP53 mutations were notably more prevalent in BM from LUAD.³⁰⁹ Alvarez-Prado et al.³³⁵ employed whole exome/genome sequencing and RNA sequencing of immune cell populations to delineate the immune genomic landscape in LCBM and BCBM. This study highlighted the impact of TP53 mutations on immune profiles, including increased tumor mutational burden and neoantigens, enhanced tumor proliferation, and altered immune cell infiltration patterns.³³⁵ Moreover, despite a modest overlap in differentially expressed genes between BM cell models and patient tissues, validating the relevance of these models remains essential for faithfully representing BM patients’ realities in preclinical settings.³⁵⁹ Exploring solutions such as organoid models may enhance fidelity and relevance in preclinical research. Ultimately, enhancing interdisciplinary communication, fostering collaborations spanning preclinical to clinical data integration, and advancing understanding of the biological underpinnings of BM are critical pursuits. These efforts aim to leverage patient-derived transcriptomic, proteomic, and metabolomic profiles to elucidate diverse prognostic subtypes and inform personalized therapeutic strategies.

6.1 Break away from the original tumor and break through the ECM

Gene expression in primary tumors plays a crucial role in the development of BM.³⁵⁹ Researchers are actively identifying cell subtypes that drive LC metastasis. High expression of RAC1 are pivotal in adhesion, ECM interactions, and VEGF signaling pathways, ultimately contributing to BM.³⁰⁰ Amplification of PMS2 influences thiamine, butyrate, and glutathione metabolism, promoting the formation of LCBM.³⁶⁰ Conversely, downregulation of CERS1 in LCBM enhances tumor growth via the PI3K/AKT/mTOR pathway.³⁶¹ Studies have shown that MUC5AC is linked with increased epithelial–mesenchymal transition (EMT), invasiveness, and metastatic potential in tumors.³⁶² Tumor cells with heightened angiogenic capacity facilitate nutrient supply to both primary LC and BM (Figure 4A①).³⁶³ Interestingly, Levallet et al.³³³ proposed that hypoxia induces activation of the RASSF1A/kinases Hippo pathway, thereby enhancing YAP and NDR2 in human bronchial epithelial cells, which exacerbates BM formation (Figure 4A②). Patients harboring EGFR mutations frequently develop LCBM. Li et al.³⁶⁴ demonstrated that EGFR mutations promote BM through the ERK1/2–E2F1–WNT5A axis (Figure 4A③). LGALS8–AS1 targets miR-885-3p to regulate the expression of fascin actin-binding protein 1 (FSCN1), ultimately contributing to BM formation (Figure 4A④).³⁶⁵

BC cells facilitate tumor metastasis by regulating EMT and ECM remodeling.^366-368 Hyaluronidase Hyal-1 contributes to tumor proliferation by forming HA coating around the cells (Figure 4B①).³⁶⁹ The secretory iron transporter Lipocalin-2 (LCN2) promotes tumor growth and enhances ECM reorganization, angiogenesis, and BBB disruption through interaction with MMP9 (Figure 4B②).^{360, 370} The loss of Kmt2c or Kmt2d in BC cells results in reduced H3K27me3, which upregulates MMP3 via KDM6A, thereby facilitating BM formation (Figure 4B③).³⁷¹ Additionally, the circRNA–miRNA–mRNA regulatory axes, such as circ_0087558/miR-604/MMP2, circbcBM1/miR-125a/BRD4, and circKIF4A–miR-637–STAT3, may influence the development of BCBM.^372-374

Tumor-initiating cells represent a rare but crucial subpopulation within tumors, characterized by stem-like properties that drive tumor initiation, metastasis, and resistance to chemotherapy.³⁷⁵ RNA sequencing have revealed common transcriptomic features among BM-initiating cells from lung, breast, and melanoma origins. In paired specimens of primary LC and BM, elevated HLA-G expression has been observed. HLA-G interacts with adjacent tumor cells via SPAG9/STAT3 signaling to enhance their self-renewal and growth capabilities (Figure 4A⑤).^{291, 376} Additionally, CD44+ LC stem cells differentiate into vascular pericytes, facilitating the formation of metastatic niches and transmigration across endothelial barriers under the influence of G protein-coupled receptor 124 (GPR124) (Figure 4A⑥).³⁷⁷ Smoking has been identified as a significant risk factor for BM. Tyagi et al.³⁷⁸ demonstrated that nicotine induces neutrophils to release exosomal miR-4466, thereby enhancing the stemness of LC cells and promoting their invasive properties in BM (Figure 4A⑦). Additionally, a high expression of NDRG1 is associated with stemness and is linked to BCBM (Figure 4B④).³⁷⁹

Although proteomics has identified ECM receptor interactions and collagen-containing ECM as crucial in LCBM, research into how tumors penetrate this ECM remains limited.^{336, 380} HA is a key component of the ECM and plays a significant role in the metastasis of various tumors. Tumor cells mediated by HSP47 deposit collagen in the metastatic niche, which can inhibit the antitumor immune response.³⁸¹ Zhao et al.³⁸² found that plasma HA levels are associated with bone metastasis in LC. CAFs are generally considered to promote tumor metastasis by influencing the ECM. However, their role in BM appears contradictory, and more robust evidence is needed to clarify their impact.^{32, 98, 289, 301, 383} Nevertheless, due to challenges in extracting and studying CAFs, mechanistic investigations involving these cells remain relatively complex.

6.2 Transfer process

The process of LC metastasis to the brain is diverse, involving a lengthy journey in terms of both time and distance. It is widely accepted that the lungs, as highly vascularized organs, facilitate the passage of tumor cells to the brain through systemic circulation.³⁸⁴ In current research, there is a predominant focus on understanding the mechanisms underlying primary and metastatic lesions, whereas tracking changes specifically in BM throughout the metastatic process remains challenging. In recent years, the advent of liquid biopsies, particularly the detection of CTCs, has not only proposed a diagnostic role for CSF CTCs in BM but also enabled the tracking of the metastatic process and prediction of treatment efficacy.^{385, 386} Technologies like single-cell sequencing have furthered our ability to identify stage-specific markers in the malignant progression and metastasis of LC, which could potentially aid in intervening at various stages of metastasis to prevent or manage BM.³⁸⁷

6.3 Interaction with the BBB and colonization growth in brain

6.3.1 LCBM

Extravasation of tumor cells across the BBB represents a critical step in the formation of BM.³⁸⁸ In the vasculature of glioblastoma and BM, genes associated with cell proliferation, angiogenesis, and ECM deposition are dysregulated, leading to significant disruptions in the integrity of physical and biochemical barriers.^{389, 390} CD44+ LC stem cells differentiate into perivascular cells, form metastatic niches, and traverse the endothelium mediated by GPR124. Mesothelin (MSLN), a tumor-associated antigen expressed in various solid tumors, has been linked to the progression of multiple cancers. Xia et al.²⁷² demonstrated that MSLN enhance tumor cells to penetrate the BBB and promote BM through activation of the MET pathway via the c-Jun N-terminal kinase signaling pathway (Figure 4C①). Exosomal LINC01356, miR-375-3p, lnc-MMP2-2, or CXCR4+ tumor cells derived from LCBM suppress tight junction proteins of the BBB (claudin-5, occludin, and ZO-1), thereby promoting BM (Figure 4C②).^391-394

In contrast to the microenvironment of extracranial metastases, the BM microenvironment contains distinct cell types, primarily astrocytes, microglia, oligodendrocytes, and neurons.³⁹⁵ The interaction between tumor cells and the brain microenvironment facilitates breaches of the BBB, colonization, and growth. Reactive astrocytes and tumor-associated macrophages (TAMs) are pivotal in NSCLC BM, influencing tumor progression and immune evasion. Gonzalez et al.³⁹⁶ proposed two functional prototypes of BM: proliferative and inflammatory. Proliferative BM archetypes are characterized by DNA replication, G2/M phase transition, and pre-mRNA maturation or spliceosome signatures, whereas inflammatory BM archetypes are marked by inflammatory responses, ECM remodeling, and stress responses (Figure 5C).³⁹⁶

Astrocytes, the predominant glial cells in brain, play dual roles in different phases of BM. Initially, as integral components of the BBB, astrocytes act to impede tumor cell invasion.³⁹⁷ However, once tumor cells breach this barrier, astrocytes are exploited by tumor cells to facilitate metastatic proliferation.^{398, 399} Through mechanisms involving gap junctions, exosomes, and other pathways, astrocytes upregulate tumor-related signaling, modulate the immune microenvironment, and enhance tumor metabolism, angiogenesis, and growth, thereby promoting BM.^{33, 397, 400, 401} Studies have confirmed that STAT3+ astrocytes foster BM in LC, BC, and melanoma by influencing both innate and adaptive immune responses.⁴⁰⁰ Recently, Dankner et al.⁴⁰⁰ demonstrated that aggressive BM proliferation correlates with CHI3L1 released from p-STAT3+ astrocytes (Figure 5A①). In SCLC, reelin secretion stimulates astrocyte recruitment, prompting astrocytes to secrete SERPINE1 and other proteins that enhance SCLC growth in brain (Figure 5A②).²⁴⁴ Emerging research indicates that tumor cells induce astrocytes to upregulate MUC5AC, thereby promoting enhanced colonization (Figure 5A③).³⁶²

TAMs and microglia play crucial roles in BM. Microglia, an integral part of the brain microenvironment, are particularly significant in LCBM compared with other forms of cancer.^{356, 402} These microglia exhibit distinct phenotypes: M1-type microglia promote inflammation, tissue damage, antigen presentation, and tumor cells killing, whereas M2-type microglia support anti-inflammatory responses, tissue repair, angiogenesis, immune suppression, and tumor progression.^{402, 403} Communication between tumor cells and microglia involves pathways such as DLL4–NOTCH4 and MIF–CD74, yet the full spectrum of their interactions in brain colonization remains elusive.^{356, 404} IL6 emerges as a pivotal mediator, with high serum levels correlating with increased BM risk by enhancing JAK2/STAT3 signaling in microglia to induce an anti-inflammatory state conducive to tumor cells colonization (Figure 5B①).⁴⁰⁵ IL6 also impedes T cell function in BM and contributes to ICIs resistance.³⁵⁸ Early BM cells expressing IFITM1 stimulate microglial activation, augmenting CD8+ T cell cytolytic activity against tumor cells (Figure 5B②).⁴⁰⁶ Furthermore, microglia express genes encoding ECM and matrix proteins, influencing both pro- and antimetastatic properties.³⁵⁸ Manipulating the M1/M2 microglia ratio represents a promising strategy for BM treatment, given that factors like extracellular vesicles and radiation can induce microglial transformation toward the M2 phenotype, facilitating metastasis.⁴⁰⁷ The HSP47–collagen axis contributes to M2 polarization via the α2β1 integrin/NF-κB pathway, increasing anti-inflammatory cytokine production and inhibiting CD8+ T cell responses (Figure 5B③).³⁸¹ Additionally, Klemm et al.³⁵⁸ underscored the involvement of cathepsins CTSB and CTSW in monocyte-derived macrophages, promoting invasion and metastasis in LCBM (Figure 5B④). Inhibiting the anti-inflammatory macrophage phenotype has shown promise in reducing tumor growth in mouse models of BM.⁴⁰⁸ In primary brain tumors like glioblastoma, macrophages exhibit a dual role as both antitumor agents and promoters of tumor progression, highlighting the complexity of their function in brain microenvironments.^409-411 The precise roles and regulatory mechanisms of macrophages in LCBM warrant further investigation.

In typical solid tumors, CAFs are often implicated in promoting metastasis through their interactions with the ECM. However, in LCBM, their role appears paradoxical and requires further investigation to elucidate their precise contributions. Studies by Wu et al.⁹⁸ have suggested a lack of inflammatory-like CAFs in LCBM, contrasting with the enrichment of pericytes. This observation indicates a distinct composition of the stromal cells within the BM microenvironment compared with other tumors. Zhang et al.³² utilized spatial transcriptomics to confirm fibrotic niches in LCBM, while Liang et al.³⁰¹ highlighted an increase in myofibroblast-like CAFs and enhanced angiogenic capacity of endothelial cells in these settings.²⁸⁹ These findings underscore the heterogeneity and dynamic nature of CAF populations in BM. Hypoxia has been identified as a key mediator inducing a lineage transition in CAFs mediated by HIF-2α, leading to enhanced angiogenesis, metabolic reprogramming, and promotion of tumor growth (Figure 5B⑤).³⁸³ Despite these insights, the difficulty in extracting and studying CAFs poses challenges for detailed mechanistic investigations. Single-cell transcriptome analyses have further illuminated the role of fibroblasts in BM, revealing their high expression of type I collagen genes and significant involvement in cell–cell interactions within the TME.³⁸³ This highlights their potential influence on ECM dynamics and cellular interactions critical for metastatic progression. In addition to CAFs, interactions between tumor cells and brain-specific microenvironmental components such as astrocytes and microglia play pivotal roles in BM. For instance, melanoma-derived Aβ has been shown to activate astrocytes, promoting a prometastatic, anti-inflammatory phenotype that supports tumor survival by inhibiting microglial phagocytosis.⁴¹²

Overall, the complexity of the BM microenvironment underscores the need for comprehensive studies to elucidate the specific molecular mechanisms and key cell populations involved in tumor cells interactions and adaptation within the brain. Clarifying these interactions holds promise for developing targeted therapies that could potentially disrupt metastatic processes specific to the brain, improving outcomes for patients with metastatic disease.

6.4 Metastatic niche and immune regulation

After traversing a long journey, tumor cells arrive at distant sites and integrate into new environments, establishing a premetastatic niche.^{380, 441} This process involves complex interactions between tumor cells and various components of the microenvironment, ultimately facilitating their survival and growth in brain. Tumor cell-derived exosomes play a crucial role in orchestrating the formation of this premetastatic niche. These exosomes can stimulate astrocytes to secrete cytokines such as IFN-γ, IL-3, IL-5, and IL-15. These cytokines contribute to creating an inflammatory microenvironment that supports tumor cells survival and fosters an immunosuppressive milieu that evades immune surveillance.⁴⁴² Research by Gonzalez et al.³⁹⁶ has highlighted similarities in the metastatic niches originating from different primary tumors, suggesting that common mechanisms may underpinning the adaptation of tumor cells to the brain microenvironment, regardless of their tissue of origin. Furthermore, myeloid cells, including metastasis-associated macrophages and dendritic cells, play pivotal roles in shaping the metastatic niche.^{396, 443} For instance, the loss of CSCL3 in myeloid cells leads to increased expression in CXCL10, promoting the recruitment of VISTA+ PD-L1+ CNS-native myeloid cells to LCBM. This recruitment creates an immunosuppressive niche that facilitates tumor immune evasion and growth.⁴⁴³ Additionally, CD44+ LC stem cells can differentiate into perivascular cells within the brain microvasculature, contributing to the formation of a metastatic niche and aiding tumor stem cells in crossing the endothelial barrier and establishing themselves within brain tissue.³⁷⁷ Overall, the formation of a premetastatic niche involves intricate interactions between tumor cells, astrocytes, myeloid cells, and other components of the brain microenvironment. Understanding these interactions at a molecular and cellular level is critical for developing targeted therapies aimed at disrupting the metastatic process and improving outcomes for LCBM patients.

The interaction between tumor cells and brain cells contributes to a microenvironment that supports BC growth by secreting factors such as ERH, RPA2, S100A9, and nerve growth factor inducible (VGF). Ahuja and Lazar⁴⁴⁴ investigated these secreted factors in vitro and confirmed that brain cells release factors that create an inflammatory environment. Concurrently, BC cells secrete factors like ERH, RPA2, and S100A9, which are not normally present in the brain, thereby enhancing tumor proliferation.⁴⁴⁴ Additionally, VGF is linked to HER2 overexpression and TNBC characteristics, which are associated with poor prognosis in BCBM patients. BC cells secrete VGF to disrupt the BBB and activate microglia.⁴⁴⁵ Turker et al.⁴⁴⁶ examined the interaction between BC cells and the ECM of brain cells using hydrogels of thiohyaluronan. Their results demonstrated that BC cells form multicellular aggregates within the ECM, which helps sustain their activity.⁴⁴⁶

Immune escape may be influenced by intratumor genetic heterogeneity.³¹⁶ For instance, EGFR-mutated NSCLC is more prone to BM, potentially due to immune escape mechanisms. Compared with EGFR/ALK-negative LCBM, the immune microenvironment of EGFR/ALK− LCBM is more suppressive.³¹ Furthermore, Tang et al.⁴⁰¹ demonstrated that EGFR-mutated LC cells induce reactive astrocytes to secrete IL-11, which leads to tumor cells PD-L1 expression and CD8+ T lymphocyte apoptosis.⁴⁰¹ Meanwhile, IL-11 also acts on GP130 to assist tumor cells in immune escape.⁴⁰¹ High expression of IFITM1 can promote CD8+ T cell killing of tumors by inducing microglia to secrete complement component 3, but BM cells can effectively hide IFITM1 to achieve immune escape.⁴⁰⁶ Additionally, brain-derived neurotrophic factor may drive an immunosuppressive TME by converting TAMs to a protumorigenic M2 phenotype. Most studies focus on specific stages of tumor cells progression, such as detachment from the primary site, circulation, and crossing the BBB. However, Chang et al.⁴⁴⁷ described the role of YTHDF3 in promoting BM at various stages of BC. Understanding how tumor cells navigate to the brain and establish themselves there remains incomplete. Clarifying the key regulatory mechanisms at each stage of BM, preventing the occurrence of BM, and adopting dynamic and forward-looking treatment strategies may reduce the occurrence of BM.

Research into the mechanisms of BM is advancing, but there are similarities in the preclinical research of both BC and LC. Primary tumors research often focuses on EMT and overcome the ECM. However, there is little studies specifically on the metastasis process itself, particularly interaction between tumor cells and the brain microenvironment. Recent research highlights interactions between tumor cells and astrocytes, microglia, and macrophages as critical areas of focus. Nevertheless, current studies have yielded inconsistent results, showing that tumor cells and the microenvironment can either promote or suppress tumor progression. Identifying cell subtypes with metastatic advantages and understanding the interactions with the microenvironment are crucial for developing potential treatments.

7.1 LCBM

Recent research has overturned the long-held belief that the brain microenvironment is immune-exempt. Karimi et al.³⁰⁷ elucidated distinct differences in immune landscapes between various primary tumors and BM.²⁹⁰ Understanding these immunological disparities is crucial for advancing the study of LCBM as a distinct field. Souza et al.,⁴⁴⁸ through a meta-analysis of transcriptomic data, identified significant transcriptional alterations in key driver genes like CD69 and GZMA, emphasizing the immune system's role in BM of LUAD. Furthermore, reaffirmation of the immunosuppressive TME in BM has been documented.⁴⁴⁸

Overall, LCBM are characterized by an immunosuppressive and fibrotic niche (Figure 6A).³⁹⁶ Key features of this microenvironment include diminished immune cell subsets, reduced cytotoxic T cells, and increased populations of Treg cells and macrophages.^{98, 291, 358} Specifically, T cells in BM exhibit several changes: CD20+ TILs are decreased, CD4+ T cells display an anergic phenotype with low reactivity, and CD8+ T cells show signs of exhaustion commonly seen with chronic activation.^{290, 358} Wischnewski et al.,⁴⁴⁹ using single-cell sequencing, identified distinct T cell subtypes associated with BM including CD39+ potential tumor-reactive T cells expressing CXCL13. Moreover, macrophages and dendritic cells in BM also display protumor and anti-inflammatory characteristics.⁹⁸ The BM environment is marked by a reduction in antigen-presenting cells, B cells, and T cells, alongside increases in neutrophils, M2 macrophages, immature microglia, and reactive astrocytes.³² Several cytokines and chemokines, such as TGF-β1, Visfatin, TNF-α, PAI-1, IL-2, IL-6, IL-7, IL-8, and elevated levels of CCL23, CXCL5, CXCL8, CCL8, CCL13, CCL17, and CCL18, play pivotal roles in the initiation and progression of BM.^{358, 450} Maurya et al.⁴⁵¹ summarized the chemokine cascade in brain tumors, highlighting its importance in tumor progression.

Proteomics studies have also revealed significant changes in protein expression related to hypoxia and oxidative stress in BM. Additionally, pathways related to metabolism, translation, and vesicle formation are overrepresented in metastatic tumors underlining their critical roles.^{338, 452} Underscoring big data, detailed descriptions of cell can be fully described, enhancing the understanding of the BM microenvironment at various stages of BM. This insight provides valuable therapeutic targets and potential strategies for intervention.^{290, 453}

7.2 BCBM

Similar to LC, the microenvironment of BCBM is characterized by both immunosuppressive and fibrotic properties (Figure 6B).²⁹⁵ Xiao et al.⁴³⁷ demonstrated that BC primarily consists of immune cells, including CD4+/CD8+ T cells and M2 macrophages, whereas CAFs are predominant in BMs. Among BCBMs, only 3.8% exhibit a “hot” TME with an active immune response, whereas 73.1% display a “cold” TME, with significantly reduced TILs.⁴⁵⁴ In LCBM, there is an increase in AT2 cells, CD4+ T cells, and exhausted CD8+ T cells. In contrast, BCBM are marked by a high presence of epithelial cells and myCAF cells.⁴⁵⁵ Advanced techniques such as scRNA-seq have identified new immune cell subsets in BMs, revealing the presence of immunosuppressive cell populations including FOXP3+ Tregs, RGS5+ CAFs, CCL18+ M2-like macrophages, and LGALS1+ microglia.⁴⁵⁵

Using multiplex immunofluorescence, Griguolo et al.⁴⁵⁶ characterized the BM microenvironment across various BC molecular subtypes. They found distinct immune profiles for different subtypes: HR−/HER2−, HER2+, HR+/HER2−, HR−/HER2−, and HR+/HER2− BCBMs. Respectively, these subtypes exhibited variations in immune cell types, with higher levels of CD8+ lymphocytes, increased CD4+ FoxP3+/CD8+ cells, and elevated CD163+ M2-polarized microglia/macrophages.⁴⁵⁶

In contrast to LC, the immune checkpoints such as PD-1 and PD-L1/2 are rarely expressed in CD8+ T cells and other associated cells in the BC microenvironment.³⁰⁴ Specifically, the expression of immune checkpoints including PD-L1, CTLA4, TIGIT, and B7H3 is notably reduced in BCBMs.⁴⁵⁴ Additionally, immune checkpoints like LAG3–LGALS3 and TIGIT–NECTIN2 might play significant roles in facilitating immune escape.³⁰⁴

Overall, LCBM and BCBM display a more immunosuppressive and fibrotic microenvironment compared with their primary tumors. The immune cell content is reduced, and their functions are often exhausted. The immune checkpoint expression is typically decreased, and in BCBM, PD-1/PD-L1 may not be the primary axis of immune escape. Recent research increasingly focuses on the roles of astrocytes, microglia, macrophages, and CAFs. This summary highlights the latest research progress on BM, particularly emphasizing transcriptome data to provide up-to-date insights for researchers.

8.1 LCBM

8.2 BCBM