To obtain spatial gene expression information, the target region can first be isolated from the sample, and then the dissected single region is placed separately in a test tube for RNA extraction and sequencing. The separation of target regions can be achieved by frozen section, RNA tomography (tomo-seq), laser capture microdissection (LCM) and other methods (15–17). Among them, LCM is more favored by researchers due to its advantages of high speed, high precision and versatility.

LCM is a technique for cutting, separating and collecting selected samples under a microscope using a micromanipulation system. It works by combining a laser with an inverted microscope and connecting it to a computer for direct visual analysis (18, 19). The combination of LCM and scRNA-seq allows RNA sequencing of individual cells or tissue regions (20). On this basis, Chen et al. constructed a geographical position sequencing (GEO-seq) method by integrating and optimizing LCM and scRNA-seq, which can obtain a small amount of cell transcriptomics information while retaining the original location information of cells, so as to achieve high-throughput gene expression (21). Geo-seq enables high-resolution ST by using a technique that tags mRNA in situ with spatially barcoded probes to map gene expression across tissue sections.

Additionally, transcriptomics in vivo analysis (TIVA) enables the examination of ST profiles in living cells. However, its analytical throughput is limited, with the capacity to analyze only a few single cells at a time (22). NICHE-seq employs photoactivation to conduct transcriptomics analysis on individual cells within a specific niche, but its applicability is limited to model organisms due to the necessity of genetic engineering (23). ProximID is another approach to examine the niche within the cellular microenvironment. It innovatively identifies protein-protein interactions within cells by using proximity labeling and high-throughput sequencing, allowing for spatially resolved mapping of protein interactions. Nevertheless, it remains constrained by the limitation of low flux (24).

In conclusion, as shown in Figure 1A , this technology can be used to obtain ST information at the cellular level; however, its analytical throughput is typically low, and the operation is complex and time-consuming.

Imaging-based ST technology employs microscopy to observe mRNA following the combination of a chromogenic group with a probe for in situ imaging. In situ hybridization (ISH) and in situ sequencing (ISS) are the two most commonly utilized methods.

ISH is continuously hybridized in the tissue by a marker probe that is complementary to the target sequence, and then imaged and quantitatively positioned under a microscope (25). Single-molecule RNA fluorescence in situ hybridization (smFISH) employs multiple shorter probes to target discrete regions of the transcriptomics, thereby enabling quantitative measurement of the transcriptomics (26). Subsequently, the advent of RNAscope technology enabled the simultaneous detection of up to 12 RNA targets, facilitating the identification of a smaller number of mRNA and protein targets (27). Continuous fluorescence in situ hybridization increases the signal intensity of FISH by performing multiple rounds of hybridization of a series of probes with transcripts to achieve high-throughput analysis (28–30). Furthermore, a novel technology designated as DNA Microscopy was put forth in 2019 (31). This method is based on thermodynamic entropy and does not necessitate the use of sophisticated, costly optical instrumentation. The sample itself, in conjunction with the requisite reagents, is capable of providing the requisite spatial information as part of the chemical reaction, thus assisting in the clarification of the spatial structure information present in cells and tissues (32).

In summary, ISH offers the benefits of high resolution and extensive flux detection; however, there are still constraints, including prolonged processing times, significant errors, and elevated costs ( Figure 1B ).

In 2013, the inaugural publication of ISS technology was released (33). In complete tissue sections, mRNA is initially transcribed into cDNA, and subsequently, a single-stranded DNA molecule, referred to as a lock-probe in situ sequencing (Lock-Seq) method, is employed to target identified genes. In comparison to ISH, ISS exhibits superior subcellular resolution; however, the number of targets that it can examine is restricted. Subsequently, FISSEQ and ExSeq enhanced the capabilities of ISS by employing two query bases to sequence circular and RCA-amplified cDNAs, as opposed to utilizing a single query base per probe to sequence gene barcodes (34, 35). The advantage of STARmap is that a second primer is added to the site next to Lock-Seq, thereby avoiding the reverse transcription step and improving efficiency and reducing noise (36). Furthermore, the advent of technologies such as BaristaSeq, INSTA-Seq, and HybISS have revolutionized ISS by improving spatial resolution, throughput, and sensitivity. BaristaSeq enables high-throughput sequencing directly from tissue sections with enhanced precision, INSTA-Seq provides rapid, single-molecule sequencing for faster analysis, and HybISS combines hybridization and ISS for precise ST (37–39). These advancements allow for more detailed, scalable, and efficient mapping of gene expression across tissues, opening new possibilities for studying complex tissue architectures and their molecular dynamics in health and disease ( Figure 1C ).

The ST techniques described above are based on the separation of known tissue regions of interest, or in situ visualization of RNA molecules by hybridization or sequencing, while NGS technology is based on in situ capture of transcripts, followed by ectopic sequencing. This method was developed from the conceptual innovation of the scRNA method.

In 2016, Ståhl PL et al. successfully obtained spatial-resolved full-transcriptomics information using ST technology. This technique captures polyadenylated RNA on the slide of the spatial barcode microarray before reverse transcription, thereby ensuring that each transcript can be mapped back to its original point through a unique positional molecular barcode (40). Subsequently, this innovative approach was acquired by 10× Genomics and commercialized as a Visium space gene expression solution (41).

In 2019, high-throughput ST sequencing emerged, with the most representative method being high definition spatial transcriptomics (HDST) (42–44). HDST involves embedding silica gel magnetic beads with a diameter of 2 μm on a slide, with a specific spatial barcode connected to the surface of the magnetic beads to capture the mRNA of the cells at the corresponding position for reverse transcription and transcriptomics sequencing.

As shown in Figure 1D , the limitation of ST is that the analysis location is a small area of multiple cells, which is not a genuine analysis of a single cell. Consequently, subsequent new technology has focused on maintaining the resolution of a single cell or subcellular level. Seq-Scope, based on the Illumina sequencing platform, allows for the visualization of transcriptomic heterogeneity at the cellular and subcellular levels in a range of tissues, with a sub-micron resolution (45). Sci-Space combines single-cell gene expression differences with spatial backgrounds, preserving single-cell resolution (46). Furthermore, Stereo-Seq integrates DNA nanospheres (DNB) and in situ RNA capture technology to facilitate high-throughput transcriptomics analysis of tissue sections at unparalleled nanoscale resolutions. The visual field can be extended to the centimeter level, exhibiting high sensitivity and uniform capture rate (47). Stereo-Seq is anticipated to become a fundamental tool for analyzing transcriptomics heterogeneity in complex tissues and organisms.

The high cost of ST precludes its extensive use in clinical applications, particularly when evaluated against the backdrop of alternative experimental methods. Currently, the advent of information technology has given rise to a novel approach for calculating spatial analysis gene expression datasets. This method employs computer algorithms to simulate the spatial morphology of reconstructed tissues based on single-cell transcriptomics data. To illustrate, Seurat, an R package that is frequently employed in single-cell data analysis (48). By integrating spatial information and gene expression data, Seurat is able to reconstruct spatial transcriptional profiles of tissues, revealing the spatial distribution and functional status of cells in tissues. It can process in situ sequencing data of spatially tagged genes for spatial clustering, differential expression analysis and cell type annotation. In addition, Seurat supports multimodal data integration to help researchers understand the dynamics of gene expression at the spatial level and promote in-depth studies of disease mechanisms and tissue structure (49, 50).

Furthermore, the optimization of deep learning calculation methods can be conducted repeatedly in the context of continuous modeling and verification. Additionally, the prediction of ST expression can be achieved through the utilization of existing histological images (51, 52). The toolboxes Starfysh, DeepST, MISTy, standR, and others are based on deep generative models. Such methods can not only analyze the spatial dynamic characteristics of complex tissues, but also identify spatial hubs composed of different subtypes of cells. They have become invaluable for investigating the structural and functional interactions of disparate spatial backgrounds in ST.

Starfysh is designed for processing spatial transcriptome data from tissue sections. It is able to automatically identify and classify spatial regions and reveal the spatial distribution of different cell populations in tissues by combining image data and gene expression information. Using deep neural networks, Starfysh accurately combines spatial information from tissue sections with gene expression patterns to achieve a comprehensive analysis of spatial structure and function, which is widely used to study spatial heterogeneity of tissues and dynamic changes in the microenvironment (53). DeepST aims to predict the spatial location of cells in tissue sections by learning complex patterns in spatial transcriptomic data. The tool is able to automatically learn spatial associations between cells from the data by combining gene expression data with tissue section image information and using a convolutional neural network (CNN) model for spatial reconstruction and prediction. DeepST is mainly used to study complex tissue structures such as cell type distribution, spatial dynamics features, and the TME, and helps to reveal the interactions between cell function and tissue morphology (54). MISTy accurately predicts the spatial location of cells and their transcriptional status in different spatial regions by analyzing high-resolution images of tissue sections in combination with spatial transcriptome data. It can reveal the multilevel structure and dynamic changes of spatial gene expression, and is particularly suitable for spatial gene expression analysis of the tumor immune microenvironment and complex tissues (55). StandR is mainly used to solve the problem of correlation between spatial structure and gene expression of complex tissues. It is capable of identifying cell subtypes and their distribution patterns in different spatial locations through deep learning modeling of spatial transcriptome data combined with tissue section image information. standR not only supports large-scale data analysis, but also effectively identifies spatial interactions among cells, and is widely used in the study of tissue differentiation, tumor heterogeneity and disease-related microenvironments (56).

The innovative calculation method can also address the issue of ST being challenging to differentiate between similar cell populations, thereby enhancing the efficiency of sample capture. Artificial intelligence (AI) can identify different cell types and tissue regions in HE staining images and assign cell type labels to each spatial location. Combined with spatial transcriptome data, this annotation information helps to resolve the spatial layout of different cell populations in tissues and their association with gene expression (57–59). Zubair A et al. employed cell type-specific immunofluorescence markers obtained from mouse brain tissue sections and utilized artificial intelligence to annotate the output of HE tissue images, which markedly enhanced the recognition rate of clinically pertinent immune cell infiltration in tissues and broadened the scope of ST applications (60).

In short, laser capture microdissection-based approaches and in situ imaging-based approaches usually require expensive equipment (e.g. laser microdissectors, high-resolution imaging microscopes) and are therefore more costly. Both approaches also require more time for data acquisition and are less efficient, especially for complex tissue sections. NGS-based approaches and computer reconstruction of spatial data, on the other hand, are relatively less costly as they rely on high-throughput equipment and computational power, but do not require expensive hardware support. They are more efficient when dealing with large numbers of samples and can quickly process and analyze large spatial data sets. In terms of accuracy of results, laser capture microdissection-based approaches and in situ imaging-based approaches provide high spatial resolution and accuracy, and are suitable for fine analysis at the cellular level. NGS-based approaches and computer reconstruction, although capable of handling large data sets, may not provide the same precise spatial localization as the former two and are therefore slightly less accurate in terms of result accuracy. Table 2 shows a comparison of the advantages and disadvantages of the different technological tools.

Initially, fresh tissue samples were fixed in isopentane and frozen in liquid nitrogen. This process was employed to safeguard the quality of RNA and the integrity of tissue morphology, thereby ensuring an accurate reflection of cellular and genetic distribution in space. Optimal cutting temperature compound (OCT) is a fixative agent. Following rapid freezing, the tissue samples must be embedded with OCT embedding agent. This has the dual benefit of reducing wrinkles and fragmentation during slicing and increasing the continuity of tissues. Prior to slicing with a freezing microtome, the tissue and 10 × slide were placed in a −20°C freezing microtome box for approximately 30 minutes. The thickness of the slice was typically 10 μm, with a maximum size limitation (61).

After the preparation of the slices, HE staining or IF staining was performed and photographed to detect the quality of the slices and patches. Subsequently, tissue permeabilization was performed and fluorescent cDNA was synthesized. The optimal permeabilization time was then identified according to the fluorescence intensity. The slices were then processed according to the optimal permeabilization time. The addition of permeabilization reagents to the slices resulted in the mRNA of the cells being released vertically down to the slide and bound to the capture sequence of the spatial microarray. cDNA was then synthesized by reverse transcription of the captured mRNA.

Following the acquisition of cDNA, the library was constructed and sequenced using an illumina high-throughput sequencing instrument, thereby obtaining the requisite sequencing data. The fundamental aspect of ST data analysis is the implementation of dimensionality reduction and clustering techniques, which are based on the gene expression variations observed across different spatial units on each chip. This is then correlated with the corresponding tissue image. Concurrently, the spatial position of each gene expression within the tissue can be determined.

In general, Seurat is employed for the analysis, visualization and integration of spatial data sets. The field of spatial analysis encompasses a range of techniques, including normalization, dimensional reduction and clustering, spatially-variable features, interactive visualization, and working with multiple slices. The steps of ST data preprocessing are analogous to those of typical scRNA-seq. To address the discrepancy in sequencing depth between data points, the data must be normalized. In contrast to scRNA-seq, the molecular plot of a spatial dataset can exhibit considerable variation, particularly in instances where there are notable differences in cell density across the tissue, resulting in a substantial degree of heterogeneity. Sctransform facilitates the normalization of data, enabling the detection of high variation features and the storage of data in SCT format (62).

10× Genomics Visium technology is a powerful ST tool that enables efficient spatial localization of gene expression in tissue sections (63, 64). However, there are some potential limitations of this technology, especially when dealing with different types of breast cancer samples. Firstly, the limited spatial resolution of the Visium technique, typically 55 µm, means that it may not be able to capture fine-scale differences between cellular subpopulations in the TME, particularly interactions between tumor cells and immune or stromal cells. Secondly, Visium is dependent on the quality and integrity of the tissue sections, which may affect the accuracy of the data for some breast cancer samples, especially those with high tissue stiffness or heterogeneity. In addition, the technique may be affected by background noise and technical bias when processing complex tumor samples, requiring multiple optimizations and algorithmic adjustments to improve data availability and resolution. Therefore, the Visium technology still needs to be combined with other technologies (e.g., laser microdissection or scRNA-seq) to obtain more accurate spatial transcriptomic data in studies of different breast cancer subtypes (65, 66).

By anchoring and integrating ST and scRNA-seq data, a three-dimensional transcriptomics map of the target tissue can be obtained. In this process, since the noise patterns of the representation space and the single-cell data set are very different, it is generally recommended to use the integral method rather than the deconvolution method (67).

The emerging ST technology, combined with microscopic imaging and scRNA-seq, can reveal the gene expression and arrangement distribution of individual cells at subcellular spatial resolution. This allows for the maximization of cellular heterogeneity, the discovery of new cell populations and cell subsets, and the clarification of the regulatory mechanisms that govern cellular status. This, in turn, provides technical support for a range of applications (68, 69).

Tumor heterogeneity refers to the changes in molecular biology or genes during tumor growth, resulting in differences in growth rate, invasion ability, drug sensitivity and prognosis of different tumor cells (74, 75). Despite the prevailing view that breast cancer has a monoclonal origin, the occurrence and development of this disease is the result of numerous divisions and proliferations, continuous evolution, epigenetics, genomics and microenvironment changes, which collectively give rise to heterogeneity (76). Heterogeneity represents a significant challenge in the treatment of breast cancer, manifesting as diverse histological subtypes, varying treatment sensitivities, and disparate clinical outcomes among patients. ST offers a valuable avenue for elucidating the genetic and phenotypic distinctions of tumor cells, analyzing the spatial heterogeneity of gene expression across diverse tissue regions, and providing a multi-dimensional and comprehensive perspective for investigating tumor heterogeneity (77–79).

The majority of studies have integrated ST with pathological annotation and deconvolution, utilizing diverse methodologies to accurately delineate the location, enrichment characteristics and differentially expressed genes (DEGs) of breast cancer subtypes, thereby elucidating the spatial distribution and transcriptional heterogeneity of breast cancer (80, 81). In 2013, the heterogeneity of 31 gene expression in breast cancer tissues was first observed using ISS technology. On this basis, Svedlund et al. conducted a combined molecular and morphological analysis of the expression of 91 genes in breast cancer tissues, revealing the intratumoral heterogeneity associated with breast cancer subtypes (82). Single-cell transcriptome sequencing of 26 primary breast cancers and spatial transcriptome sequencing of six breast cancers were performed to construct single-cell and spatial maps of breast cancers, revealing cellular heterogeneity in recurrent tumors and identifying novel PD-L1/PD-L2+ macrophage populations associated with clinical outcomes. Through this integrative analysis, the researchers were able to classify breast cancers into nine ‘ecotypes’ with distinct cellular composition and clinical outcomes, suggesting that the use of ST is statistically significant in classifying breast cancers and predicting response to therapy (83). A paucity of zinc finger protein 689 (ZNF689) has been demonstrated to facilitate LINE-1 reverse transcription, thereby exacerbating genomic instability and resulting in elevated intratumoral heterogeneity (ITH) (84).

The combined analysis of ST and single-cell omics allows for the consideration of both resolution and gene-level flux, which is an essential method for elucidating tumor heterogeneity (85). Studies have found that ER+ breast cancer has four heterogeneous groups, namely estrogen-responsive group, proliferative group, hypoxia-inducible group and inflammation-related group. The proliferative compartment has been identified as a key factor in the invasiveness of luminal B breast cancer (86). Liu et al. not only determined the heterogeneity of characteristics, origin and function of different cell populations in breast cancer tissues, but also employed the spatial distribution map of tumor subsets to describe the different stromal cell types in different tissue regions, providing a comprehensive method for describing the heterogeneity and structure of breast cancer (87). In luminal epithelial cells, Kohler et al. observed 11 distinct breast cancer clusters and differentiation trajectories originating from keratin 15+ (K15+) progenitor cells, indicating a robust correlation between normal K15+ primordium and basal-like breast cancer (88).

Copy number variation (CNV) can be used to reveal precancerous lesions (89). Lu et al. employed a combination of FISH and LCM with Smart-3SEQ to ascertain that HER2-amplified ductal carcinoma in situ (DCIS) has multiple subclones with different CNVs (90).

In addition, the spatial heterogeneity of metabolic structure is also an important part of tumor heterogeneity. ST studies have shown that glycolysis, profibrotic states, and metabolic reprogramming based on lipid metabolism play an important role in the extensive heterogeneity of breast cancer (91–93).

The high cost of ST limits its clinical application. To address this issue, some more cost-effective new methods have emerged. The deep learning algorithms ST-Net and BrST-Net can link the cell morphological characteristics and gene expression in ST, predict gene expression using HE tissue case images, and identify the heterogeneity of gene expression in tumors (94, 95). Based on the optimization and verification of a single deep convolutional neural network, the expression-morphology (EMO) analysis of the transcriptomics coverage area was established to predict gene transcription differences and proliferation markers in different regions of H&E staining images. The new method is reliable, economical and scalable (96).

Table 3 provides an overview of studies on the use of ST to study heterogeneity in breast cancer.

Intratumoral heterogeneity emphasizes the difference of cells or cell subsets in TME, which can reshape the occurrence, development, invasion, metastasis and drug sensitivity of tumors (97). TME is constituted of a heterogeneous population of cells, including tumor cells and surrounding immune cells, interstitial cells, and active mediators (98). It can be divided into two distinct categories: an immune microenvironment, which is dominated by immune cells, and a non-immune microenvironment, which is dominated by fibroblasts. Currently, cancer research and treatment have changed from a cancer-centric model to a tumor microenvironment-centric model. ST technology is instrumental in elucidating the diversity and intricacies of cell populations within the breast cancer microenvironment, and has become a powerful tool for a comprehensive investigation of the occurrence, progression and escape mechanisms of breast cancer (99–101).

Through spatial analysis of samples from HER2+ breast cancer patients before and after neoadjuvant HER2-targeted therapy, it was found that treatment resulted in decreased activation of HER2-dependent signaling, increased immune infiltration, and that increased expression of CD45 at the time of treatment more highly accurately predicted the presence of a pathological complete remission (pCR) in patients, findings which reveal the role of ST in predicting treatment response and guide clinical management (102).

Tumor-infiltrating macrophages play a crucial role in tumor progression, and their accumulation in TME is associated with poor prognosis and drug resistance in a multitude of cancers. In TME, macrophages exhibit different phenotypic alterations, leading to heterogeneous immune activation (103). At present, the treatment of macrophages focuses on inhibiting their recruitment or reprogramming their phenotype from M2-like to M1-like. The development of ST has improved the understanding of the macrophage body, phenotype and functional plasticity, and provided a new way for macrophage targeted therapy (104). In order to study the determinants of functional heterogeneity of TME macrophages, researchers employed an in vitro organ-type TME (oTME) model, an in vivo mouse model, and human samples of breast cancer. The study found a subset of F4/80^highSca-1+ self-renewing macrophages mediated by Notch4 and maintained by type I interferon (IFN), which is closely related to the growth and metastasis of breast cancer (105). In breast cancer cell-infiltrating lymph nodes, macrophages are abundant in pathways that promote tumor growth, NF-κB and NOD-like receptor signaling pathways, and elucidate the differentiation trajectory of macrophages from active chemokine production to active lymphocyte activation (106). Tzeng et al. found that LSM1, a part of the cytoplasmic protein complex Lsm1-7-Pat1, affects the energy metabolism of TME macrophages and breast cancer metastasis, suggesting its potential as a diagnostic and prognostic marker (107).

Dysregulation of lipid metabolism is also a feature of TME, especially excessive free fatty acids (FFA). Studies have found that lipid-associated macrophages (LAMs), a unique macrophage subset, can highly express macrophage markers, lipid metabolism genes and lipid receptors, have M2 macrophage phagocytosis characteristics, and can enhance lipid accumulation (108). In addition, macrophages exhibit strong tryptophan metabolic activity in breast cancer, which can promote the polarization of M1 macrophages and predict the efficacy of immunotherapy (109).

The spatial distribution of tumor infiltrating lymphocytes (TILs) can be a predictor of the prognosis of breast cancer and the response to systemic treatment. Romanens et al. employed ST-FFE to analyze the immune cell subsets in the stromal and intraepithelial regions of TNBC. Their findings indicated that, in comparison to stromal cells, intraepithelial T cells and B cells exhibited reduced diversity and stronger clonality (110).

Cancer-associated fibroblasts (CAFs) are another diverse cell population indispensable for remodeling TME. Studies have identified four distinct functional fibroblast subsets and characterized their spatial distribution patterns. Among these, iCAFs are enriched in breast cancer and play a pivotal role in promoting an immunosuppressive microenvironment, which is altered in accordance with the status of breast cancer (111, 112). In a study by Kang et al., the impact of homologous recombination defect (HRD) on TME development at varying scales was investigated. The findings revealed that HRD alters the dominant phenotype of CAFs from mCAFs to iCAFs and is associated with immune cell function (113).

Table 4 presents a summary of studies on the applications of ST to breast cancer microenvironment.

Drug resistance is the main obstacle to the treatment of breast cancer, which is related to tumor heterogeneity. Using ST to detect drug resistance-related cell subsets and explore their unique gene expression patterns is essential for predicting and reversing drug resistance. Basal epithelial subsets are located in the matrix and have the characteristics of drug resistance, and the up-regulated ACTN1 can promote chemotherapy resistance (114). MCU, as a diagnostic biomarker for breast cancer, is associated with advanced clinical status and a low overall survival rate. Inactivation of this protein has been shown to increase sensitivity to specific small molecule drugs (115).

The objective of immunotherapy is to reinstate a normal anti-tumor immune response by reactivating and sustaining the tumor-immune cycle, thereby controlling and clarifying tumors. Immunotherapy includes immune checkpoint inhibitors, therapeutic antibodies, and small molecule inhibitors. In recent years, immunotherapy has also shown robust anti-tumor activity in breast cancer, bringing long-term clinical benefits to some patients (116). Unfortunately, due to the loss of immune response and the emergence of drug resistance, this treatment is only effective for some patients. ST can facilitate a deeper understanding of the immunobiology of TME and identify biomarkers for predicting the efficacy of immunotherapy in patients (117). Tashireva et al. observed that PD-L1-positive tumors lack direct contact between PD1 receptor-expressing cells and PD-L1 ligand-expressing cells, and the lack of this specific immune response may be the reason for the low sensitivity of TNBC patients to immune checkpoint inhibitors (ICIs) (118). Indeed, only a single subgroup of TNBC exhibits responsiveness to ICIs. Investigations have examined the correlation between the spatial organization of immune cells in TNBC and T cell evasion, identifying that deposits of collagen-10, enhanced glycolysis, and activation of TGF-β/VEGF pathways are associated with resistance to anti-PD1 (119). TNBC patients with high Treg infiltration-related scores lack the stimulation of immune activation pathways, which results in resistance to immunotherapy against PD-1 (120).

OTUD4/CD73 proteolytic axis therapy represents a promising avenue for the treatment of immunosuppressive TNBC. The ubiquitination of CD73 by OTUD4 can counteract the deubiquitination of CD73 by TRIM21, resulting in the stability of CD73 and the prevention of a tumor immune response. ST80, a novel inhibitor that destroys the proteolytic interaction between OTUD4 and CD73, has the potential to enhance the efficacy of PD-L1 treatment in TNBC (121).

IMM2902 is an anti-CD47/HER2 bispecific antibody that has been developed for the treatment of patients with trastuzumab-resistant breast cancer. Furthermore, IMM2902 can also stimulate macrophages to produce CXCL9 and CXCL10, thereby increasing the level of T cells and NK cells at the tumor site (122).

Antibody-drug conjugates (ADCs) are a type of targeted therapy for cancer that connect antibodies to drugs through precisely designed chemical ligation methods, showing excellent efficacy in the treatment of breast cancer. ADCs hinder the progression of cell cycle. Cetuximab-CDK inhibitors can use the overexpression of epidermal growth factor receptor and the disordered cell cycle in invasive and drug-resistant tumors to target and effectively deliver drugs targeting cell cycle to basal-like/TNBCs, bringing new hope for the treatment of drug-resistant breast cancer patients (123).

A summary of studies applying ST to breast cancer drug resistance is presented in Table 5 .