Multidimensional Imaging of Breast Cancer

Abstract

Breast cancer is a pathological condition characterized by high morphological and molecular heterogeneity. Not only the breast cancer cells, but also their tumor micro-environment consists of a multitude of cell types and states, which continuously evolve throughout progression of the disease. To understand breast cancer evolution within this complex environment, in situ analysis of breast cancer and their co-evolving cells and structures in space and time are essential. In this review, recent technical advances in three-dimensional (3D) and intravital imaging of breast cancer are discussed. Moreover, we highlight the resulting new knowledge on breast cancer biology obtained through these innovative imaging technologies. Finally, we discuss how multidimensional imaging technologies can be integrated with molecular profiling to understand the full complexity of breast cancer and the tumor micro-environment during tumor progression and treatment response.

Recent exciting developments in multidimensional imaging enabled the large-scale analysis of the heterogeneity of breast cancer, including the spatial distribution of the multitude of cell types and states present in tumors. These analyses identified the presence of many populations of intertwined cancer and microenvironmental cells with various cellular shapes that are intricately organized in 3D. Importantly, this organization is not static and can evolve over time with cells actively moving around. Spatial and dynamic analyses are, therefore, crucial to not only profile the cellular, but also this structural composition to fully appreciate the significance of cell localization, interconnection, morphology, and behavior for cancer progression and treatment design. Advances in tissue clearing (Richardson and Lichtman 2015; Almagro et al. 2021) coupled with light-sheet (Reynaud et al. 2015), confocal (Jonkman et al. 2020), and multiphoton (Andresen et al. 2009) microscopy technologies allow for the observation of complex structures within large intact tissue specimens and led to discoveries relevant to breast development (Dawson et al. 2021), including mammary stem cell biology (Rios et al. 2014; Davis et al. 2016; Scheele et al. 2017) and interplay with the microenvironment (Dawson et al. 2020; Lloyd-Lewis 2020). By providing the 3D context of tissue architecture and cellular composition, and even subcellular organization of healthy and diseased tissue, these techniques can also uniquely unveil spatiodynamic traits of cancer (Fig. 1A; van Ineveld et al. 2021). Moreover, an increased number of sample preparation methods designed for preserving endogenous fluorescence (Li et al. 2019; Rios et al. 2019; Messal et al. 2021a), performing large sample immunolabeling (Renier et al. 2014; Ku et al. 2020) and moving toward multiplex imaging (Murray et al. 2015; Goltsev et al. 2018; Stoltzfus et al. 2020; van Ineveld et al. 2021; Seo et al. 2022), allowed adaptation of 3D imaging from application in fluorescently engineered animal models to studying biological processes in human samples. Here, we review recent advances in sample preparation and multidimensional imaging technologies and the resulting new knowledge on breast tumor biology. Furthermore, we discuss expected future contributions and novel technological developments that will transcend the application of multidimensional imaging in the field of breast cancer research.

Figure 1.

3D imaging technologies to uncover the spatial traits of breast cancer progression. (A) 3D diagram demonstrating the features of diverse commonly used imaging modalities to study intact breast cancer samples. The horizontal axis represents the image dimensionality (number of labels typically used to visualize different structures, cells, or molecules), the vertical axis represents the imaging scale ranging from nanometers (nm) to centimeters (cm), and the size of the dots represents image resolution ranging from subcellular (small dots) to macrostructural (large dots). Note that none of the currently available imaging technologies reach a combination of large-scale high dimensionality and high resolution. (IF) Immunofluorescence. (B) An example of panoramic imaging of the full intact mouse body using whole-body clearing and light-sheet imaging, which can be used to detect small cell numbers, such as micrometastases, in large volumes, such as the entire mouse. Scale bar, 2 cm. (Photo in B is reprinted from Pan et al. 2019 with permission from Elsevier © 2019.) (C) 3D light-sheet imaging of macrostructural components in primary breast tumors highlighting high spatial heterogeneity in vascular architecture including poorly perfused regions (region i) and well-vascularized regions (region ii) within the same tumor sample. Scale bars, 250 µm (overview panel) and 100 µm (zoom regions). (Photo C is reprinted from Dobosz et al. 2014 under the terms of a Creative Commons CC-BY-NC-ND license.) (D) Whole-mount 3D confocal imaging of an Elf5-rtTA/TetO-cre/Pten^fl/fl/p53^fl/fl/R26R-Confetti mammary gland with F-actin (blue) in a precancerous stage. Mutant cells are labeled with CFP, GFP, YFP, or RFP. Scale bar, 500 µm. (Photo in D is reprinted from Rios et al. 2019 with permission from Elsevier © 2019.) (E) Representative 3D image of breast tumor tissue derived from xenotransplanted human breast tumor organoids, immunolabeled with seven different markers and imaged using multispectral confocal imaging. Scale bar, 300 µm. (Photo E is reprinted from van Ineveld et al. 2021 with permission from the authors © 2021, who hold full copyright under exclusive license to Springer Nature America.)

CLARIFYING BREAST CANCER TISSUE

Most tissues and tumors contain a complex mixture of water, proteins, and lipids, each of which have different refractive indices (RIs), which makes them inherently opaque. The RI mismatch between the different constituents induces light scattering and, thereby, introduces loss of fluorescent signal when performing imaging in intact tissues. The breast is especially difficult to image because of its lipid-rich structure with epithelial mammary ducts buried deep within adipose tissue that strongly reflects light (Dawson and Visvader 2021). Furthermore, breast tumors change in density, as compared to surrounding healthy tissue, and are often characterized by a dense fibrotic tumor stroma, due to accumulation of extracellular matrix (ECM) proteins, such as collagen and fibronectin (Fabiano et al. 2022). To achieve complex 3D images of intact breast tissue and tumors, dedicated protocols are required to render breast tumor tissue transparent, while keeping the original tissue structure and composition intact. In this section, we summarize advances in tissue clearing and highlight the most suitable methods for breast cancer applications.

Over the last decade, various chemical strategies have been developed to render organs translucent to reduce light scattering and increase imaging depth. Pioneering studies have primarily focused on deep brain imaging with the development of clearing agents that preserve brain architecture and cellular composition for analysis of intact neural networks (Chung et al. 2013). Since then, over 60 different clearing methods have been described and applied to diverse organs and biomedical research fields, including developmental and stem cell biology, immunology, and oncology (Almagro et al. 2021). They depend on two main categories of clearing agents: solvent-based solutions that remove water and lipids, while homogenizing the tissue RI, and aqueous solutions that simply match the RI of the tissue, for some methods including additional lipid removal (Richardson and Lichtman 2015). Solvent-based clearing agents, such as the DISCO family (Ertürk et al. 2012; Renier et al. 2014; Pan et al. 2016; Chi et al. 2018; Perin et al. 2019; Qi et al. 2019), are most effective in achieving transparency and have, therefore, been of particular interest for macroscale 3D imaging. However, they can cause tissue shrinkage, structural deformation, and tend to quench native fluorescence (Lloyd-Lewis et al. 2016; Pan et al. 2016). Aqueous-based agents, in particular FUnGI (Rios et al. 2019), Ce3D (Li et al. 2017), FLASH (Messal et al. 2021a), CUBIC, and SeeDB (Davis et al. 2016; Lloyd-Lewis et al. 2016), are well-suited for high-resolution single-cell 3D imaging of the mammary gland, as well as breast tumors (Rios et al. 2019; Messal et al. 2021a). The sample volume and desired resolution (e.g., macroscale, [sub]cellular) will determine the clearing method of choice and subsequent imaging technique best suited for visualization of the cleared specimen (for further reading on tissue clearing and method selection, please refer to Almagro et al. 2021; van Ineveld et al. 2022).

THE MACROSCOPIC LANDSCAPE OF BREAST CANCER

Light-sheet microscopy presents an excellent system for imaging large samples, such as entire mice and recently even human organs (Zhao et al. 2020). Through selective plane illumination, fluorescent molecules are excited together in an entire sheet and detected simultaneously, allowing increased scan speed acquisition compared to classical, laser point scanning confocal microscopy. Although recent advances have achieved subcellular resolution imaging (Chakraborty et al. 2019), common light-sheet microscopy uses relatively low numerical aperture objectives in detriment of some resolution to achieve a large field of view. As a result, it is especially suited for studying macroscopic structures, such as blood and lymphatic vasculature to fully appreciate the complex organization of these networks in 3D (Fig. 1A–C). It has, thereby, provided key information on how these structures within the tumor microenvironment (TME) can be actively remodeled by the tumor cells (Liu et al. 2013; Lin et al. 2016; Chen et al. 2019). For example, it has uniquely revealed the chaotic and immature tumor angioarchitecture, identifying tumor areas with poorly perfused microvasculature providing a cause for frequently observed intratumoral hypoxia, a feature that often facilitates cancer cell evasion (Fig. 1C; Dobosz et al. 2014). Applied in animal models, volumetric imaging of blood vasculature has also helped to assess the efficacy of therapy delivery within mammary carcinoma (Lee et al. 2019a). 3D visualization of systemic antibody drug delivery, including the well-known immune checkpoint inhibitor anti-PD-L1 (programmed death-ligand 1), identified increased permeability of capillaries when compared to blood vessels that were supported by vascular fibroblasts and smooth muscle cells (Lee et al. 2019a). In addition, targeted anti-HER2 antibody was found to be sequestered into periphery nests within HER2⁺ tumors, potentially by binding to target antigens, as control nontargeted antibodies easily extravagated into the tumor parenchyma (Lee et al. 2019a). These findings might explain fluctuating responses to systemic treatment that have been linked to perfusion restraints (Dobosz et al. 2014). Light-sheet 3D imaging also uniquely identified a novel crossover pathway between lymph and blood vasculature, with breast metastasizing cells exiting the lymph node by invading local lymph node blood vessels, rather than using efferent lymphatic vessels (Brown et al. 2018). This key observation shed new light on the debate about whether lymph nodes could be an active dissemination route and which preferential route tumor cells use to reach blood circulation from lymph nodes. It might have implications for treatment decisions, as tumor cells located adjacent to, or within blood vessels, are expected to be more predictive of poor prognosis, compared to potentially larger lymph node tumor cell deposits distant from blood vessels (Dart 2018; Tjan-Heijnen and Viale 2018). With breast cancer being a key example of tumors that constantly fail to respond to angiogenesis-targeted therapy for improving survival outcomes (Ayoub et al. 2022), macroscale 3D imaging technology might be of great interest for testing novel angiogenic therapies to normalize tumor vasculature and study their impact on drug delivery as well as metastatic dissemination.

Next to the above-described evaluation of specific metastatic routes and local drug delivery, recent technology advances in panoptic 3D imaging now provide an unprecedented readout to assess metastatic tumor burden alongside therapy distribution, not only at the tumor site, but also at the whole-body level. The potent clearing methodologies, CUBIC (Susaki et al. 2014) and next generation of 3DISCO (such as uDISCO [Pan et al. 2016] and subsequently vDISCO [Pan et al. 2019]), together with further-advanced light-sheet microscopes and immunolabeling protocols, have enabled detection of cells in small quantities, while scanning massive volumes to image entire mice (Kubota et al. 2017; Cai et al. 2019). Using 13 mouse models and nine cancer cell lines, including the human metastatic breast cancer line, MDA-MB-231, Kubota et al. (2017) used CUBIC to quantify metastasis at the whole-mouse body level and assess their response to chemotherapy. The unprecedented scale of whole-body imaging poses significant challenges for handling and analysis of large imaging data sets, necessitating the development of deep learning–based quantification platforms (Fig. 1B). The DeepMACT analysis pipeline (Pan et al. 2019) allows mapping of individual metastasis in whole-animal breast cancer models. This uniquely revealed a previously unrecognized micrometastasis pattern at numerous sites across the body, but also only a partial degree of overlap with the distribution of a monoclonal antibody therapy, suggesting incomplete targeting (Pan et al. 2019). Building onto this seminal work, whole-body distribution mapping of novel targeted or cellular therapies is a promising avenue that could uniquely be addressed by these 3D panoramic imaging technologies. Further automation of these imaging pipelines could deliver an invaluable tool for translational studies to identify the most promising therapies prior to clinical testing (for an extensive overview of 3D image analyses pipelines, we refer to van Ineveld et al. 2022). Together, this body of work demonstrates the value of 3D imaging to unravel the macroscopic landscape of breast cancer across the full trajectory from tumor progression to end-stage metastasis, as well as during treatment response.

MAPPING BREAST CANCER HETEROGENEITY

Confocal or two-photon microscopes offer mainstream systems for large-scale 3D imaging with higher resolution, as compared to light-sheet technology, yet at the expense of some imaging depth and speed. Therefore, recent efforts have been directed at obtaining easy-to-use protocols for routine 3D imaging of intact tumors with clearing methodologies that preserve epitope integrity, limit distortion of tissue, and allow for rapid sample preparation (Li et al. 2017, 2019; Rios et al. 2019; Messal et al. 2021a). These protocols are compatible with visualization of native fluorescence and, thereby, enable visualization of breast cancer progression in fluorescently engineered animal or xenograft models. Multicolored fluorescent lineage-tracing approaches, such as confetti, enable tracking of individual cell fates throughout the progression of cancer (Kretzschmar and Watt 2012) and have allowed prediction of clonal relationships between tumor cells in vivo and, thereby, the modeling of tumor-wide clonal contributions (Fig. 1D; Lamprecht et al. 2017; Yanai et al. 2017; Rios et al. 2019; Tang et al. 2019; Tiede et al. 2021; Yum et al. 2021). For instance, LSR-3D (large-scale single-cell resolution-3D) imaging of confetti lineage-tracing breast cancer models revealed profound clonal restriction during breast cancer progression (Rios et al. 2019), a phenomenon also observed in other cancers (Vermeulen et al. 2013; Snippert et al. 2014; Brown et al. 2017; Ying et al. 2018; Bruens et al. 2020). Integration with clonal RNA sequencing and immunolabeling revealed the molecular heterogeneity of these predominant stem- or progenitor cell–derived clones, with cellular phenotypes varying along a spectrum of reversible states in between epithelial and mesenchymal fate (Fig. 1D; Rios et al. 2019). Intratumoral heterogeneity also manifests as differences in cell proliferation within invasive carcinomas derived from different subtypes of breast cancer, including luminal A-, luminal B-, and HER2-enriched human breast cancer. Interestingly, in vivo clonal tracing demonstrates that cancer cell populations efficiently targeted by chemotherapy are mainly slow-proliferative clones, whereas chemotherapy fails to repress fast-proliferative clones, suggesting that intratumoral heterogeneity of tumor cell proliferation may enhance the tumor's resistance to chemotherapy (Tiede et al. 2021). Altogether, this indicates inherent tumor plasticity and shows that many clonal evolution routes can lead to effective cancer progression and treatment resistance (Rios et al. 2019; Tiede et al. 2021). Multidimensional imaging is also relevant for gaining knowledge in metastasis. Deep multiphoton microscopy combined with an integrative analysis pipeline, SMART 3D, revealed the spatial landscape of experimentally induced brain metastasis derived from breast cancer cells and again identified proliferative differences in the metastases, as well as specific metastasis-associated astrogliosis (Guldner et al. 2016).

Altogether, these findings show the power of confocal/multiphoton 3D imaging to visualize and quantify global spatial and clonal changes occurring during cancer progression and reveal key cellular processes, such as proliferation and epithelial-to-mesenchymal transition (EMT), that drive tumor heterogeneity. In other types of cancers, these 3D imaging techniques have, furthermore, identified physical constraints, such as epithelial curvature, to not only dictate morphology, but also predict aggressive tumor behavior, such as invasion (Messal et al. 2019; Fiore et al. 2020). This type of analysis might be of great interest for early-stage ductal carcinoma in situ (DCIS), accounting for 25% of all diagnosed breast cancers that can be considered noninvasive or pre-invasive breast cancer. They often associate with a high survival rate, but due to their heterogeneous pathological traits are clinically challenging in terms of the therapeutic decision whether to offer less aggressive treatment (Farante et al. 2022). As there is currently no method to predict which of the DCIS will become invasive, multidimensional analysis of the biophysical properties of DCIS could potentially help to predict invasiveness and, thereby, guide treatment management in the near future.

RESOLVING TME INTERPLAY IN BREAST CANCER

Multidimensional imaging contributed significantly to the identification of tumor heterogeneity including the complex TME. This is of particular importance for the immune compartment that has been increasingly recognized for its role in tumor progression and is, thereby, an important focus of ongoing therapy development (Hegde and Chen 2020). Indeed, 3D imaging has been applied to spatially map diverse immune cell subsets in breast tumors (Lee et al. 2019b), as well as the expression of immune regulatory molecules (Lee et al. 2019a). Multispectral imaging that increases the number of markers that can be visualized by spectrally resolving numerous fluorophores offers a methodology of choice for simultaneously analyzing diverse immune and cancer cell types and states. The Opal method, although performed in 2D, allows recording of up to eight markers in a single acquisition. This methodology was used to spatially decode the TME immune composition of breast tumors, and helped identify differences in immune infiltration and immunomodulatory molecules PDL-1 and CTL4 in BRCA1-deficient triple-negative breast cancer (TNBC) tumors and provided strong rationale for clinical studies evaluating the efficacy of combined immune checkpoint inhibitors in BRCA-associated TNBC (Nolan et al. 2017). This clearly shows the clinical relevance of spatially resolving the complexity of tumors and their TME by visualizing multiple molecular markers at once. In recent years, several protocols have been developed for multispectral 3D imaging with a high number of markers (Valm et al. 2017; Coutu et al. 2018; Li et al. 2019; van Ineveld et al. 2021). mLSR-3D, for example, implements “on the fly” linear unmixing for single-scan acquisition of eight spectrally resolved fluorophores and enables fast and light-efficient high-dimensional imaging (van Ineveld et al. 2021). Combined with a deep learning–based segmentation pipeline, STAPL-3D, hundreds of molecular and spatial features can be extracted from millions of cells in 3D imaging data sets, allowing omics-like analysis of imaging data (van Ineveld et al. 2021). The power of this 3D imaging and computational advance was demonstrated by the identification of novel and rare tumor cell subsets in pediatric kidney Wilms tumor, but also characterization of intratumoral heterogeneity that can be modeled in breast cancer organoids and derived xenograft models (Fig. 1E; van Ineveld et al. 2021). This showcases the already highly relevant resolving capacity of multispectral 3D microscopy to capture the complex cellular heterogeneity of tumors by overcoming the traditional limit of four to five markers imaged at the time. This can be further expanded with ongoing advancements that are expected to continue to sharply increase the number of markers that can be imaged (Seo et al. 2022).

Another important imaging trend for increasing molecular resolution is represented by multiplex fluorescence imaging, a methodology that implements sequential rounds of protein or RNA labeling and microscopy acquisition (Junker et al. 2014; Chen et al. 2017; Karaiskos et al. 2017). RNA labeling currently enables the highest in situ molecular resolution (Murray et al. 2015; Wang et al. 2018b; Alon et al. 2021). Applied to a core biopsy from breast cancer metastasis, Exseq (Expansion sequencing), for instance, elucidated the relationship between tumor cells and fibroblasts in the hypoxic tumor environment (Alon et al. 2021). By providing spatial profiling of 296 genes within 2395 cells, this technology showed that tumor cells overexpress HIF1A, a molecule involved in tumor maintenance, when in close proximity to HSPG2 positive fibroblasts (Alon et al. 2021). Furthermore, S100A8, a regulator of inflammatory immune responses and known biomarker for potential relapse or late progression of breast cancer (Wang et al. 2018a; Zhong et al. 2018), was found to be overexpressed fourfold in a subset of B cells, when they are located near EGFR-positive tumor cells. This shows the power of enhanced spatial molecular mapping for identification of subtle changes in cellular states, for example, in immune cells, according to their distance to other cell types such as tumor cells (Alon et al. 2021). However, for larger specimens, the multiplexing experimental procedure can become incredibly time consuming and complex, as sample coregistration is required to integrate data from each sequential imaging round in 3D. STARmap, for example, reports visualization of 1020 genes in 5 days on thin sections, but this number drastically reduces to 28-in-thick samples, while almost doubling experimental duration (Wang et al. 2018b).

Overall, with both multiplexed and multispectral imaging still rapidly evolving, we see a clear application for high-dimensional 3D imaging in probing the heterogeneity of both the tumor itself, as well as its cellular environment and, importantly, the interaction between those two. Its power for drawing connections between spatial cellular organizations within tumors and how those relate to different cell types and states has already been demonstrated and might eventually be linked to clinical outcomes. To fully explore this clinical predictive power and start using it as a diagnostic tool, 3D imaging of archival samples stored in patient biobanks will be the next step forward, as discussed in the following section.

3D VISUALIZATION OF BREAST CANCER ARCHIVAL TISSUE

Pathological formalin-fixed paraffin-embedded (FFPE) samples are current practice for diagnosis and, therefore, stored alongside relevant clinical data. This represents a widely available resource that is for now not fully exploited for its clinical and research potential by routine 2D sectional imaging (Haddad et al. 2021). Indeed, standard histological practices only allow for the observation of a few slices per sample, representing a fraction of the entire sample, thereby providing limited information. As a result, the full spatial heterogeneity in complex tissues, such as breast cancer samples, cannot be studied. This deficiency is underscored by current 2D methods for histological grading of primary tumors leading to inconsistency and misdiagnosis (Kruskal et al. 1997; Brown et al. 2010; Catalona et al. 2017; Epstein 2018; Ahdoot et al. 2020). For breast cancer, it has been shown that chemotherapy-induced histological changes associate with false-negative sentinel lymph nodes and, therefore, might be misleading for monitoring treatment effects on late metastatic progression (Brown et al. 2010). Many laboratories have begun to explore 3D histopathology of FFPE primary patient material for tumor grading (van Royen et al. 2016) or to assign novel treatment options, such as immunotherapy (Si et al. 2019).

3D imaging of full FFPE tissue blocks can be achieved by 3D reconstruction of serial acquired 2D sections. The advantages of this approach include full compatibility with current tissue-staining protocols and a vast number of antibodies validated for immunohistochemistry. 3D histological reconstructions of DCIS have identified structural dissimilarities between benign lobules and DCIS, offering an improved structural analysis to potentially discriminate noncancerous-resembling lobular structures from early-stage breast cancer manifestations (Booth et al. 2015). However, 3D rendering from serial sectioning is tremendously laborious and time consuming, as for large volumes thousands of slides must be processed and imaged. Automated staining machines and slide scanners can potentially relieve some of this burden, but still artefacts can arise, due to tissue deformations associated with sectioning. Protocols for 3D imaging of intact FFPE specimens can prevent these drawbacks and are currently emerging based on similar principles of optical clearing, as described for freshly fixed tissue. Furthermore, these protocols are similarly compatible with immunolabeling and confocal or light-sheet microscopy (Rios et al. 2016; Nojima et al. 2017; Chen et al. 2019; Yoshizawa et al. 2020; van Ineveld et al., in press). 3D imaging of breast cancer FFPE tissue, revealed spatial variation in the cell-cycling profile of tumor cells throughout the tumor, which was not detectable in FFPE 2D sections (Tanaka et al. 2018). Further clinical application of 3D archival tissue imaging was demonstrated by DIPCO, an iDISCO-based protocol for FFPE tissue, shown to have better accuracy compared to 2D histological analysis for diagnosis and stratification of patient prognosis (Tanaka et al. 2017, 2018). In addition, an updated protocol including in situ hybridization for RNA detection, further improving the molecular phenotyping capabilities of the technique (Tanaka et al. 2020). This revealed in TNBC, a disease subtype strongly associated with poor prognosis, a higher density of spatial niches enriched with cancer stem-like cells, a specific cancer cell population known to fuel breast cancer progression and, so far, difficult to therapeutically target (Tanaka et al. 2020). Further implementation of such advancements for breast cancer might be of great interest for treatment guidance and monitoring outcomes, as discussed previously for DCIS and sentinel lymph nodes.

The wealth of biologically and clinically relevant data obtained from FFPE tissues can be further expanded by superresolution imaging. This allows differentiation malignant macrostructures beyond the diffraction limit (Weiss 2000) and can achieve below 1 nm or 20 nm resolution, depending on the specific method (Rust et al. 2006; Jones 2012). Applied to clinical FFPE sections of cancerous tissue, it revealed 3D macrostructures resolvable to a depth of 800 nm (Creech et al. 2017). HER2-receptor expression, an important biomarker for breast cancer, could be accurately visualized on membrane protrusions in FFPE breast cancer biopsies. These were previously shown to be implicated in persistent localization and signaling of HER2 in breast cancer cells in vitro (Jeong et al. 2016), which demonstrates that similar HER2-enriched protrusions appear to exist in vivo and may play a functional role in breast tumor biology (Creech et al. 2017). Thus, although for now limited to very thin sections, superresolution imaging may offer important functional insights into disease mechanisms and diagnosis, by precisely mapping structural molecular arrangements within cancer cells. Together with 3D imaging technology advances for archival samples, this can create a very complete picture of breast tumor characteristics associated with poor prognosis and treatment response.

This body of literature already demonstrates enhanced discriminative and, thereby, clinical potential of 3D archival tissue imaging. Its potential will further expand with the number of molecules and, thus, cell types and macrostructures, that can be visualized. Advances in label-free imaging approaches, including THG (Third Harmonic Generation), SHG (Second Harmonic Generation), and FTIR (Fourier-transform infrared spectroscopy) (Ren et al. 2017; Rivenson et al. 2019; Gavgiotaki et al. 2020), can further help to achieve this goal. They have already been applied to 2D tissue sections to discriminate morphology differences between malignant and healthy breast tissue, as well different grades of the disease (Gavgiotaki et al. 2020). These approaches might offer easy-to-implement alternatives to antibody labeling of thick FFPE tissues. Another exciting avenue is highly multiplexed molecular and cellular mapping using multidimensional imaging approaches, including 3D imaging mass cytometry (di Palma and Bodenmiller 2015; Kuett et al. 2021). It has the important potential of adding high-dimensional molecular classification for cancer and microenvironmental cells to histological H&E scoring of FFPE samples, thereby identifying novel biomarkers to refine diagnosis or guide treatment decision, especially for immunotherapy, the largest current body of development in cancer therapy.

INTRAVITAL MICROSCOPY TO STUDY BREAST TUMOR DYNAMICS

3D imaging of fixed breast tissues provides a snapshot of the architecture and the different cellular and noncellular components, but does not provide information on the dynamic interplay between the various components and their subsequent fate. To uncover cellular and architectural dynamics, intravital microscopy (IVM) approaches have been developed. To obtain high-resolution images deep inside tissues (up to 1 mm) while minimizing photo damage, multiphoton imaging technologies have been the preferred choice for IVM of breast tissues (Bakker et al. 2022; Choe et al. 2022). To get optical access beyond the imaging depth of the multiphoton microscope, the mammary gland can be surgically exposed (often referred to as the skinflap approach) to allow for longitudinal imaging over several hours (Ewald et al. 2011; Harney et al. 2016; Dawson et al. 2021; Messal et al. 2021b). To get visual access to tissues over multiple weeks, permanent imaging windows were developed, which are surgically implanted on top of the anatomical location of interest, which can be the primary mammary tumor site (mammary imaging window) (Kedrin et al. 2008; Jacquemin et al. 2021; Messal et al. 2021b; Maiorino et al. 2022; Mourao et al. 2022) or the metastatic site (bone marrow, liver, lung, brain) (Fig. 2; Ritsma et al. 2013; Alieva et al. 2014; Entenberg et al. 2018).

Figure 2.

Dynamic intravital microscopy to uncover breast cancer dynamics at a cellular level. (A) Intravital imaging of mutant mammary cells revealed the first steps of tumorigenesis in the mammary gland where β-catenin activation drives the formation of hyperplastic regions with squamous differentiation within a few days. Scale bar, 50 µm. (Photos in A are reprinted from Lloyd-Lewis et al. 2022 under the terms of a Creative Commons Attribution 4.0 International license (CC BY 4.0).) (B) In vivo imaging of mammary tumor progression revealed several modes of invasion, including collective strand invasion (panel i) and single-cell invasion (panel ii). Scale bars, 200 µm (overviews) and 25 µm (panels i and ii). (Photos in B are reprinted from Ilina et al. 2018 under the terms of a Creative Commons CC-BY license.) (C) In vivo detection of metastatic breast cancer cells in the vasculature of the lymph node elucidating a role for the lymph node as a cancer cell hub prior to further distant metastatic spreading. Scale bar, 25 µm. (Photo in C is reprinted from Pereira et al. 2018 with permission from American Association for the Advancement of Science © 2018.) (D) Time-lapse intravital imaging of the lung vasculature allows behavioral tracking of disseminated tumors cells upon arrival in the lungs, including recirculation or apoptosis (top panels) or extravasation (bottom panels). Scale bar, 15 µm. (Photos in D are reprinted from Borriello et al. 2022 under the terms of a Creative Commons Attribution 4.0 International License (CC BY 4.0).) (E) Multiday intravital microscopy of metastatic breast cancer cells in the brain revealed that the rate-limiting step for metastatic progression is to overcome the dormancy-inducing microenvironment. Scale bars, 40 µm. (Photos in E are reprinted from Dai et al. 2022 with permission from the authors who hold the exclusive copyright license for this work published by Springer Nature.)

To obtain contrast between different components in breast tissues (e.g., epithelial cells, stromal cells, immune cells, ECM), it is crucial to fluorescently label components of interest. For most IVM studies, this is done either by inducible reporter constructs in genetically engineered mouse models, transplantation of fluorescently labeled cells, intravenous or local injection of fluorescently labeled antibodies raised against cell- or matrix-specific epitopes, or injection of dyes, such as dextran or lectin, to label the vasculature. Of note, a recent report underscored that expression of fluorescent proteins or any other xenobiotic marker genes may elicit immunogenicity leading to attenuation of tumor growth and progression (Grzelak et al. 2022), underscoring the importance of evaluating the effects of experimental design in imaging studies (Day et al. 2022). When using multiphoton imaging, several label-free imaging approaches can be used to visualize structural components of the tissue environment, such as collagen I, fat cells, muscles, and nerve bundles (Friedl et al. 2007; Weigelin et al. 2016; You et al. 2018). For each process that will be imaged in vivo, the experimental setup needs to be optimized. For instance, fast events such as cancer cell migration, cancer cell–immune cell interactions, or intravasation will require IVM time-lapse imaging with a short time interval of seconds or minutes, whereas the study of cancer cell growth dynamics will require multiday IVM through a chronic imaging window with a time interval of days or even weeks. In this part of the review, we discuss how IVM (3D imaging over time) has contributed to our understanding of the formation and progression of breast cancer.

IMAGING BREAST TUMOR INITIATION AND PROGRESSION

IVM has been used to study the cellular dynamics of all stages of breast tumor formation. For example, using an inducible model of β-catenin activation combined with lineage tracing and real-time IVM, it was elegantly shown that luminal and basal epithelial cells rearrange upon sustained Wnt signaling (Fig. 2A; Lloyd-Lewis et al. 2022). These altered dynamics eventually led to the formation of hyperplastic lesions with squamous features, irrespective of the cell lineage of origin (Lloyd-Lewis et al. 2022). Using a similar longitudinal IVM approach combined with multicolor confetti lineage tracing at a later state of mammary tumor development, it was shown that primary mammary tumors display a large degree of plasticity (Zomer et al. 2013). Following the same cells and their progeny over multiple weeks indicated that the vast majority of the mammary tumor cells have a short lifetime (<wk) and do not contribute to tumor growth. Only a small subset of labeled cancer cells remained and proliferated leading to outgrowth of large clones as also observed by 3D microscopy of fixed tumor tissues (Rios et al. 2019). This indicates that these cancer cells represent the cancer stem cell (CSC) population and fuel tumor progression. Interestingly, IVM over multiple weeks showed that cells can gain or lose the CSC clonogenic properties, indicating widespread phenotypic plasticity in primary mammary tumors (Zomer et al. 2013). Together, these studies on primary breast lesion dynamics uncover a role for cell-state plasticity already starting from the first stages of tumorigenesis.

IMAGING BREAST CANCER CELL ESCAPE

Over the past years, IVM revealed several mechanisms on how cells escape from primary mammary tumors. First, cells need to detach from neighboring cells and migrate toward vessels. IVM has revealed several modes of in vivo migration including collective strand invasion (Fig. 2B; Ilina et al. 2018), cellular streaming (Patsialou et al. 2013; Leung et al. 2017), and individual cell movement (Zomer et al. 2015; Beerling et al. 2016). Cell motility can be driven by diverse intracellular mechanisms or by cues in the TME. IVM of mammary tumors combined with in silico modeling recently showed that the mode of migration may be determined by a combination of ECM confinement and cell–cell junction stability, where highly confined tumor neighborhoods drive collective migration irrespective of cell–cell junction stability and single-cell escape occurs in areas with free space (Ilina et al. 2020). A systematic analysis of migratory cells in mammary carcinoma revealed that motile mammary tumor cells show differences in their migration speed. Slow migratory mammary tumor cells expressed CSC markers and had the capacity to degrade the ECM and disseminate (Sharma et al. 2021), whereas the faster migratory mammary tumor cells lacked these capabilities (Gligorijevic et al. 2014). The important role of matrix composition and density was underscored by dynamic IVM of the collagen network (using second harmonics generation) upon down-regulation of SERPIN E2, an extracellular protease inhibitor known to be involved in promoting breast cancer metastasis. SERPIN E2 knockdown or inhibition led to a cascade of changes in the TME, eventually leading to the deposition of a dense collagen network around the primary mammary tumor (Smirnova et al. 2016). Interestingly, this excessive matrix deposition showed a clear inhibitory effect both on local invasion as well as on distant metastasis formation (Smirnova et al. 2016).

Both during intravasation (entry into the bloodstream) at the primary tumor site and extravasation (exit from the bloodstream into a secondary organ), cancer cells need to cross the endothelial cell barrier. IVM has elucidated several mechanisms by which cancer cells can enter or exit the bloodstream. IVM showed that intravasation of disseminating primary breast cancer cells into the bloodstream can be facilitated by perivascular macrophages. The close interaction between the macrophage and endothelial cells leads to transient and local disconnection of the endothelial cell–cell junctions, which in turn facilitates cancer cell intravasation and dissemination (Harney et al. 2015; Karagiannis et al. 2017). Moreover, it was shown that the cancer cells interacting with these macrophages expressed CSC markers (Sharma et al. 2021), indicating that CSCs have the capacity to intravasate. IVM time-lapse imaging of intravasation revealed that invading breast cancer cells form invadopodia, which are dynamic and actin-rich protrusions. Invadopodia are assembled and disassembled in response to diverse external stimuli, such as integrin signaling (Paz et al. 2014) or chemotactic factors (Williams et al. 2019), and inhibition of invadopodia formation blocks extravasation and metastatic outgrowth (Leong et al. 2014). Upon intravasation into the bloodstream, IVM revealed the mechanisms by which circulating tumor cell (CTC) clusters may arise. Using time-lapse imaging of diverse patient-derived xenograft and spontaneous mouse models, it was shown that migrating tumor cells either cluster together near the vasculature or intravasate as individual cells and subsequently aggregate in the bloodstream. Tumor cell clustering was shown to depend on CD44 expression, which mediates intercellular interactions within the tumor cell aggregates with increased capacity to seed lung metastases (Liu et al. 2019).

THE ROLE OF CANCER CELL PLASTICITY DURING TUMOR PROGRESSION

Primary mammary tumor cells display a high degree of plasticity between epithelial and mesenchymal states to respond to changing environments. Using IVM of primary mammary tumors with endogenously labeled E-cadherin, it was shown that mammary tumor cells dynamically lose or gain E-cadherin expression during the metastatic cascade (Beerling et al. 2016). The majority of the migratory cancer cells within the primary tumor transiently lost their membranous E-cadherin expression, while this was always retained in nonmigrating E-cadherin-positive cancer cells. Moreover, the motile cancer cells adopted a mesenchymal expression profile, suggesting that an EMT preceded local invasion and intravasation (Beerling et al. 2016). This EMT phenotype was shown to be specifically induced in close proximity to blood vessels (Zhao et al. 2016). Importantly, a recent study using novel reporters for EMT lineage tracing showed that mammary tumor cells mostly transition between an epithelial and partial EMT state (Lüönd et al. 2021). Moreover, it was shown that mammary tumor cells with partial EMT show increased collective migration, and the ability to metastasize and grow out in the lungs, whereas mammary tumor cells undergoing full EMT were static, lost their plasticity, and failed to establish metastases (Bornes et al. 2019; Lüönd et al. 2021). Indeed, upon intravasation and arrival in the lungs, it was shown that metastatic tumor cells quickly reverted to an epithelial state (Beerling et al. 2016). Together, these studies underscore the importance of EMT plasticity during the metastatic cascade in breast cancer.

IMAGING BREAST TUMOR METASTASES

IVM of spontaneous metastases is challenging as it is impossible to predict where and when primary mammary tumor cells will arrive in the secondary organ and grow out into metastatic lesions. Therefore, experimental metastasis models are often used to study the dynamics of cancer cell extravasation and metastatic outgrowth, which are obtained by direct injection of the primary breast cancer cells into the bloodstream, such as by tail vein injection to obtain lung metastases, mesenteric vein injection to obtain liver metastases, or by intracardiac injection to obtain brain metastases. Recently, a new permanent lung imaging window was designed to allow for longitudinal visualization of the lung (Entenberg et al. 2018), thereby enabling the assessment of spontaneous lung metastasis formation using diverse cell lines of breast cancer directly derived from the primary tumor. Direct comparison of the in vivo dynamics of experimental and spontaneous metastases in the lung revealed significant behavioral differences, where spontaneous disseminating cancer cells showed increased levels of retention in the lung vasculature, faster extravasation from the lung vessels, and increased survival after extravasation when compared to experimental metastatic cells (Fig. 2D; Borriello et al. 2022). Mechanistically, it was shown that spontaneously disseminating cells represent a subset of cancer cells that concomitantly expressed increased levels of stem cell markers, such as SOX9, and markers for cellular dormancy, such as NRF2F1 (Borriello et al. 2022). Using an innovative hypoxia-inducing nano-intravital device (Williams et al. 2016), it was shown that a hypoxic environment in primary mammary tumors can induce local dormancy (Fluegen et al. 2017). Although the ability to intravasate was not changed in these hypoxia-induced dormant tumor cells, their dormant state, but not their hypoxic state, was retained upon seeding in the lungs. This indicates that a hypoxic environment at the primary tumor site may prime cells for into a dormant state at the distant site for the long term.

Tumor cell dormancy was also shown to play an important role during breast cancer brain metastases initiation and outgrowth. Using multiday IVM through a cranial imaging window in an experimental model of TNBC brain metastasis, it was demonstrated that the rate-limiting step to form brain metastasis is not the capacity for extravasation, but rather the capacity to overcome a dormancy-inducing microenvironment (Fig. 2E; Dai et al. 2022). Proliferative and expanding brain metastasis were situated in a vascular niche devoid of astrocytes, whereas dormant TNBC cells were associated with vessels that directly associated with astrocytes. Further analyses revealed that astrocytes prevent TNBC proliferation by deposition of laminin-211, a component of the parenchymal basement membrane. Laminin-211 is sensed by the TNBC through dystroglycan, which in turn negatively regulates transcriptional activator Yes-associated protein (YAP), thereby inhibiting proliferation in these metastatic TNBC cells (Dai et al. 2022). A similar dormancy-sustaining mechanism by the ECM was described for disseminated tumor cells in the lung, where deposition of collagen type III was shown to be required to retain a dormant state (di Martino et al. 2022). Reactivation of dormant cells could be achieved both by disruption of the collagen III architecture and by changes in the collagen III abundance (di Martino et al. 2022). In addition to the local tissue architecture, other extrinsic factors were identified as stimuli-reactivating dormant metastatic breast cancer cells. In models of experimental and spontaneous breast cancer lung metastases, sustained inflammation, for example, induced by exposure to tobacco smoke, was identified as one of these triggers (Park et al. 2016; Albrengues et al. 2018). IVM revealed that inflammation resulted in high infiltration of neutrophils in the lungs, which in turn resulted in dormant cancer cells that reentered the cell cycle. Interestingly, the effect of the neutrophils was not exerted by the immune cells directly, but in an indirect manner through the deposition of neutrophil extracellular traps (NETs). NETs are DNA scaffolds in the extracellular space associated with proteases, which in turn were shown to facilitate proteolytic ECM remodeling and cleavage of laminin-111 to trigger cancer cell proliferation (Albrengues et al. 2018). Together, these experiments reveal a local and temporal control mechanism of tumor cell dormancy by structural and cellular components of the host tissues. Sustaining these natural barriers could potentially prevent outgrowth of disseminated cancer cells into overt metastases.

In addition to blood-borne metastases, two hallmark papers recently provided evidence for a hybrid route in which lymph node metastases serve as a hub prior to further dissemination to distant organs using either spontaneous metastases models or experimental lymph node metastases of breast cancer by intralymphatic infusion (Fig. 2C; Brown et al. 2018; Pereira et al. 2018). To study the dynamics of spontaneous lymph node metastases, diverse cancer models (including mammary tumor cells) expressing a photoconvertible protein were orthotopically transplanted (Pereira et al. 2018). Cancer cells that spontaneously metastasized to the lymph node were photoconverted (from green to red) to determine the origin of distant metastases arising in the lungs (i.e., nonconverted [green] metastatic lesions would indicate direct transit from the primary tumor, whereas converted [red] metastatic lesions would indicate a hybrid route via the lymph node). Interestingly, both red and green lung metastasis were detected indicating that both routes are used by cancer cells to reach the distant organs. Next, using time-lapse IVM, it was shown that cancer cells disseminating through the lymphatics enter the lymph nodes via the subcapsular sinus and migrate toward the cortex of the lymph nodes, where they associated with the endothelial cells and the resident dendritic cells (Pereira et al. 2018). Similar spatial dynamics were observed in static time-point analyses after intralymphatic infusion of breast cancer cells (Brown et al. 2018), where the high endothelial venules, the main point of entry for incoming lymphocytes, were identified as the main point of exit for the cancer cells to seed to distant organs (Brown et al. 2018). Intravasation of breast cancer cells into the lymphatics was shown to correlate with Ezrin expression, and IVM of breast cancer cell behavior after systemic Ezrin inhibition revealed impaired cell migration and a reduction in metastatic burden in the lymph nodes and lungs (Ghaffari et al. 2019). Recently, the role of lymph node colonization prior to distant metastasis formation was shown to extend beyond a mere transit route. Instead, it was shown that cancer cells colonizing the lymph node are able to evade the immune system, activate regulatory T cells, and actively induce systemic tumor-specific immune tolerance to facilitate further metastatic spread (Reticker-Flynn et al. 2022). Together, these studies identify the lymph nodes as an initial hub prior to systemic spreading using a powerful combination of dynamic IVM and static ex vivo imaging.

IMAGING OF DRUG RESPONSE DYNAMICS IN BREAST CANCER MODELS

To successfully treat breast cancer, it is important to know the in vivo biodistribution, kinetics, and mode of action of a potential drug. Both drug delivery and dynamics are hard to predict based on ex vivo analysis, and, specifically in the case of a heterogeneous response, a cellular resolution is required to understand where and when a potential anticancer drug exerts its function. IVM is perfectly suited to visualize drug distribution and dynamics, and several IVM studies have helped to elucidate drug dynamics and mode of action. For instance, when using IVM, it was shown that Parp inhibitors reach their cellular targets within seconds to minutes in the in vivo setting at a sufficiently high concentration, suggesting that resistance is not caused by an inefficient bioavailability of the drug (Thurber et al. 2013). Moreover, the TME is an important player in drug responses that can only be assessed in the unperturbed in vivo setting (Miller and Weissleder 2017). For example, in the case of monoclonal antibodies targeting HER2-positive breast cancers (trastuzumab), IVM revealed that, at the first 24 hours after injection, the HER2-overexpressing breast cancer cells were reached. However, soon after, this distribution shifted toward predominant accumulation of trastuzumab in the tumor-associated macrophages (Li et al. 2020). A similar drug-sequestering effect of the tumor-associated macrophages was observed in tumors treated with anti-PD1 antibodies used to target and activate intratumoral CD8⁺ T cells (Arlauckas et al. 2017). Although an immediate binding of anti-PD1 to the T cells within the tumor was observed, this interaction was transient as tumor-associated macrophages captured these antibodies from the T cells within a few minutes (Arlauckas et al. 2017). Anti-PD1 engagement could be prolonged using a blockade of the Fcγ receptors, indicating that a combination therapy of anti-PD1 with Fcγ-receptor blockade could be a way to improve immunotherapy (Arlauckas et al. 2017). Also, in doxorubicin-treated mammary tumors, IVM demonstrated that the TME plays an important role, where differential regulation of vascular permeability and myeloid cell infiltration were shown to be key determinants of effective drug response (Nakasone et al. 2012).

Many breast cancer patients eventually develop resistance to therapy. To better understand the dynamics of drug-resistance, IVM has been used as a tool. For instance, IVM combined with fluorescent analogs of eribulin, a microtubule-targeting agent, identified that resistance was spatially determined by distance of the cancer cells to the 3D tumor vasculature and the expression of drug efflux proteins (Laughney et al. 2014). A promising class of anticancer drugs are therapeutic nanoparticles, which not only lead to reduced toxicity, but also show improved bioavailability. However, efficient delivery to the tumor site to maximize their therapeutic effect is still a hurdle in clinical practice. Using a combination of IVM and modeling, it was shown that a single, low dose of local radiation can be used to boost nanoparticle influx into the breast tumor (Miller et al. 2017). Local radiation leads to a transient increase in vascular permeability, which in turn drives extravasation of nanoparticles and tumor-associated macrophages, which took up the therapeutic nanoparticles, into the tumor. Several other approaches have indeed used this tendency of tumor-associated immune cells to take up the nanoparticles. Both tumor-associated macrophages (Rodell et al. 2018), neutrophils (Chu et al. 2017; Naumenko et al. 2020), or myeloid cells were used as “Trojan horses” to improve the delivery of therapeutic nanoparticles specifically toward the tumor site (Lin et al. 2020).

To study the molecular dynamics together with the spatiotemporal dynamics of drug target vulnerabilities, several biosensor mouse models were developed (Erami et al. 2016; Nobis et al. 2017, 2018). Recently, a Rac1-FRET biosensor mouse was generated to study the potential of Rac1-inhibition in breast cancer. Rac1, a small GTPase and regulator of the actin cytoskeleton, may play a key role in the major steps of metastasis, such as cell motility, invasion, and EMT–MET (mesenchymal-to-epithelial transition) plasticity during extravasation and distant colonization (Floerchinger et al. 2021). IVM of MMTV-PyMT and MMTV-Her2 mouse models expressing the Rac1-FRET biosensor in primary breast tumors revealed that Rac1 activity is specifically up-regulated in proximity to tumor vasculature. Pharmacodynamic monitoring of Rac1 inhibition through a mammary imaging window showed a temporal decrease in Rac activity and a reduction in intratumoral migration in the short term, as well as increased survival and decreased lung metastasis on the long term (Floerchinger et al. 2021). Taken together, IVM has proven to be a valuable tool not only to visualize biodistribution of diverse anticancer drugs, but also to elucidate the spatial and temporal heterogeneity of drug efficacy in the intact in vivo environment.

CONCLUDING REMARKS AND FUTURE DIRECTIONS

Due to advances in sample preparation protocols and microscope technology, we can now achieve incredible cellular detail in large tissue volumes, demonstrating an invaluable asset for 3D and intravital imaging of breast cancer. Novel reporter model systems combined with innovative imaging technologies now allow for visualization of diverse aspects of the cancer cells and their TME beyond cell migration or proliferation, such as the molecular signaling dynamics or tumor metabolism, in a multiplexed way (Zhu et al. 2017; Madonna et al. 2021). With accompanying advances in artificial intelligence and large data handling, a multitude of parameters can be extracted from those imaging data sets. A recent study exemplified the power of combining imaging with large-scale analyses to extract the behavioral identity and states of immune cells in their native environments (Crainiciuc et al. 2022). The combination of dynamic imaging and large-scale or automated analyses offers a unique opportunity to correlate tumoral heterogeneity in space and time with clinical outcomes such as cancer progression or treatment response. Finally, recent developments allow for noninvasive monitoring of tumor growth dynamics in preclinical models of breast cancer at a cellular resolution using a noninvasive and vacuum stabilized imaging window, allowing for dynamic repositioning of the imaging field-of-view (Ozturk et al. 2021). Using a similar noninvasive approach with a miniaturized microscope, MediSCAPE, has demonstrated live imaging of oral mucosa in healthy volunteers, thereby paving the way for live imaging in humans (Patel et al. 2022). Such advancements hold promise that, in the coming years, live imaging could be further developed for real-time diagnosis and even incorporate dynamic information that might lead to new biomarkers of cancer aggressiveness.

Another way for deep exploration of breast cancer biology at the patient population scale is provided by recent advancements in human breast tissue in vitro modeling and development of patient-derived breast cancer organoid biobanks (Sachs et al. 2018; Rosenbluth et al. 2020; Dekkers et al. 2021) combined with 3D imaging. For instance, a combination of organoid culture and imaging can provide a powerful readout for testing the efficacy of immunotherapies, such as engineered T cells in coculture assays with cancer organoids (for a comprehensive review on organoid imaging, refer to Rios and Clevers 2018; Lukonin et al. 2021). This potential was demonstrated by applying patient-derived cancer organoids and 3D live imaging to reveal a cancer-metabolome-sensing T cell immunotherapy with broad targeting efficacy across several subtypes of breast cancer (Dekkers et al. 2022). With the deployment of BEHAV3D, a tailored 3D imaging-based framework for organoid technology, the study further revealed mode-of-action of diverse cellular immunotherapies and the influence of patient-tumor variability on their functioning. Moreover, efforts dedicated to integrating imaging with other omics data sets (RNAseq, DNAseq, ATACseq, etc.), will further enhance the dimensionality retrieved from a single biological specimen (Rodriques et al. 2019; Cang and Nie 2020; Liu et al. 2020; Mantri et al. 2021; Payne et al. 2021; Zanfardino et al. 2021). Thus, we can expect that combining advancements in human model engineering with tailored imaging modalities, integrated with behavioral phenotyping and cell-type identification at the single-cell level, will move the field of breast cancer toward precision medicine and accelerated drug development in the coming years (Rios and Clevers 2018; Dekkers et al. 2022). These future directions will greatly aid in understanding the underlying molecular signatures responsible for complex cellular heterogeneity of breast tumors and their microenvironment identified by imaging as well as the heterogenous composition and mode of action of cellular cancer therapies.